summaryrefslogtreecommitdiffstats
path: root/library/stdarch/crates/core_arch/src/arm_shared/neon/generated.rs
diff options
context:
space:
mode:
Diffstat (limited to 'library/stdarch/crates/core_arch/src/arm_shared/neon/generated.rs')
-rw-r--r--library/stdarch/crates/core_arch/src/arm_shared/neon/generated.rs3090
1 files changed, 1545 insertions, 1545 deletions
diff --git a/library/stdarch/crates/core_arch/src/arm_shared/neon/generated.rs b/library/stdarch/crates/core_arch/src/arm_shared/neon/generated.rs
index ac2709744..fe473c51e 100644
--- a/library/stdarch/crates/core_arch/src/arm_shared/neon/generated.rs
+++ b/library/stdarch/crates/core_arch/src/arm_shared/neon/generated.rs
@@ -17,7 +17,7 @@ use stdarch_test::assert_instr;
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vand))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(and))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vand_s8(a: int8x8_t, b: int8x8_t) -> int8x8_t {
simd_and(a, b)
}
@@ -30,7 +30,7 @@ pub unsafe fn vand_s8(a: int8x8_t, b: int8x8_t) -> int8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vand))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(and))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vandq_s8(a: int8x16_t, b: int8x16_t) -> int8x16_t {
simd_and(a, b)
}
@@ -43,7 +43,7 @@ pub unsafe fn vandq_s8(a: int8x16_t, b: int8x16_t) -> int8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vand))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(and))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vand_s16(a: int16x4_t, b: int16x4_t) -> int16x4_t {
simd_and(a, b)
}
@@ -56,7 +56,7 @@ pub unsafe fn vand_s16(a: int16x4_t, b: int16x4_t) -> int16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vand))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(and))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vandq_s16(a: int16x8_t, b: int16x8_t) -> int16x8_t {
simd_and(a, b)
}
@@ -69,7 +69,7 @@ pub unsafe fn vandq_s16(a: int16x8_t, b: int16x8_t) -> int16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vand))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(and))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vand_s32(a: int32x2_t, b: int32x2_t) -> int32x2_t {
simd_and(a, b)
}
@@ -82,7 +82,7 @@ pub unsafe fn vand_s32(a: int32x2_t, b: int32x2_t) -> int32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vand))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(and))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vandq_s32(a: int32x4_t, b: int32x4_t) -> int32x4_t {
simd_and(a, b)
}
@@ -95,7 +95,7 @@ pub unsafe fn vandq_s32(a: int32x4_t, b: int32x4_t) -> int32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vand))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(and))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vand_u8(a: uint8x8_t, b: uint8x8_t) -> uint8x8_t {
simd_and(a, b)
}
@@ -108,7 +108,7 @@ pub unsafe fn vand_u8(a: uint8x8_t, b: uint8x8_t) -> uint8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vand))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(and))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vandq_u8(a: uint8x16_t, b: uint8x16_t) -> uint8x16_t {
simd_and(a, b)
}
@@ -121,7 +121,7 @@ pub unsafe fn vandq_u8(a: uint8x16_t, b: uint8x16_t) -> uint8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vand))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(and))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vand_u16(a: uint16x4_t, b: uint16x4_t) -> uint16x4_t {
simd_and(a, b)
}
@@ -134,7 +134,7 @@ pub unsafe fn vand_u16(a: uint16x4_t, b: uint16x4_t) -> uint16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vand))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(and))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vandq_u16(a: uint16x8_t, b: uint16x8_t) -> uint16x8_t {
simd_and(a, b)
}
@@ -147,7 +147,7 @@ pub unsafe fn vandq_u16(a: uint16x8_t, b: uint16x8_t) -> uint16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vand))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(and))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vand_u32(a: uint32x2_t, b: uint32x2_t) -> uint32x2_t {
simd_and(a, b)
}
@@ -160,7 +160,7 @@ pub unsafe fn vand_u32(a: uint32x2_t, b: uint32x2_t) -> uint32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vand))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(and))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vandq_u32(a: uint32x4_t, b: uint32x4_t) -> uint32x4_t {
simd_and(a, b)
}
@@ -173,7 +173,7 @@ pub unsafe fn vandq_u32(a: uint32x4_t, b: uint32x4_t) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vand))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(and))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vand_s64(a: int64x1_t, b: int64x1_t) -> int64x1_t {
simd_and(a, b)
}
@@ -186,7 +186,7 @@ pub unsafe fn vand_s64(a: int64x1_t, b: int64x1_t) -> int64x1_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vand))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(and))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vandq_s64(a: int64x2_t, b: int64x2_t) -> int64x2_t {
simd_and(a, b)
}
@@ -199,7 +199,7 @@ pub unsafe fn vandq_s64(a: int64x2_t, b: int64x2_t) -> int64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vand))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(and))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vand_u64(a: uint64x1_t, b: uint64x1_t) -> uint64x1_t {
simd_and(a, b)
}
@@ -212,7 +212,7 @@ pub unsafe fn vand_u64(a: uint64x1_t, b: uint64x1_t) -> uint64x1_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vand))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(and))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vandq_u64(a: uint64x2_t, b: uint64x2_t) -> uint64x2_t {
simd_and(a, b)
}
@@ -225,7 +225,7 @@ pub unsafe fn vandq_u64(a: uint64x2_t, b: uint64x2_t) -> uint64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vorr))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(orr))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vorr_s8(a: int8x8_t, b: int8x8_t) -> int8x8_t {
simd_or(a, b)
}
@@ -238,7 +238,7 @@ pub unsafe fn vorr_s8(a: int8x8_t, b: int8x8_t) -> int8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vorr))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(orr))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vorrq_s8(a: int8x16_t, b: int8x16_t) -> int8x16_t {
simd_or(a, b)
}
@@ -251,7 +251,7 @@ pub unsafe fn vorrq_s8(a: int8x16_t, b: int8x16_t) -> int8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vorr))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(orr))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vorr_s16(a: int16x4_t, b: int16x4_t) -> int16x4_t {
simd_or(a, b)
}
@@ -264,7 +264,7 @@ pub unsafe fn vorr_s16(a: int16x4_t, b: int16x4_t) -> int16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vorr))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(orr))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vorrq_s16(a: int16x8_t, b: int16x8_t) -> int16x8_t {
simd_or(a, b)
}
@@ -277,7 +277,7 @@ pub unsafe fn vorrq_s16(a: int16x8_t, b: int16x8_t) -> int16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vorr))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(orr))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vorr_s32(a: int32x2_t, b: int32x2_t) -> int32x2_t {
simd_or(a, b)
}
@@ -290,7 +290,7 @@ pub unsafe fn vorr_s32(a: int32x2_t, b: int32x2_t) -> int32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vorr))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(orr))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vorrq_s32(a: int32x4_t, b: int32x4_t) -> int32x4_t {
simd_or(a, b)
}
@@ -303,7 +303,7 @@ pub unsafe fn vorrq_s32(a: int32x4_t, b: int32x4_t) -> int32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vorr))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(orr))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vorr_u8(a: uint8x8_t, b: uint8x8_t) -> uint8x8_t {
simd_or(a, b)
}
@@ -316,7 +316,7 @@ pub unsafe fn vorr_u8(a: uint8x8_t, b: uint8x8_t) -> uint8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vorr))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(orr))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vorrq_u8(a: uint8x16_t, b: uint8x16_t) -> uint8x16_t {
simd_or(a, b)
}
@@ -329,7 +329,7 @@ pub unsafe fn vorrq_u8(a: uint8x16_t, b: uint8x16_t) -> uint8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vorr))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(orr))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vorr_u16(a: uint16x4_t, b: uint16x4_t) -> uint16x4_t {
simd_or(a, b)
}
@@ -342,7 +342,7 @@ pub unsafe fn vorr_u16(a: uint16x4_t, b: uint16x4_t) -> uint16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vorr))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(orr))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vorrq_u16(a: uint16x8_t, b: uint16x8_t) -> uint16x8_t {
simd_or(a, b)
}
@@ -355,7 +355,7 @@ pub unsafe fn vorrq_u16(a: uint16x8_t, b: uint16x8_t) -> uint16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vorr))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(orr))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vorr_u32(a: uint32x2_t, b: uint32x2_t) -> uint32x2_t {
simd_or(a, b)
}
@@ -368,7 +368,7 @@ pub unsafe fn vorr_u32(a: uint32x2_t, b: uint32x2_t) -> uint32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vorr))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(orr))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vorrq_u32(a: uint32x4_t, b: uint32x4_t) -> uint32x4_t {
simd_or(a, b)
}
@@ -381,7 +381,7 @@ pub unsafe fn vorrq_u32(a: uint32x4_t, b: uint32x4_t) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vorr))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(orr))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vorr_s64(a: int64x1_t, b: int64x1_t) -> int64x1_t {
simd_or(a, b)
}
@@ -394,7 +394,7 @@ pub unsafe fn vorr_s64(a: int64x1_t, b: int64x1_t) -> int64x1_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vorr))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(orr))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vorrq_s64(a: int64x2_t, b: int64x2_t) -> int64x2_t {
simd_or(a, b)
}
@@ -407,7 +407,7 @@ pub unsafe fn vorrq_s64(a: int64x2_t, b: int64x2_t) -> int64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vorr))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(orr))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vorr_u64(a: uint64x1_t, b: uint64x1_t) -> uint64x1_t {
simd_or(a, b)
}
@@ -420,7 +420,7 @@ pub unsafe fn vorr_u64(a: uint64x1_t, b: uint64x1_t) -> uint64x1_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vorr))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(orr))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vorrq_u64(a: uint64x2_t, b: uint64x2_t) -> uint64x2_t {
simd_or(a, b)
}
@@ -433,7 +433,7 @@ pub unsafe fn vorrq_u64(a: uint64x2_t, b: uint64x2_t) -> uint64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(veor))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(eor))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn veor_s8(a: int8x8_t, b: int8x8_t) -> int8x8_t {
simd_xor(a, b)
}
@@ -446,7 +446,7 @@ pub unsafe fn veor_s8(a: int8x8_t, b: int8x8_t) -> int8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(veor))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(eor))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn veorq_s8(a: int8x16_t, b: int8x16_t) -> int8x16_t {
simd_xor(a, b)
}
@@ -459,7 +459,7 @@ pub unsafe fn veorq_s8(a: int8x16_t, b: int8x16_t) -> int8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(veor))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(eor))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn veor_s16(a: int16x4_t, b: int16x4_t) -> int16x4_t {
simd_xor(a, b)
}
@@ -472,7 +472,7 @@ pub unsafe fn veor_s16(a: int16x4_t, b: int16x4_t) -> int16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(veor))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(eor))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn veorq_s16(a: int16x8_t, b: int16x8_t) -> int16x8_t {
simd_xor(a, b)
}
@@ -485,7 +485,7 @@ pub unsafe fn veorq_s16(a: int16x8_t, b: int16x8_t) -> int16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(veor))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(eor))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn veor_s32(a: int32x2_t, b: int32x2_t) -> int32x2_t {
simd_xor(a, b)
}
@@ -498,7 +498,7 @@ pub unsafe fn veor_s32(a: int32x2_t, b: int32x2_t) -> int32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(veor))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(eor))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn veorq_s32(a: int32x4_t, b: int32x4_t) -> int32x4_t {
simd_xor(a, b)
}
@@ -511,7 +511,7 @@ pub unsafe fn veorq_s32(a: int32x4_t, b: int32x4_t) -> int32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(veor))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(eor))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn veor_u8(a: uint8x8_t, b: uint8x8_t) -> uint8x8_t {
simd_xor(a, b)
}
@@ -524,7 +524,7 @@ pub unsafe fn veor_u8(a: uint8x8_t, b: uint8x8_t) -> uint8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(veor))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(eor))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn veorq_u8(a: uint8x16_t, b: uint8x16_t) -> uint8x16_t {
simd_xor(a, b)
}
@@ -537,7 +537,7 @@ pub unsafe fn veorq_u8(a: uint8x16_t, b: uint8x16_t) -> uint8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(veor))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(eor))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn veor_u16(a: uint16x4_t, b: uint16x4_t) -> uint16x4_t {
simd_xor(a, b)
}
@@ -550,7 +550,7 @@ pub unsafe fn veor_u16(a: uint16x4_t, b: uint16x4_t) -> uint16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(veor))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(eor))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn veorq_u16(a: uint16x8_t, b: uint16x8_t) -> uint16x8_t {
simd_xor(a, b)
}
@@ -563,7 +563,7 @@ pub unsafe fn veorq_u16(a: uint16x8_t, b: uint16x8_t) -> uint16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(veor))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(eor))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn veor_u32(a: uint32x2_t, b: uint32x2_t) -> uint32x2_t {
simd_xor(a, b)
}
@@ -576,7 +576,7 @@ pub unsafe fn veor_u32(a: uint32x2_t, b: uint32x2_t) -> uint32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(veor))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(eor))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn veorq_u32(a: uint32x4_t, b: uint32x4_t) -> uint32x4_t {
simd_xor(a, b)
}
@@ -589,7 +589,7 @@ pub unsafe fn veorq_u32(a: uint32x4_t, b: uint32x4_t) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(veor))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(eor))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn veor_s64(a: int64x1_t, b: int64x1_t) -> int64x1_t {
simd_xor(a, b)
}
@@ -602,7 +602,7 @@ pub unsafe fn veor_s64(a: int64x1_t, b: int64x1_t) -> int64x1_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(veor))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(eor))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn veorq_s64(a: int64x2_t, b: int64x2_t) -> int64x2_t {
simd_xor(a, b)
}
@@ -615,7 +615,7 @@ pub unsafe fn veorq_s64(a: int64x2_t, b: int64x2_t) -> int64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(veor))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(eor))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn veor_u64(a: uint64x1_t, b: uint64x1_t) -> uint64x1_t {
simd_xor(a, b)
}
@@ -628,7 +628,7 @@ pub unsafe fn veor_u64(a: uint64x1_t, b: uint64x1_t) -> uint64x1_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(veor))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(eor))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn veorq_u64(a: uint64x2_t, b: uint64x2_t) -> uint64x2_t {
simd_xor(a, b)
}
@@ -641,7 +641,7 @@ pub unsafe fn veorq_u64(a: uint64x2_t, b: uint64x2_t) -> uint64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vabd.s8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sabd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vabd_s8(a: int8x8_t, b: int8x8_t) -> int8x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -660,7 +660,7 @@ vabd_s8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vabd.s8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sabd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vabdq_s8(a: int8x16_t, b: int8x16_t) -> int8x16_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -679,7 +679,7 @@ vabdq_s8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vabd.s16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sabd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vabd_s16(a: int16x4_t, b: int16x4_t) -> int16x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -698,7 +698,7 @@ vabd_s16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vabd.s16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sabd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vabdq_s16(a: int16x8_t, b: int16x8_t) -> int16x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -717,7 +717,7 @@ vabdq_s16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vabd.s32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sabd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vabd_s32(a: int32x2_t, b: int32x2_t) -> int32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -736,7 +736,7 @@ vabd_s32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vabd.s32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sabd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vabdq_s32(a: int32x4_t, b: int32x4_t) -> int32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -755,7 +755,7 @@ vabdq_s32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vabd.u8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uabd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vabd_u8(a: uint8x8_t, b: uint8x8_t) -> uint8x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -774,7 +774,7 @@ vabd_u8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vabd.u8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uabd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vabdq_u8(a: uint8x16_t, b: uint8x16_t) -> uint8x16_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -793,7 +793,7 @@ vabdq_u8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vabd.u16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uabd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vabd_u16(a: uint16x4_t, b: uint16x4_t) -> uint16x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -812,7 +812,7 @@ vabd_u16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vabd.u16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uabd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vabdq_u16(a: uint16x8_t, b: uint16x8_t) -> uint16x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -831,7 +831,7 @@ vabdq_u16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vabd.u32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uabd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vabd_u32(a: uint32x2_t, b: uint32x2_t) -> uint32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -850,7 +850,7 @@ vabd_u32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vabd.u32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uabd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vabdq_u32(a: uint32x4_t, b: uint32x4_t) -> uint32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -869,7 +869,7 @@ vabdq_u32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vabd.f32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fabd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vabd_f32(a: float32x2_t, b: float32x2_t) -> float32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -888,7 +888,7 @@ vabd_f32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vabd.f32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fabd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vabdq_f32(a: float32x4_t, b: float32x4_t) -> float32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -907,7 +907,7 @@ vabdq_f32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vabdl.u8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uabdl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vabdl_u8(a: uint8x8_t, b: uint8x8_t) -> uint16x8_t {
simd_cast(vabd_u8(a, b))
}
@@ -920,7 +920,7 @@ pub unsafe fn vabdl_u8(a: uint8x8_t, b: uint8x8_t) -> uint16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vabdl.u16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uabdl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vabdl_u16(a: uint16x4_t, b: uint16x4_t) -> uint32x4_t {
simd_cast(vabd_u16(a, b))
}
@@ -933,7 +933,7 @@ pub unsafe fn vabdl_u16(a: uint16x4_t, b: uint16x4_t) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vabdl.u32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uabdl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vabdl_u32(a: uint32x2_t, b: uint32x2_t) -> uint64x2_t {
simd_cast(vabd_u32(a, b))
}
@@ -946,7 +946,7 @@ pub unsafe fn vabdl_u32(a: uint32x2_t, b: uint32x2_t) -> uint64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vabdl.s8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sabdl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vabdl_s8(a: int8x8_t, b: int8x8_t) -> int16x8_t {
let c: uint8x8_t = simd_cast(vabd_s8(a, b));
simd_cast(c)
@@ -960,7 +960,7 @@ pub unsafe fn vabdl_s8(a: int8x8_t, b: int8x8_t) -> int16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vabdl.s16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sabdl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vabdl_s16(a: int16x4_t, b: int16x4_t) -> int32x4_t {
let c: uint16x4_t = simd_cast(vabd_s16(a, b));
simd_cast(c)
@@ -974,7 +974,7 @@ pub unsafe fn vabdl_s16(a: int16x4_t, b: int16x4_t) -> int32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vabdl.s32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sabdl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vabdl_s32(a: int32x2_t, b: int32x2_t) -> int64x2_t {
let c: uint32x2_t = simd_cast(vabd_s32(a, b));
simd_cast(c)
@@ -988,7 +988,7 @@ pub unsafe fn vabdl_s32(a: int32x2_t, b: int32x2_t) -> int64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vceq.i8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmeq))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vceq_u8(a: uint8x8_t, b: uint8x8_t) -> uint8x8_t {
simd_eq(a, b)
}
@@ -1001,7 +1001,7 @@ pub unsafe fn vceq_u8(a: uint8x8_t, b: uint8x8_t) -> uint8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vceq.i8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmeq))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vceqq_u8(a: uint8x16_t, b: uint8x16_t) -> uint8x16_t {
simd_eq(a, b)
}
@@ -1014,7 +1014,7 @@ pub unsafe fn vceqq_u8(a: uint8x16_t, b: uint8x16_t) -> uint8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vceq.i16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmeq))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vceq_u16(a: uint16x4_t, b: uint16x4_t) -> uint16x4_t {
simd_eq(a, b)
}
@@ -1027,7 +1027,7 @@ pub unsafe fn vceq_u16(a: uint16x4_t, b: uint16x4_t) -> uint16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vceq.i16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmeq))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vceqq_u16(a: uint16x8_t, b: uint16x8_t) -> uint16x8_t {
simd_eq(a, b)
}
@@ -1040,7 +1040,7 @@ pub unsafe fn vceqq_u16(a: uint16x8_t, b: uint16x8_t) -> uint16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vceq.i32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmeq))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vceq_u32(a: uint32x2_t, b: uint32x2_t) -> uint32x2_t {
simd_eq(a, b)
}
@@ -1053,7 +1053,7 @@ pub unsafe fn vceq_u32(a: uint32x2_t, b: uint32x2_t) -> uint32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vceq.i32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmeq))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vceqq_u32(a: uint32x4_t, b: uint32x4_t) -> uint32x4_t {
simd_eq(a, b)
}
@@ -1066,7 +1066,7 @@ pub unsafe fn vceqq_u32(a: uint32x4_t, b: uint32x4_t) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vceq.i8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmeq))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vceq_s8(a: int8x8_t, b: int8x8_t) -> uint8x8_t {
simd_eq(a, b)
}
@@ -1079,7 +1079,7 @@ pub unsafe fn vceq_s8(a: int8x8_t, b: int8x8_t) -> uint8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vceq.i8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmeq))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vceqq_s8(a: int8x16_t, b: int8x16_t) -> uint8x16_t {
simd_eq(a, b)
}
@@ -1092,7 +1092,7 @@ pub unsafe fn vceqq_s8(a: int8x16_t, b: int8x16_t) -> uint8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vceq.i16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmeq))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vceq_s16(a: int16x4_t, b: int16x4_t) -> uint16x4_t {
simd_eq(a, b)
}
@@ -1105,7 +1105,7 @@ pub unsafe fn vceq_s16(a: int16x4_t, b: int16x4_t) -> uint16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vceq.i16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmeq))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vceqq_s16(a: int16x8_t, b: int16x8_t) -> uint16x8_t {
simd_eq(a, b)
}
@@ -1118,7 +1118,7 @@ pub unsafe fn vceqq_s16(a: int16x8_t, b: int16x8_t) -> uint16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vceq.i32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmeq))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vceq_s32(a: int32x2_t, b: int32x2_t) -> uint32x2_t {
simd_eq(a, b)
}
@@ -1131,7 +1131,7 @@ pub unsafe fn vceq_s32(a: int32x2_t, b: int32x2_t) -> uint32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vceq.i32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmeq))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vceqq_s32(a: int32x4_t, b: int32x4_t) -> uint32x4_t {
simd_eq(a, b)
}
@@ -1144,7 +1144,7 @@ pub unsafe fn vceqq_s32(a: int32x4_t, b: int32x4_t) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vceq.i8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmeq))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vceq_p8(a: poly8x8_t, b: poly8x8_t) -> uint8x8_t {
simd_eq(a, b)
}
@@ -1157,7 +1157,7 @@ pub unsafe fn vceq_p8(a: poly8x8_t, b: poly8x8_t) -> uint8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vceq.i8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmeq))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vceqq_p8(a: poly8x16_t, b: poly8x16_t) -> uint8x16_t {
simd_eq(a, b)
}
@@ -1170,7 +1170,7 @@ pub unsafe fn vceqq_p8(a: poly8x16_t, b: poly8x16_t) -> uint8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vceq.f32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fcmeq))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vceq_f32(a: float32x2_t, b: float32x2_t) -> uint32x2_t {
simd_eq(a, b)
}
@@ -1183,7 +1183,7 @@ pub unsafe fn vceq_f32(a: float32x2_t, b: float32x2_t) -> uint32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vceq.f32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fcmeq))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vceqq_f32(a: float32x4_t, b: float32x4_t) -> uint32x4_t {
simd_eq(a, b)
}
@@ -1196,7 +1196,7 @@ pub unsafe fn vceqq_f32(a: float32x4_t, b: float32x4_t) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vtst))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmtst))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vtst_s8(a: int8x8_t, b: int8x8_t) -> uint8x8_t {
let c: int8x8_t = simd_and(a, b);
let d: i8x8 = i8x8::new(0, 0, 0, 0, 0, 0, 0, 0);
@@ -1211,7 +1211,7 @@ pub unsafe fn vtst_s8(a: int8x8_t, b: int8x8_t) -> uint8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vtst))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmtst))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vtstq_s8(a: int8x16_t, b: int8x16_t) -> uint8x16_t {
let c: int8x16_t = simd_and(a, b);
let d: i8x16 = i8x16::new(0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0);
@@ -1226,7 +1226,7 @@ pub unsafe fn vtstq_s8(a: int8x16_t, b: int8x16_t) -> uint8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vtst))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmtst))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vtst_s16(a: int16x4_t, b: int16x4_t) -> uint16x4_t {
let c: int16x4_t = simd_and(a, b);
let d: i16x4 = i16x4::new(0, 0, 0, 0);
@@ -1241,7 +1241,7 @@ pub unsafe fn vtst_s16(a: int16x4_t, b: int16x4_t) -> uint16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vtst))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmtst))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vtstq_s16(a: int16x8_t, b: int16x8_t) -> uint16x8_t {
let c: int16x8_t = simd_and(a, b);
let d: i16x8 = i16x8::new(0, 0, 0, 0, 0, 0, 0, 0);
@@ -1256,7 +1256,7 @@ pub unsafe fn vtstq_s16(a: int16x8_t, b: int16x8_t) -> uint16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vtst))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmtst))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vtst_s32(a: int32x2_t, b: int32x2_t) -> uint32x2_t {
let c: int32x2_t = simd_and(a, b);
let d: i32x2 = i32x2::new(0, 0);
@@ -1271,7 +1271,7 @@ pub unsafe fn vtst_s32(a: int32x2_t, b: int32x2_t) -> uint32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vtst))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmtst))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vtstq_s32(a: int32x4_t, b: int32x4_t) -> uint32x4_t {
let c: int32x4_t = simd_and(a, b);
let d: i32x4 = i32x4::new(0, 0, 0, 0);
@@ -1286,7 +1286,7 @@ pub unsafe fn vtstq_s32(a: int32x4_t, b: int32x4_t) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vtst))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmtst))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vtst_p8(a: poly8x8_t, b: poly8x8_t) -> uint8x8_t {
let c: poly8x8_t = simd_and(a, b);
let d: i8x8 = i8x8::new(0, 0, 0, 0, 0, 0, 0, 0);
@@ -1301,7 +1301,7 @@ pub unsafe fn vtst_p8(a: poly8x8_t, b: poly8x8_t) -> uint8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vtst))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmtst))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vtstq_p8(a: poly8x16_t, b: poly8x16_t) -> uint8x16_t {
let c: poly8x16_t = simd_and(a, b);
let d: i8x16 = i8x16::new(0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0);
@@ -1316,7 +1316,7 @@ pub unsafe fn vtstq_p8(a: poly8x16_t, b: poly8x16_t) -> uint8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vtst))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmtst))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vtst_p16(a: poly16x4_t, b: poly16x4_t) -> uint16x4_t {
let c: poly16x4_t = simd_and(a, b);
let d: i16x4 = i16x4::new(0, 0, 0, 0);
@@ -1331,7 +1331,7 @@ pub unsafe fn vtst_p16(a: poly16x4_t, b: poly16x4_t) -> uint16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vtst))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmtst))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vtstq_p16(a: poly16x8_t, b: poly16x8_t) -> uint16x8_t {
let c: poly16x8_t = simd_and(a, b);
let d: i16x8 = i16x8::new(0, 0, 0, 0, 0, 0, 0, 0);
@@ -1346,7 +1346,7 @@ pub unsafe fn vtstq_p16(a: poly16x8_t, b: poly16x8_t) -> uint16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vtst))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmtst))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vtst_u8(a: uint8x8_t, b: uint8x8_t) -> uint8x8_t {
let c: uint8x8_t = simd_and(a, b);
let d: u8x8 = u8x8::new(0, 0, 0, 0, 0, 0, 0, 0);
@@ -1361,7 +1361,7 @@ pub unsafe fn vtst_u8(a: uint8x8_t, b: uint8x8_t) -> uint8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vtst))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmtst))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vtstq_u8(a: uint8x16_t, b: uint8x16_t) -> uint8x16_t {
let c: uint8x16_t = simd_and(a, b);
let d: u8x16 = u8x16::new(0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0);
@@ -1376,7 +1376,7 @@ pub unsafe fn vtstq_u8(a: uint8x16_t, b: uint8x16_t) -> uint8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vtst))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmtst))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vtst_u16(a: uint16x4_t, b: uint16x4_t) -> uint16x4_t {
let c: uint16x4_t = simd_and(a, b);
let d: u16x4 = u16x4::new(0, 0, 0, 0);
@@ -1391,7 +1391,7 @@ pub unsafe fn vtst_u16(a: uint16x4_t, b: uint16x4_t) -> uint16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vtst))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmtst))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vtstq_u16(a: uint16x8_t, b: uint16x8_t) -> uint16x8_t {
let c: uint16x8_t = simd_and(a, b);
let d: u16x8 = u16x8::new(0, 0, 0, 0, 0, 0, 0, 0);
@@ -1406,7 +1406,7 @@ pub unsafe fn vtstq_u16(a: uint16x8_t, b: uint16x8_t) -> uint16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vtst))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmtst))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vtst_u32(a: uint32x2_t, b: uint32x2_t) -> uint32x2_t {
let c: uint32x2_t = simd_and(a, b);
let d: u32x2 = u32x2::new(0, 0);
@@ -1421,7 +1421,7 @@ pub unsafe fn vtst_u32(a: uint32x2_t, b: uint32x2_t) -> uint32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vtst))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmtst))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vtstq_u32(a: uint32x4_t, b: uint32x4_t) -> uint32x4_t {
let c: uint32x4_t = simd_and(a, b);
let d: u32x4 = u32x4::new(0, 0, 0, 0);
@@ -1436,7 +1436,7 @@ pub unsafe fn vtstq_u32(a: uint32x4_t, b: uint32x4_t) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vabs))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fabs))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vabs_f32(a: float32x2_t) -> float32x2_t {
simd_fabs(a)
}
@@ -1449,7 +1449,7 @@ pub unsafe fn vabs_f32(a: float32x2_t) -> float32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vabs))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fabs))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vabsq_f32(a: float32x4_t) -> float32x4_t {
simd_fabs(a)
}
@@ -1462,7 +1462,7 @@ pub unsafe fn vabsq_f32(a: float32x4_t) -> float32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcgt.s8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmgt))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcgt_s8(a: int8x8_t, b: int8x8_t) -> uint8x8_t {
simd_gt(a, b)
}
@@ -1475,7 +1475,7 @@ pub unsafe fn vcgt_s8(a: int8x8_t, b: int8x8_t) -> uint8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcgt.s8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmgt))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcgtq_s8(a: int8x16_t, b: int8x16_t) -> uint8x16_t {
simd_gt(a, b)
}
@@ -1488,7 +1488,7 @@ pub unsafe fn vcgtq_s8(a: int8x16_t, b: int8x16_t) -> uint8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcgt.s16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmgt))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcgt_s16(a: int16x4_t, b: int16x4_t) -> uint16x4_t {
simd_gt(a, b)
}
@@ -1501,7 +1501,7 @@ pub unsafe fn vcgt_s16(a: int16x4_t, b: int16x4_t) -> uint16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcgt.s16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmgt))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcgtq_s16(a: int16x8_t, b: int16x8_t) -> uint16x8_t {
simd_gt(a, b)
}
@@ -1514,7 +1514,7 @@ pub unsafe fn vcgtq_s16(a: int16x8_t, b: int16x8_t) -> uint16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcgt.s32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmgt))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcgt_s32(a: int32x2_t, b: int32x2_t) -> uint32x2_t {
simd_gt(a, b)
}
@@ -1527,7 +1527,7 @@ pub unsafe fn vcgt_s32(a: int32x2_t, b: int32x2_t) -> uint32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcgt.s32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmgt))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcgtq_s32(a: int32x4_t, b: int32x4_t) -> uint32x4_t {
simd_gt(a, b)
}
@@ -1540,7 +1540,7 @@ pub unsafe fn vcgtq_s32(a: int32x4_t, b: int32x4_t) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcgt.u8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmhi))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcgt_u8(a: uint8x8_t, b: uint8x8_t) -> uint8x8_t {
simd_gt(a, b)
}
@@ -1553,7 +1553,7 @@ pub unsafe fn vcgt_u8(a: uint8x8_t, b: uint8x8_t) -> uint8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcgt.u8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmhi))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcgtq_u8(a: uint8x16_t, b: uint8x16_t) -> uint8x16_t {
simd_gt(a, b)
}
@@ -1566,7 +1566,7 @@ pub unsafe fn vcgtq_u8(a: uint8x16_t, b: uint8x16_t) -> uint8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcgt.u16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmhi))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcgt_u16(a: uint16x4_t, b: uint16x4_t) -> uint16x4_t {
simd_gt(a, b)
}
@@ -1579,7 +1579,7 @@ pub unsafe fn vcgt_u16(a: uint16x4_t, b: uint16x4_t) -> uint16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcgt.u16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmhi))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcgtq_u16(a: uint16x8_t, b: uint16x8_t) -> uint16x8_t {
simd_gt(a, b)
}
@@ -1592,7 +1592,7 @@ pub unsafe fn vcgtq_u16(a: uint16x8_t, b: uint16x8_t) -> uint16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcgt.u32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmhi))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcgt_u32(a: uint32x2_t, b: uint32x2_t) -> uint32x2_t {
simd_gt(a, b)
}
@@ -1605,7 +1605,7 @@ pub unsafe fn vcgt_u32(a: uint32x2_t, b: uint32x2_t) -> uint32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcgt.u32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmhi))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcgtq_u32(a: uint32x4_t, b: uint32x4_t) -> uint32x4_t {
simd_gt(a, b)
}
@@ -1618,7 +1618,7 @@ pub unsafe fn vcgtq_u32(a: uint32x4_t, b: uint32x4_t) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcgt.f32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fcmgt))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcgt_f32(a: float32x2_t, b: float32x2_t) -> uint32x2_t {
simd_gt(a, b)
}
@@ -1631,7 +1631,7 @@ pub unsafe fn vcgt_f32(a: float32x2_t, b: float32x2_t) -> uint32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcgt.f32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fcmgt))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcgtq_f32(a: float32x4_t, b: float32x4_t) -> uint32x4_t {
simd_gt(a, b)
}
@@ -1644,7 +1644,7 @@ pub unsafe fn vcgtq_f32(a: float32x4_t, b: float32x4_t) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcgt.s8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmgt))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vclt_s8(a: int8x8_t, b: int8x8_t) -> uint8x8_t {
simd_lt(a, b)
}
@@ -1657,7 +1657,7 @@ pub unsafe fn vclt_s8(a: int8x8_t, b: int8x8_t) -> uint8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcgt.s8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmgt))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcltq_s8(a: int8x16_t, b: int8x16_t) -> uint8x16_t {
simd_lt(a, b)
}
@@ -1670,7 +1670,7 @@ pub unsafe fn vcltq_s8(a: int8x16_t, b: int8x16_t) -> uint8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcgt.s16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmgt))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vclt_s16(a: int16x4_t, b: int16x4_t) -> uint16x4_t {
simd_lt(a, b)
}
@@ -1683,7 +1683,7 @@ pub unsafe fn vclt_s16(a: int16x4_t, b: int16x4_t) -> uint16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcgt.s16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmgt))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcltq_s16(a: int16x8_t, b: int16x8_t) -> uint16x8_t {
simd_lt(a, b)
}
@@ -1696,7 +1696,7 @@ pub unsafe fn vcltq_s16(a: int16x8_t, b: int16x8_t) -> uint16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcgt.s32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmgt))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vclt_s32(a: int32x2_t, b: int32x2_t) -> uint32x2_t {
simd_lt(a, b)
}
@@ -1709,7 +1709,7 @@ pub unsafe fn vclt_s32(a: int32x2_t, b: int32x2_t) -> uint32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcgt.s32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmgt))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcltq_s32(a: int32x4_t, b: int32x4_t) -> uint32x4_t {
simd_lt(a, b)
}
@@ -1722,7 +1722,7 @@ pub unsafe fn vcltq_s32(a: int32x4_t, b: int32x4_t) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcgt.u8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmhi))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vclt_u8(a: uint8x8_t, b: uint8x8_t) -> uint8x8_t {
simd_lt(a, b)
}
@@ -1735,7 +1735,7 @@ pub unsafe fn vclt_u8(a: uint8x8_t, b: uint8x8_t) -> uint8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcgt.u8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmhi))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcltq_u8(a: uint8x16_t, b: uint8x16_t) -> uint8x16_t {
simd_lt(a, b)
}
@@ -1748,7 +1748,7 @@ pub unsafe fn vcltq_u8(a: uint8x16_t, b: uint8x16_t) -> uint8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcgt.u16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmhi))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vclt_u16(a: uint16x4_t, b: uint16x4_t) -> uint16x4_t {
simd_lt(a, b)
}
@@ -1761,7 +1761,7 @@ pub unsafe fn vclt_u16(a: uint16x4_t, b: uint16x4_t) -> uint16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcgt.u16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmhi))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcltq_u16(a: uint16x8_t, b: uint16x8_t) -> uint16x8_t {
simd_lt(a, b)
}
@@ -1774,7 +1774,7 @@ pub unsafe fn vcltq_u16(a: uint16x8_t, b: uint16x8_t) -> uint16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcgt.u32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmhi))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vclt_u32(a: uint32x2_t, b: uint32x2_t) -> uint32x2_t {
simd_lt(a, b)
}
@@ -1787,7 +1787,7 @@ pub unsafe fn vclt_u32(a: uint32x2_t, b: uint32x2_t) -> uint32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcgt.u32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmhi))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcltq_u32(a: uint32x4_t, b: uint32x4_t) -> uint32x4_t {
simd_lt(a, b)
}
@@ -1800,7 +1800,7 @@ pub unsafe fn vcltq_u32(a: uint32x4_t, b: uint32x4_t) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcgt.f32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fcmgt))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vclt_f32(a: float32x2_t, b: float32x2_t) -> uint32x2_t {
simd_lt(a, b)
}
@@ -1813,7 +1813,7 @@ pub unsafe fn vclt_f32(a: float32x2_t, b: float32x2_t) -> uint32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcgt.f32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fcmgt))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcltq_f32(a: float32x4_t, b: float32x4_t) -> uint32x4_t {
simd_lt(a, b)
}
@@ -1826,7 +1826,7 @@ pub unsafe fn vcltq_f32(a: float32x4_t, b: float32x4_t) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcge.s8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmge))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcle_s8(a: int8x8_t, b: int8x8_t) -> uint8x8_t {
simd_le(a, b)
}
@@ -1839,7 +1839,7 @@ pub unsafe fn vcle_s8(a: int8x8_t, b: int8x8_t) -> uint8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcge.s8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmge))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcleq_s8(a: int8x16_t, b: int8x16_t) -> uint8x16_t {
simd_le(a, b)
}
@@ -1852,7 +1852,7 @@ pub unsafe fn vcleq_s8(a: int8x16_t, b: int8x16_t) -> uint8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcge.s16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmge))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcle_s16(a: int16x4_t, b: int16x4_t) -> uint16x4_t {
simd_le(a, b)
}
@@ -1865,7 +1865,7 @@ pub unsafe fn vcle_s16(a: int16x4_t, b: int16x4_t) -> uint16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcge.s16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmge))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcleq_s16(a: int16x8_t, b: int16x8_t) -> uint16x8_t {
simd_le(a, b)
}
@@ -1878,7 +1878,7 @@ pub unsafe fn vcleq_s16(a: int16x8_t, b: int16x8_t) -> uint16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcge.s32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmge))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcle_s32(a: int32x2_t, b: int32x2_t) -> uint32x2_t {
simd_le(a, b)
}
@@ -1891,7 +1891,7 @@ pub unsafe fn vcle_s32(a: int32x2_t, b: int32x2_t) -> uint32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcge.s32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmge))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcleq_s32(a: int32x4_t, b: int32x4_t) -> uint32x4_t {
simd_le(a, b)
}
@@ -1904,7 +1904,7 @@ pub unsafe fn vcleq_s32(a: int32x4_t, b: int32x4_t) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcge.u8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmhs))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcle_u8(a: uint8x8_t, b: uint8x8_t) -> uint8x8_t {
simd_le(a, b)
}
@@ -1917,7 +1917,7 @@ pub unsafe fn vcle_u8(a: uint8x8_t, b: uint8x8_t) -> uint8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcge.u8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmhs))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcleq_u8(a: uint8x16_t, b: uint8x16_t) -> uint8x16_t {
simd_le(a, b)
}
@@ -1930,7 +1930,7 @@ pub unsafe fn vcleq_u8(a: uint8x16_t, b: uint8x16_t) -> uint8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcge.u16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmhs))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcle_u16(a: uint16x4_t, b: uint16x4_t) -> uint16x4_t {
simd_le(a, b)
}
@@ -1943,7 +1943,7 @@ pub unsafe fn vcle_u16(a: uint16x4_t, b: uint16x4_t) -> uint16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcge.u16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmhs))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcleq_u16(a: uint16x8_t, b: uint16x8_t) -> uint16x8_t {
simd_le(a, b)
}
@@ -1956,7 +1956,7 @@ pub unsafe fn vcleq_u16(a: uint16x8_t, b: uint16x8_t) -> uint16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcge.u32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmhs))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcle_u32(a: uint32x2_t, b: uint32x2_t) -> uint32x2_t {
simd_le(a, b)
}
@@ -1969,7 +1969,7 @@ pub unsafe fn vcle_u32(a: uint32x2_t, b: uint32x2_t) -> uint32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcge.u32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmhs))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcleq_u32(a: uint32x4_t, b: uint32x4_t) -> uint32x4_t {
simd_le(a, b)
}
@@ -1982,7 +1982,7 @@ pub unsafe fn vcleq_u32(a: uint32x4_t, b: uint32x4_t) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcge.f32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fcmge))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcle_f32(a: float32x2_t, b: float32x2_t) -> uint32x2_t {
simd_le(a, b)
}
@@ -1995,7 +1995,7 @@ pub unsafe fn vcle_f32(a: float32x2_t, b: float32x2_t) -> uint32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcge.f32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fcmge))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcleq_f32(a: float32x4_t, b: float32x4_t) -> uint32x4_t {
simd_le(a, b)
}
@@ -2008,7 +2008,7 @@ pub unsafe fn vcleq_f32(a: float32x4_t, b: float32x4_t) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcge.s8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmge))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcge_s8(a: int8x8_t, b: int8x8_t) -> uint8x8_t {
simd_ge(a, b)
}
@@ -2021,7 +2021,7 @@ pub unsafe fn vcge_s8(a: int8x8_t, b: int8x8_t) -> uint8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcge.s8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmge))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcgeq_s8(a: int8x16_t, b: int8x16_t) -> uint8x16_t {
simd_ge(a, b)
}
@@ -2034,7 +2034,7 @@ pub unsafe fn vcgeq_s8(a: int8x16_t, b: int8x16_t) -> uint8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcge.s16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmge))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcge_s16(a: int16x4_t, b: int16x4_t) -> uint16x4_t {
simd_ge(a, b)
}
@@ -2047,7 +2047,7 @@ pub unsafe fn vcge_s16(a: int16x4_t, b: int16x4_t) -> uint16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcge.s16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmge))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcgeq_s16(a: int16x8_t, b: int16x8_t) -> uint16x8_t {
simd_ge(a, b)
}
@@ -2060,7 +2060,7 @@ pub unsafe fn vcgeq_s16(a: int16x8_t, b: int16x8_t) -> uint16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcge.s32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmge))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcge_s32(a: int32x2_t, b: int32x2_t) -> uint32x2_t {
simd_ge(a, b)
}
@@ -2073,7 +2073,7 @@ pub unsafe fn vcge_s32(a: int32x2_t, b: int32x2_t) -> uint32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcge.s32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmge))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcgeq_s32(a: int32x4_t, b: int32x4_t) -> uint32x4_t {
simd_ge(a, b)
}
@@ -2086,7 +2086,7 @@ pub unsafe fn vcgeq_s32(a: int32x4_t, b: int32x4_t) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcge.u8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmhs))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcge_u8(a: uint8x8_t, b: uint8x8_t) -> uint8x8_t {
simd_ge(a, b)
}
@@ -2099,7 +2099,7 @@ pub unsafe fn vcge_u8(a: uint8x8_t, b: uint8x8_t) -> uint8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcge.u8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmhs))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcgeq_u8(a: uint8x16_t, b: uint8x16_t) -> uint8x16_t {
simd_ge(a, b)
}
@@ -2112,7 +2112,7 @@ pub unsafe fn vcgeq_u8(a: uint8x16_t, b: uint8x16_t) -> uint8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcge.u16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmhs))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcge_u16(a: uint16x4_t, b: uint16x4_t) -> uint16x4_t {
simd_ge(a, b)
}
@@ -2125,7 +2125,7 @@ pub unsafe fn vcge_u16(a: uint16x4_t, b: uint16x4_t) -> uint16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcge.u16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmhs))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcgeq_u16(a: uint16x8_t, b: uint16x8_t) -> uint16x8_t {
simd_ge(a, b)
}
@@ -2138,7 +2138,7 @@ pub unsafe fn vcgeq_u16(a: uint16x8_t, b: uint16x8_t) -> uint16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcge.u32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmhs))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcge_u32(a: uint32x2_t, b: uint32x2_t) -> uint32x2_t {
simd_ge(a, b)
}
@@ -2151,7 +2151,7 @@ pub unsafe fn vcge_u32(a: uint32x2_t, b: uint32x2_t) -> uint32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcge.u32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmhs))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcgeq_u32(a: uint32x4_t, b: uint32x4_t) -> uint32x4_t {
simd_ge(a, b)
}
@@ -2164,7 +2164,7 @@ pub unsafe fn vcgeq_u32(a: uint32x4_t, b: uint32x4_t) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcge.f32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fcmge))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcge_f32(a: float32x2_t, b: float32x2_t) -> uint32x2_t {
simd_ge(a, b)
}
@@ -2177,7 +2177,7 @@ pub unsafe fn vcge_f32(a: float32x2_t, b: float32x2_t) -> uint32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcge.f32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fcmge))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcgeq_f32(a: float32x4_t, b: float32x4_t) -> uint32x4_t {
simd_ge(a, b)
}
@@ -2190,7 +2190,7 @@ pub unsafe fn vcgeq_f32(a: float32x4_t, b: float32x4_t) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcls.s8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cls))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcls_s8(a: int8x8_t) -> int8x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -2209,7 +2209,7 @@ vcls_s8_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcls.s8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cls))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vclsq_s8(a: int8x16_t) -> int8x16_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -2228,7 +2228,7 @@ vclsq_s8_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcls.s16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cls))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcls_s16(a: int16x4_t) -> int16x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -2247,7 +2247,7 @@ vcls_s16_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcls.s16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cls))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vclsq_s16(a: int16x8_t) -> int16x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -2266,7 +2266,7 @@ vclsq_s16_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcls.s32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cls))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcls_s32(a: int32x2_t) -> int32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -2285,7 +2285,7 @@ vcls_s32_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vcls.s32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cls))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vclsq_s32(a: int32x4_t) -> int32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -2304,7 +2304,7 @@ vclsq_s32_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vcls))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cls))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcls_u8(a: uint8x8_t) -> int8x8_t {
transmute(vcls_s8(transmute(a)))
}
@@ -2317,7 +2317,7 @@ pub unsafe fn vcls_u8(a: uint8x8_t) -> int8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vcls))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cls))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vclsq_u8(a: uint8x16_t) -> int8x16_t {
transmute(vclsq_s8(transmute(a)))
}
@@ -2330,7 +2330,7 @@ pub unsafe fn vclsq_u8(a: uint8x16_t) -> int8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vcls))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cls))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcls_u16(a: uint16x4_t) -> int16x4_t {
transmute(vcls_s16(transmute(a)))
}
@@ -2343,7 +2343,7 @@ pub unsafe fn vcls_u16(a: uint16x4_t) -> int16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vcls))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cls))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vclsq_u16(a: uint16x8_t) -> int16x8_t {
transmute(vclsq_s16(transmute(a)))
}
@@ -2356,7 +2356,7 @@ pub unsafe fn vclsq_u16(a: uint16x8_t) -> int16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vcls))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cls))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcls_u32(a: uint32x2_t) -> int32x2_t {
transmute(vcls_s32(transmute(a)))
}
@@ -2369,7 +2369,7 @@ pub unsafe fn vcls_u32(a: uint32x2_t) -> int32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vcls))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cls))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vclsq_u32(a: uint32x4_t) -> int32x4_t {
transmute(vclsq_s32(transmute(a)))
}
@@ -2382,7 +2382,7 @@ pub unsafe fn vclsq_u32(a: uint32x4_t) -> int32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vclz.i8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(clz))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vclz_s8(a: int8x8_t) -> int8x8_t {
vclz_s8_(a)
}
@@ -2395,7 +2395,7 @@ pub unsafe fn vclz_s8(a: int8x8_t) -> int8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vclz.i8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(clz))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vclzq_s8(a: int8x16_t) -> int8x16_t {
vclzq_s8_(a)
}
@@ -2408,7 +2408,7 @@ pub unsafe fn vclzq_s8(a: int8x16_t) -> int8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vclz.i16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(clz))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vclz_s16(a: int16x4_t) -> int16x4_t {
vclz_s16_(a)
}
@@ -2421,7 +2421,7 @@ pub unsafe fn vclz_s16(a: int16x4_t) -> int16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vclz.i16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(clz))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vclzq_s16(a: int16x8_t) -> int16x8_t {
vclzq_s16_(a)
}
@@ -2434,7 +2434,7 @@ pub unsafe fn vclzq_s16(a: int16x8_t) -> int16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vclz.i32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(clz))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vclz_s32(a: int32x2_t) -> int32x2_t {
vclz_s32_(a)
}
@@ -2447,7 +2447,7 @@ pub unsafe fn vclz_s32(a: int32x2_t) -> int32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vclz.i32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(clz))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vclzq_s32(a: int32x4_t) -> int32x4_t {
vclzq_s32_(a)
}
@@ -2460,7 +2460,7 @@ pub unsafe fn vclzq_s32(a: int32x4_t) -> int32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vclz.i8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(clz))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vclz_u8(a: uint8x8_t) -> uint8x8_t {
transmute(vclz_s8_(transmute(a)))
}
@@ -2473,7 +2473,7 @@ pub unsafe fn vclz_u8(a: uint8x8_t) -> uint8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vclz.i8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(clz))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vclzq_u8(a: uint8x16_t) -> uint8x16_t {
transmute(vclzq_s8_(transmute(a)))
}
@@ -2486,7 +2486,7 @@ pub unsafe fn vclzq_u8(a: uint8x16_t) -> uint8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vclz.i16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(clz))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vclz_u16(a: uint16x4_t) -> uint16x4_t {
transmute(vclz_s16_(transmute(a)))
}
@@ -2499,7 +2499,7 @@ pub unsafe fn vclz_u16(a: uint16x4_t) -> uint16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vclz.i16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(clz))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vclzq_u16(a: uint16x8_t) -> uint16x8_t {
transmute(vclzq_s16_(transmute(a)))
}
@@ -2512,7 +2512,7 @@ pub unsafe fn vclzq_u16(a: uint16x8_t) -> uint16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vclz.i32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(clz))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vclz_u32(a: uint32x2_t) -> uint32x2_t {
transmute(vclz_s32_(transmute(a)))
}
@@ -2525,7 +2525,7 @@ pub unsafe fn vclz_u32(a: uint32x2_t) -> uint32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vclz.i32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(clz))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vclzq_u32(a: uint32x4_t) -> uint32x4_t {
transmute(vclzq_s32_(transmute(a)))
}
@@ -2538,7 +2538,7 @@ pub unsafe fn vclzq_u32(a: uint32x4_t) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vacgt.f32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(facgt))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcagt_f32(a: float32x2_t, b: float32x2_t) -> uint32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -2557,7 +2557,7 @@ vcagt_f32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vacgt.f32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(facgt))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcagtq_f32(a: float32x4_t, b: float32x4_t) -> uint32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -2576,7 +2576,7 @@ vcagtq_f32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vacge.f32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(facge))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcage_f32(a: float32x2_t, b: float32x2_t) -> uint32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -2595,7 +2595,7 @@ vcage_f32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vacge.f32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(facge))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcageq_f32(a: float32x4_t, b: float32x4_t) -> uint32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -2614,7 +2614,7 @@ vcageq_f32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vacgt.f32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(facgt))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcalt_f32(a: float32x2_t, b: float32x2_t) -> uint32x2_t {
vcagt_f32(b, a)
}
@@ -2627,7 +2627,7 @@ pub unsafe fn vcalt_f32(a: float32x2_t, b: float32x2_t) -> uint32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vacgt.f32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(facgt))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcaltq_f32(a: float32x4_t, b: float32x4_t) -> uint32x4_t {
vcagtq_f32(b, a)
}
@@ -2640,7 +2640,7 @@ pub unsafe fn vcaltq_f32(a: float32x4_t, b: float32x4_t) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vacge.f32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(facge))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcale_f32(a: float32x2_t, b: float32x2_t) -> uint32x2_t {
vcage_f32(b, a)
}
@@ -2653,7 +2653,7 @@ pub unsafe fn vcale_f32(a: float32x2_t, b: float32x2_t) -> uint32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vacge.f32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(facge))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcaleq_f32(a: float32x4_t, b: float32x4_t) -> uint32x4_t {
vcageq_f32(b, a)
}
@@ -2666,7 +2666,7 @@ pub unsafe fn vcaleq_f32(a: float32x4_t, b: float32x4_t) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcreate_s8(a: u64) -> int8x8_t {
transmute(a)
}
@@ -2679,7 +2679,7 @@ pub unsafe fn vcreate_s8(a: u64) -> int8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcreate_s16(a: u64) -> int16x4_t {
transmute(a)
}
@@ -2692,7 +2692,7 @@ pub unsafe fn vcreate_s16(a: u64) -> int16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcreate_s32(a: u64) -> int32x2_t {
transmute(a)
}
@@ -2705,7 +2705,7 @@ pub unsafe fn vcreate_s32(a: u64) -> int32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcreate_s64(a: u64) -> int64x1_t {
transmute(a)
}
@@ -2718,7 +2718,7 @@ pub unsafe fn vcreate_s64(a: u64) -> int64x1_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcreate_u8(a: u64) -> uint8x8_t {
transmute(a)
}
@@ -2731,7 +2731,7 @@ pub unsafe fn vcreate_u8(a: u64) -> uint8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcreate_u16(a: u64) -> uint16x4_t {
transmute(a)
}
@@ -2744,7 +2744,7 @@ pub unsafe fn vcreate_u16(a: u64) -> uint16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcreate_u32(a: u64) -> uint32x2_t {
transmute(a)
}
@@ -2757,7 +2757,7 @@ pub unsafe fn vcreate_u32(a: u64) -> uint32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcreate_u64(a: u64) -> uint64x1_t {
transmute(a)
}
@@ -2770,7 +2770,7 @@ pub unsafe fn vcreate_u64(a: u64) -> uint64x1_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcreate_p8(a: u64) -> poly8x8_t {
transmute(a)
}
@@ -2783,7 +2783,7 @@ pub unsafe fn vcreate_p8(a: u64) -> poly8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcreate_p16(a: u64) -> poly16x4_t {
transmute(a)
}
@@ -2796,7 +2796,7 @@ pub unsafe fn vcreate_p16(a: u64) -> poly16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcreate_p64(a: u64) -> poly64x1_t {
transmute(a)
}
@@ -2809,7 +2809,7 @@ pub unsafe fn vcreate_p64(a: u64) -> poly64x1_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcreate_f32(a: u64) -> float32x2_t {
transmute(a)
}
@@ -2822,7 +2822,7 @@ pub unsafe fn vcreate_f32(a: u64) -> float32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vcvt))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(scvtf))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcvt_f32_s32(a: int32x2_t) -> float32x2_t {
simd_cast(a)
}
@@ -2835,7 +2835,7 @@ pub unsafe fn vcvt_f32_s32(a: int32x2_t) -> float32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vcvt))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(scvtf))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcvtq_f32_s32(a: int32x4_t) -> float32x4_t {
simd_cast(a)
}
@@ -2848,7 +2848,7 @@ pub unsafe fn vcvtq_f32_s32(a: int32x4_t) -> float32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vcvt))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ucvtf))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcvt_f32_u32(a: uint32x2_t) -> float32x2_t {
simd_cast(a)
}
@@ -2861,7 +2861,7 @@ pub unsafe fn vcvt_f32_u32(a: uint32x2_t) -> float32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vcvt))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ucvtf))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcvtq_f32_u32(a: uint32x4_t) -> float32x4_t {
simd_cast(a)
}
@@ -3170,7 +3170,7 @@ vcvtq_n_u32_f32_(a, N)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vcvt))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fcvtzs))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcvt_s32_f32(a: float32x2_t) -> int32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -3189,7 +3189,7 @@ vcvt_s32_f32_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vcvt))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fcvtzs))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcvtq_s32_f32(a: float32x4_t) -> int32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -3208,7 +3208,7 @@ vcvtq_s32_f32_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vcvt))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fcvtzu))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcvt_u32_f32(a: float32x2_t) -> uint32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -3227,7 +3227,7 @@ vcvt_u32_f32_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vcvt))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fcvtzu))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vcvtq_u32_f32(a: float32x4_t) -> uint32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -3247,7 +3247,7 @@ vcvtq_u32_f32_(a)
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vdup.8", N = 4))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(dup, N = 4))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vdup_lane_s8<const N: i32>(a: int8x8_t) -> int8x8_t {
static_assert_imm3!(N);
simd_shuffle8!(a, a, <const N: i32> [N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32])
@@ -3262,7 +3262,7 @@ pub unsafe fn vdup_lane_s8<const N: i32>(a: int8x8_t) -> int8x8_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vdup.8", N = 8))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(dup, N = 8))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vdupq_laneq_s8<const N: i32>(a: int8x16_t) -> int8x16_t {
static_assert_imm4!(N);
simd_shuffle16!(a, a, <const N: i32> [N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32])
@@ -3277,7 +3277,7 @@ pub unsafe fn vdupq_laneq_s8<const N: i32>(a: int8x16_t) -> int8x16_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vdup.16", N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(dup, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vdup_lane_s16<const N: i32>(a: int16x4_t) -> int16x4_t {
static_assert_imm2!(N);
simd_shuffle4!(a, a, <const N: i32> [N as u32, N as u32, N as u32, N as u32])
@@ -3292,7 +3292,7 @@ pub unsafe fn vdup_lane_s16<const N: i32>(a: int16x4_t) -> int16x4_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vdup.16", N = 4))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(dup, N = 4))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vdupq_laneq_s16<const N: i32>(a: int16x8_t) -> int16x8_t {
static_assert_imm3!(N);
simd_shuffle8!(a, a, <const N: i32> [N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32])
@@ -3307,7 +3307,7 @@ pub unsafe fn vdupq_laneq_s16<const N: i32>(a: int16x8_t) -> int16x8_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vdup.32", N = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(dup, N = 1))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vdup_lane_s32<const N: i32>(a: int32x2_t) -> int32x2_t {
static_assert_imm1!(N);
simd_shuffle2!(a, a, <const N: i32> [N as u32, N as u32])
@@ -3322,7 +3322,7 @@ pub unsafe fn vdup_lane_s32<const N: i32>(a: int32x2_t) -> int32x2_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vdup.32", N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(dup, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vdupq_laneq_s32<const N: i32>(a: int32x4_t) -> int32x4_t {
static_assert_imm2!(N);
simd_shuffle4!(a, a, <const N: i32> [N as u32, N as u32, N as u32, N as u32])
@@ -3337,7 +3337,7 @@ pub unsafe fn vdupq_laneq_s32<const N: i32>(a: int32x4_t) -> int32x4_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vdup.8", N = 8))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(dup, N = 8))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vdup_laneq_s8<const N: i32>(a: int8x16_t) -> int8x8_t {
static_assert_imm4!(N);
simd_shuffle8!(a, a, <const N: i32> [N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32])
@@ -3352,7 +3352,7 @@ pub unsafe fn vdup_laneq_s8<const N: i32>(a: int8x16_t) -> int8x8_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vdup.16", N = 4))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(dup, N = 4))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vdup_laneq_s16<const N: i32>(a: int16x8_t) -> int16x4_t {
static_assert_imm3!(N);
simd_shuffle4!(a, a, <const N: i32> [N as u32, N as u32, N as u32, N as u32])
@@ -3367,7 +3367,7 @@ pub unsafe fn vdup_laneq_s16<const N: i32>(a: int16x8_t) -> int16x4_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vdup.32", N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(dup, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vdup_laneq_s32<const N: i32>(a: int32x4_t) -> int32x2_t {
static_assert_imm2!(N);
simd_shuffle2!(a, a, <const N: i32> [N as u32, N as u32])
@@ -3382,7 +3382,7 @@ pub unsafe fn vdup_laneq_s32<const N: i32>(a: int32x4_t) -> int32x2_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vdup.8", N = 4))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(dup, N = 4))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vdupq_lane_s8<const N: i32>(a: int8x8_t) -> int8x16_t {
static_assert_imm3!(N);
simd_shuffle16!(a, a, <const N: i32> [N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32])
@@ -3397,7 +3397,7 @@ pub unsafe fn vdupq_lane_s8<const N: i32>(a: int8x8_t) -> int8x16_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vdup.16", N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(dup, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vdupq_lane_s16<const N: i32>(a: int16x4_t) -> int16x8_t {
static_assert_imm2!(N);
simd_shuffle8!(a, a, <const N: i32> [N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32])
@@ -3412,7 +3412,7 @@ pub unsafe fn vdupq_lane_s16<const N: i32>(a: int16x4_t) -> int16x8_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vdup.32", N = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(dup, N = 1))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vdupq_lane_s32<const N: i32>(a: int32x2_t) -> int32x4_t {
static_assert_imm1!(N);
simd_shuffle4!(a, a, <const N: i32> [N as u32, N as u32, N as u32, N as u32])
@@ -3427,7 +3427,7 @@ pub unsafe fn vdupq_lane_s32<const N: i32>(a: int32x2_t) -> int32x4_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vdup.8", N = 4))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(dup, N = 4))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vdup_lane_u8<const N: i32>(a: uint8x8_t) -> uint8x8_t {
static_assert_imm3!(N);
simd_shuffle8!(a, a, <const N: i32> [N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32])
@@ -3442,7 +3442,7 @@ pub unsafe fn vdup_lane_u8<const N: i32>(a: uint8x8_t) -> uint8x8_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vdup.8", N = 8))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(dup, N = 8))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vdupq_laneq_u8<const N: i32>(a: uint8x16_t) -> uint8x16_t {
static_assert_imm4!(N);
simd_shuffle16!(a, a, <const N: i32> [N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32])
@@ -3457,7 +3457,7 @@ pub unsafe fn vdupq_laneq_u8<const N: i32>(a: uint8x16_t) -> uint8x16_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vdup.16", N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(dup, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vdup_lane_u16<const N: i32>(a: uint16x4_t) -> uint16x4_t {
static_assert_imm2!(N);
simd_shuffle4!(a, a, <const N: i32> [N as u32, N as u32, N as u32, N as u32])
@@ -3472,7 +3472,7 @@ pub unsafe fn vdup_lane_u16<const N: i32>(a: uint16x4_t) -> uint16x4_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vdup.16", N = 4))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(dup, N = 4))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vdupq_laneq_u16<const N: i32>(a: uint16x8_t) -> uint16x8_t {
static_assert_imm3!(N);
simd_shuffle8!(a, a, <const N: i32> [N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32])
@@ -3487,7 +3487,7 @@ pub unsafe fn vdupq_laneq_u16<const N: i32>(a: uint16x8_t) -> uint16x8_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vdup.32", N = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(dup, N = 1))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vdup_lane_u32<const N: i32>(a: uint32x2_t) -> uint32x2_t {
static_assert_imm1!(N);
simd_shuffle2!(a, a, <const N: i32> [N as u32, N as u32])
@@ -3502,7 +3502,7 @@ pub unsafe fn vdup_lane_u32<const N: i32>(a: uint32x2_t) -> uint32x2_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vdup.32", N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(dup, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vdupq_laneq_u32<const N: i32>(a: uint32x4_t) -> uint32x4_t {
static_assert_imm2!(N);
simd_shuffle4!(a, a, <const N: i32> [N as u32, N as u32, N as u32, N as u32])
@@ -3517,7 +3517,7 @@ pub unsafe fn vdupq_laneq_u32<const N: i32>(a: uint32x4_t) -> uint32x4_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vdup.8", N = 8))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(dup, N = 8))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vdup_laneq_u8<const N: i32>(a: uint8x16_t) -> uint8x8_t {
static_assert_imm4!(N);
simd_shuffle8!(a, a, <const N: i32> [N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32])
@@ -3532,7 +3532,7 @@ pub unsafe fn vdup_laneq_u8<const N: i32>(a: uint8x16_t) -> uint8x8_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vdup.16", N = 4))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(dup, N = 4))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vdup_laneq_u16<const N: i32>(a: uint16x8_t) -> uint16x4_t {
static_assert_imm3!(N);
simd_shuffle4!(a, a, <const N: i32> [N as u32, N as u32, N as u32, N as u32])
@@ -3547,7 +3547,7 @@ pub unsafe fn vdup_laneq_u16<const N: i32>(a: uint16x8_t) -> uint16x4_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vdup.32", N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(dup, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vdup_laneq_u32<const N: i32>(a: uint32x4_t) -> uint32x2_t {
static_assert_imm2!(N);
simd_shuffle2!(a, a, <const N: i32> [N as u32, N as u32])
@@ -3562,7 +3562,7 @@ pub unsafe fn vdup_laneq_u32<const N: i32>(a: uint32x4_t) -> uint32x2_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vdup.8", N = 4))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(dup, N = 4))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vdupq_lane_u8<const N: i32>(a: uint8x8_t) -> uint8x16_t {
static_assert_imm3!(N);
simd_shuffle16!(a, a, <const N: i32> [N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32])
@@ -3577,7 +3577,7 @@ pub unsafe fn vdupq_lane_u8<const N: i32>(a: uint8x8_t) -> uint8x16_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vdup.16", N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(dup, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vdupq_lane_u16<const N: i32>(a: uint16x4_t) -> uint16x8_t {
static_assert_imm2!(N);
simd_shuffle8!(a, a, <const N: i32> [N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32])
@@ -3592,7 +3592,7 @@ pub unsafe fn vdupq_lane_u16<const N: i32>(a: uint16x4_t) -> uint16x8_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vdup.32", N = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(dup, N = 1))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vdupq_lane_u32<const N: i32>(a: uint32x2_t) -> uint32x4_t {
static_assert_imm1!(N);
simd_shuffle4!(a, a, <const N: i32> [N as u32, N as u32, N as u32, N as u32])
@@ -3607,7 +3607,7 @@ pub unsafe fn vdupq_lane_u32<const N: i32>(a: uint32x2_t) -> uint32x4_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vdup.8", N = 4))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(dup, N = 4))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vdup_lane_p8<const N: i32>(a: poly8x8_t) -> poly8x8_t {
static_assert_imm3!(N);
simd_shuffle8!(a, a, <const N: i32> [N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32])
@@ -3622,7 +3622,7 @@ pub unsafe fn vdup_lane_p8<const N: i32>(a: poly8x8_t) -> poly8x8_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vdup.8", N = 8))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(dup, N = 8))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vdupq_laneq_p8<const N: i32>(a: poly8x16_t) -> poly8x16_t {
static_assert_imm4!(N);
simd_shuffle16!(a, a, <const N: i32> [N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32])
@@ -3637,7 +3637,7 @@ pub unsafe fn vdupq_laneq_p8<const N: i32>(a: poly8x16_t) -> poly8x16_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vdup.16", N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(dup, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vdup_lane_p16<const N: i32>(a: poly16x4_t) -> poly16x4_t {
static_assert_imm2!(N);
simd_shuffle4!(a, a, <const N: i32> [N as u32, N as u32, N as u32, N as u32])
@@ -3652,7 +3652,7 @@ pub unsafe fn vdup_lane_p16<const N: i32>(a: poly16x4_t) -> poly16x4_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vdup.16", N = 4))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(dup, N = 4))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vdupq_laneq_p16<const N: i32>(a: poly16x8_t) -> poly16x8_t {
static_assert_imm3!(N);
simd_shuffle8!(a, a, <const N: i32> [N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32])
@@ -3667,7 +3667,7 @@ pub unsafe fn vdupq_laneq_p16<const N: i32>(a: poly16x8_t) -> poly16x8_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vdup.8", N = 8))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(dup, N = 8))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vdup_laneq_p8<const N: i32>(a: poly8x16_t) -> poly8x8_t {
static_assert_imm4!(N);
simd_shuffle8!(a, a, <const N: i32> [N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32])
@@ -3682,7 +3682,7 @@ pub unsafe fn vdup_laneq_p8<const N: i32>(a: poly8x16_t) -> poly8x8_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vdup.16", N = 4))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(dup, N = 4))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vdup_laneq_p16<const N: i32>(a: poly16x8_t) -> poly16x4_t {
static_assert_imm3!(N);
simd_shuffle4!(a, a, <const N: i32> [N as u32, N as u32, N as u32, N as u32])
@@ -3697,7 +3697,7 @@ pub unsafe fn vdup_laneq_p16<const N: i32>(a: poly16x8_t) -> poly16x4_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vdup.8", N = 4))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(dup, N = 4))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vdupq_lane_p8<const N: i32>(a: poly8x8_t) -> poly8x16_t {
static_assert_imm3!(N);
simd_shuffle16!(a, a, <const N: i32> [N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32])
@@ -3712,7 +3712,7 @@ pub unsafe fn vdupq_lane_p8<const N: i32>(a: poly8x8_t) -> poly8x16_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vdup.16", N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(dup, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vdupq_lane_p16<const N: i32>(a: poly16x4_t) -> poly16x8_t {
static_assert_imm2!(N);
simd_shuffle8!(a, a, <const N: i32> [N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32, N as u32])
@@ -3727,7 +3727,7 @@ pub unsafe fn vdupq_lane_p16<const N: i32>(a: poly16x4_t) -> poly16x8_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmov, N = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(dup, N = 1))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vdupq_laneq_s64<const N: i32>(a: int64x2_t) -> int64x2_t {
static_assert_imm1!(N);
simd_shuffle2!(a, a, <const N: i32> [N as u32, N as u32])
@@ -3742,7 +3742,7 @@ pub unsafe fn vdupq_laneq_s64<const N: i32>(a: int64x2_t) -> int64x2_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmov, N = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(dup, N = 0))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vdupq_lane_s64<const N: i32>(a: int64x1_t) -> int64x2_t {
static_assert!(N : i32 where N == 0);
simd_shuffle2!(a, a, <const N: i32> [N as u32, N as u32])
@@ -3757,7 +3757,7 @@ pub unsafe fn vdupq_lane_s64<const N: i32>(a: int64x1_t) -> int64x2_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmov, N = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(dup, N = 1))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vdupq_laneq_u64<const N: i32>(a: uint64x2_t) -> uint64x2_t {
static_assert_imm1!(N);
simd_shuffle2!(a, a, <const N: i32> [N as u32, N as u32])
@@ -3772,7 +3772,7 @@ pub unsafe fn vdupq_laneq_u64<const N: i32>(a: uint64x2_t) -> uint64x2_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmov, N = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(dup, N = 0))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vdupq_lane_u64<const N: i32>(a: uint64x1_t) -> uint64x2_t {
static_assert!(N : i32 where N == 0);
simd_shuffle2!(a, a, <const N: i32> [N as u32, N as u32])
@@ -3787,7 +3787,7 @@ pub unsafe fn vdupq_lane_u64<const N: i32>(a: uint64x1_t) -> uint64x2_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vdup.32", N = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(dup, N = 1))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vdup_lane_f32<const N: i32>(a: float32x2_t) -> float32x2_t {
static_assert_imm1!(N);
simd_shuffle2!(a, a, <const N: i32> [N as u32, N as u32])
@@ -3802,7 +3802,7 @@ pub unsafe fn vdup_lane_f32<const N: i32>(a: float32x2_t) -> float32x2_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vdup.32", N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(dup, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vdupq_laneq_f32<const N: i32>(a: float32x4_t) -> float32x4_t {
static_assert_imm2!(N);
simd_shuffle4!(a, a, <const N: i32> [N as u32, N as u32, N as u32, N as u32])
@@ -3817,7 +3817,7 @@ pub unsafe fn vdupq_laneq_f32<const N: i32>(a: float32x4_t) -> float32x4_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vdup.32", N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(dup, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vdup_laneq_f32<const N: i32>(a: float32x4_t) -> float32x2_t {
static_assert_imm2!(N);
simd_shuffle2!(a, a, <const N: i32> [N as u32, N as u32])
@@ -3832,7 +3832,7 @@ pub unsafe fn vdup_laneq_f32<const N: i32>(a: float32x4_t) -> float32x2_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vdup.32", N = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(dup, N = 1))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vdupq_lane_f32<const N: i32>(a: float32x2_t) -> float32x4_t {
static_assert_imm1!(N);
simd_shuffle4!(a, a, <const N: i32> [N as u32, N as u32, N as u32, N as u32])
@@ -3847,7 +3847,7 @@ pub unsafe fn vdupq_lane_f32<const N: i32>(a: float32x2_t) -> float32x4_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, N = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, N = 0))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vdup_lane_s64<const N: i32>(a: int64x1_t) -> int64x1_t {
static_assert!(N : i32 where N == 0);
a
@@ -3862,7 +3862,7 @@ pub unsafe fn vdup_lane_s64<const N: i32>(a: int64x1_t) -> int64x1_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, N = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, N = 0))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vdup_lane_u64<const N: i32>(a: uint64x1_t) -> uint64x1_t {
static_assert!(N : i32 where N == 0);
a
@@ -3877,7 +3877,7 @@ pub unsafe fn vdup_lane_u64<const N: i32>(a: uint64x1_t) -> uint64x1_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmov, N = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, N = 1))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vdup_laneq_s64<const N: i32>(a: int64x2_t) -> int64x1_t {
static_assert_imm1!(N);
transmute::<i64, _>(simd_extract(a, N as u32))
@@ -3892,7 +3892,7 @@ pub unsafe fn vdup_laneq_s64<const N: i32>(a: int64x2_t) -> int64x1_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmov, N = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, N = 1))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vdup_laneq_u64<const N: i32>(a: uint64x2_t) -> uint64x1_t {
static_assert_imm1!(N);
transmute::<u64, _>(simd_extract(a, N as u32))
@@ -3907,7 +3907,7 @@ pub unsafe fn vdup_laneq_u64<const N: i32>(a: uint64x2_t) -> uint64x1_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vext.8", N = 7))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ext, N = 7))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vext_s8<const N: i32>(a: int8x8_t, b: int8x8_t) -> int8x8_t {
static_assert_imm3!(N);
match N & 0b111 {
@@ -3932,7 +3932,7 @@ pub unsafe fn vext_s8<const N: i32>(a: int8x8_t, b: int8x8_t) -> int8x8_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vext.8", N = 15))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ext, N = 15))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vextq_s8<const N: i32>(a: int8x16_t, b: int8x16_t) -> int8x16_t {
static_assert_imm4!(N);
match N & 0b1111 {
@@ -3965,7 +3965,7 @@ pub unsafe fn vextq_s8<const N: i32>(a: int8x16_t, b: int8x16_t) -> int8x16_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vext.8", N = 3))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ext, N = 3))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vext_s16<const N: i32>(a: int16x4_t, b: int16x4_t) -> int16x4_t {
static_assert_imm2!(N);
match N & 0b11 {
@@ -3986,7 +3986,7 @@ pub unsafe fn vext_s16<const N: i32>(a: int16x4_t, b: int16x4_t) -> int16x4_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vext.8", N = 7))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ext, N = 7))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vextq_s16<const N: i32>(a: int16x8_t, b: int16x8_t) -> int16x8_t {
static_assert_imm3!(N);
match N & 0b111 {
@@ -4011,7 +4011,7 @@ pub unsafe fn vextq_s16<const N: i32>(a: int16x8_t, b: int16x8_t) -> int16x8_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vext.8", N = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ext, N = 1))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vext_s32<const N: i32>(a: int32x2_t, b: int32x2_t) -> int32x2_t {
static_assert_imm1!(N);
match N & 0b1 {
@@ -4030,7 +4030,7 @@ pub unsafe fn vext_s32<const N: i32>(a: int32x2_t, b: int32x2_t) -> int32x2_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vext.8", N = 3))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ext, N = 3))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vextq_s32<const N: i32>(a: int32x4_t, b: int32x4_t) -> int32x4_t {
static_assert_imm2!(N);
match N & 0b11 {
@@ -4051,7 +4051,7 @@ pub unsafe fn vextq_s32<const N: i32>(a: int32x4_t, b: int32x4_t) -> int32x4_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vext.8", N = 7))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ext, N = 7))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vext_u8<const N: i32>(a: uint8x8_t, b: uint8x8_t) -> uint8x8_t {
static_assert_imm3!(N);
match N & 0b111 {
@@ -4076,7 +4076,7 @@ pub unsafe fn vext_u8<const N: i32>(a: uint8x8_t, b: uint8x8_t) -> uint8x8_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vext.8", N = 15))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ext, N = 15))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vextq_u8<const N: i32>(a: uint8x16_t, b: uint8x16_t) -> uint8x16_t {
static_assert_imm4!(N);
match N & 0b1111 {
@@ -4109,7 +4109,7 @@ pub unsafe fn vextq_u8<const N: i32>(a: uint8x16_t, b: uint8x16_t) -> uint8x16_t
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vext.8", N = 3))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ext, N = 3))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vext_u16<const N: i32>(a: uint16x4_t, b: uint16x4_t) -> uint16x4_t {
static_assert_imm2!(N);
match N & 0b11 {
@@ -4130,7 +4130,7 @@ pub unsafe fn vext_u16<const N: i32>(a: uint16x4_t, b: uint16x4_t) -> uint16x4_t
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vext.8", N = 7))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ext, N = 7))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vextq_u16<const N: i32>(a: uint16x8_t, b: uint16x8_t) -> uint16x8_t {
static_assert_imm3!(N);
match N & 0b111 {
@@ -4155,7 +4155,7 @@ pub unsafe fn vextq_u16<const N: i32>(a: uint16x8_t, b: uint16x8_t) -> uint16x8_
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vext.8", N = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ext, N = 1))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vext_u32<const N: i32>(a: uint32x2_t, b: uint32x2_t) -> uint32x2_t {
static_assert_imm1!(N);
match N & 0b1 {
@@ -4174,7 +4174,7 @@ pub unsafe fn vext_u32<const N: i32>(a: uint32x2_t, b: uint32x2_t) -> uint32x2_t
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vext.8", N = 3))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ext, N = 3))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vextq_u32<const N: i32>(a: uint32x4_t, b: uint32x4_t) -> uint32x4_t {
static_assert_imm2!(N);
match N & 0b11 {
@@ -4195,7 +4195,7 @@ pub unsafe fn vextq_u32<const N: i32>(a: uint32x4_t, b: uint32x4_t) -> uint32x4_
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vext.8", N = 7))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ext, N = 7))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vext_p8<const N: i32>(a: poly8x8_t, b: poly8x8_t) -> poly8x8_t {
static_assert_imm3!(N);
match N & 0b111 {
@@ -4220,7 +4220,7 @@ pub unsafe fn vext_p8<const N: i32>(a: poly8x8_t, b: poly8x8_t) -> poly8x8_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vext.8", N = 15))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ext, N = 15))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vextq_p8<const N: i32>(a: poly8x16_t, b: poly8x16_t) -> poly8x16_t {
static_assert_imm4!(N);
match N & 0b1111 {
@@ -4253,7 +4253,7 @@ pub unsafe fn vextq_p8<const N: i32>(a: poly8x16_t, b: poly8x16_t) -> poly8x16_t
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vext.8", N = 3))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ext, N = 3))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vext_p16<const N: i32>(a: poly16x4_t, b: poly16x4_t) -> poly16x4_t {
static_assert_imm2!(N);
match N & 0b11 {
@@ -4274,7 +4274,7 @@ pub unsafe fn vext_p16<const N: i32>(a: poly16x4_t, b: poly16x4_t) -> poly16x4_t
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vext.8", N = 7))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ext, N = 7))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vextq_p16<const N: i32>(a: poly16x8_t, b: poly16x8_t) -> poly16x8_t {
static_assert_imm3!(N);
match N & 0b111 {
@@ -4299,7 +4299,7 @@ pub unsafe fn vextq_p16<const N: i32>(a: poly16x8_t, b: poly16x8_t) -> poly16x8_
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmov, N = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ext, N = 1))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vextq_s64<const N: i32>(a: int64x2_t, b: int64x2_t) -> int64x2_t {
static_assert_imm1!(N);
match N & 0b1 {
@@ -4318,7 +4318,7 @@ pub unsafe fn vextq_s64<const N: i32>(a: int64x2_t, b: int64x2_t) -> int64x2_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmov, N = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ext, N = 1))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vextq_u64<const N: i32>(a: uint64x2_t, b: uint64x2_t) -> uint64x2_t {
static_assert_imm1!(N);
match N & 0b1 {
@@ -4337,7 +4337,7 @@ pub unsafe fn vextq_u64<const N: i32>(a: uint64x2_t, b: uint64x2_t) -> uint64x2_
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vext.8", N = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ext, N = 1))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vext_f32<const N: i32>(a: float32x2_t, b: float32x2_t) -> float32x2_t {
static_assert_imm1!(N);
match N & 0b1 {
@@ -4356,7 +4356,7 @@ pub unsafe fn vext_f32<const N: i32>(a: float32x2_t, b: float32x2_t) -> float32x
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vext.8", N = 3))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ext, N = 3))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vextq_f32<const N: i32>(a: float32x4_t, b: float32x4_t) -> float32x4_t {
static_assert_imm2!(N);
match N & 0b11 {
@@ -4376,7 +4376,7 @@ pub unsafe fn vextq_f32<const N: i32>(a: float32x4_t, b: float32x4_t) -> float32
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmla.i8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mla))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmla_s8(a: int8x8_t, b: int8x8_t, c: int8x8_t) -> int8x8_t {
simd_add(a, simd_mul(b, c))
}
@@ -4389,7 +4389,7 @@ pub unsafe fn vmla_s8(a: int8x8_t, b: int8x8_t, c: int8x8_t) -> int8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmla.i8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mla))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlaq_s8(a: int8x16_t, b: int8x16_t, c: int8x16_t) -> int8x16_t {
simd_add(a, simd_mul(b, c))
}
@@ -4402,7 +4402,7 @@ pub unsafe fn vmlaq_s8(a: int8x16_t, b: int8x16_t, c: int8x16_t) -> int8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmla.i16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mla))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmla_s16(a: int16x4_t, b: int16x4_t, c: int16x4_t) -> int16x4_t {
simd_add(a, simd_mul(b, c))
}
@@ -4415,7 +4415,7 @@ pub unsafe fn vmla_s16(a: int16x4_t, b: int16x4_t, c: int16x4_t) -> int16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmla.i16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mla))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlaq_s16(a: int16x8_t, b: int16x8_t, c: int16x8_t) -> int16x8_t {
simd_add(a, simd_mul(b, c))
}
@@ -4428,7 +4428,7 @@ pub unsafe fn vmlaq_s16(a: int16x8_t, b: int16x8_t, c: int16x8_t) -> int16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmla.i32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mla))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmla_s32(a: int32x2_t, b: int32x2_t, c: int32x2_t) -> int32x2_t {
simd_add(a, simd_mul(b, c))
}
@@ -4441,7 +4441,7 @@ pub unsafe fn vmla_s32(a: int32x2_t, b: int32x2_t, c: int32x2_t) -> int32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmla.i32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mla))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlaq_s32(a: int32x4_t, b: int32x4_t, c: int32x4_t) -> int32x4_t {
simd_add(a, simd_mul(b, c))
}
@@ -4454,7 +4454,7 @@ pub unsafe fn vmlaq_s32(a: int32x4_t, b: int32x4_t, c: int32x4_t) -> int32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmla.i8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mla))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmla_u8(a: uint8x8_t, b: uint8x8_t, c: uint8x8_t) -> uint8x8_t {
simd_add(a, simd_mul(b, c))
}
@@ -4467,7 +4467,7 @@ pub unsafe fn vmla_u8(a: uint8x8_t, b: uint8x8_t, c: uint8x8_t) -> uint8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmla.i8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mla))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlaq_u8(a: uint8x16_t, b: uint8x16_t, c: uint8x16_t) -> uint8x16_t {
simd_add(a, simd_mul(b, c))
}
@@ -4480,7 +4480,7 @@ pub unsafe fn vmlaq_u8(a: uint8x16_t, b: uint8x16_t, c: uint8x16_t) -> uint8x16_
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmla.i16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mla))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmla_u16(a: uint16x4_t, b: uint16x4_t, c: uint16x4_t) -> uint16x4_t {
simd_add(a, simd_mul(b, c))
}
@@ -4493,7 +4493,7 @@ pub unsafe fn vmla_u16(a: uint16x4_t, b: uint16x4_t, c: uint16x4_t) -> uint16x4_
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmla.i16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mla))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlaq_u16(a: uint16x8_t, b: uint16x8_t, c: uint16x8_t) -> uint16x8_t {
simd_add(a, simd_mul(b, c))
}
@@ -4506,7 +4506,7 @@ pub unsafe fn vmlaq_u16(a: uint16x8_t, b: uint16x8_t, c: uint16x8_t) -> uint16x8
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmla.i32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mla))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmla_u32(a: uint32x2_t, b: uint32x2_t, c: uint32x2_t) -> uint32x2_t {
simd_add(a, simd_mul(b, c))
}
@@ -4519,7 +4519,7 @@ pub unsafe fn vmla_u32(a: uint32x2_t, b: uint32x2_t, c: uint32x2_t) -> uint32x2_
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmla.i32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mla))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlaq_u32(a: uint32x4_t, b: uint32x4_t, c: uint32x4_t) -> uint32x4_t {
simd_add(a, simd_mul(b, c))
}
@@ -4532,7 +4532,7 @@ pub unsafe fn vmlaq_u32(a: uint32x4_t, b: uint32x4_t, c: uint32x4_t) -> uint32x4
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmla.f32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fmul))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmla_f32(a: float32x2_t, b: float32x2_t, c: float32x2_t) -> float32x2_t {
simd_add(a, simd_mul(b, c))
}
@@ -4545,7 +4545,7 @@ pub unsafe fn vmla_f32(a: float32x2_t, b: float32x2_t, c: float32x2_t) -> float3
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmla.f32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fmul))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlaq_f32(a: float32x4_t, b: float32x4_t, c: float32x4_t) -> float32x4_t {
simd_add(a, simd_mul(b, c))
}
@@ -4558,7 +4558,7 @@ pub unsafe fn vmlaq_f32(a: float32x4_t, b: float32x4_t, c: float32x4_t) -> float
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmla.i16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mla))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmla_n_s16(a: int16x4_t, b: int16x4_t, c: i16) -> int16x4_t {
vmla_s16(a, b, vdup_n_s16(c))
}
@@ -4571,7 +4571,7 @@ pub unsafe fn vmla_n_s16(a: int16x4_t, b: int16x4_t, c: i16) -> int16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmla.i16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mla))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlaq_n_s16(a: int16x8_t, b: int16x8_t, c: i16) -> int16x8_t {
vmlaq_s16(a, b, vdupq_n_s16(c))
}
@@ -4584,7 +4584,7 @@ pub unsafe fn vmlaq_n_s16(a: int16x8_t, b: int16x8_t, c: i16) -> int16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmla.i32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mla))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmla_n_s32(a: int32x2_t, b: int32x2_t, c: i32) -> int32x2_t {
vmla_s32(a, b, vdup_n_s32(c))
}
@@ -4597,7 +4597,7 @@ pub unsafe fn vmla_n_s32(a: int32x2_t, b: int32x2_t, c: i32) -> int32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmla.i32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mla))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlaq_n_s32(a: int32x4_t, b: int32x4_t, c: i32) -> int32x4_t {
vmlaq_s32(a, b, vdupq_n_s32(c))
}
@@ -4610,7 +4610,7 @@ pub unsafe fn vmlaq_n_s32(a: int32x4_t, b: int32x4_t, c: i32) -> int32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmla.i16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mla))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmla_n_u16(a: uint16x4_t, b: uint16x4_t, c: u16) -> uint16x4_t {
vmla_u16(a, b, vdup_n_u16(c))
}
@@ -4623,7 +4623,7 @@ pub unsafe fn vmla_n_u16(a: uint16x4_t, b: uint16x4_t, c: u16) -> uint16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmla.i16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mla))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlaq_n_u16(a: uint16x8_t, b: uint16x8_t, c: u16) -> uint16x8_t {
vmlaq_u16(a, b, vdupq_n_u16(c))
}
@@ -4636,7 +4636,7 @@ pub unsafe fn vmlaq_n_u16(a: uint16x8_t, b: uint16x8_t, c: u16) -> uint16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmla.i32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mla))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmla_n_u32(a: uint32x2_t, b: uint32x2_t, c: u32) -> uint32x2_t {
vmla_u32(a, b, vdup_n_u32(c))
}
@@ -4649,7 +4649,7 @@ pub unsafe fn vmla_n_u32(a: uint32x2_t, b: uint32x2_t, c: u32) -> uint32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmla.i32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mla))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlaq_n_u32(a: uint32x4_t, b: uint32x4_t, c: u32) -> uint32x4_t {
vmlaq_u32(a, b, vdupq_n_u32(c))
}
@@ -4662,7 +4662,7 @@ pub unsafe fn vmlaq_n_u32(a: uint32x4_t, b: uint32x4_t, c: u32) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmla.f32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fmul))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmla_n_f32(a: float32x2_t, b: float32x2_t, c: f32) -> float32x2_t {
vmla_f32(a, b, vdup_n_f32(c))
}
@@ -4675,7 +4675,7 @@ pub unsafe fn vmla_n_f32(a: float32x2_t, b: float32x2_t, c: f32) -> float32x2_t
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmla.f32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fmul))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlaq_n_f32(a: float32x4_t, b: float32x4_t, c: f32) -> float32x4_t {
vmlaq_f32(a, b, vdupq_n_f32(c))
}
@@ -4689,7 +4689,7 @@ pub unsafe fn vmlaq_n_f32(a: float32x4_t, b: float32x4_t, c: f32) -> float32x4_t
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmla.i16", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mla, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmla_lane_s16<const LANE: i32>(a: int16x4_t, b: int16x4_t, c: int16x4_t) -> int16x4_t {
static_assert_imm2!(LANE);
vmla_s16(a, b, simd_shuffle4!(c, c, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -4704,7 +4704,7 @@ pub unsafe fn vmla_lane_s16<const LANE: i32>(a: int16x4_t, b: int16x4_t, c: int1
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmla.i16", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mla, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmla_laneq_s16<const LANE: i32>(a: int16x4_t, b: int16x4_t, c: int16x8_t) -> int16x4_t {
static_assert_imm3!(LANE);
vmla_s16(a, b, simd_shuffle4!(c, c, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -4719,7 +4719,7 @@ pub unsafe fn vmla_laneq_s16<const LANE: i32>(a: int16x4_t, b: int16x4_t, c: int
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmla.i16", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mla, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlaq_lane_s16<const LANE: i32>(a: int16x8_t, b: int16x8_t, c: int16x4_t) -> int16x8_t {
static_assert_imm2!(LANE);
vmlaq_s16(a, b, simd_shuffle8!(c, c, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -4734,7 +4734,7 @@ pub unsafe fn vmlaq_lane_s16<const LANE: i32>(a: int16x8_t, b: int16x8_t, c: int
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmla.i16", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mla, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlaq_laneq_s16<const LANE: i32>(a: int16x8_t, b: int16x8_t, c: int16x8_t) -> int16x8_t {
static_assert_imm3!(LANE);
vmlaq_s16(a, b, simd_shuffle8!(c, c, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -4749,7 +4749,7 @@ pub unsafe fn vmlaq_laneq_s16<const LANE: i32>(a: int16x8_t, b: int16x8_t, c: in
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmla.i32", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mla, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmla_lane_s32<const LANE: i32>(a: int32x2_t, b: int32x2_t, c: int32x2_t) -> int32x2_t {
static_assert_imm1!(LANE);
vmla_s32(a, b, simd_shuffle2!(c, c, <const LANE: i32> [LANE as u32, LANE as u32]))
@@ -4764,7 +4764,7 @@ pub unsafe fn vmla_lane_s32<const LANE: i32>(a: int32x2_t, b: int32x2_t, c: int3
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmla.i32", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mla, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmla_laneq_s32<const LANE: i32>(a: int32x2_t, b: int32x2_t, c: int32x4_t) -> int32x2_t {
static_assert_imm2!(LANE);
vmla_s32(a, b, simd_shuffle2!(c, c, <const LANE: i32> [LANE as u32, LANE as u32]))
@@ -4779,7 +4779,7 @@ pub unsafe fn vmla_laneq_s32<const LANE: i32>(a: int32x2_t, b: int32x2_t, c: int
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmla.i32", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mla, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlaq_lane_s32<const LANE: i32>(a: int32x4_t, b: int32x4_t, c: int32x2_t) -> int32x4_t {
static_assert_imm1!(LANE);
vmlaq_s32(a, b, simd_shuffle4!(c, c, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -4794,7 +4794,7 @@ pub unsafe fn vmlaq_lane_s32<const LANE: i32>(a: int32x4_t, b: int32x4_t, c: int
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmla.i32", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mla, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlaq_laneq_s32<const LANE: i32>(a: int32x4_t, b: int32x4_t, c: int32x4_t) -> int32x4_t {
static_assert_imm2!(LANE);
vmlaq_s32(a, b, simd_shuffle4!(c, c, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -4809,7 +4809,7 @@ pub unsafe fn vmlaq_laneq_s32<const LANE: i32>(a: int32x4_t, b: int32x4_t, c: in
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmla.i16", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mla, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmla_lane_u16<const LANE: i32>(a: uint16x4_t, b: uint16x4_t, c: uint16x4_t) -> uint16x4_t {
static_assert_imm2!(LANE);
vmla_u16(a, b, simd_shuffle4!(c, c, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -4824,7 +4824,7 @@ pub unsafe fn vmla_lane_u16<const LANE: i32>(a: uint16x4_t, b: uint16x4_t, c: ui
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmla.i16", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mla, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmla_laneq_u16<const LANE: i32>(a: uint16x4_t, b: uint16x4_t, c: uint16x8_t) -> uint16x4_t {
static_assert_imm3!(LANE);
vmla_u16(a, b, simd_shuffle4!(c, c, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -4839,7 +4839,7 @@ pub unsafe fn vmla_laneq_u16<const LANE: i32>(a: uint16x4_t, b: uint16x4_t, c: u
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmla.i16", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mla, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlaq_lane_u16<const LANE: i32>(a: uint16x8_t, b: uint16x8_t, c: uint16x4_t) -> uint16x8_t {
static_assert_imm2!(LANE);
vmlaq_u16(a, b, simd_shuffle8!(c, c, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -4854,7 +4854,7 @@ pub unsafe fn vmlaq_lane_u16<const LANE: i32>(a: uint16x8_t, b: uint16x8_t, c: u
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmla.i16", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mla, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlaq_laneq_u16<const LANE: i32>(a: uint16x8_t, b: uint16x8_t, c: uint16x8_t) -> uint16x8_t {
static_assert_imm3!(LANE);
vmlaq_u16(a, b, simd_shuffle8!(c, c, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -4869,7 +4869,7 @@ pub unsafe fn vmlaq_laneq_u16<const LANE: i32>(a: uint16x8_t, b: uint16x8_t, c:
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmla.i32", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mla, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmla_lane_u32<const LANE: i32>(a: uint32x2_t, b: uint32x2_t, c: uint32x2_t) -> uint32x2_t {
static_assert_imm1!(LANE);
vmla_u32(a, b, simd_shuffle2!(c, c, <const LANE: i32> [LANE as u32, LANE as u32]))
@@ -4884,7 +4884,7 @@ pub unsafe fn vmla_lane_u32<const LANE: i32>(a: uint32x2_t, b: uint32x2_t, c: ui
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmla.i32", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mla, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmla_laneq_u32<const LANE: i32>(a: uint32x2_t, b: uint32x2_t, c: uint32x4_t) -> uint32x2_t {
static_assert_imm2!(LANE);
vmla_u32(a, b, simd_shuffle2!(c, c, <const LANE: i32> [LANE as u32, LANE as u32]))
@@ -4899,7 +4899,7 @@ pub unsafe fn vmla_laneq_u32<const LANE: i32>(a: uint32x2_t, b: uint32x2_t, c: u
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmla.i32", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mla, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlaq_lane_u32<const LANE: i32>(a: uint32x4_t, b: uint32x4_t, c: uint32x2_t) -> uint32x4_t {
static_assert_imm1!(LANE);
vmlaq_u32(a, b, simd_shuffle4!(c, c, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -4914,7 +4914,7 @@ pub unsafe fn vmlaq_lane_u32<const LANE: i32>(a: uint32x4_t, b: uint32x4_t, c: u
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmla.i32", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mla, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlaq_laneq_u32<const LANE: i32>(a: uint32x4_t, b: uint32x4_t, c: uint32x4_t) -> uint32x4_t {
static_assert_imm2!(LANE);
vmlaq_u32(a, b, simd_shuffle4!(c, c, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -4929,7 +4929,7 @@ pub unsafe fn vmlaq_laneq_u32<const LANE: i32>(a: uint32x4_t, b: uint32x4_t, c:
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmla.f32", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fmul, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmla_lane_f32<const LANE: i32>(a: float32x2_t, b: float32x2_t, c: float32x2_t) -> float32x2_t {
static_assert_imm1!(LANE);
vmla_f32(a, b, simd_shuffle2!(c, c, <const LANE: i32> [LANE as u32, LANE as u32]))
@@ -4944,7 +4944,7 @@ pub unsafe fn vmla_lane_f32<const LANE: i32>(a: float32x2_t, b: float32x2_t, c:
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmla.f32", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fmul, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmla_laneq_f32<const LANE: i32>(a: float32x2_t, b: float32x2_t, c: float32x4_t) -> float32x2_t {
static_assert_imm2!(LANE);
vmla_f32(a, b, simd_shuffle2!(c, c, <const LANE: i32> [LANE as u32, LANE as u32]))
@@ -4959,7 +4959,7 @@ pub unsafe fn vmla_laneq_f32<const LANE: i32>(a: float32x2_t, b: float32x2_t, c:
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmla.f32", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fmul, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlaq_lane_f32<const LANE: i32>(a: float32x4_t, b: float32x4_t, c: float32x2_t) -> float32x4_t {
static_assert_imm1!(LANE);
vmlaq_f32(a, b, simd_shuffle4!(c, c, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -4974,7 +4974,7 @@ pub unsafe fn vmlaq_lane_f32<const LANE: i32>(a: float32x4_t, b: float32x4_t, c:
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmla.f32", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fmul, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlaq_laneq_f32<const LANE: i32>(a: float32x4_t, b: float32x4_t, c: float32x4_t) -> float32x4_t {
static_assert_imm2!(LANE);
vmlaq_f32(a, b, simd_shuffle4!(c, c, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -4988,7 +4988,7 @@ pub unsafe fn vmlaq_laneq_f32<const LANE: i32>(a: float32x4_t, b: float32x4_t, c
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmlal.s8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(smlal))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlal_s8(a: int16x8_t, b: int8x8_t, c: int8x8_t) -> int16x8_t {
simd_add(a, vmull_s8(b, c))
}
@@ -5001,7 +5001,7 @@ pub unsafe fn vmlal_s8(a: int16x8_t, b: int8x8_t, c: int8x8_t) -> int16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmlal.s16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(smlal))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlal_s16(a: int32x4_t, b: int16x4_t, c: int16x4_t) -> int32x4_t {
simd_add(a, vmull_s16(b, c))
}
@@ -5014,7 +5014,7 @@ pub unsafe fn vmlal_s16(a: int32x4_t, b: int16x4_t, c: int16x4_t) -> int32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmlal.s32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(smlal))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlal_s32(a: int64x2_t, b: int32x2_t, c: int32x2_t) -> int64x2_t {
simd_add(a, vmull_s32(b, c))
}
@@ -5027,7 +5027,7 @@ pub unsafe fn vmlal_s32(a: int64x2_t, b: int32x2_t, c: int32x2_t) -> int64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmlal.u8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(umlal))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlal_u8(a: uint16x8_t, b: uint8x8_t, c: uint8x8_t) -> uint16x8_t {
simd_add(a, vmull_u8(b, c))
}
@@ -5040,7 +5040,7 @@ pub unsafe fn vmlal_u8(a: uint16x8_t, b: uint8x8_t, c: uint8x8_t) -> uint16x8_t
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmlal.u16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(umlal))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlal_u16(a: uint32x4_t, b: uint16x4_t, c: uint16x4_t) -> uint32x4_t {
simd_add(a, vmull_u16(b, c))
}
@@ -5053,7 +5053,7 @@ pub unsafe fn vmlal_u16(a: uint32x4_t, b: uint16x4_t, c: uint16x4_t) -> uint32x4
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmlal.u32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(umlal))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlal_u32(a: uint64x2_t, b: uint32x2_t, c: uint32x2_t) -> uint64x2_t {
simd_add(a, vmull_u32(b, c))
}
@@ -5066,7 +5066,7 @@ pub unsafe fn vmlal_u32(a: uint64x2_t, b: uint32x2_t, c: uint32x2_t) -> uint64x2
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmlal.s16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(smlal))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlal_n_s16(a: int32x4_t, b: int16x4_t, c: i16) -> int32x4_t {
vmlal_s16(a, b, vdup_n_s16(c))
}
@@ -5079,7 +5079,7 @@ pub unsafe fn vmlal_n_s16(a: int32x4_t, b: int16x4_t, c: i16) -> int32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmlal.s32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(smlal))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlal_n_s32(a: int64x2_t, b: int32x2_t, c: i32) -> int64x2_t {
vmlal_s32(a, b, vdup_n_s32(c))
}
@@ -5092,7 +5092,7 @@ pub unsafe fn vmlal_n_s32(a: int64x2_t, b: int32x2_t, c: i32) -> int64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmlal.u16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(umlal))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlal_n_u16(a: uint32x4_t, b: uint16x4_t, c: u16) -> uint32x4_t {
vmlal_u16(a, b, vdup_n_u16(c))
}
@@ -5105,7 +5105,7 @@ pub unsafe fn vmlal_n_u16(a: uint32x4_t, b: uint16x4_t, c: u16) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmlal.u32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(umlal))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlal_n_u32(a: uint64x2_t, b: uint32x2_t, c: u32) -> uint64x2_t {
vmlal_u32(a, b, vdup_n_u32(c))
}
@@ -5119,7 +5119,7 @@ pub unsafe fn vmlal_n_u32(a: uint64x2_t, b: uint32x2_t, c: u32) -> uint64x2_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmlal.s16", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(smlal, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlal_lane_s16<const LANE: i32>(a: int32x4_t, b: int16x4_t, c: int16x4_t) -> int32x4_t {
static_assert_imm2!(LANE);
vmlal_s16(a, b, simd_shuffle4!(c, c, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -5134,7 +5134,7 @@ pub unsafe fn vmlal_lane_s16<const LANE: i32>(a: int32x4_t, b: int16x4_t, c: int
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmlal.s16", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(smlal, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlal_laneq_s16<const LANE: i32>(a: int32x4_t, b: int16x4_t, c: int16x8_t) -> int32x4_t {
static_assert_imm3!(LANE);
vmlal_s16(a, b, simd_shuffle4!(c, c, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -5149,7 +5149,7 @@ pub unsafe fn vmlal_laneq_s16<const LANE: i32>(a: int32x4_t, b: int16x4_t, c: in
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmlal.s32", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(smlal, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlal_lane_s32<const LANE: i32>(a: int64x2_t, b: int32x2_t, c: int32x2_t) -> int64x2_t {
static_assert_imm1!(LANE);
vmlal_s32(a, b, simd_shuffle2!(c, c, <const LANE: i32> [LANE as u32, LANE as u32]))
@@ -5164,7 +5164,7 @@ pub unsafe fn vmlal_lane_s32<const LANE: i32>(a: int64x2_t, b: int32x2_t, c: int
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmlal.s32", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(smlal, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlal_laneq_s32<const LANE: i32>(a: int64x2_t, b: int32x2_t, c: int32x4_t) -> int64x2_t {
static_assert_imm2!(LANE);
vmlal_s32(a, b, simd_shuffle2!(c, c, <const LANE: i32> [LANE as u32, LANE as u32]))
@@ -5179,7 +5179,7 @@ pub unsafe fn vmlal_laneq_s32<const LANE: i32>(a: int64x2_t, b: int32x2_t, c: in
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmlal.u16", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(umlal, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlal_lane_u16<const LANE: i32>(a: uint32x4_t, b: uint16x4_t, c: uint16x4_t) -> uint32x4_t {
static_assert_imm2!(LANE);
vmlal_u16(a, b, simd_shuffle4!(c, c, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -5194,7 +5194,7 @@ pub unsafe fn vmlal_lane_u16<const LANE: i32>(a: uint32x4_t, b: uint16x4_t, c: u
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmlal.u16", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(umlal, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlal_laneq_u16<const LANE: i32>(a: uint32x4_t, b: uint16x4_t, c: uint16x8_t) -> uint32x4_t {
static_assert_imm3!(LANE);
vmlal_u16(a, b, simd_shuffle4!(c, c, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -5209,7 +5209,7 @@ pub unsafe fn vmlal_laneq_u16<const LANE: i32>(a: uint32x4_t, b: uint16x4_t, c:
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmlal.u32", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(umlal, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlal_lane_u32<const LANE: i32>(a: uint64x2_t, b: uint32x2_t, c: uint32x2_t) -> uint64x2_t {
static_assert_imm1!(LANE);
vmlal_u32(a, b, simd_shuffle2!(c, c, <const LANE: i32> [LANE as u32, LANE as u32]))
@@ -5224,7 +5224,7 @@ pub unsafe fn vmlal_lane_u32<const LANE: i32>(a: uint64x2_t, b: uint32x2_t, c: u
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmlal.u32", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(umlal, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlal_laneq_u32<const LANE: i32>(a: uint64x2_t, b: uint32x2_t, c: uint32x4_t) -> uint64x2_t {
static_assert_imm2!(LANE);
vmlal_u32(a, b, simd_shuffle2!(c, c, <const LANE: i32> [LANE as u32, LANE as u32]))
@@ -5238,7 +5238,7 @@ pub unsafe fn vmlal_laneq_u32<const LANE: i32>(a: uint64x2_t, b: uint32x2_t, c:
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmls.i8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mls))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmls_s8(a: int8x8_t, b: int8x8_t, c: int8x8_t) -> int8x8_t {
simd_sub(a, simd_mul(b, c))
}
@@ -5251,7 +5251,7 @@ pub unsafe fn vmls_s8(a: int8x8_t, b: int8x8_t, c: int8x8_t) -> int8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmls.i8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mls))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlsq_s8(a: int8x16_t, b: int8x16_t, c: int8x16_t) -> int8x16_t {
simd_sub(a, simd_mul(b, c))
}
@@ -5264,7 +5264,7 @@ pub unsafe fn vmlsq_s8(a: int8x16_t, b: int8x16_t, c: int8x16_t) -> int8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmls.i16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mls))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmls_s16(a: int16x4_t, b: int16x4_t, c: int16x4_t) -> int16x4_t {
simd_sub(a, simd_mul(b, c))
}
@@ -5277,7 +5277,7 @@ pub unsafe fn vmls_s16(a: int16x4_t, b: int16x4_t, c: int16x4_t) -> int16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmls.i16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mls))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlsq_s16(a: int16x8_t, b: int16x8_t, c: int16x8_t) -> int16x8_t {
simd_sub(a, simd_mul(b, c))
}
@@ -5290,7 +5290,7 @@ pub unsafe fn vmlsq_s16(a: int16x8_t, b: int16x8_t, c: int16x8_t) -> int16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmls.i32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mls))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmls_s32(a: int32x2_t, b: int32x2_t, c: int32x2_t) -> int32x2_t {
simd_sub(a, simd_mul(b, c))
}
@@ -5303,7 +5303,7 @@ pub unsafe fn vmls_s32(a: int32x2_t, b: int32x2_t, c: int32x2_t) -> int32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmls.i32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mls))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlsq_s32(a: int32x4_t, b: int32x4_t, c: int32x4_t) -> int32x4_t {
simd_sub(a, simd_mul(b, c))
}
@@ -5316,7 +5316,7 @@ pub unsafe fn vmlsq_s32(a: int32x4_t, b: int32x4_t, c: int32x4_t) -> int32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmls.i8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mls))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmls_u8(a: uint8x8_t, b: uint8x8_t, c: uint8x8_t) -> uint8x8_t {
simd_sub(a, simd_mul(b, c))
}
@@ -5329,7 +5329,7 @@ pub unsafe fn vmls_u8(a: uint8x8_t, b: uint8x8_t, c: uint8x8_t) -> uint8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmls.i8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mls))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlsq_u8(a: uint8x16_t, b: uint8x16_t, c: uint8x16_t) -> uint8x16_t {
simd_sub(a, simd_mul(b, c))
}
@@ -5342,7 +5342,7 @@ pub unsafe fn vmlsq_u8(a: uint8x16_t, b: uint8x16_t, c: uint8x16_t) -> uint8x16_
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmls.i16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mls))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmls_u16(a: uint16x4_t, b: uint16x4_t, c: uint16x4_t) -> uint16x4_t {
simd_sub(a, simd_mul(b, c))
}
@@ -5355,7 +5355,7 @@ pub unsafe fn vmls_u16(a: uint16x4_t, b: uint16x4_t, c: uint16x4_t) -> uint16x4_
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmls.i16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mls))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlsq_u16(a: uint16x8_t, b: uint16x8_t, c: uint16x8_t) -> uint16x8_t {
simd_sub(a, simd_mul(b, c))
}
@@ -5368,7 +5368,7 @@ pub unsafe fn vmlsq_u16(a: uint16x8_t, b: uint16x8_t, c: uint16x8_t) -> uint16x8
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmls.i32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mls))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmls_u32(a: uint32x2_t, b: uint32x2_t, c: uint32x2_t) -> uint32x2_t {
simd_sub(a, simd_mul(b, c))
}
@@ -5381,7 +5381,7 @@ pub unsafe fn vmls_u32(a: uint32x2_t, b: uint32x2_t, c: uint32x2_t) -> uint32x2_
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmls.i32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mls))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlsq_u32(a: uint32x4_t, b: uint32x4_t, c: uint32x4_t) -> uint32x4_t {
simd_sub(a, simd_mul(b, c))
}
@@ -5394,7 +5394,7 @@ pub unsafe fn vmlsq_u32(a: uint32x4_t, b: uint32x4_t, c: uint32x4_t) -> uint32x4
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmls.f32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fmul))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmls_f32(a: float32x2_t, b: float32x2_t, c: float32x2_t) -> float32x2_t {
simd_sub(a, simd_mul(b, c))
}
@@ -5407,7 +5407,7 @@ pub unsafe fn vmls_f32(a: float32x2_t, b: float32x2_t, c: float32x2_t) -> float3
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmls.f32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fmul))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlsq_f32(a: float32x4_t, b: float32x4_t, c: float32x4_t) -> float32x4_t {
simd_sub(a, simd_mul(b, c))
}
@@ -5420,7 +5420,7 @@ pub unsafe fn vmlsq_f32(a: float32x4_t, b: float32x4_t, c: float32x4_t) -> float
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmls.i16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mls))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmls_n_s16(a: int16x4_t, b: int16x4_t, c: i16) -> int16x4_t {
vmls_s16(a, b, vdup_n_s16(c))
}
@@ -5433,7 +5433,7 @@ pub unsafe fn vmls_n_s16(a: int16x4_t, b: int16x4_t, c: i16) -> int16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmls.i16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mls))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlsq_n_s16(a: int16x8_t, b: int16x8_t, c: i16) -> int16x8_t {
vmlsq_s16(a, b, vdupq_n_s16(c))
}
@@ -5446,7 +5446,7 @@ pub unsafe fn vmlsq_n_s16(a: int16x8_t, b: int16x8_t, c: i16) -> int16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmls.i32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mls))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmls_n_s32(a: int32x2_t, b: int32x2_t, c: i32) -> int32x2_t {
vmls_s32(a, b, vdup_n_s32(c))
}
@@ -5459,7 +5459,7 @@ pub unsafe fn vmls_n_s32(a: int32x2_t, b: int32x2_t, c: i32) -> int32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmls.i32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mls))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlsq_n_s32(a: int32x4_t, b: int32x4_t, c: i32) -> int32x4_t {
vmlsq_s32(a, b, vdupq_n_s32(c))
}
@@ -5472,7 +5472,7 @@ pub unsafe fn vmlsq_n_s32(a: int32x4_t, b: int32x4_t, c: i32) -> int32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmls.i16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mls))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmls_n_u16(a: uint16x4_t, b: uint16x4_t, c: u16) -> uint16x4_t {
vmls_u16(a, b, vdup_n_u16(c))
}
@@ -5485,7 +5485,7 @@ pub unsafe fn vmls_n_u16(a: uint16x4_t, b: uint16x4_t, c: u16) -> uint16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmls.i16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mls))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlsq_n_u16(a: uint16x8_t, b: uint16x8_t, c: u16) -> uint16x8_t {
vmlsq_u16(a, b, vdupq_n_u16(c))
}
@@ -5498,7 +5498,7 @@ pub unsafe fn vmlsq_n_u16(a: uint16x8_t, b: uint16x8_t, c: u16) -> uint16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmls.i32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mls))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmls_n_u32(a: uint32x2_t, b: uint32x2_t, c: u32) -> uint32x2_t {
vmls_u32(a, b, vdup_n_u32(c))
}
@@ -5511,7 +5511,7 @@ pub unsafe fn vmls_n_u32(a: uint32x2_t, b: uint32x2_t, c: u32) -> uint32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmls.i32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mls))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlsq_n_u32(a: uint32x4_t, b: uint32x4_t, c: u32) -> uint32x4_t {
vmlsq_u32(a, b, vdupq_n_u32(c))
}
@@ -5524,7 +5524,7 @@ pub unsafe fn vmlsq_n_u32(a: uint32x4_t, b: uint32x4_t, c: u32) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmls.f32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fmul))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmls_n_f32(a: float32x2_t, b: float32x2_t, c: f32) -> float32x2_t {
vmls_f32(a, b, vdup_n_f32(c))
}
@@ -5537,7 +5537,7 @@ pub unsafe fn vmls_n_f32(a: float32x2_t, b: float32x2_t, c: f32) -> float32x2_t
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmls.f32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fmul))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlsq_n_f32(a: float32x4_t, b: float32x4_t, c: f32) -> float32x4_t {
vmlsq_f32(a, b, vdupq_n_f32(c))
}
@@ -5551,7 +5551,7 @@ pub unsafe fn vmlsq_n_f32(a: float32x4_t, b: float32x4_t, c: f32) -> float32x4_t
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmls.i16", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mls, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmls_lane_s16<const LANE: i32>(a: int16x4_t, b: int16x4_t, c: int16x4_t) -> int16x4_t {
static_assert_imm2!(LANE);
vmls_s16(a, b, simd_shuffle4!(c, c, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -5566,7 +5566,7 @@ pub unsafe fn vmls_lane_s16<const LANE: i32>(a: int16x4_t, b: int16x4_t, c: int1
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmls.i16", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mls, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmls_laneq_s16<const LANE: i32>(a: int16x4_t, b: int16x4_t, c: int16x8_t) -> int16x4_t {
static_assert_imm3!(LANE);
vmls_s16(a, b, simd_shuffle4!(c, c, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -5581,7 +5581,7 @@ pub unsafe fn vmls_laneq_s16<const LANE: i32>(a: int16x4_t, b: int16x4_t, c: int
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmls.i16", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mls, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlsq_lane_s16<const LANE: i32>(a: int16x8_t, b: int16x8_t, c: int16x4_t) -> int16x8_t {
static_assert_imm2!(LANE);
vmlsq_s16(a, b, simd_shuffle8!(c, c, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -5596,7 +5596,7 @@ pub unsafe fn vmlsq_lane_s16<const LANE: i32>(a: int16x8_t, b: int16x8_t, c: int
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmls.i16", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mls, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlsq_laneq_s16<const LANE: i32>(a: int16x8_t, b: int16x8_t, c: int16x8_t) -> int16x8_t {
static_assert_imm3!(LANE);
vmlsq_s16(a, b, simd_shuffle8!(c, c, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -5611,7 +5611,7 @@ pub unsafe fn vmlsq_laneq_s16<const LANE: i32>(a: int16x8_t, b: int16x8_t, c: in
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmls.i32", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mls, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmls_lane_s32<const LANE: i32>(a: int32x2_t, b: int32x2_t, c: int32x2_t) -> int32x2_t {
static_assert_imm1!(LANE);
vmls_s32(a, b, simd_shuffle2!(c, c, <const LANE: i32> [LANE as u32, LANE as u32]))
@@ -5626,7 +5626,7 @@ pub unsafe fn vmls_lane_s32<const LANE: i32>(a: int32x2_t, b: int32x2_t, c: int3
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmls.i32", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mls, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmls_laneq_s32<const LANE: i32>(a: int32x2_t, b: int32x2_t, c: int32x4_t) -> int32x2_t {
static_assert_imm2!(LANE);
vmls_s32(a, b, simd_shuffle2!(c, c, <const LANE: i32> [LANE as u32, LANE as u32]))
@@ -5641,7 +5641,7 @@ pub unsafe fn vmls_laneq_s32<const LANE: i32>(a: int32x2_t, b: int32x2_t, c: int
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmls.i32", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mls, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlsq_lane_s32<const LANE: i32>(a: int32x4_t, b: int32x4_t, c: int32x2_t) -> int32x4_t {
static_assert_imm1!(LANE);
vmlsq_s32(a, b, simd_shuffle4!(c, c, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -5656,7 +5656,7 @@ pub unsafe fn vmlsq_lane_s32<const LANE: i32>(a: int32x4_t, b: int32x4_t, c: int
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmls.i32", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mls, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlsq_laneq_s32<const LANE: i32>(a: int32x4_t, b: int32x4_t, c: int32x4_t) -> int32x4_t {
static_assert_imm2!(LANE);
vmlsq_s32(a, b, simd_shuffle4!(c, c, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -5671,7 +5671,7 @@ pub unsafe fn vmlsq_laneq_s32<const LANE: i32>(a: int32x4_t, b: int32x4_t, c: in
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmls.i16", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mls, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmls_lane_u16<const LANE: i32>(a: uint16x4_t, b: uint16x4_t, c: uint16x4_t) -> uint16x4_t {
static_assert_imm2!(LANE);
vmls_u16(a, b, simd_shuffle4!(c, c, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -5686,7 +5686,7 @@ pub unsafe fn vmls_lane_u16<const LANE: i32>(a: uint16x4_t, b: uint16x4_t, c: ui
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmls.i16", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mls, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmls_laneq_u16<const LANE: i32>(a: uint16x4_t, b: uint16x4_t, c: uint16x8_t) -> uint16x4_t {
static_assert_imm3!(LANE);
vmls_u16(a, b, simd_shuffle4!(c, c, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -5701,7 +5701,7 @@ pub unsafe fn vmls_laneq_u16<const LANE: i32>(a: uint16x4_t, b: uint16x4_t, c: u
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmls.i16", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mls, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlsq_lane_u16<const LANE: i32>(a: uint16x8_t, b: uint16x8_t, c: uint16x4_t) -> uint16x8_t {
static_assert_imm2!(LANE);
vmlsq_u16(a, b, simd_shuffle8!(c, c, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -5716,7 +5716,7 @@ pub unsafe fn vmlsq_lane_u16<const LANE: i32>(a: uint16x8_t, b: uint16x8_t, c: u
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmls.i16", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mls, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlsq_laneq_u16<const LANE: i32>(a: uint16x8_t, b: uint16x8_t, c: uint16x8_t) -> uint16x8_t {
static_assert_imm3!(LANE);
vmlsq_u16(a, b, simd_shuffle8!(c, c, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -5731,7 +5731,7 @@ pub unsafe fn vmlsq_laneq_u16<const LANE: i32>(a: uint16x8_t, b: uint16x8_t, c:
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmls.i32", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mls, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmls_lane_u32<const LANE: i32>(a: uint32x2_t, b: uint32x2_t, c: uint32x2_t) -> uint32x2_t {
static_assert_imm1!(LANE);
vmls_u32(a, b, simd_shuffle2!(c, c, <const LANE: i32> [LANE as u32, LANE as u32]))
@@ -5746,7 +5746,7 @@ pub unsafe fn vmls_lane_u32<const LANE: i32>(a: uint32x2_t, b: uint32x2_t, c: ui
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmls.i32", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mls, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmls_laneq_u32<const LANE: i32>(a: uint32x2_t, b: uint32x2_t, c: uint32x4_t) -> uint32x2_t {
static_assert_imm2!(LANE);
vmls_u32(a, b, simd_shuffle2!(c, c, <const LANE: i32> [LANE as u32, LANE as u32]))
@@ -5761,7 +5761,7 @@ pub unsafe fn vmls_laneq_u32<const LANE: i32>(a: uint32x2_t, b: uint32x2_t, c: u
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmls.i32", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mls, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlsq_lane_u32<const LANE: i32>(a: uint32x4_t, b: uint32x4_t, c: uint32x2_t) -> uint32x4_t {
static_assert_imm1!(LANE);
vmlsq_u32(a, b, simd_shuffle4!(c, c, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -5776,7 +5776,7 @@ pub unsafe fn vmlsq_lane_u32<const LANE: i32>(a: uint32x4_t, b: uint32x4_t, c: u
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmls.i32", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mls, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlsq_laneq_u32<const LANE: i32>(a: uint32x4_t, b: uint32x4_t, c: uint32x4_t) -> uint32x4_t {
static_assert_imm2!(LANE);
vmlsq_u32(a, b, simd_shuffle4!(c, c, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -5791,7 +5791,7 @@ pub unsafe fn vmlsq_laneq_u32<const LANE: i32>(a: uint32x4_t, b: uint32x4_t, c:
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmls.f32", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fmul, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmls_lane_f32<const LANE: i32>(a: float32x2_t, b: float32x2_t, c: float32x2_t) -> float32x2_t {
static_assert_imm1!(LANE);
vmls_f32(a, b, simd_shuffle2!(c, c, <const LANE: i32> [LANE as u32, LANE as u32]))
@@ -5806,7 +5806,7 @@ pub unsafe fn vmls_lane_f32<const LANE: i32>(a: float32x2_t, b: float32x2_t, c:
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmls.f32", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fmul, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmls_laneq_f32<const LANE: i32>(a: float32x2_t, b: float32x2_t, c: float32x4_t) -> float32x2_t {
static_assert_imm2!(LANE);
vmls_f32(a, b, simd_shuffle2!(c, c, <const LANE: i32> [LANE as u32, LANE as u32]))
@@ -5821,7 +5821,7 @@ pub unsafe fn vmls_laneq_f32<const LANE: i32>(a: float32x2_t, b: float32x2_t, c:
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmls.f32", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fmul, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlsq_lane_f32<const LANE: i32>(a: float32x4_t, b: float32x4_t, c: float32x2_t) -> float32x4_t {
static_assert_imm1!(LANE);
vmlsq_f32(a, b, simd_shuffle4!(c, c, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -5836,7 +5836,7 @@ pub unsafe fn vmlsq_lane_f32<const LANE: i32>(a: float32x4_t, b: float32x4_t, c:
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmls.f32", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fmul, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlsq_laneq_f32<const LANE: i32>(a: float32x4_t, b: float32x4_t, c: float32x4_t) -> float32x4_t {
static_assert_imm2!(LANE);
vmlsq_f32(a, b, simd_shuffle4!(c, c, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -5850,7 +5850,7 @@ pub unsafe fn vmlsq_laneq_f32<const LANE: i32>(a: float32x4_t, b: float32x4_t, c
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmlsl.s8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(smlsl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlsl_s8(a: int16x8_t, b: int8x8_t, c: int8x8_t) -> int16x8_t {
simd_sub(a, vmull_s8(b, c))
}
@@ -5863,7 +5863,7 @@ pub unsafe fn vmlsl_s8(a: int16x8_t, b: int8x8_t, c: int8x8_t) -> int16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmlsl.s16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(smlsl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlsl_s16(a: int32x4_t, b: int16x4_t, c: int16x4_t) -> int32x4_t {
simd_sub(a, vmull_s16(b, c))
}
@@ -5876,7 +5876,7 @@ pub unsafe fn vmlsl_s16(a: int32x4_t, b: int16x4_t, c: int16x4_t) -> int32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmlsl.s32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(smlsl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlsl_s32(a: int64x2_t, b: int32x2_t, c: int32x2_t) -> int64x2_t {
simd_sub(a, vmull_s32(b, c))
}
@@ -5889,7 +5889,7 @@ pub unsafe fn vmlsl_s32(a: int64x2_t, b: int32x2_t, c: int32x2_t) -> int64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmlsl.u8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(umlsl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlsl_u8(a: uint16x8_t, b: uint8x8_t, c: uint8x8_t) -> uint16x8_t {
simd_sub(a, vmull_u8(b, c))
}
@@ -5902,7 +5902,7 @@ pub unsafe fn vmlsl_u8(a: uint16x8_t, b: uint8x8_t, c: uint8x8_t) -> uint16x8_t
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmlsl.u16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(umlsl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlsl_u16(a: uint32x4_t, b: uint16x4_t, c: uint16x4_t) -> uint32x4_t {
simd_sub(a, vmull_u16(b, c))
}
@@ -5915,7 +5915,7 @@ pub unsafe fn vmlsl_u16(a: uint32x4_t, b: uint16x4_t, c: uint16x4_t) -> uint32x4
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmlsl.u32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(umlsl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlsl_u32(a: uint64x2_t, b: uint32x2_t, c: uint32x2_t) -> uint64x2_t {
simd_sub(a, vmull_u32(b, c))
}
@@ -5928,7 +5928,7 @@ pub unsafe fn vmlsl_u32(a: uint64x2_t, b: uint32x2_t, c: uint32x2_t) -> uint64x2
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmlsl.s16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(smlsl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlsl_n_s16(a: int32x4_t, b: int16x4_t, c: i16) -> int32x4_t {
vmlsl_s16(a, b, vdup_n_s16(c))
}
@@ -5941,7 +5941,7 @@ pub unsafe fn vmlsl_n_s16(a: int32x4_t, b: int16x4_t, c: i16) -> int32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmlsl.s32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(smlsl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlsl_n_s32(a: int64x2_t, b: int32x2_t, c: i32) -> int64x2_t {
vmlsl_s32(a, b, vdup_n_s32(c))
}
@@ -5954,7 +5954,7 @@ pub unsafe fn vmlsl_n_s32(a: int64x2_t, b: int32x2_t, c: i32) -> int64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmlsl.u16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(umlsl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlsl_n_u16(a: uint32x4_t, b: uint16x4_t, c: u16) -> uint32x4_t {
vmlsl_u16(a, b, vdup_n_u16(c))
}
@@ -5967,7 +5967,7 @@ pub unsafe fn vmlsl_n_u16(a: uint32x4_t, b: uint16x4_t, c: u16) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmlsl.u32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(umlsl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlsl_n_u32(a: uint64x2_t, b: uint32x2_t, c: u32) -> uint64x2_t {
vmlsl_u32(a, b, vdup_n_u32(c))
}
@@ -5981,7 +5981,7 @@ pub unsafe fn vmlsl_n_u32(a: uint64x2_t, b: uint32x2_t, c: u32) -> uint64x2_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmlsl.s16", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(smlsl, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlsl_lane_s16<const LANE: i32>(a: int32x4_t, b: int16x4_t, c: int16x4_t) -> int32x4_t {
static_assert_imm2!(LANE);
vmlsl_s16(a, b, simd_shuffle4!(c, c, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -5996,7 +5996,7 @@ pub unsafe fn vmlsl_lane_s16<const LANE: i32>(a: int32x4_t, b: int16x4_t, c: int
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmlsl.s16", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(smlsl, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlsl_laneq_s16<const LANE: i32>(a: int32x4_t, b: int16x4_t, c: int16x8_t) -> int32x4_t {
static_assert_imm3!(LANE);
vmlsl_s16(a, b, simd_shuffle4!(c, c, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -6011,7 +6011,7 @@ pub unsafe fn vmlsl_laneq_s16<const LANE: i32>(a: int32x4_t, b: int16x4_t, c: in
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmlsl.s32", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(smlsl, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlsl_lane_s32<const LANE: i32>(a: int64x2_t, b: int32x2_t, c: int32x2_t) -> int64x2_t {
static_assert_imm1!(LANE);
vmlsl_s32(a, b, simd_shuffle2!(c, c, <const LANE: i32> [LANE as u32, LANE as u32]))
@@ -6026,7 +6026,7 @@ pub unsafe fn vmlsl_lane_s32<const LANE: i32>(a: int64x2_t, b: int32x2_t, c: int
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmlsl.s32", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(smlsl, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlsl_laneq_s32<const LANE: i32>(a: int64x2_t, b: int32x2_t, c: int32x4_t) -> int64x2_t {
static_assert_imm2!(LANE);
vmlsl_s32(a, b, simd_shuffle2!(c, c, <const LANE: i32> [LANE as u32, LANE as u32]))
@@ -6041,7 +6041,7 @@ pub unsafe fn vmlsl_laneq_s32<const LANE: i32>(a: int64x2_t, b: int32x2_t, c: in
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmlsl.u16", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(umlsl, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlsl_lane_u16<const LANE: i32>(a: uint32x4_t, b: uint16x4_t, c: uint16x4_t) -> uint32x4_t {
static_assert_imm2!(LANE);
vmlsl_u16(a, b, simd_shuffle4!(c, c, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -6056,7 +6056,7 @@ pub unsafe fn vmlsl_lane_u16<const LANE: i32>(a: uint32x4_t, b: uint16x4_t, c: u
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmlsl.u16", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(umlsl, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlsl_laneq_u16<const LANE: i32>(a: uint32x4_t, b: uint16x4_t, c: uint16x8_t) -> uint32x4_t {
static_assert_imm3!(LANE);
vmlsl_u16(a, b, simd_shuffle4!(c, c, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -6071,7 +6071,7 @@ pub unsafe fn vmlsl_laneq_u16<const LANE: i32>(a: uint32x4_t, b: uint16x4_t, c:
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmlsl.u32", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(umlsl, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlsl_lane_u32<const LANE: i32>(a: uint64x2_t, b: uint32x2_t, c: uint32x2_t) -> uint64x2_t {
static_assert_imm1!(LANE);
vmlsl_u32(a, b, simd_shuffle2!(c, c, <const LANE: i32> [LANE as u32, LANE as u32]))
@@ -6086,7 +6086,7 @@ pub unsafe fn vmlsl_lane_u32<const LANE: i32>(a: uint64x2_t, b: uint32x2_t, c: u
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmlsl.u32", LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(umlsl, LANE = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmlsl_laneq_u32<const LANE: i32>(a: uint64x2_t, b: uint32x2_t, c: uint32x4_t) -> uint64x2_t {
static_assert_imm2!(LANE);
vmlsl_u32(a, b, simd_shuffle2!(c, c, <const LANE: i32> [LANE as u32, LANE as u32]))
@@ -6100,7 +6100,7 @@ pub unsafe fn vmlsl_laneq_u32<const LANE: i32>(a: uint64x2_t, b: uint32x2_t, c:
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vneg.s8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(neg))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vneg_s8(a: int8x8_t) -> int8x8_t {
simd_neg(a)
}
@@ -6113,7 +6113,7 @@ pub unsafe fn vneg_s8(a: int8x8_t) -> int8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vneg.s8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(neg))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vnegq_s8(a: int8x16_t) -> int8x16_t {
simd_neg(a)
}
@@ -6126,7 +6126,7 @@ pub unsafe fn vnegq_s8(a: int8x16_t) -> int8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vneg.s16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(neg))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vneg_s16(a: int16x4_t) -> int16x4_t {
simd_neg(a)
}
@@ -6139,7 +6139,7 @@ pub unsafe fn vneg_s16(a: int16x4_t) -> int16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vneg.s16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(neg))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vnegq_s16(a: int16x8_t) -> int16x8_t {
simd_neg(a)
}
@@ -6152,7 +6152,7 @@ pub unsafe fn vnegq_s16(a: int16x8_t) -> int16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vneg.s32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(neg))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vneg_s32(a: int32x2_t) -> int32x2_t {
simd_neg(a)
}
@@ -6165,7 +6165,7 @@ pub unsafe fn vneg_s32(a: int32x2_t) -> int32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vneg.s32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(neg))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vnegq_s32(a: int32x4_t) -> int32x4_t {
simd_neg(a)
}
@@ -6178,7 +6178,7 @@ pub unsafe fn vnegq_s32(a: int32x4_t) -> int32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vneg.f32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fneg))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vneg_f32(a: float32x2_t) -> float32x2_t {
simd_neg(a)
}
@@ -6191,7 +6191,7 @@ pub unsafe fn vneg_f32(a: float32x2_t) -> float32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vneg.f32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fneg))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vnegq_f32(a: float32x4_t) -> float32x4_t {
simd_neg(a)
}
@@ -6204,7 +6204,7 @@ pub unsafe fn vnegq_f32(a: float32x4_t) -> float32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vqneg.s8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqneg))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqneg_s8(a: int8x8_t) -> int8x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -6223,7 +6223,7 @@ vqneg_s8_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vqneg.s8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqneg))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqnegq_s8(a: int8x16_t) -> int8x16_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -6242,7 +6242,7 @@ vqnegq_s8_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vqneg.s16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqneg))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqneg_s16(a: int16x4_t) -> int16x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -6261,7 +6261,7 @@ vqneg_s16_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vqneg.s16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqneg))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqnegq_s16(a: int16x8_t) -> int16x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -6280,7 +6280,7 @@ vqnegq_s16_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vqneg.s32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqneg))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqneg_s32(a: int32x2_t) -> int32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -6299,7 +6299,7 @@ vqneg_s32_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vqneg.s32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqneg))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqnegq_s32(a: int32x4_t) -> int32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -6318,7 +6318,7 @@ vqnegq_s32_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vqsub.u8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uqsub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqsub_u8(a: uint8x8_t, b: uint8x8_t) -> uint8x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -6337,7 +6337,7 @@ vqsub_u8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vqsub.u8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uqsub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqsubq_u8(a: uint8x16_t, b: uint8x16_t) -> uint8x16_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -6356,7 +6356,7 @@ vqsubq_u8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vqsub.u16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uqsub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqsub_u16(a: uint16x4_t, b: uint16x4_t) -> uint16x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -6375,7 +6375,7 @@ vqsub_u16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vqsub.u16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uqsub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqsubq_u16(a: uint16x8_t, b: uint16x8_t) -> uint16x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -6394,7 +6394,7 @@ vqsubq_u16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vqsub.u32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uqsub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqsub_u32(a: uint32x2_t, b: uint32x2_t) -> uint32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -6413,7 +6413,7 @@ vqsub_u32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vqsub.u32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uqsub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqsubq_u32(a: uint32x4_t, b: uint32x4_t) -> uint32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -6432,7 +6432,7 @@ vqsubq_u32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vqsub.u64"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uqsub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqsub_u64(a: uint64x1_t, b: uint64x1_t) -> uint64x1_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -6451,7 +6451,7 @@ vqsub_u64_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vqsub.u64"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uqsub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqsubq_u64(a: uint64x2_t, b: uint64x2_t) -> uint64x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -6470,7 +6470,7 @@ vqsubq_u64_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vqsub.s8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqsub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqsub_s8(a: int8x8_t, b: int8x8_t) -> int8x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -6489,7 +6489,7 @@ vqsub_s8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vqsub.s8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqsub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqsubq_s8(a: int8x16_t, b: int8x16_t) -> int8x16_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -6508,7 +6508,7 @@ vqsubq_s8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vqsub.s16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqsub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqsub_s16(a: int16x4_t, b: int16x4_t) -> int16x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -6527,7 +6527,7 @@ vqsub_s16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vqsub.s16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqsub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqsubq_s16(a: int16x8_t, b: int16x8_t) -> int16x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -6546,7 +6546,7 @@ vqsubq_s16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vqsub.s32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqsub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqsub_s32(a: int32x2_t, b: int32x2_t) -> int32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -6565,7 +6565,7 @@ vqsub_s32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vqsub.s32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqsub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqsubq_s32(a: int32x4_t, b: int32x4_t) -> int32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -6584,7 +6584,7 @@ vqsubq_s32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vqsub.s64"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqsub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqsub_s64(a: int64x1_t, b: int64x1_t) -> int64x1_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -6603,7 +6603,7 @@ vqsub_s64_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vqsub.s64"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqsub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqsubq_s64(a: int64x2_t, b: int64x2_t) -> int64x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -6622,7 +6622,7 @@ vqsubq_s64_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vhadd.u8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uhadd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vhadd_u8(a: uint8x8_t, b: uint8x8_t) -> uint8x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -6641,7 +6641,7 @@ vhadd_u8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vhadd.u8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uhadd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vhaddq_u8(a: uint8x16_t, b: uint8x16_t) -> uint8x16_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -6660,7 +6660,7 @@ vhaddq_u8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vhadd.u16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uhadd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vhadd_u16(a: uint16x4_t, b: uint16x4_t) -> uint16x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -6679,7 +6679,7 @@ vhadd_u16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vhadd.u16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uhadd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vhaddq_u16(a: uint16x8_t, b: uint16x8_t) -> uint16x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -6698,7 +6698,7 @@ vhaddq_u16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vhadd.u32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uhadd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vhadd_u32(a: uint32x2_t, b: uint32x2_t) -> uint32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -6717,7 +6717,7 @@ vhadd_u32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vhadd.u32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uhadd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vhaddq_u32(a: uint32x4_t, b: uint32x4_t) -> uint32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -6736,7 +6736,7 @@ vhaddq_u32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vhadd.s8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(shadd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vhadd_s8(a: int8x8_t, b: int8x8_t) -> int8x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -6755,7 +6755,7 @@ vhadd_s8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vhadd.s8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(shadd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vhaddq_s8(a: int8x16_t, b: int8x16_t) -> int8x16_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -6774,7 +6774,7 @@ vhaddq_s8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vhadd.s16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(shadd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vhadd_s16(a: int16x4_t, b: int16x4_t) -> int16x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -6793,7 +6793,7 @@ vhadd_s16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vhadd.s16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(shadd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vhaddq_s16(a: int16x8_t, b: int16x8_t) -> int16x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -6812,7 +6812,7 @@ vhaddq_s16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vhadd.s32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(shadd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vhadd_s32(a: int32x2_t, b: int32x2_t) -> int32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -6831,7 +6831,7 @@ vhadd_s32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vhadd.s32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(shadd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vhaddq_s32(a: int32x4_t, b: int32x4_t) -> int32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -6850,7 +6850,7 @@ vhaddq_s32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vrhadd.u8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(urhadd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrhadd_u8(a: uint8x8_t, b: uint8x8_t) -> uint8x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -6869,7 +6869,7 @@ vrhadd_u8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vrhadd.u8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(urhadd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrhaddq_u8(a: uint8x16_t, b: uint8x16_t) -> uint8x16_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -6888,7 +6888,7 @@ vrhaddq_u8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vrhadd.u16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(urhadd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrhadd_u16(a: uint16x4_t, b: uint16x4_t) -> uint16x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -6907,7 +6907,7 @@ vrhadd_u16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vrhadd.u16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(urhadd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrhaddq_u16(a: uint16x8_t, b: uint16x8_t) -> uint16x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -6926,7 +6926,7 @@ vrhaddq_u16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vrhadd.u32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(urhadd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrhadd_u32(a: uint32x2_t, b: uint32x2_t) -> uint32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -6945,7 +6945,7 @@ vrhadd_u32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vrhadd.u32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(urhadd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrhaddq_u32(a: uint32x4_t, b: uint32x4_t) -> uint32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -6964,7 +6964,7 @@ vrhaddq_u32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vrhadd.s8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(srhadd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrhadd_s8(a: int8x8_t, b: int8x8_t) -> int8x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -6983,7 +6983,7 @@ vrhadd_s8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vrhadd.s8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(srhadd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrhaddq_s8(a: int8x16_t, b: int8x16_t) -> int8x16_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7002,7 +7002,7 @@ vrhaddq_s8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vrhadd.s16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(srhadd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrhadd_s16(a: int16x4_t, b: int16x4_t) -> int16x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7021,7 +7021,7 @@ vrhadd_s16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vrhadd.s16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(srhadd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrhaddq_s16(a: int16x8_t, b: int16x8_t) -> int16x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7040,7 +7040,7 @@ vrhaddq_s16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vrhadd.s32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(srhadd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrhadd_s32(a: int32x2_t, b: int32x2_t) -> int32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7059,7 +7059,7 @@ vrhadd_s32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vrhadd.s32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(srhadd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrhaddq_s32(a: int32x4_t, b: int32x4_t) -> int32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7078,7 +7078,7 @@ vrhaddq_s32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "fp-armv8,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrintn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(frintn))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrndn_f32(a: float32x2_t) -> float32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7097,7 +7097,7 @@ vrndn_f32_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "fp-armv8,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrintn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(frintn))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrndnq_f32(a: float32x4_t) -> float32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7116,7 +7116,7 @@ vrndnq_f32_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vqadd.u8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uqadd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqadd_u8(a: uint8x8_t, b: uint8x8_t) -> uint8x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7135,7 +7135,7 @@ vqadd_u8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vqadd.u8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uqadd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqaddq_u8(a: uint8x16_t, b: uint8x16_t) -> uint8x16_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7154,7 +7154,7 @@ vqaddq_u8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vqadd.u16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uqadd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqadd_u16(a: uint16x4_t, b: uint16x4_t) -> uint16x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7173,7 +7173,7 @@ vqadd_u16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vqadd.u16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uqadd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqaddq_u16(a: uint16x8_t, b: uint16x8_t) -> uint16x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7192,7 +7192,7 @@ vqaddq_u16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vqadd.u32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uqadd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqadd_u32(a: uint32x2_t, b: uint32x2_t) -> uint32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7211,7 +7211,7 @@ vqadd_u32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vqadd.u32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uqadd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqaddq_u32(a: uint32x4_t, b: uint32x4_t) -> uint32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7230,7 +7230,7 @@ vqaddq_u32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vqadd.u64"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uqadd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqadd_u64(a: uint64x1_t, b: uint64x1_t) -> uint64x1_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7249,7 +7249,7 @@ vqadd_u64_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vqadd.u64"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uqadd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqaddq_u64(a: uint64x2_t, b: uint64x2_t) -> uint64x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7268,7 +7268,7 @@ vqaddq_u64_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vqadd.s8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqadd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqadd_s8(a: int8x8_t, b: int8x8_t) -> int8x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7287,7 +7287,7 @@ vqadd_s8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vqadd.s8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqadd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqaddq_s8(a: int8x16_t, b: int8x16_t) -> int8x16_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7306,7 +7306,7 @@ vqaddq_s8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vqadd.s16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqadd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqadd_s16(a: int16x4_t, b: int16x4_t) -> int16x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7325,7 +7325,7 @@ vqadd_s16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vqadd.s16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqadd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqaddq_s16(a: int16x8_t, b: int16x8_t) -> int16x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7344,7 +7344,7 @@ vqaddq_s16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vqadd.s32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqadd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqadd_s32(a: int32x2_t, b: int32x2_t) -> int32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7363,7 +7363,7 @@ vqadd_s32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vqadd.s32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqadd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqaddq_s32(a: int32x4_t, b: int32x4_t) -> int32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7382,7 +7382,7 @@ vqaddq_s32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vqadd.s64"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqadd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqadd_s64(a: int64x1_t, b: int64x1_t) -> int64x1_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7401,7 +7401,7 @@ vqadd_s64_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vqadd.s64"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqadd))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqaddq_s64(a: int64x2_t, b: int64x2_t) -> int64x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7420,7 +7420,7 @@ vqaddq_s64_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1_s8_x2(a: *const i8) -> int8x8x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7439,7 +7439,7 @@ vld1_s8_x2_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1_s16_x2(a: *const i16) -> int16x4x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7458,7 +7458,7 @@ vld1_s16_x2_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1_s32_x2(a: *const i32) -> int32x2x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7477,7 +7477,7 @@ vld1_s32_x2_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1_s64_x2(a: *const i64) -> int64x1x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7496,7 +7496,7 @@ vld1_s64_x2_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1q_s8_x2(a: *const i8) -> int8x16x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7515,7 +7515,7 @@ vld1q_s8_x2_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1q_s16_x2(a: *const i16) -> int16x8x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7534,7 +7534,7 @@ vld1q_s16_x2_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1q_s32_x2(a: *const i32) -> int32x4x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7553,7 +7553,7 @@ vld1q_s32_x2_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1q_s64_x2(a: *const i64) -> int64x2x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7572,7 +7572,7 @@ vld1q_s64_x2_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1_s8_x3(a: *const i8) -> int8x8x3_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7591,7 +7591,7 @@ vld1_s8_x3_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1_s16_x3(a: *const i16) -> int16x4x3_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7610,7 +7610,7 @@ vld1_s16_x3_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1_s32_x3(a: *const i32) -> int32x2x3_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7629,7 +7629,7 @@ vld1_s32_x3_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1_s64_x3(a: *const i64) -> int64x1x3_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7648,7 +7648,7 @@ vld1_s64_x3_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1q_s8_x3(a: *const i8) -> int8x16x3_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7667,7 +7667,7 @@ vld1q_s8_x3_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1q_s16_x3(a: *const i16) -> int16x8x3_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7686,7 +7686,7 @@ vld1q_s16_x3_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1q_s32_x3(a: *const i32) -> int32x4x3_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7705,7 +7705,7 @@ vld1q_s32_x3_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1q_s64_x3(a: *const i64) -> int64x2x3_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7724,7 +7724,7 @@ vld1q_s64_x3_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1_s8_x4(a: *const i8) -> int8x8x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7743,7 +7743,7 @@ vld1_s8_x4_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1_s16_x4(a: *const i16) -> int16x4x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7762,7 +7762,7 @@ vld1_s16_x4_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1_s32_x4(a: *const i32) -> int32x2x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7781,7 +7781,7 @@ vld1_s32_x4_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1_s64_x4(a: *const i64) -> int64x1x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7800,7 +7800,7 @@ vld1_s64_x4_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1q_s8_x4(a: *const i8) -> int8x16x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7819,7 +7819,7 @@ vld1q_s8_x4_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1q_s16_x4(a: *const i16) -> int16x8x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7838,7 +7838,7 @@ vld1q_s16_x4_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1q_s32_x4(a: *const i32) -> int32x4x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7857,7 +7857,7 @@ vld1q_s32_x4_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1q_s64_x4(a: *const i64) -> int64x2x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -7876,7 +7876,7 @@ vld1q_s64_x4_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1_u8_x2(a: *const u8) -> uint8x8x2_t {
transmute(vld1_s8_x2(transmute(a)))
}
@@ -7889,7 +7889,7 @@ pub unsafe fn vld1_u8_x2(a: *const u8) -> uint8x8x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1_u16_x2(a: *const u16) -> uint16x4x2_t {
transmute(vld1_s16_x2(transmute(a)))
}
@@ -7902,7 +7902,7 @@ pub unsafe fn vld1_u16_x2(a: *const u16) -> uint16x4x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1_u32_x2(a: *const u32) -> uint32x2x2_t {
transmute(vld1_s32_x2(transmute(a)))
}
@@ -7915,7 +7915,7 @@ pub unsafe fn vld1_u32_x2(a: *const u32) -> uint32x2x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1_u64_x2(a: *const u64) -> uint64x1x2_t {
transmute(vld1_s64_x2(transmute(a)))
}
@@ -7928,7 +7928,7 @@ pub unsafe fn vld1_u64_x2(a: *const u64) -> uint64x1x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1q_u8_x2(a: *const u8) -> uint8x16x2_t {
transmute(vld1q_s8_x2(transmute(a)))
}
@@ -7941,7 +7941,7 @@ pub unsafe fn vld1q_u8_x2(a: *const u8) -> uint8x16x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1q_u16_x2(a: *const u16) -> uint16x8x2_t {
transmute(vld1q_s16_x2(transmute(a)))
}
@@ -7954,7 +7954,7 @@ pub unsafe fn vld1q_u16_x2(a: *const u16) -> uint16x8x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1q_u32_x2(a: *const u32) -> uint32x4x2_t {
transmute(vld1q_s32_x2(transmute(a)))
}
@@ -7967,7 +7967,7 @@ pub unsafe fn vld1q_u32_x2(a: *const u32) -> uint32x4x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1q_u64_x2(a: *const u64) -> uint64x2x2_t {
transmute(vld1q_s64_x2(transmute(a)))
}
@@ -7980,7 +7980,7 @@ pub unsafe fn vld1q_u64_x2(a: *const u64) -> uint64x2x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1_u8_x3(a: *const u8) -> uint8x8x3_t {
transmute(vld1_s8_x3(transmute(a)))
}
@@ -7993,7 +7993,7 @@ pub unsafe fn vld1_u8_x3(a: *const u8) -> uint8x8x3_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1_u16_x3(a: *const u16) -> uint16x4x3_t {
transmute(vld1_s16_x3(transmute(a)))
}
@@ -8006,7 +8006,7 @@ pub unsafe fn vld1_u16_x3(a: *const u16) -> uint16x4x3_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1_u32_x3(a: *const u32) -> uint32x2x3_t {
transmute(vld1_s32_x3(transmute(a)))
}
@@ -8019,7 +8019,7 @@ pub unsafe fn vld1_u32_x3(a: *const u32) -> uint32x2x3_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1_u64_x3(a: *const u64) -> uint64x1x3_t {
transmute(vld1_s64_x3(transmute(a)))
}
@@ -8032,7 +8032,7 @@ pub unsafe fn vld1_u64_x3(a: *const u64) -> uint64x1x3_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1q_u8_x3(a: *const u8) -> uint8x16x3_t {
transmute(vld1q_s8_x3(transmute(a)))
}
@@ -8045,7 +8045,7 @@ pub unsafe fn vld1q_u8_x3(a: *const u8) -> uint8x16x3_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1q_u16_x3(a: *const u16) -> uint16x8x3_t {
transmute(vld1q_s16_x3(transmute(a)))
}
@@ -8058,7 +8058,7 @@ pub unsafe fn vld1q_u16_x3(a: *const u16) -> uint16x8x3_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1q_u32_x3(a: *const u32) -> uint32x4x3_t {
transmute(vld1q_s32_x3(transmute(a)))
}
@@ -8071,7 +8071,7 @@ pub unsafe fn vld1q_u32_x3(a: *const u32) -> uint32x4x3_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1q_u64_x3(a: *const u64) -> uint64x2x3_t {
transmute(vld1q_s64_x3(transmute(a)))
}
@@ -8084,7 +8084,7 @@ pub unsafe fn vld1q_u64_x3(a: *const u64) -> uint64x2x3_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1_u8_x4(a: *const u8) -> uint8x8x4_t {
transmute(vld1_s8_x4(transmute(a)))
}
@@ -8097,7 +8097,7 @@ pub unsafe fn vld1_u8_x4(a: *const u8) -> uint8x8x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1_u16_x4(a: *const u16) -> uint16x4x4_t {
transmute(vld1_s16_x4(transmute(a)))
}
@@ -8110,7 +8110,7 @@ pub unsafe fn vld1_u16_x4(a: *const u16) -> uint16x4x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1_u32_x4(a: *const u32) -> uint32x2x4_t {
transmute(vld1_s32_x4(transmute(a)))
}
@@ -8123,7 +8123,7 @@ pub unsafe fn vld1_u32_x4(a: *const u32) -> uint32x2x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1_u64_x4(a: *const u64) -> uint64x1x4_t {
transmute(vld1_s64_x4(transmute(a)))
}
@@ -8136,7 +8136,7 @@ pub unsafe fn vld1_u64_x4(a: *const u64) -> uint64x1x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1q_u8_x4(a: *const u8) -> uint8x16x4_t {
transmute(vld1q_s8_x4(transmute(a)))
}
@@ -8149,7 +8149,7 @@ pub unsafe fn vld1q_u8_x4(a: *const u8) -> uint8x16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1q_u16_x4(a: *const u16) -> uint16x8x4_t {
transmute(vld1q_s16_x4(transmute(a)))
}
@@ -8162,7 +8162,7 @@ pub unsafe fn vld1q_u16_x4(a: *const u16) -> uint16x8x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1q_u32_x4(a: *const u32) -> uint32x4x4_t {
transmute(vld1q_s32_x4(transmute(a)))
}
@@ -8175,7 +8175,7 @@ pub unsafe fn vld1q_u32_x4(a: *const u32) -> uint32x4x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1q_u64_x4(a: *const u64) -> uint64x2x4_t {
transmute(vld1q_s64_x4(transmute(a)))
}
@@ -8188,7 +8188,7 @@ pub unsafe fn vld1q_u64_x4(a: *const u64) -> uint64x2x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1_p8_x2(a: *const p8) -> poly8x8x2_t {
transmute(vld1_s8_x2(transmute(a)))
}
@@ -8201,7 +8201,7 @@ pub unsafe fn vld1_p8_x2(a: *const p8) -> poly8x8x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1_p8_x3(a: *const p8) -> poly8x8x3_t {
transmute(vld1_s8_x3(transmute(a)))
}
@@ -8214,7 +8214,7 @@ pub unsafe fn vld1_p8_x3(a: *const p8) -> poly8x8x3_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1_p8_x4(a: *const p8) -> poly8x8x4_t {
transmute(vld1_s8_x4(transmute(a)))
}
@@ -8227,7 +8227,7 @@ pub unsafe fn vld1_p8_x4(a: *const p8) -> poly8x8x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1q_p8_x2(a: *const p8) -> poly8x16x2_t {
transmute(vld1q_s8_x2(transmute(a)))
}
@@ -8240,7 +8240,7 @@ pub unsafe fn vld1q_p8_x2(a: *const p8) -> poly8x16x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1q_p8_x3(a: *const p8) -> poly8x16x3_t {
transmute(vld1q_s8_x3(transmute(a)))
}
@@ -8253,7 +8253,7 @@ pub unsafe fn vld1q_p8_x3(a: *const p8) -> poly8x16x3_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1q_p8_x4(a: *const p8) -> poly8x16x4_t {
transmute(vld1q_s8_x4(transmute(a)))
}
@@ -8266,7 +8266,7 @@ pub unsafe fn vld1q_p8_x4(a: *const p8) -> poly8x16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1_p16_x2(a: *const p16) -> poly16x4x2_t {
transmute(vld1_s16_x2(transmute(a)))
}
@@ -8279,7 +8279,7 @@ pub unsafe fn vld1_p16_x2(a: *const p16) -> poly16x4x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1_p16_x3(a: *const p16) -> poly16x4x3_t {
transmute(vld1_s16_x3(transmute(a)))
}
@@ -8292,7 +8292,7 @@ pub unsafe fn vld1_p16_x3(a: *const p16) -> poly16x4x3_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1_p16_x4(a: *const p16) -> poly16x4x4_t {
transmute(vld1_s16_x4(transmute(a)))
}
@@ -8305,7 +8305,7 @@ pub unsafe fn vld1_p16_x4(a: *const p16) -> poly16x4x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1q_p16_x2(a: *const p16) -> poly16x8x2_t {
transmute(vld1q_s16_x2(transmute(a)))
}
@@ -8318,7 +8318,7 @@ pub unsafe fn vld1q_p16_x2(a: *const p16) -> poly16x8x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1q_p16_x3(a: *const p16) -> poly16x8x3_t {
transmute(vld1q_s16_x3(transmute(a)))
}
@@ -8331,7 +8331,7 @@ pub unsafe fn vld1q_p16_x3(a: *const p16) -> poly16x8x3_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1q_p16_x4(a: *const p16) -> poly16x8x4_t {
transmute(vld1q_s16_x4(transmute(a)))
}
@@ -8344,7 +8344,7 @@ pub unsafe fn vld1q_p16_x4(a: *const p16) -> poly16x8x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1_p64_x2(a: *const p64) -> poly64x1x2_t {
transmute(vld1_s64_x2(transmute(a)))
}
@@ -8357,7 +8357,7 @@ pub unsafe fn vld1_p64_x2(a: *const p64) -> poly64x1x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1_p64_x3(a: *const p64) -> poly64x1x3_t {
transmute(vld1_s64_x3(transmute(a)))
}
@@ -8370,7 +8370,7 @@ pub unsafe fn vld1_p64_x3(a: *const p64) -> poly64x1x3_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1_p64_x4(a: *const p64) -> poly64x1x4_t {
transmute(vld1_s64_x4(transmute(a)))
}
@@ -8383,7 +8383,7 @@ pub unsafe fn vld1_p64_x4(a: *const p64) -> poly64x1x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1q_p64_x2(a: *const p64) -> poly64x2x2_t {
transmute(vld1q_s64_x2(transmute(a)))
}
@@ -8396,7 +8396,7 @@ pub unsafe fn vld1q_p64_x2(a: *const p64) -> poly64x2x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1q_p64_x3(a: *const p64) -> poly64x2x3_t {
transmute(vld1q_s64_x3(transmute(a)))
}
@@ -8409,7 +8409,7 @@ pub unsafe fn vld1q_p64_x3(a: *const p64) -> poly64x2x3_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1q_p64_x4(a: *const p64) -> poly64x2x4_t {
transmute(vld1q_s64_x4(transmute(a)))
}
@@ -8422,7 +8422,7 @@ pub unsafe fn vld1q_p64_x4(a: *const p64) -> poly64x2x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1_f32_x2(a: *const f32) -> float32x2x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -8441,7 +8441,7 @@ vld1_f32_x2_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1q_f32_x2(a: *const f32) -> float32x4x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -8460,7 +8460,7 @@ vld1q_f32_x2_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1_f32_x3(a: *const f32) -> float32x2x3_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -8479,7 +8479,7 @@ vld1_f32_x3_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1q_f32_x3(a: *const f32) -> float32x4x3_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -8498,7 +8498,7 @@ vld1q_f32_x3_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1_f32_x4(a: *const f32) -> float32x2x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -8517,7 +8517,7 @@ vld1_f32_x4_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld1q_f32_x4(a: *const f32) -> float32x4x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -8767,7 +8767,7 @@ vld2_s64_(a as _)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld2))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld2_u8(a: *const u8) -> uint8x8x2_t {
transmute(vld2_s8(transmute(a)))
}
@@ -8780,7 +8780,7 @@ pub unsafe fn vld2_u8(a: *const u8) -> uint8x8x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld2))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld2_u16(a: *const u16) -> uint16x4x2_t {
transmute(vld2_s16(transmute(a)))
}
@@ -8793,7 +8793,7 @@ pub unsafe fn vld2_u16(a: *const u16) -> uint16x4x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld2))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld2_u32(a: *const u32) -> uint32x2x2_t {
transmute(vld2_s32(transmute(a)))
}
@@ -8806,7 +8806,7 @@ pub unsafe fn vld2_u32(a: *const u32) -> uint32x2x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld2))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld2q_u8(a: *const u8) -> uint8x16x2_t {
transmute(vld2q_s8(transmute(a)))
}
@@ -8819,7 +8819,7 @@ pub unsafe fn vld2q_u8(a: *const u8) -> uint8x16x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld2))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld2q_u16(a: *const u16) -> uint16x8x2_t {
transmute(vld2q_s16(transmute(a)))
}
@@ -8832,7 +8832,7 @@ pub unsafe fn vld2q_u16(a: *const u16) -> uint16x8x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld2))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld2q_u32(a: *const u32) -> uint32x4x2_t {
transmute(vld2q_s32(transmute(a)))
}
@@ -8845,7 +8845,7 @@ pub unsafe fn vld2q_u32(a: *const u32) -> uint32x4x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld2))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld2_p8(a: *const p8) -> poly8x8x2_t {
transmute(vld2_s8(transmute(a)))
}
@@ -8858,7 +8858,7 @@ pub unsafe fn vld2_p8(a: *const p8) -> poly8x8x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld2))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld2_p16(a: *const p16) -> poly16x4x2_t {
transmute(vld2_s16(transmute(a)))
}
@@ -8871,7 +8871,7 @@ pub unsafe fn vld2_p16(a: *const p16) -> poly16x4x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld2))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld2q_p8(a: *const p8) -> poly8x16x2_t {
transmute(vld2q_s8(transmute(a)))
}
@@ -8884,7 +8884,7 @@ pub unsafe fn vld2q_p8(a: *const p8) -> poly8x16x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld2))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld2q_p16(a: *const p16) -> poly16x8x2_t {
transmute(vld2q_s16(transmute(a)))
}
@@ -8897,7 +8897,7 @@ pub unsafe fn vld2q_p16(a: *const p16) -> poly16x8x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld2_u64(a: *const u64) -> uint64x1x2_t {
transmute(vld2_s64(transmute(a)))
}
@@ -8910,7 +8910,7 @@ pub unsafe fn vld2_u64(a: *const u64) -> uint64x1x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld2_p64(a: *const p64) -> poly64x1x2_t {
transmute(vld2_s64(transmute(a)))
}
@@ -9220,7 +9220,7 @@ vld2_dup_s64_(a as _)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld2r))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld2_dup_u8(a: *const u8) -> uint8x8x2_t {
transmute(vld2_dup_s8(transmute(a)))
}
@@ -9233,7 +9233,7 @@ pub unsafe fn vld2_dup_u8(a: *const u8) -> uint8x8x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld2r))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld2_dup_u16(a: *const u16) -> uint16x4x2_t {
transmute(vld2_dup_s16(transmute(a)))
}
@@ -9246,7 +9246,7 @@ pub unsafe fn vld2_dup_u16(a: *const u16) -> uint16x4x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld2r))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld2_dup_u32(a: *const u32) -> uint32x2x2_t {
transmute(vld2_dup_s32(transmute(a)))
}
@@ -9259,7 +9259,7 @@ pub unsafe fn vld2_dup_u32(a: *const u32) -> uint32x2x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld2r))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld2q_dup_u8(a: *const u8) -> uint8x16x2_t {
transmute(vld2q_dup_s8(transmute(a)))
}
@@ -9272,7 +9272,7 @@ pub unsafe fn vld2q_dup_u8(a: *const u8) -> uint8x16x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld2r))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld2q_dup_u16(a: *const u16) -> uint16x8x2_t {
transmute(vld2q_dup_s16(transmute(a)))
}
@@ -9285,7 +9285,7 @@ pub unsafe fn vld2q_dup_u16(a: *const u16) -> uint16x8x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld2r))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld2q_dup_u32(a: *const u32) -> uint32x4x2_t {
transmute(vld2q_dup_s32(transmute(a)))
}
@@ -9298,7 +9298,7 @@ pub unsafe fn vld2q_dup_u32(a: *const u32) -> uint32x4x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld2r))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld2_dup_p8(a: *const p8) -> poly8x8x2_t {
transmute(vld2_dup_s8(transmute(a)))
}
@@ -9311,7 +9311,7 @@ pub unsafe fn vld2_dup_p8(a: *const p8) -> poly8x8x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld2r))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld2_dup_p16(a: *const p16) -> poly16x4x2_t {
transmute(vld2_dup_s16(transmute(a)))
}
@@ -9324,7 +9324,7 @@ pub unsafe fn vld2_dup_p16(a: *const p16) -> poly16x4x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld2r))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld2q_dup_p8(a: *const p8) -> poly8x16x2_t {
transmute(vld2q_dup_s8(transmute(a)))
}
@@ -9337,7 +9337,7 @@ pub unsafe fn vld2q_dup_p8(a: *const p8) -> poly8x16x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld2r))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld2q_dup_p16(a: *const p16) -> poly16x8x2_t {
transmute(vld2q_dup_s16(transmute(a)))
}
@@ -9350,7 +9350,7 @@ pub unsafe fn vld2q_dup_p16(a: *const p16) -> poly16x8x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld2r))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld2_dup_u64(a: *const u64) -> uint64x1x2_t {
transmute(vld2_dup_s64(transmute(a)))
}
@@ -9363,7 +9363,7 @@ pub unsafe fn vld2_dup_u64(a: *const u64) -> uint64x1x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld2r))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld2_dup_p64(a: *const p64) -> poly64x1x2_t {
transmute(vld2_dup_s64(transmute(a)))
}
@@ -9628,7 +9628,7 @@ vld2q_lane_s32_(b.0, b.1, LANE as i64, a as _)
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld2, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld2, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld2_lane_u8<const LANE: i32>(a: *const u8, b: uint8x8x2_t) -> uint8x8x2_t {
static_assert_imm3!(LANE);
transmute(vld2_lane_s8::<LANE>(transmute(a), transmute(b)))
@@ -9643,7 +9643,7 @@ pub unsafe fn vld2_lane_u8<const LANE: i32>(a: *const u8, b: uint8x8x2_t) -> uin
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld2, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld2, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld2_lane_u16<const LANE: i32>(a: *const u16, b: uint16x4x2_t) -> uint16x4x2_t {
static_assert_imm2!(LANE);
transmute(vld2_lane_s16::<LANE>(transmute(a), transmute(b)))
@@ -9658,7 +9658,7 @@ pub unsafe fn vld2_lane_u16<const LANE: i32>(a: *const u16, b: uint16x4x2_t) ->
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld2, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld2, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld2_lane_u32<const LANE: i32>(a: *const u32, b: uint32x2x2_t) -> uint32x2x2_t {
static_assert_imm1!(LANE);
transmute(vld2_lane_s32::<LANE>(transmute(a), transmute(b)))
@@ -9673,7 +9673,7 @@ pub unsafe fn vld2_lane_u32<const LANE: i32>(a: *const u32, b: uint32x2x2_t) ->
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld2, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld2, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld2q_lane_u16<const LANE: i32>(a: *const u16, b: uint16x8x2_t) -> uint16x8x2_t {
static_assert_imm3!(LANE);
transmute(vld2q_lane_s16::<LANE>(transmute(a), transmute(b)))
@@ -9688,7 +9688,7 @@ pub unsafe fn vld2q_lane_u16<const LANE: i32>(a: *const u16, b: uint16x8x2_t) ->
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld2, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld2, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld2q_lane_u32<const LANE: i32>(a: *const u32, b: uint32x4x2_t) -> uint32x4x2_t {
static_assert_imm2!(LANE);
transmute(vld2q_lane_s32::<LANE>(transmute(a), transmute(b)))
@@ -9703,7 +9703,7 @@ pub unsafe fn vld2q_lane_u32<const LANE: i32>(a: *const u32, b: uint32x4x2_t) ->
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld2, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld2, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld2_lane_p8<const LANE: i32>(a: *const p8, b: poly8x8x2_t) -> poly8x8x2_t {
static_assert_imm3!(LANE);
transmute(vld2_lane_s8::<LANE>(transmute(a), transmute(b)))
@@ -9718,7 +9718,7 @@ pub unsafe fn vld2_lane_p8<const LANE: i32>(a: *const p8, b: poly8x8x2_t) -> pol
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld2, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld2, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld2_lane_p16<const LANE: i32>(a: *const p16, b: poly16x4x2_t) -> poly16x4x2_t {
static_assert_imm2!(LANE);
transmute(vld2_lane_s16::<LANE>(transmute(a), transmute(b)))
@@ -9733,7 +9733,7 @@ pub unsafe fn vld2_lane_p16<const LANE: i32>(a: *const p16, b: poly16x4x2_t) ->
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld2, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld2, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld2q_lane_p16<const LANE: i32>(a: *const p16, b: poly16x8x2_t) -> poly16x8x2_t {
static_assert_imm3!(LANE);
transmute(vld2q_lane_s16::<LANE>(transmute(a), transmute(b)))
@@ -10052,7 +10052,7 @@ vld3_s64_(a as _)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld3))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld3))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld3_u8(a: *const u8) -> uint8x8x3_t {
transmute(vld3_s8(transmute(a)))
}
@@ -10065,7 +10065,7 @@ pub unsafe fn vld3_u8(a: *const u8) -> uint8x8x3_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld3))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld3))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld3_u16(a: *const u16) -> uint16x4x3_t {
transmute(vld3_s16(transmute(a)))
}
@@ -10078,7 +10078,7 @@ pub unsafe fn vld3_u16(a: *const u16) -> uint16x4x3_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld3))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld3))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld3_u32(a: *const u32) -> uint32x2x3_t {
transmute(vld3_s32(transmute(a)))
}
@@ -10091,7 +10091,7 @@ pub unsafe fn vld3_u32(a: *const u32) -> uint32x2x3_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld3))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld3))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld3q_u8(a: *const u8) -> uint8x16x3_t {
transmute(vld3q_s8(transmute(a)))
}
@@ -10104,7 +10104,7 @@ pub unsafe fn vld3q_u8(a: *const u8) -> uint8x16x3_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld3))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld3))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld3q_u16(a: *const u16) -> uint16x8x3_t {
transmute(vld3q_s16(transmute(a)))
}
@@ -10117,7 +10117,7 @@ pub unsafe fn vld3q_u16(a: *const u16) -> uint16x8x3_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld3))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld3))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld3q_u32(a: *const u32) -> uint32x4x3_t {
transmute(vld3q_s32(transmute(a)))
}
@@ -10130,7 +10130,7 @@ pub unsafe fn vld3q_u32(a: *const u32) -> uint32x4x3_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld3))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld3))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld3_p8(a: *const p8) -> poly8x8x3_t {
transmute(vld3_s8(transmute(a)))
}
@@ -10143,7 +10143,7 @@ pub unsafe fn vld3_p8(a: *const p8) -> poly8x8x3_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld3))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld3))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld3_p16(a: *const p16) -> poly16x4x3_t {
transmute(vld3_s16(transmute(a)))
}
@@ -10156,7 +10156,7 @@ pub unsafe fn vld3_p16(a: *const p16) -> poly16x4x3_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld3))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld3))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld3q_p8(a: *const p8) -> poly8x16x3_t {
transmute(vld3q_s8(transmute(a)))
}
@@ -10169,7 +10169,7 @@ pub unsafe fn vld3q_p8(a: *const p8) -> poly8x16x3_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld3))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld3))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld3q_p16(a: *const p16) -> poly16x8x3_t {
transmute(vld3q_s16(transmute(a)))
}
@@ -10182,7 +10182,7 @@ pub unsafe fn vld3q_p16(a: *const p16) -> poly16x8x3_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld3_u64(a: *const u64) -> uint64x1x3_t {
transmute(vld3_s64(transmute(a)))
}
@@ -10195,7 +10195,7 @@ pub unsafe fn vld3_u64(a: *const u64) -> uint64x1x3_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld3_p64(a: *const p64) -> poly64x1x3_t {
transmute(vld3_s64(transmute(a)))
}
@@ -10505,7 +10505,7 @@ vld3_dup_s64_(a as _)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld3))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld3r))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld3_dup_u8(a: *const u8) -> uint8x8x3_t {
transmute(vld3_dup_s8(transmute(a)))
}
@@ -10518,7 +10518,7 @@ pub unsafe fn vld3_dup_u8(a: *const u8) -> uint8x8x3_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld3))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld3r))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld3_dup_u16(a: *const u16) -> uint16x4x3_t {
transmute(vld3_dup_s16(transmute(a)))
}
@@ -10531,7 +10531,7 @@ pub unsafe fn vld3_dup_u16(a: *const u16) -> uint16x4x3_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld3))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld3r))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld3_dup_u32(a: *const u32) -> uint32x2x3_t {
transmute(vld3_dup_s32(transmute(a)))
}
@@ -10544,7 +10544,7 @@ pub unsafe fn vld3_dup_u32(a: *const u32) -> uint32x2x3_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld3))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld3r))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld3q_dup_u8(a: *const u8) -> uint8x16x3_t {
transmute(vld3q_dup_s8(transmute(a)))
}
@@ -10557,7 +10557,7 @@ pub unsafe fn vld3q_dup_u8(a: *const u8) -> uint8x16x3_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld3))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld3r))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld3q_dup_u16(a: *const u16) -> uint16x8x3_t {
transmute(vld3q_dup_s16(transmute(a)))
}
@@ -10570,7 +10570,7 @@ pub unsafe fn vld3q_dup_u16(a: *const u16) -> uint16x8x3_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld3))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld3r))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld3q_dup_u32(a: *const u32) -> uint32x4x3_t {
transmute(vld3q_dup_s32(transmute(a)))
}
@@ -10583,7 +10583,7 @@ pub unsafe fn vld3q_dup_u32(a: *const u32) -> uint32x4x3_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld3))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld3r))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld3_dup_p8(a: *const p8) -> poly8x8x3_t {
transmute(vld3_dup_s8(transmute(a)))
}
@@ -10596,7 +10596,7 @@ pub unsafe fn vld3_dup_p8(a: *const p8) -> poly8x8x3_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld3))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld3r))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld3_dup_p16(a: *const p16) -> poly16x4x3_t {
transmute(vld3_dup_s16(transmute(a)))
}
@@ -10609,7 +10609,7 @@ pub unsafe fn vld3_dup_p16(a: *const p16) -> poly16x4x3_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld3))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld3r))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld3q_dup_p8(a: *const p8) -> poly8x16x3_t {
transmute(vld3q_dup_s8(transmute(a)))
}
@@ -10622,7 +10622,7 @@ pub unsafe fn vld3q_dup_p8(a: *const p8) -> poly8x16x3_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld3))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld3r))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld3q_dup_p16(a: *const p16) -> poly16x8x3_t {
transmute(vld3q_dup_s16(transmute(a)))
}
@@ -10635,7 +10635,7 @@ pub unsafe fn vld3q_dup_p16(a: *const p16) -> poly16x8x3_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld3r))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld3_dup_u64(a: *const u64) -> uint64x1x3_t {
transmute(vld3_dup_s64(transmute(a)))
}
@@ -10648,7 +10648,7 @@ pub unsafe fn vld3_dup_u64(a: *const u64) -> uint64x1x3_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld3r))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld3_dup_p64(a: *const p64) -> poly64x1x3_t {
transmute(vld3_dup_s64(transmute(a)))
}
@@ -10913,7 +10913,7 @@ vld3q_lane_s32_(b.0, b.1, b.2, LANE as i64, a as _)
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld3, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld3, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld3_lane_u8<const LANE: i32>(a: *const u8, b: uint8x8x3_t) -> uint8x8x3_t {
static_assert_imm3!(LANE);
transmute(vld3_lane_s8::<LANE>(transmute(a), transmute(b)))
@@ -10928,7 +10928,7 @@ pub unsafe fn vld3_lane_u8<const LANE: i32>(a: *const u8, b: uint8x8x3_t) -> uin
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld3, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld3, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld3_lane_u16<const LANE: i32>(a: *const u16, b: uint16x4x3_t) -> uint16x4x3_t {
static_assert_imm2!(LANE);
transmute(vld3_lane_s16::<LANE>(transmute(a), transmute(b)))
@@ -10943,7 +10943,7 @@ pub unsafe fn vld3_lane_u16<const LANE: i32>(a: *const u16, b: uint16x4x3_t) ->
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld3, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld3, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld3_lane_u32<const LANE: i32>(a: *const u32, b: uint32x2x3_t) -> uint32x2x3_t {
static_assert_imm1!(LANE);
transmute(vld3_lane_s32::<LANE>(transmute(a), transmute(b)))
@@ -10958,7 +10958,7 @@ pub unsafe fn vld3_lane_u32<const LANE: i32>(a: *const u32, b: uint32x2x3_t) ->
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld3, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld3, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld3q_lane_u16<const LANE: i32>(a: *const u16, b: uint16x8x3_t) -> uint16x8x3_t {
static_assert_imm3!(LANE);
transmute(vld3q_lane_s16::<LANE>(transmute(a), transmute(b)))
@@ -10973,7 +10973,7 @@ pub unsafe fn vld3q_lane_u16<const LANE: i32>(a: *const u16, b: uint16x8x3_t) ->
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld3, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld3, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld3q_lane_u32<const LANE: i32>(a: *const u32, b: uint32x4x3_t) -> uint32x4x3_t {
static_assert_imm2!(LANE);
transmute(vld3q_lane_s32::<LANE>(transmute(a), transmute(b)))
@@ -10988,7 +10988,7 @@ pub unsafe fn vld3q_lane_u32<const LANE: i32>(a: *const u32, b: uint32x4x3_t) ->
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld3, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld3, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld3_lane_p8<const LANE: i32>(a: *const p8, b: poly8x8x3_t) -> poly8x8x3_t {
static_assert_imm3!(LANE);
transmute(vld3_lane_s8::<LANE>(transmute(a), transmute(b)))
@@ -11003,7 +11003,7 @@ pub unsafe fn vld3_lane_p8<const LANE: i32>(a: *const p8, b: poly8x8x3_t) -> pol
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld3, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld3, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld3_lane_p16<const LANE: i32>(a: *const p16, b: poly16x4x3_t) -> poly16x4x3_t {
static_assert_imm2!(LANE);
transmute(vld3_lane_s16::<LANE>(transmute(a), transmute(b)))
@@ -11018,7 +11018,7 @@ pub unsafe fn vld3_lane_p16<const LANE: i32>(a: *const p16, b: poly16x4x3_t) ->
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld3, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld3, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld3q_lane_p16<const LANE: i32>(a: *const p16, b: poly16x8x3_t) -> poly16x8x3_t {
static_assert_imm3!(LANE);
transmute(vld3q_lane_s16::<LANE>(transmute(a), transmute(b)))
@@ -11337,7 +11337,7 @@ vld4_s64_(a as _)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld4))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld4))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld4_u8(a: *const u8) -> uint8x8x4_t {
transmute(vld4_s8(transmute(a)))
}
@@ -11350,7 +11350,7 @@ pub unsafe fn vld4_u8(a: *const u8) -> uint8x8x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld4))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld4))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld4_u16(a: *const u16) -> uint16x4x4_t {
transmute(vld4_s16(transmute(a)))
}
@@ -11363,7 +11363,7 @@ pub unsafe fn vld4_u16(a: *const u16) -> uint16x4x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld4))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld4))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld4_u32(a: *const u32) -> uint32x2x4_t {
transmute(vld4_s32(transmute(a)))
}
@@ -11376,7 +11376,7 @@ pub unsafe fn vld4_u32(a: *const u32) -> uint32x2x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld4))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld4))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld4q_u8(a: *const u8) -> uint8x16x4_t {
transmute(vld4q_s8(transmute(a)))
}
@@ -11389,7 +11389,7 @@ pub unsafe fn vld4q_u8(a: *const u8) -> uint8x16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld4))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld4))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld4q_u16(a: *const u16) -> uint16x8x4_t {
transmute(vld4q_s16(transmute(a)))
}
@@ -11402,7 +11402,7 @@ pub unsafe fn vld4q_u16(a: *const u16) -> uint16x8x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld4))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld4))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld4q_u32(a: *const u32) -> uint32x4x4_t {
transmute(vld4q_s32(transmute(a)))
}
@@ -11415,7 +11415,7 @@ pub unsafe fn vld4q_u32(a: *const u32) -> uint32x4x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld4))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld4))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld4_p8(a: *const p8) -> poly8x8x4_t {
transmute(vld4_s8(transmute(a)))
}
@@ -11428,7 +11428,7 @@ pub unsafe fn vld4_p8(a: *const p8) -> poly8x8x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld4))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld4))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld4_p16(a: *const p16) -> poly16x4x4_t {
transmute(vld4_s16(transmute(a)))
}
@@ -11441,7 +11441,7 @@ pub unsafe fn vld4_p16(a: *const p16) -> poly16x4x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld4))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld4))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld4q_p8(a: *const p8) -> poly8x16x4_t {
transmute(vld4q_s8(transmute(a)))
}
@@ -11454,7 +11454,7 @@ pub unsafe fn vld4q_p8(a: *const p8) -> poly8x16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld4))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld4))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld4q_p16(a: *const p16) -> poly16x8x4_t {
transmute(vld4q_s16(transmute(a)))
}
@@ -11467,7 +11467,7 @@ pub unsafe fn vld4q_p16(a: *const p16) -> poly16x8x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld4_u64(a: *const u64) -> uint64x1x4_t {
transmute(vld4_s64(transmute(a)))
}
@@ -11480,7 +11480,7 @@ pub unsafe fn vld4_u64(a: *const u64) -> uint64x1x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld4_p64(a: *const p64) -> poly64x1x4_t {
transmute(vld4_s64(transmute(a)))
}
@@ -11790,7 +11790,7 @@ vld4_dup_s64_(a as _)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld4))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld4r))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld4_dup_u8(a: *const u8) -> uint8x8x4_t {
transmute(vld4_dup_s8(transmute(a)))
}
@@ -11803,7 +11803,7 @@ pub unsafe fn vld4_dup_u8(a: *const u8) -> uint8x8x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld4))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld4r))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld4_dup_u16(a: *const u16) -> uint16x4x4_t {
transmute(vld4_dup_s16(transmute(a)))
}
@@ -11816,7 +11816,7 @@ pub unsafe fn vld4_dup_u16(a: *const u16) -> uint16x4x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld4))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld4r))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld4_dup_u32(a: *const u32) -> uint32x2x4_t {
transmute(vld4_dup_s32(transmute(a)))
}
@@ -11829,7 +11829,7 @@ pub unsafe fn vld4_dup_u32(a: *const u32) -> uint32x2x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld4))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld4r))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld4q_dup_u8(a: *const u8) -> uint8x16x4_t {
transmute(vld4q_dup_s8(transmute(a)))
}
@@ -11842,7 +11842,7 @@ pub unsafe fn vld4q_dup_u8(a: *const u8) -> uint8x16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld4))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld4r))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld4q_dup_u16(a: *const u16) -> uint16x8x4_t {
transmute(vld4q_dup_s16(transmute(a)))
}
@@ -11855,7 +11855,7 @@ pub unsafe fn vld4q_dup_u16(a: *const u16) -> uint16x8x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld4))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld4r))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld4q_dup_u32(a: *const u32) -> uint32x4x4_t {
transmute(vld4q_dup_s32(transmute(a)))
}
@@ -11868,7 +11868,7 @@ pub unsafe fn vld4q_dup_u32(a: *const u32) -> uint32x4x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld4))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld4r))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld4_dup_p8(a: *const p8) -> poly8x8x4_t {
transmute(vld4_dup_s8(transmute(a)))
}
@@ -11881,7 +11881,7 @@ pub unsafe fn vld4_dup_p8(a: *const p8) -> poly8x8x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld4))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld4r))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld4_dup_p16(a: *const p16) -> poly16x4x4_t {
transmute(vld4_dup_s16(transmute(a)))
}
@@ -11894,7 +11894,7 @@ pub unsafe fn vld4_dup_p16(a: *const p16) -> poly16x4x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld4))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld4r))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld4q_dup_p8(a: *const p8) -> poly8x16x4_t {
transmute(vld4q_dup_s8(transmute(a)))
}
@@ -11907,7 +11907,7 @@ pub unsafe fn vld4q_dup_p8(a: *const p8) -> poly8x16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld4))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld4r))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld4q_dup_p16(a: *const p16) -> poly16x8x4_t {
transmute(vld4q_dup_s16(transmute(a)))
}
@@ -11920,7 +11920,7 @@ pub unsafe fn vld4q_dup_p16(a: *const p16) -> poly16x8x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld4r))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld4_dup_u64(a: *const u64) -> uint64x1x4_t {
transmute(vld4_dup_s64(transmute(a)))
}
@@ -11933,7 +11933,7 @@ pub unsafe fn vld4_dup_u64(a: *const u64) -> uint64x1x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld4r))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld4_dup_p64(a: *const p64) -> poly64x1x4_t {
transmute(vld4_dup_s64(transmute(a)))
}
@@ -12198,7 +12198,7 @@ vld4q_lane_s32_(b.0, b.1, b.2, b.3, LANE as i64, a as _)
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld4, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld4, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld4_lane_u8<const LANE: i32>(a: *const u8, b: uint8x8x4_t) -> uint8x8x4_t {
static_assert_imm3!(LANE);
transmute(vld4_lane_s8::<LANE>(transmute(a), transmute(b)))
@@ -12213,7 +12213,7 @@ pub unsafe fn vld4_lane_u8<const LANE: i32>(a: *const u8, b: uint8x8x4_t) -> uin
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld4, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld4, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld4_lane_u16<const LANE: i32>(a: *const u16, b: uint16x4x4_t) -> uint16x4x4_t {
static_assert_imm2!(LANE);
transmute(vld4_lane_s16::<LANE>(transmute(a), transmute(b)))
@@ -12228,7 +12228,7 @@ pub unsafe fn vld4_lane_u16<const LANE: i32>(a: *const u16, b: uint16x4x4_t) ->
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld4, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld4, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld4_lane_u32<const LANE: i32>(a: *const u32, b: uint32x2x4_t) -> uint32x2x4_t {
static_assert_imm1!(LANE);
transmute(vld4_lane_s32::<LANE>(transmute(a), transmute(b)))
@@ -12243,7 +12243,7 @@ pub unsafe fn vld4_lane_u32<const LANE: i32>(a: *const u32, b: uint32x2x4_t) ->
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld4, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld4, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld4q_lane_u16<const LANE: i32>(a: *const u16, b: uint16x8x4_t) -> uint16x8x4_t {
static_assert_imm3!(LANE);
transmute(vld4q_lane_s16::<LANE>(transmute(a), transmute(b)))
@@ -12258,7 +12258,7 @@ pub unsafe fn vld4q_lane_u16<const LANE: i32>(a: *const u16, b: uint16x8x4_t) ->
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld4, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld4, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld4q_lane_u32<const LANE: i32>(a: *const u32, b: uint32x4x4_t) -> uint32x4x4_t {
static_assert_imm2!(LANE);
transmute(vld4q_lane_s32::<LANE>(transmute(a), transmute(b)))
@@ -12273,7 +12273,7 @@ pub unsafe fn vld4q_lane_u32<const LANE: i32>(a: *const u32, b: uint32x4x4_t) ->
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld4, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld4, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld4_lane_p8<const LANE: i32>(a: *const p8, b: poly8x8x4_t) -> poly8x8x4_t {
static_assert_imm3!(LANE);
transmute(vld4_lane_s8::<LANE>(transmute(a), transmute(b)))
@@ -12288,7 +12288,7 @@ pub unsafe fn vld4_lane_p8<const LANE: i32>(a: *const p8, b: poly8x8x4_t) -> pol
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld4, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld4, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld4_lane_p16<const LANE: i32>(a: *const p16, b: poly16x4x4_t) -> poly16x4x4_t {
static_assert_imm2!(LANE);
transmute(vld4_lane_s16::<LANE>(transmute(a), transmute(b)))
@@ -12303,7 +12303,7 @@ pub unsafe fn vld4_lane_p16<const LANE: i32>(a: *const p16, b: poly16x4x4_t) ->
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vld4, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ld4, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vld4q_lane_p16<const LANE: i32>(a: *const p16, b: poly16x8x4_t) -> poly16x8x4_t {
static_assert_imm3!(LANE);
transmute(vld4q_lane_s16::<LANE>(transmute(a), transmute(b)))
@@ -12392,7 +12392,7 @@ vld4q_lane_f32_(b.0, b.1, b.2, b.3, LANE as i64, a as _)
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1_lane_s8<const LANE: i32>(a: *mut i8, b: int8x8_t) {
static_assert_imm3!(LANE);
*a = simd_extract(b, LANE as u32);
@@ -12407,7 +12407,7 @@ pub unsafe fn vst1_lane_s8<const LANE: i32>(a: *mut i8, b: int8x8_t) {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1_lane_s16<const LANE: i32>(a: *mut i16, b: int16x4_t) {
static_assert_imm2!(LANE);
*a = simd_extract(b, LANE as u32);
@@ -12422,7 +12422,7 @@ pub unsafe fn vst1_lane_s16<const LANE: i32>(a: *mut i16, b: int16x4_t) {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1_lane_s32<const LANE: i32>(a: *mut i32, b: int32x2_t) {
static_assert_imm1!(LANE);
*a = simd_extract(b, LANE as u32);
@@ -12437,7 +12437,7 @@ pub unsafe fn vst1_lane_s32<const LANE: i32>(a: *mut i32, b: int32x2_t) {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1_lane_s64<const LANE: i32>(a: *mut i64, b: int64x1_t) {
static_assert!(LANE : i32 where LANE == 0);
*a = simd_extract(b, LANE as u32);
@@ -12452,7 +12452,7 @@ pub unsafe fn vst1_lane_s64<const LANE: i32>(a: *mut i64, b: int64x1_t) {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1q_lane_s8<const LANE: i32>(a: *mut i8, b: int8x16_t) {
static_assert_imm4!(LANE);
*a = simd_extract(b, LANE as u32);
@@ -12467,7 +12467,7 @@ pub unsafe fn vst1q_lane_s8<const LANE: i32>(a: *mut i8, b: int8x16_t) {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1q_lane_s16<const LANE: i32>(a: *mut i16, b: int16x8_t) {
static_assert_imm3!(LANE);
*a = simd_extract(b, LANE as u32);
@@ -12482,7 +12482,7 @@ pub unsafe fn vst1q_lane_s16<const LANE: i32>(a: *mut i16, b: int16x8_t) {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1q_lane_s32<const LANE: i32>(a: *mut i32, b: int32x4_t) {
static_assert_imm2!(LANE);
*a = simd_extract(b, LANE as u32);
@@ -12497,7 +12497,7 @@ pub unsafe fn vst1q_lane_s32<const LANE: i32>(a: *mut i32, b: int32x4_t) {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1q_lane_s64<const LANE: i32>(a: *mut i64, b: int64x2_t) {
static_assert_imm1!(LANE);
*a = simd_extract(b, LANE as u32);
@@ -12512,7 +12512,7 @@ pub unsafe fn vst1q_lane_s64<const LANE: i32>(a: *mut i64, b: int64x2_t) {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1_lane_u8<const LANE: i32>(a: *mut u8, b: uint8x8_t) {
static_assert_imm3!(LANE);
*a = simd_extract(b, LANE as u32);
@@ -12527,7 +12527,7 @@ pub unsafe fn vst1_lane_u8<const LANE: i32>(a: *mut u8, b: uint8x8_t) {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1_lane_u16<const LANE: i32>(a: *mut u16, b: uint16x4_t) {
static_assert_imm2!(LANE);
*a = simd_extract(b, LANE as u32);
@@ -12542,7 +12542,7 @@ pub unsafe fn vst1_lane_u16<const LANE: i32>(a: *mut u16, b: uint16x4_t) {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1_lane_u32<const LANE: i32>(a: *mut u32, b: uint32x2_t) {
static_assert_imm1!(LANE);
*a = simd_extract(b, LANE as u32);
@@ -12557,7 +12557,7 @@ pub unsafe fn vst1_lane_u32<const LANE: i32>(a: *mut u32, b: uint32x2_t) {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1_lane_u64<const LANE: i32>(a: *mut u64, b: uint64x1_t) {
static_assert!(LANE : i32 where LANE == 0);
*a = simd_extract(b, LANE as u32);
@@ -12572,7 +12572,7 @@ pub unsafe fn vst1_lane_u64<const LANE: i32>(a: *mut u64, b: uint64x1_t) {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1q_lane_u8<const LANE: i32>(a: *mut u8, b: uint8x16_t) {
static_assert_imm4!(LANE);
*a = simd_extract(b, LANE as u32);
@@ -12587,7 +12587,7 @@ pub unsafe fn vst1q_lane_u8<const LANE: i32>(a: *mut u8, b: uint8x16_t) {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1q_lane_u16<const LANE: i32>(a: *mut u16, b: uint16x8_t) {
static_assert_imm3!(LANE);
*a = simd_extract(b, LANE as u32);
@@ -12602,7 +12602,7 @@ pub unsafe fn vst1q_lane_u16<const LANE: i32>(a: *mut u16, b: uint16x8_t) {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1q_lane_u32<const LANE: i32>(a: *mut u32, b: uint32x4_t) {
static_assert_imm2!(LANE);
*a = simd_extract(b, LANE as u32);
@@ -12617,7 +12617,7 @@ pub unsafe fn vst1q_lane_u32<const LANE: i32>(a: *mut u32, b: uint32x4_t) {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1q_lane_u64<const LANE: i32>(a: *mut u64, b: uint64x2_t) {
static_assert_imm1!(LANE);
*a = simd_extract(b, LANE as u32);
@@ -12632,7 +12632,7 @@ pub unsafe fn vst1q_lane_u64<const LANE: i32>(a: *mut u64, b: uint64x2_t) {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1_lane_p8<const LANE: i32>(a: *mut p8, b: poly8x8_t) {
static_assert_imm3!(LANE);
*a = simd_extract(b, LANE as u32);
@@ -12647,7 +12647,7 @@ pub unsafe fn vst1_lane_p8<const LANE: i32>(a: *mut p8, b: poly8x8_t) {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1_lane_p16<const LANE: i32>(a: *mut p16, b: poly16x4_t) {
static_assert_imm2!(LANE);
*a = simd_extract(b, LANE as u32);
@@ -12662,7 +12662,7 @@ pub unsafe fn vst1_lane_p16<const LANE: i32>(a: *mut p16, b: poly16x4_t) {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1q_lane_p8<const LANE: i32>(a: *mut p8, b: poly8x16_t) {
static_assert_imm4!(LANE);
*a = simd_extract(b, LANE as u32);
@@ -12677,7 +12677,7 @@ pub unsafe fn vst1q_lane_p8<const LANE: i32>(a: *mut p8, b: poly8x16_t) {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1q_lane_p16<const LANE: i32>(a: *mut p16, b: poly16x8_t) {
static_assert_imm3!(LANE);
*a = simd_extract(b, LANE as u32);
@@ -12692,7 +12692,7 @@ pub unsafe fn vst1q_lane_p16<const LANE: i32>(a: *mut p16, b: poly16x8_t) {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1_lane_p64<const LANE: i32>(a: *mut p64, b: poly64x1_t) {
static_assert!(LANE : i32 where LANE == 0);
*a = simd_extract(b, LANE as u32);
@@ -12707,7 +12707,7 @@ pub unsafe fn vst1_lane_p64<const LANE: i32>(a: *mut p64, b: poly64x1_t) {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1q_lane_p64<const LANE: i32>(a: *mut p64, b: poly64x2_t) {
static_assert_imm1!(LANE);
*a = simd_extract(b, LANE as u32);
@@ -12722,7 +12722,7 @@ pub unsafe fn vst1q_lane_p64<const LANE: i32>(a: *mut p64, b: poly64x2_t) {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1_lane_f32<const LANE: i32>(a: *mut f32, b: float32x2_t) {
static_assert_imm1!(LANE);
*a = simd_extract(b, LANE as u32);
@@ -12737,7 +12737,7 @@ pub unsafe fn vst1_lane_f32<const LANE: i32>(a: *mut f32, b: float32x2_t) {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1q_lane_f32<const LANE: i32>(a: *mut f32, b: float32x4_t) {
static_assert_imm2!(LANE);
*a = simd_extract(b, LANE as u32);
@@ -13543,7 +13543,7 @@ vst1q_s64_x4_(b.0, b.1, b.2, b.3, a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1_u8_x2(a: *mut u8, b: uint8x8x2_t) {
vst1_s8_x2(transmute(a), transmute(b))
}
@@ -13556,7 +13556,7 @@ pub unsafe fn vst1_u8_x2(a: *mut u8, b: uint8x8x2_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1_u16_x2(a: *mut u16, b: uint16x4x2_t) {
vst1_s16_x2(transmute(a), transmute(b))
}
@@ -13569,7 +13569,7 @@ pub unsafe fn vst1_u16_x2(a: *mut u16, b: uint16x4x2_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1_u32_x2(a: *mut u32, b: uint32x2x2_t) {
vst1_s32_x2(transmute(a), transmute(b))
}
@@ -13582,7 +13582,7 @@ pub unsafe fn vst1_u32_x2(a: *mut u32, b: uint32x2x2_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1_u64_x2(a: *mut u64, b: uint64x1x2_t) {
vst1_s64_x2(transmute(a), transmute(b))
}
@@ -13595,7 +13595,7 @@ pub unsafe fn vst1_u64_x2(a: *mut u64, b: uint64x1x2_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1q_u8_x2(a: *mut u8, b: uint8x16x2_t) {
vst1q_s8_x2(transmute(a), transmute(b))
}
@@ -13608,7 +13608,7 @@ pub unsafe fn vst1q_u8_x2(a: *mut u8, b: uint8x16x2_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1q_u16_x2(a: *mut u16, b: uint16x8x2_t) {
vst1q_s16_x2(transmute(a), transmute(b))
}
@@ -13621,7 +13621,7 @@ pub unsafe fn vst1q_u16_x2(a: *mut u16, b: uint16x8x2_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1q_u32_x2(a: *mut u32, b: uint32x4x2_t) {
vst1q_s32_x2(transmute(a), transmute(b))
}
@@ -13634,7 +13634,7 @@ pub unsafe fn vst1q_u32_x2(a: *mut u32, b: uint32x4x2_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1q_u64_x2(a: *mut u64, b: uint64x2x2_t) {
vst1q_s64_x2(transmute(a), transmute(b))
}
@@ -13647,7 +13647,7 @@ pub unsafe fn vst1q_u64_x2(a: *mut u64, b: uint64x2x2_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1_u8_x3(a: *mut u8, b: uint8x8x3_t) {
vst1_s8_x3(transmute(a), transmute(b))
}
@@ -13660,7 +13660,7 @@ pub unsafe fn vst1_u8_x3(a: *mut u8, b: uint8x8x3_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1_u16_x3(a: *mut u16, b: uint16x4x3_t) {
vst1_s16_x3(transmute(a), transmute(b))
}
@@ -13673,7 +13673,7 @@ pub unsafe fn vst1_u16_x3(a: *mut u16, b: uint16x4x3_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1_u32_x3(a: *mut u32, b: uint32x2x3_t) {
vst1_s32_x3(transmute(a), transmute(b))
}
@@ -13686,7 +13686,7 @@ pub unsafe fn vst1_u32_x3(a: *mut u32, b: uint32x2x3_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1_u64_x3(a: *mut u64, b: uint64x1x3_t) {
vst1_s64_x3(transmute(a), transmute(b))
}
@@ -13699,7 +13699,7 @@ pub unsafe fn vst1_u64_x3(a: *mut u64, b: uint64x1x3_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1q_u8_x3(a: *mut u8, b: uint8x16x3_t) {
vst1q_s8_x3(transmute(a), transmute(b))
}
@@ -13712,7 +13712,7 @@ pub unsafe fn vst1q_u8_x3(a: *mut u8, b: uint8x16x3_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1q_u16_x3(a: *mut u16, b: uint16x8x3_t) {
vst1q_s16_x3(transmute(a), transmute(b))
}
@@ -13725,7 +13725,7 @@ pub unsafe fn vst1q_u16_x3(a: *mut u16, b: uint16x8x3_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1q_u32_x3(a: *mut u32, b: uint32x4x3_t) {
vst1q_s32_x3(transmute(a), transmute(b))
}
@@ -13738,7 +13738,7 @@ pub unsafe fn vst1q_u32_x3(a: *mut u32, b: uint32x4x3_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1q_u64_x3(a: *mut u64, b: uint64x2x3_t) {
vst1q_s64_x3(transmute(a), transmute(b))
}
@@ -13751,7 +13751,7 @@ pub unsafe fn vst1q_u64_x3(a: *mut u64, b: uint64x2x3_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1_u8_x4(a: *mut u8, b: uint8x8x4_t) {
vst1_s8_x4(transmute(a), transmute(b))
}
@@ -13764,7 +13764,7 @@ pub unsafe fn vst1_u8_x4(a: *mut u8, b: uint8x8x4_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1_u16_x4(a: *mut u16, b: uint16x4x4_t) {
vst1_s16_x4(transmute(a), transmute(b))
}
@@ -13777,7 +13777,7 @@ pub unsafe fn vst1_u16_x4(a: *mut u16, b: uint16x4x4_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1_u32_x4(a: *mut u32, b: uint32x2x4_t) {
vst1_s32_x4(transmute(a), transmute(b))
}
@@ -13790,7 +13790,7 @@ pub unsafe fn vst1_u32_x4(a: *mut u32, b: uint32x2x4_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1_u64_x4(a: *mut u64, b: uint64x1x4_t) {
vst1_s64_x4(transmute(a), transmute(b))
}
@@ -13803,7 +13803,7 @@ pub unsafe fn vst1_u64_x4(a: *mut u64, b: uint64x1x4_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1q_u8_x4(a: *mut u8, b: uint8x16x4_t) {
vst1q_s8_x4(transmute(a), transmute(b))
}
@@ -13816,7 +13816,7 @@ pub unsafe fn vst1q_u8_x4(a: *mut u8, b: uint8x16x4_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1q_u16_x4(a: *mut u16, b: uint16x8x4_t) {
vst1q_s16_x4(transmute(a), transmute(b))
}
@@ -13829,7 +13829,7 @@ pub unsafe fn vst1q_u16_x4(a: *mut u16, b: uint16x8x4_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1q_u32_x4(a: *mut u32, b: uint32x4x4_t) {
vst1q_s32_x4(transmute(a), transmute(b))
}
@@ -13842,7 +13842,7 @@ pub unsafe fn vst1q_u32_x4(a: *mut u32, b: uint32x4x4_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1q_u64_x4(a: *mut u64, b: uint64x2x4_t) {
vst1q_s64_x4(transmute(a), transmute(b))
}
@@ -13855,7 +13855,7 @@ pub unsafe fn vst1q_u64_x4(a: *mut u64, b: uint64x2x4_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1_p8_x2(a: *mut p8, b: poly8x8x2_t) {
vst1_s8_x2(transmute(a), transmute(b))
}
@@ -13868,7 +13868,7 @@ pub unsafe fn vst1_p8_x2(a: *mut p8, b: poly8x8x2_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1_p8_x3(a: *mut p8, b: poly8x8x3_t) {
vst1_s8_x3(transmute(a), transmute(b))
}
@@ -13881,7 +13881,7 @@ pub unsafe fn vst1_p8_x3(a: *mut p8, b: poly8x8x3_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1_p8_x4(a: *mut p8, b: poly8x8x4_t) {
vst1_s8_x4(transmute(a), transmute(b))
}
@@ -13894,7 +13894,7 @@ pub unsafe fn vst1_p8_x4(a: *mut p8, b: poly8x8x4_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1q_p8_x2(a: *mut p8, b: poly8x16x2_t) {
vst1q_s8_x2(transmute(a), transmute(b))
}
@@ -13907,7 +13907,7 @@ pub unsafe fn vst1q_p8_x2(a: *mut p8, b: poly8x16x2_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1q_p8_x3(a: *mut p8, b: poly8x16x3_t) {
vst1q_s8_x3(transmute(a), transmute(b))
}
@@ -13920,7 +13920,7 @@ pub unsafe fn vst1q_p8_x3(a: *mut p8, b: poly8x16x3_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1q_p8_x4(a: *mut p8, b: poly8x16x4_t) {
vst1q_s8_x4(transmute(a), transmute(b))
}
@@ -13933,7 +13933,7 @@ pub unsafe fn vst1q_p8_x4(a: *mut p8, b: poly8x16x4_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1_p16_x2(a: *mut p16, b: poly16x4x2_t) {
vst1_s16_x2(transmute(a), transmute(b))
}
@@ -13946,7 +13946,7 @@ pub unsafe fn vst1_p16_x2(a: *mut p16, b: poly16x4x2_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1_p16_x3(a: *mut p16, b: poly16x4x3_t) {
vst1_s16_x3(transmute(a), transmute(b))
}
@@ -13959,7 +13959,7 @@ pub unsafe fn vst1_p16_x3(a: *mut p16, b: poly16x4x3_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1_p16_x4(a: *mut p16, b: poly16x4x4_t) {
vst1_s16_x4(transmute(a), transmute(b))
}
@@ -13972,7 +13972,7 @@ pub unsafe fn vst1_p16_x4(a: *mut p16, b: poly16x4x4_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1q_p16_x2(a: *mut p16, b: poly16x8x2_t) {
vst1q_s16_x2(transmute(a), transmute(b))
}
@@ -13985,7 +13985,7 @@ pub unsafe fn vst1q_p16_x2(a: *mut p16, b: poly16x8x2_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1q_p16_x3(a: *mut p16, b: poly16x8x3_t) {
vst1q_s16_x3(transmute(a), transmute(b))
}
@@ -13998,7 +13998,7 @@ pub unsafe fn vst1q_p16_x3(a: *mut p16, b: poly16x8x3_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1q_p16_x4(a: *mut p16, b: poly16x8x4_t) {
vst1q_s16_x4(transmute(a), transmute(b))
}
@@ -14011,7 +14011,7 @@ pub unsafe fn vst1q_p16_x4(a: *mut p16, b: poly16x8x4_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1_p64_x2(a: *mut p64, b: poly64x1x2_t) {
vst1_s64_x2(transmute(a), transmute(b))
}
@@ -14024,7 +14024,7 @@ pub unsafe fn vst1_p64_x2(a: *mut p64, b: poly64x1x2_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1_p64_x3(a: *mut p64, b: poly64x1x3_t) {
vst1_s64_x3(transmute(a), transmute(b))
}
@@ -14037,7 +14037,7 @@ pub unsafe fn vst1_p64_x3(a: *mut p64, b: poly64x1x3_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1_p64_x4(a: *mut p64, b: poly64x1x4_t) {
vst1_s64_x4(transmute(a), transmute(b))
}
@@ -14050,7 +14050,7 @@ pub unsafe fn vst1_p64_x4(a: *mut p64, b: poly64x1x4_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1q_p64_x2(a: *mut p64, b: poly64x2x2_t) {
vst1q_s64_x2(transmute(a), transmute(b))
}
@@ -14063,7 +14063,7 @@ pub unsafe fn vst1q_p64_x2(a: *mut p64, b: poly64x2x2_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1q_p64_x3(a: *mut p64, b: poly64x2x3_t) {
vst1q_s64_x3(transmute(a), transmute(b))
}
@@ -14076,7 +14076,7 @@ pub unsafe fn vst1q_p64_x3(a: *mut p64, b: poly64x2x3_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st1))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst1q_p64_x4(a: *mut p64, b: poly64x2x4_t) {
vst1q_s64_x4(transmute(a), transmute(b))
}
@@ -14518,7 +14518,7 @@ vst2_s64_(b.0, b.1, a as _)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st2))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst2_u8(a: *mut u8, b: uint8x8x2_t) {
transmute(vst2_s8(transmute(a), transmute(b)))
}
@@ -14531,7 +14531,7 @@ pub unsafe fn vst2_u8(a: *mut u8, b: uint8x8x2_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st2))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst2_u16(a: *mut u16, b: uint16x4x2_t) {
transmute(vst2_s16(transmute(a), transmute(b)))
}
@@ -14544,7 +14544,7 @@ pub unsafe fn vst2_u16(a: *mut u16, b: uint16x4x2_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st2))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst2_u32(a: *mut u32, b: uint32x2x2_t) {
transmute(vst2_s32(transmute(a), transmute(b)))
}
@@ -14557,7 +14557,7 @@ pub unsafe fn vst2_u32(a: *mut u32, b: uint32x2x2_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st2))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst2q_u8(a: *mut u8, b: uint8x16x2_t) {
transmute(vst2q_s8(transmute(a), transmute(b)))
}
@@ -14570,7 +14570,7 @@ pub unsafe fn vst2q_u8(a: *mut u8, b: uint8x16x2_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st2))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst2q_u16(a: *mut u16, b: uint16x8x2_t) {
transmute(vst2q_s16(transmute(a), transmute(b)))
}
@@ -14583,7 +14583,7 @@ pub unsafe fn vst2q_u16(a: *mut u16, b: uint16x8x2_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st2))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst2q_u32(a: *mut u32, b: uint32x4x2_t) {
transmute(vst2q_s32(transmute(a), transmute(b)))
}
@@ -14596,7 +14596,7 @@ pub unsafe fn vst2q_u32(a: *mut u32, b: uint32x4x2_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st2))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst2_p8(a: *mut p8, b: poly8x8x2_t) {
transmute(vst2_s8(transmute(a), transmute(b)))
}
@@ -14609,7 +14609,7 @@ pub unsafe fn vst2_p8(a: *mut p8, b: poly8x8x2_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st2))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst2_p16(a: *mut p16, b: poly16x4x2_t) {
transmute(vst2_s16(transmute(a), transmute(b)))
}
@@ -14622,7 +14622,7 @@ pub unsafe fn vst2_p16(a: *mut p16, b: poly16x4x2_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st2))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst2q_p8(a: *mut p8, b: poly8x16x2_t) {
transmute(vst2q_s8(transmute(a), transmute(b)))
}
@@ -14635,7 +14635,7 @@ pub unsafe fn vst2q_p8(a: *mut p8, b: poly8x16x2_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st2))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst2q_p16(a: *mut p16, b: poly16x8x2_t) {
transmute(vst2q_s16(transmute(a), transmute(b)))
}
@@ -14648,7 +14648,7 @@ pub unsafe fn vst2q_p16(a: *mut p16, b: poly16x8x2_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst2_u64(a: *mut u64, b: uint64x1x2_t) {
transmute(vst2_s64(transmute(a), transmute(b)))
}
@@ -14661,7 +14661,7 @@ pub unsafe fn vst2_u64(a: *mut u64, b: uint64x1x2_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst2_p64(a: *mut p64, b: poly64x1x2_t) {
transmute(vst2_s64(transmute(a), transmute(b)))
}
@@ -14926,7 +14926,7 @@ vst2q_lane_s32_(b.0, b.1, LANE as i64, a as _)
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst2, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st2, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst2_lane_u8<const LANE: i32>(a: *mut u8, b: uint8x8x2_t) {
static_assert_imm3!(LANE);
transmute(vst2_lane_s8::<LANE>(transmute(a), transmute(b)))
@@ -14941,7 +14941,7 @@ pub unsafe fn vst2_lane_u8<const LANE: i32>(a: *mut u8, b: uint8x8x2_t) {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst2, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st2, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst2_lane_u16<const LANE: i32>(a: *mut u16, b: uint16x4x2_t) {
static_assert_imm2!(LANE);
transmute(vst2_lane_s16::<LANE>(transmute(a), transmute(b)))
@@ -14956,7 +14956,7 @@ pub unsafe fn vst2_lane_u16<const LANE: i32>(a: *mut u16, b: uint16x4x2_t) {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst2, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st2, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst2_lane_u32<const LANE: i32>(a: *mut u32, b: uint32x2x2_t) {
static_assert_imm1!(LANE);
transmute(vst2_lane_s32::<LANE>(transmute(a), transmute(b)))
@@ -14971,7 +14971,7 @@ pub unsafe fn vst2_lane_u32<const LANE: i32>(a: *mut u32, b: uint32x2x2_t) {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst2, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st2, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst2q_lane_u16<const LANE: i32>(a: *mut u16, b: uint16x8x2_t) {
static_assert_imm3!(LANE);
transmute(vst2q_lane_s16::<LANE>(transmute(a), transmute(b)))
@@ -14986,7 +14986,7 @@ pub unsafe fn vst2q_lane_u16<const LANE: i32>(a: *mut u16, b: uint16x8x2_t) {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst2, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st2, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst2q_lane_u32<const LANE: i32>(a: *mut u32, b: uint32x4x2_t) {
static_assert_imm2!(LANE);
transmute(vst2q_lane_s32::<LANE>(transmute(a), transmute(b)))
@@ -15001,7 +15001,7 @@ pub unsafe fn vst2q_lane_u32<const LANE: i32>(a: *mut u32, b: uint32x4x2_t) {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst2, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st2, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst2_lane_p8<const LANE: i32>(a: *mut p8, b: poly8x8x2_t) {
static_assert_imm3!(LANE);
transmute(vst2_lane_s8::<LANE>(transmute(a), transmute(b)))
@@ -15016,7 +15016,7 @@ pub unsafe fn vst2_lane_p8<const LANE: i32>(a: *mut p8, b: poly8x8x2_t) {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst2, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st2, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst2_lane_p16<const LANE: i32>(a: *mut p16, b: poly16x4x2_t) {
static_assert_imm2!(LANE);
transmute(vst2_lane_s16::<LANE>(transmute(a), transmute(b)))
@@ -15031,7 +15031,7 @@ pub unsafe fn vst2_lane_p16<const LANE: i32>(a: *mut p16, b: poly16x4x2_t) {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst2, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st2, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst2q_lane_p16<const LANE: i32>(a: *mut p16, b: poly16x8x2_t) {
static_assert_imm3!(LANE);
transmute(vst2q_lane_s16::<LANE>(transmute(a), transmute(b)))
@@ -15350,7 +15350,7 @@ vst3_s64_(b.0, b.1, b.2, a as _)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst3))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st3))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst3_u8(a: *mut u8, b: uint8x8x3_t) {
transmute(vst3_s8(transmute(a), transmute(b)))
}
@@ -15363,7 +15363,7 @@ pub unsafe fn vst3_u8(a: *mut u8, b: uint8x8x3_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst3))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st3))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst3_u16(a: *mut u16, b: uint16x4x3_t) {
transmute(vst3_s16(transmute(a), transmute(b)))
}
@@ -15376,7 +15376,7 @@ pub unsafe fn vst3_u16(a: *mut u16, b: uint16x4x3_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst3))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st3))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst3_u32(a: *mut u32, b: uint32x2x3_t) {
transmute(vst3_s32(transmute(a), transmute(b)))
}
@@ -15389,7 +15389,7 @@ pub unsafe fn vst3_u32(a: *mut u32, b: uint32x2x3_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst3))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st3))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst3q_u8(a: *mut u8, b: uint8x16x3_t) {
transmute(vst3q_s8(transmute(a), transmute(b)))
}
@@ -15402,7 +15402,7 @@ pub unsafe fn vst3q_u8(a: *mut u8, b: uint8x16x3_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst3))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st3))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst3q_u16(a: *mut u16, b: uint16x8x3_t) {
transmute(vst3q_s16(transmute(a), transmute(b)))
}
@@ -15415,7 +15415,7 @@ pub unsafe fn vst3q_u16(a: *mut u16, b: uint16x8x3_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst3))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st3))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst3q_u32(a: *mut u32, b: uint32x4x3_t) {
transmute(vst3q_s32(transmute(a), transmute(b)))
}
@@ -15428,7 +15428,7 @@ pub unsafe fn vst3q_u32(a: *mut u32, b: uint32x4x3_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst3))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st3))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst3_p8(a: *mut p8, b: poly8x8x3_t) {
transmute(vst3_s8(transmute(a), transmute(b)))
}
@@ -15441,7 +15441,7 @@ pub unsafe fn vst3_p8(a: *mut p8, b: poly8x8x3_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst3))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st3))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst3_p16(a: *mut p16, b: poly16x4x3_t) {
transmute(vst3_s16(transmute(a), transmute(b)))
}
@@ -15454,7 +15454,7 @@ pub unsafe fn vst3_p16(a: *mut p16, b: poly16x4x3_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst3))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st3))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst3q_p8(a: *mut p8, b: poly8x16x3_t) {
transmute(vst3q_s8(transmute(a), transmute(b)))
}
@@ -15467,7 +15467,7 @@ pub unsafe fn vst3q_p8(a: *mut p8, b: poly8x16x3_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst3))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st3))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst3q_p16(a: *mut p16, b: poly16x8x3_t) {
transmute(vst3q_s16(transmute(a), transmute(b)))
}
@@ -15480,7 +15480,7 @@ pub unsafe fn vst3q_p16(a: *mut p16, b: poly16x8x3_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst3_u64(a: *mut u64, b: uint64x1x3_t) {
transmute(vst3_s64(transmute(a), transmute(b)))
}
@@ -15493,7 +15493,7 @@ pub unsafe fn vst3_u64(a: *mut u64, b: uint64x1x3_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst3_p64(a: *mut p64, b: poly64x1x3_t) {
transmute(vst3_s64(transmute(a), transmute(b)))
}
@@ -15758,7 +15758,7 @@ vst3q_lane_s32_(b.0, b.1, b.2, LANE as i64, a as _)
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst3, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st3, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst3_lane_u8<const LANE: i32>(a: *mut u8, b: uint8x8x3_t) {
static_assert_imm3!(LANE);
transmute(vst3_lane_s8::<LANE>(transmute(a), transmute(b)))
@@ -15773,7 +15773,7 @@ pub unsafe fn vst3_lane_u8<const LANE: i32>(a: *mut u8, b: uint8x8x3_t) {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst3, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st3, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst3_lane_u16<const LANE: i32>(a: *mut u16, b: uint16x4x3_t) {
static_assert_imm2!(LANE);
transmute(vst3_lane_s16::<LANE>(transmute(a), transmute(b)))
@@ -15788,7 +15788,7 @@ pub unsafe fn vst3_lane_u16<const LANE: i32>(a: *mut u16, b: uint16x4x3_t) {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst3, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st3, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst3_lane_u32<const LANE: i32>(a: *mut u32, b: uint32x2x3_t) {
static_assert_imm1!(LANE);
transmute(vst3_lane_s32::<LANE>(transmute(a), transmute(b)))
@@ -15803,7 +15803,7 @@ pub unsafe fn vst3_lane_u32<const LANE: i32>(a: *mut u32, b: uint32x2x3_t) {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst3, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st3, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst3q_lane_u16<const LANE: i32>(a: *mut u16, b: uint16x8x3_t) {
static_assert_imm3!(LANE);
transmute(vst3q_lane_s16::<LANE>(transmute(a), transmute(b)))
@@ -15818,7 +15818,7 @@ pub unsafe fn vst3q_lane_u16<const LANE: i32>(a: *mut u16, b: uint16x8x3_t) {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst3, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st3, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst3q_lane_u32<const LANE: i32>(a: *mut u32, b: uint32x4x3_t) {
static_assert_imm2!(LANE);
transmute(vst3q_lane_s32::<LANE>(transmute(a), transmute(b)))
@@ -15833,7 +15833,7 @@ pub unsafe fn vst3q_lane_u32<const LANE: i32>(a: *mut u32, b: uint32x4x3_t) {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst3, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st3, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst3_lane_p8<const LANE: i32>(a: *mut p8, b: poly8x8x3_t) {
static_assert_imm3!(LANE);
transmute(vst3_lane_s8::<LANE>(transmute(a), transmute(b)))
@@ -15848,7 +15848,7 @@ pub unsafe fn vst3_lane_p8<const LANE: i32>(a: *mut p8, b: poly8x8x3_t) {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst3, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st3, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst3_lane_p16<const LANE: i32>(a: *mut p16, b: poly16x4x3_t) {
static_assert_imm2!(LANE);
transmute(vst3_lane_s16::<LANE>(transmute(a), transmute(b)))
@@ -15863,7 +15863,7 @@ pub unsafe fn vst3_lane_p16<const LANE: i32>(a: *mut p16, b: poly16x4x3_t) {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst3, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st3, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst3q_lane_p16<const LANE: i32>(a: *mut p16, b: poly16x8x3_t) {
static_assert_imm3!(LANE);
transmute(vst3q_lane_s16::<LANE>(transmute(a), transmute(b)))
@@ -16182,7 +16182,7 @@ vst4_s64_(b.0, b.1, b.2, b.3, a as _)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst4))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st4))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst4_u8(a: *mut u8, b: uint8x8x4_t) {
transmute(vst4_s8(transmute(a), transmute(b)))
}
@@ -16195,7 +16195,7 @@ pub unsafe fn vst4_u8(a: *mut u8, b: uint8x8x4_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst4))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st4))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst4_u16(a: *mut u16, b: uint16x4x4_t) {
transmute(vst4_s16(transmute(a), transmute(b)))
}
@@ -16208,7 +16208,7 @@ pub unsafe fn vst4_u16(a: *mut u16, b: uint16x4x4_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst4))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st4))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst4_u32(a: *mut u32, b: uint32x2x4_t) {
transmute(vst4_s32(transmute(a), transmute(b)))
}
@@ -16221,7 +16221,7 @@ pub unsafe fn vst4_u32(a: *mut u32, b: uint32x2x4_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst4))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st4))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst4q_u8(a: *mut u8, b: uint8x16x4_t) {
transmute(vst4q_s8(transmute(a), transmute(b)))
}
@@ -16234,7 +16234,7 @@ pub unsafe fn vst4q_u8(a: *mut u8, b: uint8x16x4_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst4))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st4))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst4q_u16(a: *mut u16, b: uint16x8x4_t) {
transmute(vst4q_s16(transmute(a), transmute(b)))
}
@@ -16247,7 +16247,7 @@ pub unsafe fn vst4q_u16(a: *mut u16, b: uint16x8x4_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst4))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st4))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst4q_u32(a: *mut u32, b: uint32x4x4_t) {
transmute(vst4q_s32(transmute(a), transmute(b)))
}
@@ -16260,7 +16260,7 @@ pub unsafe fn vst4q_u32(a: *mut u32, b: uint32x4x4_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst4))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st4))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst4_p8(a: *mut p8, b: poly8x8x4_t) {
transmute(vst4_s8(transmute(a), transmute(b)))
}
@@ -16273,7 +16273,7 @@ pub unsafe fn vst4_p8(a: *mut p8, b: poly8x8x4_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst4))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st4))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst4_p16(a: *mut p16, b: poly16x4x4_t) {
transmute(vst4_s16(transmute(a), transmute(b)))
}
@@ -16286,7 +16286,7 @@ pub unsafe fn vst4_p16(a: *mut p16, b: poly16x4x4_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst4))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st4))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst4q_p8(a: *mut p8, b: poly8x16x4_t) {
transmute(vst4q_s8(transmute(a), transmute(b)))
}
@@ -16299,7 +16299,7 @@ pub unsafe fn vst4q_p8(a: *mut p8, b: poly8x16x4_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst4))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st4))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst4q_p16(a: *mut p16, b: poly16x8x4_t) {
transmute(vst4q_s16(transmute(a), transmute(b)))
}
@@ -16312,7 +16312,7 @@ pub unsafe fn vst4q_p16(a: *mut p16, b: poly16x8x4_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst4_u64(a: *mut u64, b: uint64x1x4_t) {
transmute(vst4_s64(transmute(a), transmute(b)))
}
@@ -16325,7 +16325,7 @@ pub unsafe fn vst4_u64(a: *mut u64, b: uint64x1x4_t) {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst4_p64(a: *mut p64, b: poly64x1x4_t) {
transmute(vst4_s64(transmute(a), transmute(b)))
}
@@ -16590,7 +16590,7 @@ vst4q_lane_s32_(b.0, b.1, b.2, b.3, LANE as i64, a as _)
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst4, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st4, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst4_lane_u8<const LANE: i32>(a: *mut u8, b: uint8x8x4_t) {
static_assert_imm3!(LANE);
transmute(vst4_lane_s8::<LANE>(transmute(a), transmute(b)))
@@ -16605,7 +16605,7 @@ pub unsafe fn vst4_lane_u8<const LANE: i32>(a: *mut u8, b: uint8x8x4_t) {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst4, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st4, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst4_lane_u16<const LANE: i32>(a: *mut u16, b: uint16x4x4_t) {
static_assert_imm2!(LANE);
transmute(vst4_lane_s16::<LANE>(transmute(a), transmute(b)))
@@ -16620,7 +16620,7 @@ pub unsafe fn vst4_lane_u16<const LANE: i32>(a: *mut u16, b: uint16x4x4_t) {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst4, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st4, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst4_lane_u32<const LANE: i32>(a: *mut u32, b: uint32x2x4_t) {
static_assert_imm1!(LANE);
transmute(vst4_lane_s32::<LANE>(transmute(a), transmute(b)))
@@ -16635,7 +16635,7 @@ pub unsafe fn vst4_lane_u32<const LANE: i32>(a: *mut u32, b: uint32x2x4_t) {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst4, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st4, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst4q_lane_u16<const LANE: i32>(a: *mut u16, b: uint16x8x4_t) {
static_assert_imm3!(LANE);
transmute(vst4q_lane_s16::<LANE>(transmute(a), transmute(b)))
@@ -16650,7 +16650,7 @@ pub unsafe fn vst4q_lane_u16<const LANE: i32>(a: *mut u16, b: uint16x8x4_t) {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst4, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st4, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst4q_lane_u32<const LANE: i32>(a: *mut u32, b: uint32x4x4_t) {
static_assert_imm2!(LANE);
transmute(vst4q_lane_s32::<LANE>(transmute(a), transmute(b)))
@@ -16665,7 +16665,7 @@ pub unsafe fn vst4q_lane_u32<const LANE: i32>(a: *mut u32, b: uint32x4x4_t) {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst4, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st4, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst4_lane_p8<const LANE: i32>(a: *mut p8, b: poly8x8x4_t) {
static_assert_imm3!(LANE);
transmute(vst4_lane_s8::<LANE>(transmute(a), transmute(b)))
@@ -16680,7 +16680,7 @@ pub unsafe fn vst4_lane_p8<const LANE: i32>(a: *mut p8, b: poly8x8x4_t) {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst4, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st4, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst4_lane_p16<const LANE: i32>(a: *mut p16, b: poly16x4x4_t) {
static_assert_imm2!(LANE);
transmute(vst4_lane_s16::<LANE>(transmute(a), transmute(b)))
@@ -16695,7 +16695,7 @@ pub unsafe fn vst4_lane_p16<const LANE: i32>(a: *mut p16, b: poly16x4x4_t) {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vst4, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(st4, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vst4q_lane_p16<const LANE: i32>(a: *mut p16, b: poly16x8x4_t) {
static_assert_imm3!(LANE);
transmute(vst4q_lane_s16::<LANE>(transmute(a), transmute(b)))
@@ -16783,7 +16783,7 @@ vst4q_lane_f32_(b.0, b.1, b.2, b.3, LANE as i64, a as _)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmul.i8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mul))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmul_s8(a: int8x8_t, b: int8x8_t) -> int8x8_t {
simd_mul(a, b)
}
@@ -16796,7 +16796,7 @@ pub unsafe fn vmul_s8(a: int8x8_t, b: int8x8_t) -> int8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmul.i8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mul))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmulq_s8(a: int8x16_t, b: int8x16_t) -> int8x16_t {
simd_mul(a, b)
}
@@ -16809,7 +16809,7 @@ pub unsafe fn vmulq_s8(a: int8x16_t, b: int8x16_t) -> int8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmul.i16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mul))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmul_s16(a: int16x4_t, b: int16x4_t) -> int16x4_t {
simd_mul(a, b)
}
@@ -16822,7 +16822,7 @@ pub unsafe fn vmul_s16(a: int16x4_t, b: int16x4_t) -> int16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmul.i16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mul))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmulq_s16(a: int16x8_t, b: int16x8_t) -> int16x8_t {
simd_mul(a, b)
}
@@ -16835,7 +16835,7 @@ pub unsafe fn vmulq_s16(a: int16x8_t, b: int16x8_t) -> int16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmul.i32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mul))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmul_s32(a: int32x2_t, b: int32x2_t) -> int32x2_t {
simd_mul(a, b)
}
@@ -16848,7 +16848,7 @@ pub unsafe fn vmul_s32(a: int32x2_t, b: int32x2_t) -> int32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmul.i32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mul))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmulq_s32(a: int32x4_t, b: int32x4_t) -> int32x4_t {
simd_mul(a, b)
}
@@ -16861,7 +16861,7 @@ pub unsafe fn vmulq_s32(a: int32x4_t, b: int32x4_t) -> int32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmul.i8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mul))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmul_u8(a: uint8x8_t, b: uint8x8_t) -> uint8x8_t {
simd_mul(a, b)
}
@@ -16874,7 +16874,7 @@ pub unsafe fn vmul_u8(a: uint8x8_t, b: uint8x8_t) -> uint8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmul.i8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mul))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmulq_u8(a: uint8x16_t, b: uint8x16_t) -> uint8x16_t {
simd_mul(a, b)
}
@@ -16887,7 +16887,7 @@ pub unsafe fn vmulq_u8(a: uint8x16_t, b: uint8x16_t) -> uint8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmul.i16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mul))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmul_u16(a: uint16x4_t, b: uint16x4_t) -> uint16x4_t {
simd_mul(a, b)
}
@@ -16900,7 +16900,7 @@ pub unsafe fn vmul_u16(a: uint16x4_t, b: uint16x4_t) -> uint16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmul.i16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mul))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmulq_u16(a: uint16x8_t, b: uint16x8_t) -> uint16x8_t {
simd_mul(a, b)
}
@@ -16913,7 +16913,7 @@ pub unsafe fn vmulq_u16(a: uint16x8_t, b: uint16x8_t) -> uint16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmul.i32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mul))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmul_u32(a: uint32x2_t, b: uint32x2_t) -> uint32x2_t {
simd_mul(a, b)
}
@@ -16926,7 +16926,7 @@ pub unsafe fn vmul_u32(a: uint32x2_t, b: uint32x2_t) -> uint32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmul.i32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mul))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmulq_u32(a: uint32x4_t, b: uint32x4_t) -> uint32x4_t {
simd_mul(a, b)
}
@@ -16939,7 +16939,7 @@ pub unsafe fn vmulq_u32(a: uint32x4_t, b: uint32x4_t) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmul))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(pmul))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmul_p8(a: poly8x8_t, b: poly8x8_t) -> poly8x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -16958,7 +16958,7 @@ vmul_p8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmul))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(pmul))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmulq_p8(a: poly8x16_t, b: poly8x16_t) -> poly8x16_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -16977,7 +16977,7 @@ vmulq_p8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmul.f32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fmul))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmul_f32(a: float32x2_t, b: float32x2_t) -> float32x2_t {
simd_mul(a, b)
}
@@ -16990,7 +16990,7 @@ pub unsafe fn vmul_f32(a: float32x2_t, b: float32x2_t) -> float32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmul.f32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fmul))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmulq_f32(a: float32x4_t, b: float32x4_t) -> float32x4_t {
simd_mul(a, b)
}
@@ -17003,7 +17003,7 @@ pub unsafe fn vmulq_f32(a: float32x4_t, b: float32x4_t) -> float32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmul))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mul))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmul_n_s16(a: int16x4_t, b: i16) -> int16x4_t {
simd_mul(a, vdup_n_s16(b))
}
@@ -17016,7 +17016,7 @@ pub unsafe fn vmul_n_s16(a: int16x4_t, b: i16) -> int16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmul))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mul))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmulq_n_s16(a: int16x8_t, b: i16) -> int16x8_t {
simd_mul(a, vdupq_n_s16(b))
}
@@ -17029,7 +17029,7 @@ pub unsafe fn vmulq_n_s16(a: int16x8_t, b: i16) -> int16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmul))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mul))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmul_n_s32(a: int32x2_t, b: i32) -> int32x2_t {
simd_mul(a, vdup_n_s32(b))
}
@@ -17042,7 +17042,7 @@ pub unsafe fn vmul_n_s32(a: int32x2_t, b: i32) -> int32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmul))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mul))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmulq_n_s32(a: int32x4_t, b: i32) -> int32x4_t {
simd_mul(a, vdupq_n_s32(b))
}
@@ -17055,7 +17055,7 @@ pub unsafe fn vmulq_n_s32(a: int32x4_t, b: i32) -> int32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmul))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mul))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmul_n_u16(a: uint16x4_t, b: u16) -> uint16x4_t {
simd_mul(a, vdup_n_u16(b))
}
@@ -17068,7 +17068,7 @@ pub unsafe fn vmul_n_u16(a: uint16x4_t, b: u16) -> uint16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmul))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mul))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmulq_n_u16(a: uint16x8_t, b: u16) -> uint16x8_t {
simd_mul(a, vdupq_n_u16(b))
}
@@ -17081,7 +17081,7 @@ pub unsafe fn vmulq_n_u16(a: uint16x8_t, b: u16) -> uint16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmul))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mul))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmul_n_u32(a: uint32x2_t, b: u32) -> uint32x2_t {
simd_mul(a, vdup_n_u32(b))
}
@@ -17094,7 +17094,7 @@ pub unsafe fn vmul_n_u32(a: uint32x2_t, b: u32) -> uint32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmul))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mul))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmulq_n_u32(a: uint32x4_t, b: u32) -> uint32x4_t {
simd_mul(a, vdupq_n_u32(b))
}
@@ -17107,7 +17107,7 @@ pub unsafe fn vmulq_n_u32(a: uint32x4_t, b: u32) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmul))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fmul))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmul_n_f32(a: float32x2_t, b: f32) -> float32x2_t {
simd_mul(a, vdup_n_f32(b))
}
@@ -17120,7 +17120,7 @@ pub unsafe fn vmul_n_f32(a: float32x2_t, b: f32) -> float32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmul))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fmul))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmulq_n_f32(a: float32x4_t, b: f32) -> float32x4_t {
simd_mul(a, vdupq_n_f32(b))
}
@@ -17134,7 +17134,7 @@ pub unsafe fn vmulq_n_f32(a: float32x4_t, b: f32) -> float32x4_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmul, LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mul, LANE = 1))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmul_lane_s16<const LANE: i32>(a: int16x4_t, b: int16x4_t) -> int16x4_t {
static_assert_imm2!(LANE);
simd_mul(a, simd_shuffle4!(b, b, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -17149,7 +17149,7 @@ pub unsafe fn vmul_lane_s16<const LANE: i32>(a: int16x4_t, b: int16x4_t) -> int1
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmul, LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mul, LANE = 1))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmul_laneq_s16<const LANE: i32>(a: int16x4_t, b: int16x8_t) -> int16x4_t {
static_assert_imm3!(LANE);
simd_mul(a, simd_shuffle4!(b, b, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -17164,7 +17164,7 @@ pub unsafe fn vmul_laneq_s16<const LANE: i32>(a: int16x4_t, b: int16x8_t) -> int
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmul, LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mul, LANE = 1))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmulq_lane_s16<const LANE: i32>(a: int16x8_t, b: int16x4_t) -> int16x8_t {
static_assert_imm2!(LANE);
simd_mul(a, simd_shuffle8!(b, b, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -17179,7 +17179,7 @@ pub unsafe fn vmulq_lane_s16<const LANE: i32>(a: int16x8_t, b: int16x4_t) -> int
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmul, LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mul, LANE = 1))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmulq_laneq_s16<const LANE: i32>(a: int16x8_t, b: int16x8_t) -> int16x8_t {
static_assert_imm3!(LANE);
simd_mul(a, simd_shuffle8!(b, b, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -17194,7 +17194,7 @@ pub unsafe fn vmulq_laneq_s16<const LANE: i32>(a: int16x8_t, b: int16x8_t) -> in
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmul, LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mul, LANE = 1))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmul_lane_s32<const LANE: i32>(a: int32x2_t, b: int32x2_t) -> int32x2_t {
static_assert_imm1!(LANE);
simd_mul(a, simd_shuffle2!(b, b, <const LANE: i32> [LANE as u32, LANE as u32]))
@@ -17209,7 +17209,7 @@ pub unsafe fn vmul_lane_s32<const LANE: i32>(a: int32x2_t, b: int32x2_t) -> int3
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmul, LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mul, LANE = 1))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmul_laneq_s32<const LANE: i32>(a: int32x2_t, b: int32x4_t) -> int32x2_t {
static_assert_imm2!(LANE);
simd_mul(a, simd_shuffle2!(b, b, <const LANE: i32> [LANE as u32, LANE as u32]))
@@ -17224,7 +17224,7 @@ pub unsafe fn vmul_laneq_s32<const LANE: i32>(a: int32x2_t, b: int32x4_t) -> int
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmul, LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mul, LANE = 1))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmulq_lane_s32<const LANE: i32>(a: int32x4_t, b: int32x2_t) -> int32x4_t {
static_assert_imm1!(LANE);
simd_mul(a, simd_shuffle4!(b, b, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -17239,7 +17239,7 @@ pub unsafe fn vmulq_lane_s32<const LANE: i32>(a: int32x4_t, b: int32x2_t) -> int
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmul, LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mul, LANE = 1))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmulq_laneq_s32<const LANE: i32>(a: int32x4_t, b: int32x4_t) -> int32x4_t {
static_assert_imm2!(LANE);
simd_mul(a, simd_shuffle4!(b, b, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -17254,7 +17254,7 @@ pub unsafe fn vmulq_laneq_s32<const LANE: i32>(a: int32x4_t, b: int32x4_t) -> in
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmul, LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mul, LANE = 1))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmul_lane_u16<const LANE: i32>(a: uint16x4_t, b: uint16x4_t) -> uint16x4_t {
static_assert_imm2!(LANE);
simd_mul(a, simd_shuffle4!(b, b, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -17269,7 +17269,7 @@ pub unsafe fn vmul_lane_u16<const LANE: i32>(a: uint16x4_t, b: uint16x4_t) -> ui
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmul, LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mul, LANE = 1))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmul_laneq_u16<const LANE: i32>(a: uint16x4_t, b: uint16x8_t) -> uint16x4_t {
static_assert_imm3!(LANE);
simd_mul(a, simd_shuffle4!(b, b, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -17284,7 +17284,7 @@ pub unsafe fn vmul_laneq_u16<const LANE: i32>(a: uint16x4_t, b: uint16x8_t) -> u
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmul, LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mul, LANE = 1))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmulq_lane_u16<const LANE: i32>(a: uint16x8_t, b: uint16x4_t) -> uint16x8_t {
static_assert_imm2!(LANE);
simd_mul(a, simd_shuffle8!(b, b, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -17299,7 +17299,7 @@ pub unsafe fn vmulq_lane_u16<const LANE: i32>(a: uint16x8_t, b: uint16x4_t) -> u
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmul, LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mul, LANE = 1))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmulq_laneq_u16<const LANE: i32>(a: uint16x8_t, b: uint16x8_t) -> uint16x8_t {
static_assert_imm3!(LANE);
simd_mul(a, simd_shuffle8!(b, b, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -17314,7 +17314,7 @@ pub unsafe fn vmulq_laneq_u16<const LANE: i32>(a: uint16x8_t, b: uint16x8_t) ->
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmul, LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mul, LANE = 1))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmul_lane_u32<const LANE: i32>(a: uint32x2_t, b: uint32x2_t) -> uint32x2_t {
static_assert_imm1!(LANE);
simd_mul(a, simd_shuffle2!(b, b, <const LANE: i32> [LANE as u32, LANE as u32]))
@@ -17329,7 +17329,7 @@ pub unsafe fn vmul_lane_u32<const LANE: i32>(a: uint32x2_t, b: uint32x2_t) -> ui
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmul, LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mul, LANE = 1))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmul_laneq_u32<const LANE: i32>(a: uint32x2_t, b: uint32x4_t) -> uint32x2_t {
static_assert_imm2!(LANE);
simd_mul(a, simd_shuffle2!(b, b, <const LANE: i32> [LANE as u32, LANE as u32]))
@@ -17344,7 +17344,7 @@ pub unsafe fn vmul_laneq_u32<const LANE: i32>(a: uint32x2_t, b: uint32x4_t) -> u
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmul, LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mul, LANE = 1))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmulq_lane_u32<const LANE: i32>(a: uint32x4_t, b: uint32x2_t) -> uint32x4_t {
static_assert_imm1!(LANE);
simd_mul(a, simd_shuffle4!(b, b, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -17359,7 +17359,7 @@ pub unsafe fn vmulq_lane_u32<const LANE: i32>(a: uint32x4_t, b: uint32x2_t) -> u
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmul, LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(mul, LANE = 1))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmulq_laneq_u32<const LANE: i32>(a: uint32x4_t, b: uint32x4_t) -> uint32x4_t {
static_assert_imm2!(LANE);
simd_mul(a, simd_shuffle4!(b, b, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -17374,7 +17374,7 @@ pub unsafe fn vmulq_laneq_u32<const LANE: i32>(a: uint32x4_t, b: uint32x4_t) ->
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmul, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fmul, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmul_lane_f32<const LANE: i32>(a: float32x2_t, b: float32x2_t) -> float32x2_t {
static_assert_imm1!(LANE);
simd_mul(a, simd_shuffle2!(b, b, <const LANE: i32> [LANE as u32, LANE as u32]))
@@ -17389,7 +17389,7 @@ pub unsafe fn vmul_lane_f32<const LANE: i32>(a: float32x2_t, b: float32x2_t) ->
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmul, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fmul, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmul_laneq_f32<const LANE: i32>(a: float32x2_t, b: float32x4_t) -> float32x2_t {
static_assert_imm2!(LANE);
simd_mul(a, simd_shuffle2!(b, b, <const LANE: i32> [LANE as u32, LANE as u32]))
@@ -17404,7 +17404,7 @@ pub unsafe fn vmul_laneq_f32<const LANE: i32>(a: float32x2_t, b: float32x4_t) ->
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmul, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fmul, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmulq_lane_f32<const LANE: i32>(a: float32x4_t, b: float32x2_t) -> float32x4_t {
static_assert_imm1!(LANE);
simd_mul(a, simd_shuffle4!(b, b, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -17419,7 +17419,7 @@ pub unsafe fn vmulq_lane_f32<const LANE: i32>(a: float32x4_t, b: float32x2_t) ->
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmul, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fmul, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmulq_laneq_f32<const LANE: i32>(a: float32x4_t, b: float32x4_t) -> float32x4_t {
static_assert_imm2!(LANE);
simd_mul(a, simd_shuffle4!(b, b, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -17433,7 +17433,7 @@ pub unsafe fn vmulq_laneq_f32<const LANE: i32>(a: float32x4_t, b: float32x4_t) -
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmull.s8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(smull))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmull_s8(a: int8x8_t, b: int8x8_t) -> int16x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -17452,7 +17452,7 @@ vmull_s8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmull.s16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(smull))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmull_s16(a: int16x4_t, b: int16x4_t) -> int32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -17471,7 +17471,7 @@ vmull_s16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmull.s32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(smull))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmull_s32(a: int32x2_t, b: int32x2_t) -> int64x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -17490,7 +17490,7 @@ vmull_s32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmull.u8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(umull))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmull_u8(a: uint8x8_t, b: uint8x8_t) -> uint16x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -17509,7 +17509,7 @@ vmull_u8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmull.u16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(umull))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmull_u16(a: uint16x4_t, b: uint16x4_t) -> uint32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -17528,7 +17528,7 @@ vmull_u16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmull.u32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(umull))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmull_u32(a: uint32x2_t, b: uint32x2_t) -> uint64x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -17547,7 +17547,7 @@ vmull_u32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmull.p8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(pmull))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmull_p8(a: poly8x8_t, b: poly8x8_t) -> poly16x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -17566,7 +17566,7 @@ vmull_p8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmull))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(smull))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmull_n_s16(a: int16x4_t, b: i16) -> int32x4_t {
vmull_s16(a, vdup_n_s16(b))
}
@@ -17579,7 +17579,7 @@ pub unsafe fn vmull_n_s16(a: int16x4_t, b: i16) -> int32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmull))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(smull))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmull_n_s32(a: int32x2_t, b: i32) -> int64x2_t {
vmull_s32(a, vdup_n_s32(b))
}
@@ -17592,7 +17592,7 @@ pub unsafe fn vmull_n_s32(a: int32x2_t, b: i32) -> int64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmull))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(umull))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmull_n_u16(a: uint16x4_t, b: u16) -> uint32x4_t {
vmull_u16(a, vdup_n_u16(b))
}
@@ -17605,7 +17605,7 @@ pub unsafe fn vmull_n_u16(a: uint16x4_t, b: u16) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmull))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(umull))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmull_n_u32(a: uint32x2_t, b: u32) -> uint64x2_t {
vmull_u32(a, vdup_n_u32(b))
}
@@ -17619,7 +17619,7 @@ pub unsafe fn vmull_n_u32(a: uint32x2_t, b: u32) -> uint64x2_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmull, LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(smull, LANE = 1))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmull_lane_s16<const LANE: i32>(a: int16x4_t, b: int16x4_t) -> int32x4_t {
static_assert_imm2!(LANE);
vmull_s16(a, simd_shuffle4!(b, b, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -17634,7 +17634,7 @@ pub unsafe fn vmull_lane_s16<const LANE: i32>(a: int16x4_t, b: int16x4_t) -> int
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmull, LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(smull, LANE = 1))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmull_laneq_s16<const LANE: i32>(a: int16x4_t, b: int16x8_t) -> int32x4_t {
static_assert_imm3!(LANE);
vmull_s16(a, simd_shuffle4!(b, b, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -17649,7 +17649,7 @@ pub unsafe fn vmull_laneq_s16<const LANE: i32>(a: int16x4_t, b: int16x8_t) -> in
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmull, LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(smull, LANE = 1))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmull_lane_s32<const LANE: i32>(a: int32x2_t, b: int32x2_t) -> int64x2_t {
static_assert_imm1!(LANE);
vmull_s32(a, simd_shuffle2!(b, b, <const LANE: i32> [LANE as u32, LANE as u32]))
@@ -17664,7 +17664,7 @@ pub unsafe fn vmull_lane_s32<const LANE: i32>(a: int32x2_t, b: int32x2_t) -> int
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmull, LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(smull, LANE = 1))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmull_laneq_s32<const LANE: i32>(a: int32x2_t, b: int32x4_t) -> int64x2_t {
static_assert_imm2!(LANE);
vmull_s32(a, simd_shuffle2!(b, b, <const LANE: i32> [LANE as u32, LANE as u32]))
@@ -17679,7 +17679,7 @@ pub unsafe fn vmull_laneq_s32<const LANE: i32>(a: int32x2_t, b: int32x4_t) -> in
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmull, LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(umull, LANE = 1))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmull_lane_u16<const LANE: i32>(a: uint16x4_t, b: uint16x4_t) -> uint32x4_t {
static_assert_imm2!(LANE);
vmull_u16(a, simd_shuffle4!(b, b, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -17694,7 +17694,7 @@ pub unsafe fn vmull_lane_u16<const LANE: i32>(a: uint16x4_t, b: uint16x4_t) -> u
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmull, LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(umull, LANE = 1))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmull_laneq_u16<const LANE: i32>(a: uint16x4_t, b: uint16x8_t) -> uint32x4_t {
static_assert_imm3!(LANE);
vmull_u16(a, simd_shuffle4!(b, b, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]))
@@ -17709,7 +17709,7 @@ pub unsafe fn vmull_laneq_u16<const LANE: i32>(a: uint16x4_t, b: uint16x8_t) ->
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmull, LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(umull, LANE = 1))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmull_lane_u32<const LANE: i32>(a: uint32x2_t, b: uint32x2_t) -> uint64x2_t {
static_assert_imm1!(LANE);
vmull_u32(a, simd_shuffle2!(b, b, <const LANE: i32> [LANE as u32, LANE as u32]))
@@ -17724,7 +17724,7 @@ pub unsafe fn vmull_lane_u32<const LANE: i32>(a: uint32x2_t, b: uint32x2_t) -> u
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmull, LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(umull, LANE = 1))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmull_laneq_u32<const LANE: i32>(a: uint32x2_t, b: uint32x4_t) -> uint64x2_t {
static_assert_imm2!(LANE);
vmull_u32(a, simd_shuffle2!(b, b, <const LANE: i32> [LANE as u32, LANE as u32]))
@@ -17738,7 +17738,7 @@ pub unsafe fn vmull_laneq_u32<const LANE: i32>(a: uint32x2_t, b: uint32x4_t) ->
#[cfg_attr(target_arch = "arm", target_feature(enable = "vfp4"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vfma))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fmla))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vfma_f32(a: float32x2_t, b: float32x2_t, c: float32x2_t) -> float32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -17757,7 +17757,7 @@ vfma_f32_(b, c, a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "vfp4"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vfma))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fmla))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vfmaq_f32(a: float32x4_t, b: float32x4_t, c: float32x4_t) -> float32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -17776,7 +17776,7 @@ vfmaq_f32_(b, c, a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "vfp4"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vfma))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fmla))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vfma_n_f32(a: float32x2_t, b: float32x2_t, c: f32) -> float32x2_t {
vfma_f32(a, b, vdup_n_f32_vfp4(c))
}
@@ -17789,7 +17789,7 @@ pub unsafe fn vfma_n_f32(a: float32x2_t, b: float32x2_t, c: f32) -> float32x2_t
#[cfg_attr(target_arch = "arm", target_feature(enable = "vfp4"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vfma))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fmla))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vfmaq_n_f32(a: float32x4_t, b: float32x4_t, c: f32) -> float32x4_t {
vfmaq_f32(a, b, vdupq_n_f32_vfp4(c))
}
@@ -17802,7 +17802,7 @@ pub unsafe fn vfmaq_n_f32(a: float32x4_t, b: float32x4_t, c: f32) -> float32x4_t
#[cfg_attr(target_arch = "arm", target_feature(enable = "vfp4"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vfms))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fmls))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vfms_f32(a: float32x2_t, b: float32x2_t, c: float32x2_t) -> float32x2_t {
let b: float32x2_t = simd_neg(b);
vfma_f32(a, b, c)
@@ -17816,7 +17816,7 @@ pub unsafe fn vfms_f32(a: float32x2_t, b: float32x2_t, c: float32x2_t) -> float3
#[cfg_attr(target_arch = "arm", target_feature(enable = "vfp4"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vfms))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fmls))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vfmsq_f32(a: float32x4_t, b: float32x4_t, c: float32x4_t) -> float32x4_t {
let b: float32x4_t = simd_neg(b);
vfmaq_f32(a, b, c)
@@ -17830,7 +17830,7 @@ pub unsafe fn vfmsq_f32(a: float32x4_t, b: float32x4_t, c: float32x4_t) -> float
#[cfg_attr(target_arch = "arm", target_feature(enable = "vfp4"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vfms))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fmls))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vfms_n_f32(a: float32x2_t, b: float32x2_t, c: f32) -> float32x2_t {
vfms_f32(a, b, vdup_n_f32_vfp4(c))
}
@@ -17843,7 +17843,7 @@ pub unsafe fn vfms_n_f32(a: float32x2_t, b: float32x2_t, c: f32) -> float32x2_t
#[cfg_attr(target_arch = "arm", target_feature(enable = "vfp4"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vfms))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fmls))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vfmsq_n_f32(a: float32x4_t, b: float32x4_t, c: f32) -> float32x4_t {
vfmsq_f32(a, b, vdupq_n_f32_vfp4(c))
}
@@ -17856,7 +17856,7 @@ pub unsafe fn vfmsq_n_f32(a: float32x4_t, b: float32x4_t, c: f32) -> float32x4_t
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vsub.i8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsub_s8(a: int8x8_t, b: int8x8_t) -> int8x8_t {
simd_sub(a, b)
}
@@ -17869,7 +17869,7 @@ pub unsafe fn vsub_s8(a: int8x8_t, b: int8x8_t) -> int8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vsub.i8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsubq_s8(a: int8x16_t, b: int8x16_t) -> int8x16_t {
simd_sub(a, b)
}
@@ -17882,7 +17882,7 @@ pub unsafe fn vsubq_s8(a: int8x16_t, b: int8x16_t) -> int8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vsub.i16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsub_s16(a: int16x4_t, b: int16x4_t) -> int16x4_t {
simd_sub(a, b)
}
@@ -17895,7 +17895,7 @@ pub unsafe fn vsub_s16(a: int16x4_t, b: int16x4_t) -> int16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vsub.i16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsubq_s16(a: int16x8_t, b: int16x8_t) -> int16x8_t {
simd_sub(a, b)
}
@@ -17908,7 +17908,7 @@ pub unsafe fn vsubq_s16(a: int16x8_t, b: int16x8_t) -> int16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vsub.i32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsub_s32(a: int32x2_t, b: int32x2_t) -> int32x2_t {
simd_sub(a, b)
}
@@ -17921,7 +17921,7 @@ pub unsafe fn vsub_s32(a: int32x2_t, b: int32x2_t) -> int32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vsub.i32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsubq_s32(a: int32x4_t, b: int32x4_t) -> int32x4_t {
simd_sub(a, b)
}
@@ -17934,7 +17934,7 @@ pub unsafe fn vsubq_s32(a: int32x4_t, b: int32x4_t) -> int32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vsub.i8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsub_u8(a: uint8x8_t, b: uint8x8_t) -> uint8x8_t {
simd_sub(a, b)
}
@@ -17947,7 +17947,7 @@ pub unsafe fn vsub_u8(a: uint8x8_t, b: uint8x8_t) -> uint8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vsub.i8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsubq_u8(a: uint8x16_t, b: uint8x16_t) -> uint8x16_t {
simd_sub(a, b)
}
@@ -17960,7 +17960,7 @@ pub unsafe fn vsubq_u8(a: uint8x16_t, b: uint8x16_t) -> uint8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vsub.i16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsub_u16(a: uint16x4_t, b: uint16x4_t) -> uint16x4_t {
simd_sub(a, b)
}
@@ -17973,7 +17973,7 @@ pub unsafe fn vsub_u16(a: uint16x4_t, b: uint16x4_t) -> uint16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vsub.i16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsubq_u16(a: uint16x8_t, b: uint16x8_t) -> uint16x8_t {
simd_sub(a, b)
}
@@ -17986,7 +17986,7 @@ pub unsafe fn vsubq_u16(a: uint16x8_t, b: uint16x8_t) -> uint16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vsub.i32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsub_u32(a: uint32x2_t, b: uint32x2_t) -> uint32x2_t {
simd_sub(a, b)
}
@@ -17999,7 +17999,7 @@ pub unsafe fn vsub_u32(a: uint32x2_t, b: uint32x2_t) -> uint32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vsub.i32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsubq_u32(a: uint32x4_t, b: uint32x4_t) -> uint32x4_t {
simd_sub(a, b)
}
@@ -18012,7 +18012,7 @@ pub unsafe fn vsubq_u32(a: uint32x4_t, b: uint32x4_t) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vsub.i64"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsub_s64(a: int64x1_t, b: int64x1_t) -> int64x1_t {
simd_sub(a, b)
}
@@ -18025,7 +18025,7 @@ pub unsafe fn vsub_s64(a: int64x1_t, b: int64x1_t) -> int64x1_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vsub.i64"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsubq_s64(a: int64x2_t, b: int64x2_t) -> int64x2_t {
simd_sub(a, b)
}
@@ -18038,7 +18038,7 @@ pub unsafe fn vsubq_s64(a: int64x2_t, b: int64x2_t) -> int64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vsub.i64"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsub_u64(a: uint64x1_t, b: uint64x1_t) -> uint64x1_t {
simd_sub(a, b)
}
@@ -18051,7 +18051,7 @@ pub unsafe fn vsub_u64(a: uint64x1_t, b: uint64x1_t) -> uint64x1_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vsub.i64"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsubq_u64(a: uint64x2_t, b: uint64x2_t) -> uint64x2_t {
simd_sub(a, b)
}
@@ -18064,7 +18064,7 @@ pub unsafe fn vsubq_u64(a: uint64x2_t, b: uint64x2_t) -> uint64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vsub.f32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fsub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsub_f32(a: float32x2_t, b: float32x2_t) -> float32x2_t {
simd_sub(a, b)
}
@@ -18077,7 +18077,7 @@ pub unsafe fn vsub_f32(a: float32x2_t, b: float32x2_t) -> float32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vsub.f32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fsub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsubq_f32(a: float32x4_t, b: float32x4_t) -> float32x4_t {
simd_sub(a, b)
}
@@ -18090,7 +18090,7 @@ pub unsafe fn vsubq_f32(a: float32x4_t, b: float32x4_t) -> float32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vadd_p8(a: poly8x8_t, b: poly8x8_t) -> poly8x8_t {
simd_xor(a, b)
}
@@ -18103,7 +18103,7 @@ pub unsafe fn vadd_p8(a: poly8x8_t, b: poly8x8_t) -> poly8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vadd_p16(a: poly16x4_t, b: poly16x4_t) -> poly16x4_t {
simd_xor(a, b)
}
@@ -18116,7 +18116,7 @@ pub unsafe fn vadd_p16(a: poly16x4_t, b: poly16x4_t) -> poly16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vaddq_p8(a: poly8x16_t, b: poly8x16_t) -> poly8x16_t {
simd_xor(a, b)
}
@@ -18129,7 +18129,7 @@ pub unsafe fn vaddq_p8(a: poly8x16_t, b: poly8x16_t) -> poly8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vaddq_p16(a: poly16x8_t, b: poly16x8_t) -> poly16x8_t {
simd_xor(a, b)
}
@@ -18142,7 +18142,7 @@ pub unsafe fn vaddq_p16(a: poly16x8_t, b: poly16x8_t) -> poly16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vadd_p64(a: poly64x1_t, b: poly64x1_t) -> poly64x1_t {
simd_xor(a, b)
}
@@ -18155,7 +18155,7 @@ pub unsafe fn vadd_p64(a: poly64x1_t, b: poly64x1_t) -> poly64x1_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vaddq_p64(a: poly64x2_t, b: poly64x2_t) -> poly64x2_t {
simd_xor(a, b)
}
@@ -18168,7 +18168,7 @@ pub unsafe fn vaddq_p64(a: poly64x2_t, b: poly64x2_t) -> poly64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vaddq_p128(a: p128, b: p128) -> p128 {
a ^ b
}
@@ -18181,7 +18181,7 @@ pub unsafe fn vaddq_p128(a: p128, b: p128) -> p128 {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vsubhn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(subhn))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsubhn_s16(a: int16x8_t, b: int16x8_t) -> int8x8_t {
let c: i16x8 = i16x8::new(8, 8, 8, 8, 8, 8, 8, 8);
simd_cast(simd_shr(simd_sub(a, b), transmute(c)))
@@ -18195,7 +18195,7 @@ pub unsafe fn vsubhn_s16(a: int16x8_t, b: int16x8_t) -> int8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vsubhn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(subhn))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsubhn_s32(a: int32x4_t, b: int32x4_t) -> int16x4_t {
let c: i32x4 = i32x4::new(16, 16, 16, 16);
simd_cast(simd_shr(simd_sub(a, b), transmute(c)))
@@ -18209,7 +18209,7 @@ pub unsafe fn vsubhn_s32(a: int32x4_t, b: int32x4_t) -> int16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vsubhn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(subhn))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsubhn_s64(a: int64x2_t, b: int64x2_t) -> int32x2_t {
let c: i64x2 = i64x2::new(32, 32);
simd_cast(simd_shr(simd_sub(a, b), transmute(c)))
@@ -18223,7 +18223,7 @@ pub unsafe fn vsubhn_s64(a: int64x2_t, b: int64x2_t) -> int32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vsubhn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(subhn))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsubhn_u16(a: uint16x8_t, b: uint16x8_t) -> uint8x8_t {
let c: u16x8 = u16x8::new(8, 8, 8, 8, 8, 8, 8, 8);
simd_cast(simd_shr(simd_sub(a, b), transmute(c)))
@@ -18237,7 +18237,7 @@ pub unsafe fn vsubhn_u16(a: uint16x8_t, b: uint16x8_t) -> uint8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vsubhn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(subhn))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsubhn_u32(a: uint32x4_t, b: uint32x4_t) -> uint16x4_t {
let c: u32x4 = u32x4::new(16, 16, 16, 16);
simd_cast(simd_shr(simd_sub(a, b), transmute(c)))
@@ -18251,7 +18251,7 @@ pub unsafe fn vsubhn_u32(a: uint32x4_t, b: uint32x4_t) -> uint16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vsubhn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(subhn))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsubhn_u64(a: uint64x2_t, b: uint64x2_t) -> uint32x2_t {
let c: u64x2 = u64x2::new(32, 32);
simd_cast(simd_shr(simd_sub(a, b), transmute(c)))
@@ -18265,7 +18265,7 @@ pub unsafe fn vsubhn_u64(a: uint64x2_t, b: uint64x2_t) -> uint32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vsubhn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(subhn2))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsubhn_high_s16(a: int8x8_t, b: int16x8_t, c: int16x8_t) -> int8x16_t {
let d: int8x8_t = vsubhn_s16(b, c);
simd_shuffle16!(a, d, [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15])
@@ -18279,7 +18279,7 @@ pub unsafe fn vsubhn_high_s16(a: int8x8_t, b: int16x8_t, c: int16x8_t) -> int8x1
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vsubhn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(subhn2))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsubhn_high_s32(a: int16x4_t, b: int32x4_t, c: int32x4_t) -> int16x8_t {
let d: int16x4_t = vsubhn_s32(b, c);
simd_shuffle8!(a, d, [0, 1, 2, 3, 4, 5, 6, 7])
@@ -18293,7 +18293,7 @@ pub unsafe fn vsubhn_high_s32(a: int16x4_t, b: int32x4_t, c: int32x4_t) -> int16
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vsubhn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(subhn2))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsubhn_high_s64(a: int32x2_t, b: int64x2_t, c: int64x2_t) -> int32x4_t {
let d: int32x2_t = vsubhn_s64(b, c);
simd_shuffle4!(a, d, [0, 1, 2, 3])
@@ -18307,7 +18307,7 @@ pub unsafe fn vsubhn_high_s64(a: int32x2_t, b: int64x2_t, c: int64x2_t) -> int32
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vsubhn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(subhn2))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsubhn_high_u16(a: uint8x8_t, b: uint16x8_t, c: uint16x8_t) -> uint8x16_t {
let d: uint8x8_t = vsubhn_u16(b, c);
simd_shuffle16!(a, d, [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15])
@@ -18321,7 +18321,7 @@ pub unsafe fn vsubhn_high_u16(a: uint8x8_t, b: uint16x8_t, c: uint16x8_t) -> uin
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vsubhn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(subhn2))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsubhn_high_u32(a: uint16x4_t, b: uint32x4_t, c: uint32x4_t) -> uint16x8_t {
let d: uint16x4_t = vsubhn_u32(b, c);
simd_shuffle8!(a, d, [0, 1, 2, 3, 4, 5, 6, 7])
@@ -18335,7 +18335,7 @@ pub unsafe fn vsubhn_high_u32(a: uint16x4_t, b: uint32x4_t, c: uint32x4_t) -> ui
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vsubhn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(subhn2))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsubhn_high_u64(a: uint32x2_t, b: uint64x2_t, c: uint64x2_t) -> uint32x4_t {
let d: uint32x2_t = vsubhn_u64(b, c);
simd_shuffle4!(a, d, [0, 1, 2, 3])
@@ -18349,7 +18349,7 @@ pub unsafe fn vsubhn_high_u64(a: uint32x2_t, b: uint64x2_t, c: uint64x2_t) -> ui
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vhsub.u8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uhsub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vhsub_u8(a: uint8x8_t, b: uint8x8_t) -> uint8x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -18368,7 +18368,7 @@ vhsub_u8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vhsub.u8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uhsub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vhsubq_u8(a: uint8x16_t, b: uint8x16_t) -> uint8x16_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -18387,7 +18387,7 @@ vhsubq_u8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vhsub.u16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uhsub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vhsub_u16(a: uint16x4_t, b: uint16x4_t) -> uint16x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -18406,7 +18406,7 @@ vhsub_u16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vhsub.u16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uhsub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vhsubq_u16(a: uint16x8_t, b: uint16x8_t) -> uint16x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -18425,7 +18425,7 @@ vhsubq_u16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vhsub.u32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uhsub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vhsub_u32(a: uint32x2_t, b: uint32x2_t) -> uint32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -18444,7 +18444,7 @@ vhsub_u32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vhsub.u32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uhsub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vhsubq_u32(a: uint32x4_t, b: uint32x4_t) -> uint32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -18463,7 +18463,7 @@ vhsubq_u32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vhsub.s8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(shsub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vhsub_s8(a: int8x8_t, b: int8x8_t) -> int8x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -18482,7 +18482,7 @@ vhsub_s8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vhsub.s8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(shsub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vhsubq_s8(a: int8x16_t, b: int8x16_t) -> int8x16_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -18501,7 +18501,7 @@ vhsubq_s8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vhsub.s16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(shsub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vhsub_s16(a: int16x4_t, b: int16x4_t) -> int16x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -18520,7 +18520,7 @@ vhsub_s16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vhsub.s16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(shsub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vhsubq_s16(a: int16x8_t, b: int16x8_t) -> int16x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -18539,7 +18539,7 @@ vhsubq_s16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vhsub.s32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(shsub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vhsub_s32(a: int32x2_t, b: int32x2_t) -> int32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -18558,7 +18558,7 @@ vhsub_s32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vhsub.s32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(shsub))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vhsubq_s32(a: int32x4_t, b: int32x4_t) -> int32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -18577,7 +18577,7 @@ vhsubq_s32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vsubw))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ssubw))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsubw_s8(a: int16x8_t, b: int8x8_t) -> int16x8_t {
simd_sub(a, simd_cast(b))
}
@@ -18590,7 +18590,7 @@ pub unsafe fn vsubw_s8(a: int16x8_t, b: int8x8_t) -> int16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vsubw))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ssubw))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsubw_s16(a: int32x4_t, b: int16x4_t) -> int32x4_t {
simd_sub(a, simd_cast(b))
}
@@ -18603,7 +18603,7 @@ pub unsafe fn vsubw_s16(a: int32x4_t, b: int16x4_t) -> int32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vsubw))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ssubw))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsubw_s32(a: int64x2_t, b: int32x2_t) -> int64x2_t {
simd_sub(a, simd_cast(b))
}
@@ -18616,7 +18616,7 @@ pub unsafe fn vsubw_s32(a: int64x2_t, b: int32x2_t) -> int64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vsubw))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(usubw))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsubw_u8(a: uint16x8_t, b: uint8x8_t) -> uint16x8_t {
simd_sub(a, simd_cast(b))
}
@@ -18629,7 +18629,7 @@ pub unsafe fn vsubw_u8(a: uint16x8_t, b: uint8x8_t) -> uint16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vsubw))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(usubw))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsubw_u16(a: uint32x4_t, b: uint16x4_t) -> uint32x4_t {
simd_sub(a, simd_cast(b))
}
@@ -18642,7 +18642,7 @@ pub unsafe fn vsubw_u16(a: uint32x4_t, b: uint16x4_t) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vsubw))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(usubw))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsubw_u32(a: uint64x2_t, b: uint32x2_t) -> uint64x2_t {
simd_sub(a, simd_cast(b))
}
@@ -18655,7 +18655,7 @@ pub unsafe fn vsubw_u32(a: uint64x2_t, b: uint32x2_t) -> uint64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vsubl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ssubl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsubl_s8(a: int8x8_t, b: int8x8_t) -> int16x8_t {
let c: int16x8_t = simd_cast(a);
let d: int16x8_t = simd_cast(b);
@@ -18670,7 +18670,7 @@ pub unsafe fn vsubl_s8(a: int8x8_t, b: int8x8_t) -> int16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vsubl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ssubl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsubl_s16(a: int16x4_t, b: int16x4_t) -> int32x4_t {
let c: int32x4_t = simd_cast(a);
let d: int32x4_t = simd_cast(b);
@@ -18685,7 +18685,7 @@ pub unsafe fn vsubl_s16(a: int16x4_t, b: int16x4_t) -> int32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vsubl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ssubl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsubl_s32(a: int32x2_t, b: int32x2_t) -> int64x2_t {
let c: int64x2_t = simd_cast(a);
let d: int64x2_t = simd_cast(b);
@@ -18700,7 +18700,7 @@ pub unsafe fn vsubl_s32(a: int32x2_t, b: int32x2_t) -> int64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vsubl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(usubl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsubl_u8(a: uint8x8_t, b: uint8x8_t) -> uint16x8_t {
let c: uint16x8_t = simd_cast(a);
let d: uint16x8_t = simd_cast(b);
@@ -18715,7 +18715,7 @@ pub unsafe fn vsubl_u8(a: uint8x8_t, b: uint8x8_t) -> uint16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vsubl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(usubl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsubl_u16(a: uint16x4_t, b: uint16x4_t) -> uint32x4_t {
let c: uint32x4_t = simd_cast(a);
let d: uint32x4_t = simd_cast(b);
@@ -18730,7 +18730,7 @@ pub unsafe fn vsubl_u16(a: uint16x4_t, b: uint16x4_t) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vsubl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(usubl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsubl_u32(a: uint32x2_t, b: uint32x2_t) -> uint64x2_t {
let c: uint64x2_t = simd_cast(a);
let d: uint64x2_t = simd_cast(b);
@@ -18745,7 +18745,7 @@ pub unsafe fn vsubl_u32(a: uint32x2_t, b: uint32x2_t) -> uint64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmax))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(smax))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmax_s8(a: int8x8_t, b: int8x8_t) -> int8x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -18764,7 +18764,7 @@ vmax_s8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmax))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(smax))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmaxq_s8(a: int8x16_t, b: int8x16_t) -> int8x16_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -18783,7 +18783,7 @@ vmaxq_s8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmax))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(smax))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmax_s16(a: int16x4_t, b: int16x4_t) -> int16x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -18802,7 +18802,7 @@ vmax_s16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmax))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(smax))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmaxq_s16(a: int16x8_t, b: int16x8_t) -> int16x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -18821,7 +18821,7 @@ vmaxq_s16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmax))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(smax))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmax_s32(a: int32x2_t, b: int32x2_t) -> int32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -18840,7 +18840,7 @@ vmax_s32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmax))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(smax))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmaxq_s32(a: int32x4_t, b: int32x4_t) -> int32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -18859,7 +18859,7 @@ vmaxq_s32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmax))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(umax))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmax_u8(a: uint8x8_t, b: uint8x8_t) -> uint8x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -18878,7 +18878,7 @@ vmax_u8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmax))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(umax))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmaxq_u8(a: uint8x16_t, b: uint8x16_t) -> uint8x16_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -18897,7 +18897,7 @@ vmaxq_u8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmax))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(umax))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmax_u16(a: uint16x4_t, b: uint16x4_t) -> uint16x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -18916,7 +18916,7 @@ vmax_u16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmax))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(umax))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmaxq_u16(a: uint16x8_t, b: uint16x8_t) -> uint16x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -18935,7 +18935,7 @@ vmaxq_u16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmax))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(umax))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmax_u32(a: uint32x2_t, b: uint32x2_t) -> uint32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -18954,7 +18954,7 @@ vmax_u32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmax))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(umax))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmaxq_u32(a: uint32x4_t, b: uint32x4_t) -> uint32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -18973,7 +18973,7 @@ vmaxq_u32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmax))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fmax))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmax_f32(a: float32x2_t, b: float32x2_t) -> float32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -18992,7 +18992,7 @@ vmax_f32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmax))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fmax))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmaxq_f32(a: float32x4_t, b: float32x4_t) -> float32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -19011,7 +19011,7 @@ vmaxq_f32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "fp-armv8,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmaxnm))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fmaxnm))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmaxnm_f32(a: float32x2_t, b: float32x2_t) -> float32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -19030,7 +19030,7 @@ vmaxnm_f32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "fp-armv8,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmaxnm))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fmaxnm))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmaxnmq_f32(a: float32x4_t, b: float32x4_t) -> float32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -19049,7 +19049,7 @@ vmaxnmq_f32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmin))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(smin))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmin_s8(a: int8x8_t, b: int8x8_t) -> int8x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -19068,7 +19068,7 @@ vmin_s8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmin))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(smin))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vminq_s8(a: int8x16_t, b: int8x16_t) -> int8x16_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -19087,7 +19087,7 @@ vminq_s8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmin))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(smin))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmin_s16(a: int16x4_t, b: int16x4_t) -> int16x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -19106,7 +19106,7 @@ vmin_s16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmin))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(smin))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vminq_s16(a: int16x8_t, b: int16x8_t) -> int16x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -19125,7 +19125,7 @@ vminq_s16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmin))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(smin))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmin_s32(a: int32x2_t, b: int32x2_t) -> int32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -19144,7 +19144,7 @@ vmin_s32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmin))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(smin))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vminq_s32(a: int32x4_t, b: int32x4_t) -> int32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -19163,7 +19163,7 @@ vminq_s32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmin))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(umin))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmin_u8(a: uint8x8_t, b: uint8x8_t) -> uint8x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -19182,7 +19182,7 @@ vmin_u8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmin))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(umin))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vminq_u8(a: uint8x16_t, b: uint8x16_t) -> uint8x16_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -19201,7 +19201,7 @@ vminq_u8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmin))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(umin))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmin_u16(a: uint16x4_t, b: uint16x4_t) -> uint16x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -19220,7 +19220,7 @@ vmin_u16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmin))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(umin))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vminq_u16(a: uint16x8_t, b: uint16x8_t) -> uint16x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -19239,7 +19239,7 @@ vminq_u16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmin))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(umin))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmin_u32(a: uint32x2_t, b: uint32x2_t) -> uint32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -19258,7 +19258,7 @@ vmin_u32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmin))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(umin))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vminq_u32(a: uint32x4_t, b: uint32x4_t) -> uint32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -19277,7 +19277,7 @@ vminq_u32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmin))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fmin))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vmin_f32(a: float32x2_t, b: float32x2_t) -> float32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -19296,7 +19296,7 @@ vmin_f32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vmin))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fmin))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vminq_f32(a: float32x4_t, b: float32x4_t) -> float32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -19315,7 +19315,7 @@ vminq_f32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "fp-armv8,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vminnm))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fminnm))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vminnm_f32(a: float32x2_t, b: float32x2_t) -> float32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -19334,7 +19334,7 @@ vminnm_f32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "fp-armv8,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vminnm))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(fminnm))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vminnmq_f32(a: float32x4_t, b: float32x4_t) -> float32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -19353,7 +19353,7 @@ vminnmq_f32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vpadd))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(faddp))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vpadd_f32(a: float32x2_t, b: float32x2_t) -> float32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -19372,7 +19372,7 @@ vpadd_f32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqdmull))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqdmull))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqdmull_s16(a: int16x4_t, b: int16x4_t) -> int32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -19391,7 +19391,7 @@ vqdmull_s16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqdmull))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqdmull))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqdmull_s32(a: int32x2_t, b: int32x2_t) -> int64x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -19410,7 +19410,7 @@ vqdmull_s32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqdmull))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqdmull))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqdmull_n_s16(a: int16x4_t, b: i16) -> int32x4_t {
vqdmull_s16(a, vdup_n_s16(b))
}
@@ -19423,7 +19423,7 @@ pub unsafe fn vqdmull_n_s16(a: int16x4_t, b: i16) -> int32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqdmull))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqdmull))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqdmull_n_s32(a: int32x2_t, b: i32) -> int64x2_t {
vqdmull_s32(a, vdup_n_s32(b))
}
@@ -19437,7 +19437,7 @@ pub unsafe fn vqdmull_n_s32(a: int32x2_t, b: i32) -> int64x2_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqdmull, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqdmull, N = 2))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqdmull_lane_s16<const N: i32>(a: int16x4_t, b: int16x4_t) -> int32x4_t {
static_assert_imm2!(N);
let b: int16x4_t = simd_shuffle4!(b, b, <const N: i32> [N as u32, N as u32, N as u32, N as u32]);
@@ -19453,7 +19453,7 @@ pub unsafe fn vqdmull_lane_s16<const N: i32>(a: int16x4_t, b: int16x4_t) -> int3
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqdmull, N = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqdmull, N = 1))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqdmull_lane_s32<const N: i32>(a: int32x2_t, b: int32x2_t) -> int64x2_t {
static_assert_imm1!(N);
let b: int32x2_t = simd_shuffle2!(b, b, <const N: i32> [N as u32, N as u32]);
@@ -19468,7 +19468,7 @@ pub unsafe fn vqdmull_lane_s32<const N: i32>(a: int32x2_t, b: int32x2_t) -> int6
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqdmlal))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqdmlal))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqdmlal_s16(a: int32x4_t, b: int16x4_t, c: int16x4_t) -> int32x4_t {
vqaddq_s32(a, vqdmull_s16(b, c))
}
@@ -19481,7 +19481,7 @@ pub unsafe fn vqdmlal_s16(a: int32x4_t, b: int16x4_t, c: int16x4_t) -> int32x4_t
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqdmlal))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqdmlal))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqdmlal_s32(a: int64x2_t, b: int32x2_t, c: int32x2_t) -> int64x2_t {
vqaddq_s64(a, vqdmull_s32(b, c))
}
@@ -19494,7 +19494,7 @@ pub unsafe fn vqdmlal_s32(a: int64x2_t, b: int32x2_t, c: int32x2_t) -> int64x2_t
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqdmlal))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqdmlal))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqdmlal_n_s16(a: int32x4_t, b: int16x4_t, c: i16) -> int32x4_t {
vqaddq_s32(a, vqdmull_n_s16(b, c))
}
@@ -19507,7 +19507,7 @@ pub unsafe fn vqdmlal_n_s16(a: int32x4_t, b: int16x4_t, c: i16) -> int32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqdmlal))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqdmlal))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqdmlal_n_s32(a: int64x2_t, b: int32x2_t, c: i32) -> int64x2_t {
vqaddq_s64(a, vqdmull_n_s32(b, c))
}
@@ -19521,7 +19521,7 @@ pub unsafe fn vqdmlal_n_s32(a: int64x2_t, b: int32x2_t, c: i32) -> int64x2_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqdmlal, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqdmlal, N = 2))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqdmlal_lane_s16<const N: i32>(a: int32x4_t, b: int16x4_t, c: int16x4_t) -> int32x4_t {
static_assert_imm2!(N);
vqaddq_s32(a, vqdmull_lane_s16::<N>(b, c))
@@ -19536,7 +19536,7 @@ pub unsafe fn vqdmlal_lane_s16<const N: i32>(a: int32x4_t, b: int16x4_t, c: int1
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqdmlal, N = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqdmlal, N = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqdmlal_lane_s32<const N: i32>(a: int64x2_t, b: int32x2_t, c: int32x2_t) -> int64x2_t {
static_assert_imm1!(N);
vqaddq_s64(a, vqdmull_lane_s32::<N>(b, c))
@@ -19550,7 +19550,7 @@ pub unsafe fn vqdmlal_lane_s32<const N: i32>(a: int64x2_t, b: int32x2_t, c: int3
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqdmlsl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqdmlsl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqdmlsl_s16(a: int32x4_t, b: int16x4_t, c: int16x4_t) -> int32x4_t {
vqsubq_s32(a, vqdmull_s16(b, c))
}
@@ -19563,7 +19563,7 @@ pub unsafe fn vqdmlsl_s16(a: int32x4_t, b: int16x4_t, c: int16x4_t) -> int32x4_t
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqdmlsl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqdmlsl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqdmlsl_s32(a: int64x2_t, b: int32x2_t, c: int32x2_t) -> int64x2_t {
vqsubq_s64(a, vqdmull_s32(b, c))
}
@@ -19576,7 +19576,7 @@ pub unsafe fn vqdmlsl_s32(a: int64x2_t, b: int32x2_t, c: int32x2_t) -> int64x2_t
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqdmlsl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqdmlsl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqdmlsl_n_s16(a: int32x4_t, b: int16x4_t, c: i16) -> int32x4_t {
vqsubq_s32(a, vqdmull_n_s16(b, c))
}
@@ -19589,7 +19589,7 @@ pub unsafe fn vqdmlsl_n_s16(a: int32x4_t, b: int16x4_t, c: i16) -> int32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqdmlsl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqdmlsl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqdmlsl_n_s32(a: int64x2_t, b: int32x2_t, c: i32) -> int64x2_t {
vqsubq_s64(a, vqdmull_n_s32(b, c))
}
@@ -19603,7 +19603,7 @@ pub unsafe fn vqdmlsl_n_s32(a: int64x2_t, b: int32x2_t, c: i32) -> int64x2_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqdmlsl, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqdmlsl, N = 2))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqdmlsl_lane_s16<const N: i32>(a: int32x4_t, b: int16x4_t, c: int16x4_t) -> int32x4_t {
static_assert_imm2!(N);
vqsubq_s32(a, vqdmull_lane_s16::<N>(b, c))
@@ -19618,7 +19618,7 @@ pub unsafe fn vqdmlsl_lane_s16<const N: i32>(a: int32x4_t, b: int16x4_t, c: int1
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqdmlsl, N = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqdmlsl, N = 1))]
#[rustc_legacy_const_generics(3)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqdmlsl_lane_s32<const N: i32>(a: int64x2_t, b: int32x2_t, c: int32x2_t) -> int64x2_t {
static_assert_imm1!(N);
vqsubq_s64(a, vqdmull_lane_s32::<N>(b, c))
@@ -19632,7 +19632,7 @@ pub unsafe fn vqdmlsl_lane_s32<const N: i32>(a: int64x2_t, b: int32x2_t, c: int3
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqdmulh))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqdmulh))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqdmulh_s16(a: int16x4_t, b: int16x4_t) -> int16x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -19651,7 +19651,7 @@ vqdmulh_s16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqdmulh))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqdmulh))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqdmulhq_s16(a: int16x8_t, b: int16x8_t) -> int16x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -19670,7 +19670,7 @@ vqdmulhq_s16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqdmulh))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqdmulh))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqdmulh_s32(a: int32x2_t, b: int32x2_t) -> int32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -19689,7 +19689,7 @@ vqdmulh_s32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqdmulh))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqdmulh))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqdmulhq_s32(a: int32x4_t, b: int32x4_t) -> int32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -19708,7 +19708,7 @@ vqdmulhq_s32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqdmulh))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqdmulh))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqdmulh_n_s16(a: int16x4_t, b: i16) -> int16x4_t {
let b: int16x4_t = vdup_n_s16(b);
vqdmulh_s16(a, b)
@@ -19722,7 +19722,7 @@ pub unsafe fn vqdmulh_n_s16(a: int16x4_t, b: i16) -> int16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqdmulh))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqdmulh))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqdmulh_n_s32(a: int32x2_t, b: i32) -> int32x2_t {
let b: int32x2_t = vdup_n_s32(b);
vqdmulh_s32(a, b)
@@ -19736,7 +19736,7 @@ pub unsafe fn vqdmulh_n_s32(a: int32x2_t, b: i32) -> int32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqdmulh))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqdmulh))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqdmulhq_n_s16(a: int16x8_t, b: i16) -> int16x8_t {
let b: int16x8_t = vdupq_n_s16(b);
vqdmulhq_s16(a, b)
@@ -19750,7 +19750,7 @@ pub unsafe fn vqdmulhq_n_s16(a: int16x8_t, b: i16) -> int16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqdmulh))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqdmulh))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqdmulhq_n_s32(a: int32x4_t, b: i32) -> int32x4_t {
let b: int32x4_t = vdupq_n_s32(b);
vqdmulhq_s32(a, b)
@@ -19765,7 +19765,7 @@ pub unsafe fn vqdmulhq_n_s32(a: int32x4_t, b: i32) -> int32x4_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqdmulh, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqdmulh, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqdmulhq_laneq_s16<const LANE: i32>(a: int16x8_t, b: int16x8_t) -> int16x8_t {
static_assert_imm3!(LANE);
vqdmulhq_s16(a, vdupq_n_s16(simd_extract(b, LANE as u32)))
@@ -19780,7 +19780,7 @@ pub unsafe fn vqdmulhq_laneq_s16<const LANE: i32>(a: int16x8_t, b: int16x8_t) ->
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqdmulh, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqdmulh, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqdmulh_laneq_s16<const LANE: i32>(a: int16x4_t, b: int16x8_t) -> int16x4_t {
static_assert_imm3!(LANE);
vqdmulh_s16(a, vdup_n_s16(simd_extract(b, LANE as u32)))
@@ -19795,7 +19795,7 @@ pub unsafe fn vqdmulh_laneq_s16<const LANE: i32>(a: int16x4_t, b: int16x8_t) ->
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqdmulh, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqdmulh, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqdmulhq_laneq_s32<const LANE: i32>(a: int32x4_t, b: int32x4_t) -> int32x4_t {
static_assert_imm2!(LANE);
vqdmulhq_s32(a, vdupq_n_s32(simd_extract(b, LANE as u32)))
@@ -19810,7 +19810,7 @@ pub unsafe fn vqdmulhq_laneq_s32<const LANE: i32>(a: int32x4_t, b: int32x4_t) ->
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqdmulh, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqdmulh, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqdmulh_laneq_s32<const LANE: i32>(a: int32x2_t, b: int32x4_t) -> int32x2_t {
static_assert_imm2!(LANE);
vqdmulh_s32(a, vdup_n_s32(simd_extract(b, LANE as u32)))
@@ -19824,7 +19824,7 @@ pub unsafe fn vqdmulh_laneq_s32<const LANE: i32>(a: int32x2_t, b: int32x4_t) ->
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqmovn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqxtn))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqmovn_s16(a: int16x8_t) -> int8x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -19843,7 +19843,7 @@ vqmovn_s16_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqmovn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqxtn))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqmovn_s32(a: int32x4_t) -> int16x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -19862,7 +19862,7 @@ vqmovn_s32_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqmovn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqxtn))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqmovn_s64(a: int64x2_t) -> int32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -19881,7 +19881,7 @@ vqmovn_s64_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqmovn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uqxtn))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqmovn_u16(a: uint16x8_t) -> uint8x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -19900,7 +19900,7 @@ vqmovn_u16_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqmovn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uqxtn))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqmovn_u32(a: uint32x4_t) -> uint16x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -19919,7 +19919,7 @@ vqmovn_u32_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqmovn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uqxtn))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqmovn_u64(a: uint64x2_t) -> uint32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -19938,7 +19938,7 @@ vqmovn_u64_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqmovun))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqxtun))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqmovun_s16(a: int16x8_t) -> uint8x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -19957,7 +19957,7 @@ vqmovun_s16_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqmovun))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqxtun))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqmovun_s32(a: int32x4_t) -> uint16x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -19976,7 +19976,7 @@ vqmovun_s32_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqmovun))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqxtun))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqmovun_s64(a: int64x2_t) -> uint32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -19995,7 +19995,7 @@ vqmovun_s64_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqrdmulh))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqrdmulh))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqrdmulh_s16(a: int16x4_t, b: int16x4_t) -> int16x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -20014,7 +20014,7 @@ vqrdmulh_s16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqrdmulh))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqrdmulh))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqrdmulhq_s16(a: int16x8_t, b: int16x8_t) -> int16x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -20033,7 +20033,7 @@ vqrdmulhq_s16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqrdmulh))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqrdmulh))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqrdmulh_s32(a: int32x2_t, b: int32x2_t) -> int32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -20052,7 +20052,7 @@ vqrdmulh_s32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqrdmulh))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqrdmulh))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqrdmulhq_s32(a: int32x4_t, b: int32x4_t) -> int32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -20071,7 +20071,7 @@ vqrdmulhq_s32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqrdmulh))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqrdmulh))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqrdmulh_n_s16(a: int16x4_t, b: i16) -> int16x4_t {
vqrdmulh_s16(a, vdup_n_s16(b))
}
@@ -20084,7 +20084,7 @@ pub unsafe fn vqrdmulh_n_s16(a: int16x4_t, b: i16) -> int16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqrdmulh))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqrdmulh))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqrdmulhq_n_s16(a: int16x8_t, b: i16) -> int16x8_t {
vqrdmulhq_s16(a, vdupq_n_s16(b))
}
@@ -20097,7 +20097,7 @@ pub unsafe fn vqrdmulhq_n_s16(a: int16x8_t, b: i16) -> int16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqrdmulh))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqrdmulh))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqrdmulh_n_s32(a: int32x2_t, b: i32) -> int32x2_t {
vqrdmulh_s32(a, vdup_n_s32(b))
}
@@ -20110,7 +20110,7 @@ pub unsafe fn vqrdmulh_n_s32(a: int32x2_t, b: i32) -> int32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqrdmulh))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqrdmulh))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqrdmulhq_n_s32(a: int32x4_t, b: i32) -> int32x4_t {
vqrdmulhq_s32(a, vdupq_n_s32(b))
}
@@ -20124,7 +20124,7 @@ pub unsafe fn vqrdmulhq_n_s32(a: int32x4_t, b: i32) -> int32x4_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqrdmulh, LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqrdmulh, LANE = 1))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqrdmulh_lane_s16<const LANE: i32>(a: int16x4_t, b: int16x4_t) -> int16x4_t {
static_assert_imm2!(LANE);
let b: int16x4_t = simd_shuffle4!(b, b, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]);
@@ -20140,7 +20140,7 @@ pub unsafe fn vqrdmulh_lane_s16<const LANE: i32>(a: int16x4_t, b: int16x4_t) ->
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqrdmulh, LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqrdmulh, LANE = 1))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqrdmulh_laneq_s16<const LANE: i32>(a: int16x4_t, b: int16x8_t) -> int16x4_t {
static_assert_imm3!(LANE);
let b: int16x4_t = simd_shuffle4!(b, b, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]);
@@ -20156,7 +20156,7 @@ pub unsafe fn vqrdmulh_laneq_s16<const LANE: i32>(a: int16x4_t, b: int16x8_t) ->
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqrdmulh, LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqrdmulh, LANE = 1))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqrdmulhq_lane_s16<const LANE: i32>(a: int16x8_t, b: int16x4_t) -> int16x8_t {
static_assert_imm2!(LANE);
let b: int16x8_t = simd_shuffle8!(b, b, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32]);
@@ -20172,7 +20172,7 @@ pub unsafe fn vqrdmulhq_lane_s16<const LANE: i32>(a: int16x8_t, b: int16x4_t) ->
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqrdmulh, LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqrdmulh, LANE = 1))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqrdmulhq_laneq_s16<const LANE: i32>(a: int16x8_t, b: int16x8_t) -> int16x8_t {
static_assert_imm3!(LANE);
let b: int16x8_t = simd_shuffle8!(b, b, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32, LANE as u32]);
@@ -20188,7 +20188,7 @@ pub unsafe fn vqrdmulhq_laneq_s16<const LANE: i32>(a: int16x8_t, b: int16x8_t) -
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqrdmulh, LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqrdmulh, LANE = 1))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqrdmulh_lane_s32<const LANE: i32>(a: int32x2_t, b: int32x2_t) -> int32x2_t {
static_assert_imm1!(LANE);
let b: int32x2_t = simd_shuffle2!(b, b, <const LANE: i32> [LANE as u32, LANE as u32]);
@@ -20204,7 +20204,7 @@ pub unsafe fn vqrdmulh_lane_s32<const LANE: i32>(a: int32x2_t, b: int32x2_t) ->
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqrdmulh, LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqrdmulh, LANE = 1))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqrdmulh_laneq_s32<const LANE: i32>(a: int32x2_t, b: int32x4_t) -> int32x2_t {
static_assert_imm2!(LANE);
let b: int32x2_t = simd_shuffle2!(b, b, <const LANE: i32> [LANE as u32, LANE as u32]);
@@ -20220,7 +20220,7 @@ pub unsafe fn vqrdmulh_laneq_s32<const LANE: i32>(a: int32x2_t, b: int32x4_t) ->
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqrdmulh, LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqrdmulh, LANE = 1))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqrdmulhq_lane_s32<const LANE: i32>(a: int32x4_t, b: int32x2_t) -> int32x4_t {
static_assert_imm1!(LANE);
let b: int32x4_t = simd_shuffle4!(b, b, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]);
@@ -20236,7 +20236,7 @@ pub unsafe fn vqrdmulhq_lane_s32<const LANE: i32>(a: int32x4_t, b: int32x2_t) ->
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqrdmulh, LANE = 1))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqrdmulh, LANE = 1))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqrdmulhq_laneq_s32<const LANE: i32>(a: int32x4_t, b: int32x4_t) -> int32x4_t {
static_assert_imm2!(LANE);
let b: int32x4_t = simd_shuffle4!(b, b, <const LANE: i32> [LANE as u32, LANE as u32, LANE as u32, LANE as u32]);
@@ -20251,7 +20251,7 @@ pub unsafe fn vqrdmulhq_laneq_s32<const LANE: i32>(a: int32x4_t, b: int32x4_t) -
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqrshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqrshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqrshl_s8(a: int8x8_t, b: int8x8_t) -> int8x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -20270,7 +20270,7 @@ vqrshl_s8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqrshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqrshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqrshlq_s8(a: int8x16_t, b: int8x16_t) -> int8x16_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -20289,7 +20289,7 @@ vqrshlq_s8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqrshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqrshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqrshl_s16(a: int16x4_t, b: int16x4_t) -> int16x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -20308,7 +20308,7 @@ vqrshl_s16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqrshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqrshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqrshlq_s16(a: int16x8_t, b: int16x8_t) -> int16x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -20327,7 +20327,7 @@ vqrshlq_s16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqrshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqrshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqrshl_s32(a: int32x2_t, b: int32x2_t) -> int32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -20346,7 +20346,7 @@ vqrshl_s32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqrshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqrshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqrshlq_s32(a: int32x4_t, b: int32x4_t) -> int32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -20365,7 +20365,7 @@ vqrshlq_s32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqrshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqrshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqrshl_s64(a: int64x1_t, b: int64x1_t) -> int64x1_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -20384,7 +20384,7 @@ vqrshl_s64_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqrshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqrshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqrshlq_s64(a: int64x2_t, b: int64x2_t) -> int64x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -20403,7 +20403,7 @@ vqrshlq_s64_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqrshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uqrshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqrshl_u8(a: uint8x8_t, b: int8x8_t) -> uint8x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -20422,7 +20422,7 @@ vqrshl_u8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqrshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uqrshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqrshlq_u8(a: uint8x16_t, b: int8x16_t) -> uint8x16_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -20441,7 +20441,7 @@ vqrshlq_u8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqrshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uqrshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqrshl_u16(a: uint16x4_t, b: int16x4_t) -> uint16x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -20460,7 +20460,7 @@ vqrshl_u16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqrshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uqrshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqrshlq_u16(a: uint16x8_t, b: int16x8_t) -> uint16x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -20479,7 +20479,7 @@ vqrshlq_u16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqrshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uqrshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqrshl_u32(a: uint32x2_t, b: int32x2_t) -> uint32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -20498,7 +20498,7 @@ vqrshl_u32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqrshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uqrshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqrshlq_u32(a: uint32x4_t, b: int32x4_t) -> uint32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -20517,7 +20517,7 @@ vqrshlq_u32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqrshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uqrshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqrshl_u64(a: uint64x1_t, b: int64x1_t) -> uint64x1_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -20536,7 +20536,7 @@ vqrshl_u64_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqrshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uqrshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqrshlq_u64(a: uint64x2_t, b: int64x2_t) -> uint64x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -20888,7 +20888,7 @@ vqrshrun_n_s64_(a, N)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqshl_s8(a: int8x8_t, b: int8x8_t) -> int8x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -20907,7 +20907,7 @@ vqshl_s8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqshlq_s8(a: int8x16_t, b: int8x16_t) -> int8x16_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -20926,7 +20926,7 @@ vqshlq_s8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqshl_s16(a: int16x4_t, b: int16x4_t) -> int16x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -20945,7 +20945,7 @@ vqshl_s16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqshlq_s16(a: int16x8_t, b: int16x8_t) -> int16x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -20964,7 +20964,7 @@ vqshlq_s16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqshl_s32(a: int32x2_t, b: int32x2_t) -> int32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -20983,7 +20983,7 @@ vqshl_s32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqshlq_s32(a: int32x4_t, b: int32x4_t) -> int32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -21002,7 +21002,7 @@ vqshlq_s32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqshl_s64(a: int64x1_t, b: int64x1_t) -> int64x1_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -21021,7 +21021,7 @@ vqshl_s64_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqshlq_s64(a: int64x2_t, b: int64x2_t) -> int64x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -21040,7 +21040,7 @@ vqshlq_s64_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uqshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqshl_u8(a: uint8x8_t, b: int8x8_t) -> uint8x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -21059,7 +21059,7 @@ vqshl_u8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uqshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqshlq_u8(a: uint8x16_t, b: int8x16_t) -> uint8x16_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -21078,7 +21078,7 @@ vqshlq_u8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uqshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqshl_u16(a: uint16x4_t, b: int16x4_t) -> uint16x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -21097,7 +21097,7 @@ vqshl_u16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uqshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqshlq_u16(a: uint16x8_t, b: int16x8_t) -> uint16x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -21116,7 +21116,7 @@ vqshlq_u16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uqshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqshl_u32(a: uint32x2_t, b: int32x2_t) -> uint32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -21135,7 +21135,7 @@ vqshl_u32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uqshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqshlq_u32(a: uint32x4_t, b: int32x4_t) -> uint32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -21154,7 +21154,7 @@ vqshlq_u32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uqshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqshl_u64(a: uint64x1_t, b: int64x1_t) -> uint64x1_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -21173,7 +21173,7 @@ vqshl_u64_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uqshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqshlq_u64(a: uint64x2_t, b: int64x2_t) -> uint64x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -21193,7 +21193,7 @@ vqshlq_u64_(a, b)
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqshl, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqshl, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqshl_n_s8<const N: i32>(a: int8x8_t) -> int8x8_t {
static_assert_imm3!(N);
vqshl_s8(a, vdup_n_s8(N as _))
@@ -21208,7 +21208,7 @@ pub unsafe fn vqshl_n_s8<const N: i32>(a: int8x8_t) -> int8x8_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqshl, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqshl, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqshlq_n_s8<const N: i32>(a: int8x16_t) -> int8x16_t {
static_assert_imm3!(N);
vqshlq_s8(a, vdupq_n_s8(N as _))
@@ -21223,7 +21223,7 @@ pub unsafe fn vqshlq_n_s8<const N: i32>(a: int8x16_t) -> int8x16_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqshl, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqshl, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqshl_n_s16<const N: i32>(a: int16x4_t) -> int16x4_t {
static_assert_imm4!(N);
vqshl_s16(a, vdup_n_s16(N as _))
@@ -21238,7 +21238,7 @@ pub unsafe fn vqshl_n_s16<const N: i32>(a: int16x4_t) -> int16x4_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqshl, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqshl, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqshlq_n_s16<const N: i32>(a: int16x8_t) -> int16x8_t {
static_assert_imm4!(N);
vqshlq_s16(a, vdupq_n_s16(N as _))
@@ -21253,7 +21253,7 @@ pub unsafe fn vqshlq_n_s16<const N: i32>(a: int16x8_t) -> int16x8_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqshl, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqshl, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqshl_n_s32<const N: i32>(a: int32x2_t) -> int32x2_t {
static_assert_imm5!(N);
vqshl_s32(a, vdup_n_s32(N as _))
@@ -21268,7 +21268,7 @@ pub unsafe fn vqshl_n_s32<const N: i32>(a: int32x2_t) -> int32x2_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqshl, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqshl, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqshlq_n_s32<const N: i32>(a: int32x4_t) -> int32x4_t {
static_assert_imm5!(N);
vqshlq_s32(a, vdupq_n_s32(N as _))
@@ -21283,7 +21283,7 @@ pub unsafe fn vqshlq_n_s32<const N: i32>(a: int32x4_t) -> int32x4_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqshl, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqshl, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqshl_n_s64<const N: i32>(a: int64x1_t) -> int64x1_t {
static_assert_imm6!(N);
vqshl_s64(a, vdup_n_s64(N as _))
@@ -21298,7 +21298,7 @@ pub unsafe fn vqshl_n_s64<const N: i32>(a: int64x1_t) -> int64x1_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqshl, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqshl, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqshlq_n_s64<const N: i32>(a: int64x2_t) -> int64x2_t {
static_assert_imm6!(N);
vqshlq_s64(a, vdupq_n_s64(N as _))
@@ -21313,7 +21313,7 @@ pub unsafe fn vqshlq_n_s64<const N: i32>(a: int64x2_t) -> int64x2_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqshl, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uqshl, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqshl_n_u8<const N: i32>(a: uint8x8_t) -> uint8x8_t {
static_assert_imm3!(N);
vqshl_u8(a, vdup_n_s8(N as _))
@@ -21328,7 +21328,7 @@ pub unsafe fn vqshl_n_u8<const N: i32>(a: uint8x8_t) -> uint8x8_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqshl, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uqshl, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqshlq_n_u8<const N: i32>(a: uint8x16_t) -> uint8x16_t {
static_assert_imm3!(N);
vqshlq_u8(a, vdupq_n_s8(N as _))
@@ -21343,7 +21343,7 @@ pub unsafe fn vqshlq_n_u8<const N: i32>(a: uint8x16_t) -> uint8x16_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqshl, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uqshl, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqshl_n_u16<const N: i32>(a: uint16x4_t) -> uint16x4_t {
static_assert_imm4!(N);
vqshl_u16(a, vdup_n_s16(N as _))
@@ -21358,7 +21358,7 @@ pub unsafe fn vqshl_n_u16<const N: i32>(a: uint16x4_t) -> uint16x4_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqshl, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uqshl, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqshlq_n_u16<const N: i32>(a: uint16x8_t) -> uint16x8_t {
static_assert_imm4!(N);
vqshlq_u16(a, vdupq_n_s16(N as _))
@@ -21373,7 +21373,7 @@ pub unsafe fn vqshlq_n_u16<const N: i32>(a: uint16x8_t) -> uint16x8_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqshl, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uqshl, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqshl_n_u32<const N: i32>(a: uint32x2_t) -> uint32x2_t {
static_assert_imm5!(N);
vqshl_u32(a, vdup_n_s32(N as _))
@@ -21388,7 +21388,7 @@ pub unsafe fn vqshl_n_u32<const N: i32>(a: uint32x2_t) -> uint32x2_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqshl, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uqshl, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqshlq_n_u32<const N: i32>(a: uint32x4_t) -> uint32x4_t {
static_assert_imm5!(N);
vqshlq_u32(a, vdupq_n_s32(N as _))
@@ -21403,7 +21403,7 @@ pub unsafe fn vqshlq_n_u32<const N: i32>(a: uint32x4_t) -> uint32x4_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqshl, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uqshl, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqshl_n_u64<const N: i32>(a: uint64x1_t) -> uint64x1_t {
static_assert_imm6!(N);
vqshl_u64(a, vdup_n_s64(N as _))
@@ -21418,7 +21418,7 @@ pub unsafe fn vqshl_n_u64<const N: i32>(a: uint64x1_t) -> uint64x1_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vqshl, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uqshl, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqshlq_n_u64<const N: i32>(a: uint64x2_t) -> uint64x2_t {
static_assert_imm6!(N);
vqshlq_u64(a, vdupq_n_s64(N as _))
@@ -22061,7 +22061,7 @@ vqshrun_n_s64_(a, N)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrsqrte))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(frsqrte))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrsqrte_f32(a: float32x2_t) -> float32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -22080,7 +22080,7 @@ vrsqrte_f32_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrsqrte))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(frsqrte))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrsqrteq_f32(a: float32x4_t) -> float32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -22099,7 +22099,7 @@ vrsqrteq_f32_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrsqrte))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ursqrte))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrsqrte_u32(a: uint32x2_t) -> uint32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -22118,7 +22118,7 @@ vrsqrte_u32_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrsqrte))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ursqrte))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrsqrteq_u32(a: uint32x4_t) -> uint32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -22137,7 +22137,7 @@ vrsqrteq_u32_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrsqrts))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(frsqrts))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrsqrts_f32(a: float32x2_t, b: float32x2_t) -> float32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -22156,7 +22156,7 @@ vrsqrts_f32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrsqrts))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(frsqrts))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrsqrtsq_f32(a: float32x4_t, b: float32x4_t) -> float32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -22175,7 +22175,7 @@ vrsqrtsq_f32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrecpe))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(frecpe))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrecpe_f32(a: float32x2_t) -> float32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -22194,7 +22194,7 @@ vrecpe_f32_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrecpe))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(frecpe))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrecpeq_f32(a: float32x4_t) -> float32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -22213,7 +22213,7 @@ vrecpeq_f32_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrecpe))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(urecpe))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrecpe_u32(a: uint32x2_t) -> uint32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -22232,7 +22232,7 @@ vrecpe_u32_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrecpe))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(urecpe))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrecpeq_u32(a: uint32x4_t) -> uint32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -22251,7 +22251,7 @@ vrecpeq_u32_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrecps))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(frecps))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrecps_f32(a: float32x2_t, b: float32x2_t) -> float32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -22270,7 +22270,7 @@ vrecps_f32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrecps))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(frecps))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrecpsq_f32(a: float32x4_t, b: float32x4_t) -> float32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -22289,7 +22289,7 @@ vrecpsq_f32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_s8_u8(a: uint8x8_t) -> int8x8_t {
transmute(a)
}
@@ -22302,7 +22302,7 @@ pub unsafe fn vreinterpret_s8_u8(a: uint8x8_t) -> int8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_s8_p8(a: poly8x8_t) -> int8x8_t {
transmute(a)
}
@@ -22315,7 +22315,7 @@ pub unsafe fn vreinterpret_s8_p8(a: poly8x8_t) -> int8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_s16_p16(a: poly16x4_t) -> int16x4_t {
transmute(a)
}
@@ -22328,7 +22328,7 @@ pub unsafe fn vreinterpret_s16_p16(a: poly16x4_t) -> int16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_s16_u16(a: uint16x4_t) -> int16x4_t {
transmute(a)
}
@@ -22341,7 +22341,7 @@ pub unsafe fn vreinterpret_s16_u16(a: uint16x4_t) -> int16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_s32_u32(a: uint32x2_t) -> int32x2_t {
transmute(a)
}
@@ -22354,7 +22354,7 @@ pub unsafe fn vreinterpret_s32_u32(a: uint32x2_t) -> int32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_s64_u64(a: uint64x1_t) -> int64x1_t {
transmute(a)
}
@@ -22367,7 +22367,7 @@ pub unsafe fn vreinterpret_s64_u64(a: uint64x1_t) -> int64x1_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s8_u8(a: uint8x16_t) -> int8x16_t {
transmute(a)
}
@@ -22380,7 +22380,7 @@ pub unsafe fn vreinterpretq_s8_u8(a: uint8x16_t) -> int8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s8_p8(a: poly8x16_t) -> int8x16_t {
transmute(a)
}
@@ -22393,7 +22393,7 @@ pub unsafe fn vreinterpretq_s8_p8(a: poly8x16_t) -> int8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s16_p16(a: poly16x8_t) -> int16x8_t {
transmute(a)
}
@@ -22406,7 +22406,7 @@ pub unsafe fn vreinterpretq_s16_p16(a: poly16x8_t) -> int16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s16_u16(a: uint16x8_t) -> int16x8_t {
transmute(a)
}
@@ -22419,7 +22419,7 @@ pub unsafe fn vreinterpretq_s16_u16(a: uint16x8_t) -> int16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s32_u32(a: uint32x4_t) -> int32x4_t {
transmute(a)
}
@@ -22432,7 +22432,7 @@ pub unsafe fn vreinterpretq_s32_u32(a: uint32x4_t) -> int32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s64_u64(a: uint64x2_t) -> int64x2_t {
transmute(a)
}
@@ -22445,7 +22445,7 @@ pub unsafe fn vreinterpretq_s64_u64(a: uint64x2_t) -> int64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_u8_p8(a: poly8x8_t) -> uint8x8_t {
transmute(a)
}
@@ -22458,7 +22458,7 @@ pub unsafe fn vreinterpret_u8_p8(a: poly8x8_t) -> uint8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_u8_s8(a: int8x8_t) -> uint8x8_t {
transmute(a)
}
@@ -22471,7 +22471,7 @@ pub unsafe fn vreinterpret_u8_s8(a: int8x8_t) -> uint8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_u16_p16(a: poly16x4_t) -> uint16x4_t {
transmute(a)
}
@@ -22484,7 +22484,7 @@ pub unsafe fn vreinterpret_u16_p16(a: poly16x4_t) -> uint16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_u16_s16(a: int16x4_t) -> uint16x4_t {
transmute(a)
}
@@ -22497,7 +22497,7 @@ pub unsafe fn vreinterpret_u16_s16(a: int16x4_t) -> uint16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_u32_s32(a: int32x2_t) -> uint32x2_t {
transmute(a)
}
@@ -22510,7 +22510,7 @@ pub unsafe fn vreinterpret_u32_s32(a: int32x2_t) -> uint32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_u64_s64(a: int64x1_t) -> uint64x1_t {
transmute(a)
}
@@ -22523,7 +22523,7 @@ pub unsafe fn vreinterpret_u64_s64(a: int64x1_t) -> uint64x1_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u8_p8(a: poly8x16_t) -> uint8x16_t {
transmute(a)
}
@@ -22536,7 +22536,7 @@ pub unsafe fn vreinterpretq_u8_p8(a: poly8x16_t) -> uint8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u8_s8(a: int8x16_t) -> uint8x16_t {
transmute(a)
}
@@ -22549,7 +22549,7 @@ pub unsafe fn vreinterpretq_u8_s8(a: int8x16_t) -> uint8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u16_p16(a: poly16x8_t) -> uint16x8_t {
transmute(a)
}
@@ -22562,7 +22562,7 @@ pub unsafe fn vreinterpretq_u16_p16(a: poly16x8_t) -> uint16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u16_s16(a: int16x8_t) -> uint16x8_t {
transmute(a)
}
@@ -22575,7 +22575,7 @@ pub unsafe fn vreinterpretq_u16_s16(a: int16x8_t) -> uint16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u32_s32(a: int32x4_t) -> uint32x4_t {
transmute(a)
}
@@ -22588,7 +22588,7 @@ pub unsafe fn vreinterpretq_u32_s32(a: int32x4_t) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u64_s64(a: int64x2_t) -> uint64x2_t {
transmute(a)
}
@@ -22601,7 +22601,7 @@ pub unsafe fn vreinterpretq_u64_s64(a: int64x2_t) -> uint64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_p8_s8(a: int8x8_t) -> poly8x8_t {
transmute(a)
}
@@ -22614,7 +22614,7 @@ pub unsafe fn vreinterpret_p8_s8(a: int8x8_t) -> poly8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_p8_u8(a: uint8x8_t) -> poly8x8_t {
transmute(a)
}
@@ -22627,7 +22627,7 @@ pub unsafe fn vreinterpret_p8_u8(a: uint8x8_t) -> poly8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_p16_s16(a: int16x4_t) -> poly16x4_t {
transmute(a)
}
@@ -22640,7 +22640,7 @@ pub unsafe fn vreinterpret_p16_s16(a: int16x4_t) -> poly16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_p16_u16(a: uint16x4_t) -> poly16x4_t {
transmute(a)
}
@@ -22653,7 +22653,7 @@ pub unsafe fn vreinterpret_p16_u16(a: uint16x4_t) -> poly16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p8_s8(a: int8x16_t) -> poly8x16_t {
transmute(a)
}
@@ -22666,7 +22666,7 @@ pub unsafe fn vreinterpretq_p8_s8(a: int8x16_t) -> poly8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p8_u8(a: uint8x16_t) -> poly8x16_t {
transmute(a)
}
@@ -22679,7 +22679,7 @@ pub unsafe fn vreinterpretq_p8_u8(a: uint8x16_t) -> poly8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p16_s16(a: int16x8_t) -> poly16x8_t {
transmute(a)
}
@@ -22692,7 +22692,7 @@ pub unsafe fn vreinterpretq_p16_s16(a: int16x8_t) -> poly16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p16_u16(a: uint16x8_t) -> poly16x8_t {
transmute(a)
}
@@ -22705,7 +22705,7 @@ pub unsafe fn vreinterpretq_p16_u16(a: uint16x8_t) -> poly16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_s8_s16(a: int16x4_t) -> int8x8_t {
transmute(a)
}
@@ -22718,7 +22718,7 @@ pub unsafe fn vreinterpret_s8_s16(a: int16x4_t) -> int8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_s8_u16(a: uint16x4_t) -> int8x8_t {
transmute(a)
}
@@ -22731,7 +22731,7 @@ pub unsafe fn vreinterpret_s8_u16(a: uint16x4_t) -> int8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_s8_p16(a: poly16x4_t) -> int8x8_t {
transmute(a)
}
@@ -22744,7 +22744,7 @@ pub unsafe fn vreinterpret_s8_p16(a: poly16x4_t) -> int8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_s16_s32(a: int32x2_t) -> int16x4_t {
transmute(a)
}
@@ -22757,7 +22757,7 @@ pub unsafe fn vreinterpret_s16_s32(a: int32x2_t) -> int16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_s16_u32(a: uint32x2_t) -> int16x4_t {
transmute(a)
}
@@ -22770,7 +22770,7 @@ pub unsafe fn vreinterpret_s16_u32(a: uint32x2_t) -> int16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_s32_s64(a: int64x1_t) -> int32x2_t {
transmute(a)
}
@@ -22783,7 +22783,7 @@ pub unsafe fn vreinterpret_s32_s64(a: int64x1_t) -> int32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_s32_u64(a: uint64x1_t) -> int32x2_t {
transmute(a)
}
@@ -22796,7 +22796,7 @@ pub unsafe fn vreinterpret_s32_u64(a: uint64x1_t) -> int32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s8_s16(a: int16x8_t) -> int8x16_t {
transmute(a)
}
@@ -22809,7 +22809,7 @@ pub unsafe fn vreinterpretq_s8_s16(a: int16x8_t) -> int8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s8_u16(a: uint16x8_t) -> int8x16_t {
transmute(a)
}
@@ -22822,7 +22822,7 @@ pub unsafe fn vreinterpretq_s8_u16(a: uint16x8_t) -> int8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s8_p16(a: poly16x8_t) -> int8x16_t {
transmute(a)
}
@@ -22835,7 +22835,7 @@ pub unsafe fn vreinterpretq_s8_p16(a: poly16x8_t) -> int8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s16_s32(a: int32x4_t) -> int16x8_t {
transmute(a)
}
@@ -22848,7 +22848,7 @@ pub unsafe fn vreinterpretq_s16_s32(a: int32x4_t) -> int16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s16_u32(a: uint32x4_t) -> int16x8_t {
transmute(a)
}
@@ -22861,7 +22861,7 @@ pub unsafe fn vreinterpretq_s16_u32(a: uint32x4_t) -> int16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s32_s64(a: int64x2_t) -> int32x4_t {
transmute(a)
}
@@ -22874,7 +22874,7 @@ pub unsafe fn vreinterpretq_s32_s64(a: int64x2_t) -> int32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s32_u64(a: uint64x2_t) -> int32x4_t {
transmute(a)
}
@@ -22887,7 +22887,7 @@ pub unsafe fn vreinterpretq_s32_u64(a: uint64x2_t) -> int32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_u8_p16(a: poly16x4_t) -> uint8x8_t {
transmute(a)
}
@@ -22900,7 +22900,7 @@ pub unsafe fn vreinterpret_u8_p16(a: poly16x4_t) -> uint8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_u8_s16(a: int16x4_t) -> uint8x8_t {
transmute(a)
}
@@ -22913,7 +22913,7 @@ pub unsafe fn vreinterpret_u8_s16(a: int16x4_t) -> uint8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_u8_u16(a: uint16x4_t) -> uint8x8_t {
transmute(a)
}
@@ -22926,7 +22926,7 @@ pub unsafe fn vreinterpret_u8_u16(a: uint16x4_t) -> uint8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_u16_s32(a: int32x2_t) -> uint16x4_t {
transmute(a)
}
@@ -22939,7 +22939,7 @@ pub unsafe fn vreinterpret_u16_s32(a: int32x2_t) -> uint16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_u16_u32(a: uint32x2_t) -> uint16x4_t {
transmute(a)
}
@@ -22952,7 +22952,7 @@ pub unsafe fn vreinterpret_u16_u32(a: uint32x2_t) -> uint16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_u32_s64(a: int64x1_t) -> uint32x2_t {
transmute(a)
}
@@ -22965,7 +22965,7 @@ pub unsafe fn vreinterpret_u32_s64(a: int64x1_t) -> uint32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_u32_u64(a: uint64x1_t) -> uint32x2_t {
transmute(a)
}
@@ -22978,7 +22978,7 @@ pub unsafe fn vreinterpret_u32_u64(a: uint64x1_t) -> uint32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u8_p16(a: poly16x8_t) -> uint8x16_t {
transmute(a)
}
@@ -22991,7 +22991,7 @@ pub unsafe fn vreinterpretq_u8_p16(a: poly16x8_t) -> uint8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u8_s16(a: int16x8_t) -> uint8x16_t {
transmute(a)
}
@@ -23004,7 +23004,7 @@ pub unsafe fn vreinterpretq_u8_s16(a: int16x8_t) -> uint8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u8_u16(a: uint16x8_t) -> uint8x16_t {
transmute(a)
}
@@ -23017,7 +23017,7 @@ pub unsafe fn vreinterpretq_u8_u16(a: uint16x8_t) -> uint8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u16_s32(a: int32x4_t) -> uint16x8_t {
transmute(a)
}
@@ -23030,7 +23030,7 @@ pub unsafe fn vreinterpretq_u16_s32(a: int32x4_t) -> uint16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u16_u32(a: uint32x4_t) -> uint16x8_t {
transmute(a)
}
@@ -23043,7 +23043,7 @@ pub unsafe fn vreinterpretq_u16_u32(a: uint32x4_t) -> uint16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u32_s64(a: int64x2_t) -> uint32x4_t {
transmute(a)
}
@@ -23056,7 +23056,7 @@ pub unsafe fn vreinterpretq_u32_s64(a: int64x2_t) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u32_u64(a: uint64x2_t) -> uint32x4_t {
transmute(a)
}
@@ -23069,7 +23069,7 @@ pub unsafe fn vreinterpretq_u32_u64(a: uint64x2_t) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_p8_p16(a: poly16x4_t) -> poly8x8_t {
transmute(a)
}
@@ -23082,7 +23082,7 @@ pub unsafe fn vreinterpret_p8_p16(a: poly16x4_t) -> poly8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_p8_s16(a: int16x4_t) -> poly8x8_t {
transmute(a)
}
@@ -23095,7 +23095,7 @@ pub unsafe fn vreinterpret_p8_s16(a: int16x4_t) -> poly8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_p8_u16(a: uint16x4_t) -> poly8x8_t {
transmute(a)
}
@@ -23108,7 +23108,7 @@ pub unsafe fn vreinterpret_p8_u16(a: uint16x4_t) -> poly8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_p16_s32(a: int32x2_t) -> poly16x4_t {
transmute(a)
}
@@ -23121,7 +23121,7 @@ pub unsafe fn vreinterpret_p16_s32(a: int32x2_t) -> poly16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_p16_u32(a: uint32x2_t) -> poly16x4_t {
transmute(a)
}
@@ -23134,7 +23134,7 @@ pub unsafe fn vreinterpret_p16_u32(a: uint32x2_t) -> poly16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p8_p16(a: poly16x8_t) -> poly8x16_t {
transmute(a)
}
@@ -23147,7 +23147,7 @@ pub unsafe fn vreinterpretq_p8_p16(a: poly16x8_t) -> poly8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p8_s16(a: int16x8_t) -> poly8x16_t {
transmute(a)
}
@@ -23160,7 +23160,7 @@ pub unsafe fn vreinterpretq_p8_s16(a: int16x8_t) -> poly8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p8_u16(a: uint16x8_t) -> poly8x16_t {
transmute(a)
}
@@ -23173,7 +23173,7 @@ pub unsafe fn vreinterpretq_p8_u16(a: uint16x8_t) -> poly8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p16_s32(a: int32x4_t) -> poly16x8_t {
transmute(a)
}
@@ -23186,7 +23186,7 @@ pub unsafe fn vreinterpretq_p16_s32(a: int32x4_t) -> poly16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p16_u32(a: uint32x4_t) -> poly16x8_t {
transmute(a)
}
@@ -23199,7 +23199,7 @@ pub unsafe fn vreinterpretq_p16_u32(a: uint32x4_t) -> poly16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_s32_p64(a: poly64x1_t) -> int32x2_t {
transmute(a)
}
@@ -23212,7 +23212,7 @@ pub unsafe fn vreinterpret_s32_p64(a: poly64x1_t) -> int32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_u32_p64(a: poly64x1_t) -> uint32x2_t {
transmute(a)
}
@@ -23225,7 +23225,7 @@ pub unsafe fn vreinterpret_u32_p64(a: poly64x1_t) -> uint32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s32_p64(a: poly64x2_t) -> int32x4_t {
transmute(a)
}
@@ -23238,7 +23238,7 @@ pub unsafe fn vreinterpretq_s32_p64(a: poly64x2_t) -> int32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u32_p64(a: poly64x2_t) -> uint32x4_t {
transmute(a)
}
@@ -23251,7 +23251,7 @@ pub unsafe fn vreinterpretq_u32_p64(a: poly64x2_t) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s64_p128(a: p128) -> int64x2_t {
transmute(a)
}
@@ -23264,7 +23264,7 @@ pub unsafe fn vreinterpretq_s64_p128(a: p128) -> int64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u64_p128(a: p128) -> uint64x2_t {
transmute(a)
}
@@ -23277,7 +23277,7 @@ pub unsafe fn vreinterpretq_u64_p128(a: p128) -> uint64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p64_p128(a: p128) -> poly64x2_t {
transmute(a)
}
@@ -23290,7 +23290,7 @@ pub unsafe fn vreinterpretq_p64_p128(a: p128) -> poly64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_s16_p8(a: poly8x8_t) -> int16x4_t {
transmute(a)
}
@@ -23303,7 +23303,7 @@ pub unsafe fn vreinterpret_s16_p8(a: poly8x8_t) -> int16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_s16_s8(a: int8x8_t) -> int16x4_t {
transmute(a)
}
@@ -23316,7 +23316,7 @@ pub unsafe fn vreinterpret_s16_s8(a: int8x8_t) -> int16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_s16_u8(a: uint8x8_t) -> int16x4_t {
transmute(a)
}
@@ -23329,7 +23329,7 @@ pub unsafe fn vreinterpret_s16_u8(a: uint8x8_t) -> int16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_s32_p16(a: poly16x4_t) -> int32x2_t {
transmute(a)
}
@@ -23342,7 +23342,7 @@ pub unsafe fn vreinterpret_s32_p16(a: poly16x4_t) -> int32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_s32_s16(a: int16x4_t) -> int32x2_t {
transmute(a)
}
@@ -23355,7 +23355,7 @@ pub unsafe fn vreinterpret_s32_s16(a: int16x4_t) -> int32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_s32_u16(a: uint16x4_t) -> int32x2_t {
transmute(a)
}
@@ -23368,7 +23368,7 @@ pub unsafe fn vreinterpret_s32_u16(a: uint16x4_t) -> int32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_s64_s32(a: int32x2_t) -> int64x1_t {
transmute(a)
}
@@ -23381,7 +23381,7 @@ pub unsafe fn vreinterpret_s64_s32(a: int32x2_t) -> int64x1_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_s64_u32(a: uint32x2_t) -> int64x1_t {
transmute(a)
}
@@ -23394,7 +23394,7 @@ pub unsafe fn vreinterpret_s64_u32(a: uint32x2_t) -> int64x1_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s16_p8(a: poly8x16_t) -> int16x8_t {
transmute(a)
}
@@ -23407,7 +23407,7 @@ pub unsafe fn vreinterpretq_s16_p8(a: poly8x16_t) -> int16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s16_s8(a: int8x16_t) -> int16x8_t {
transmute(a)
}
@@ -23420,7 +23420,7 @@ pub unsafe fn vreinterpretq_s16_s8(a: int8x16_t) -> int16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s16_u8(a: uint8x16_t) -> int16x8_t {
transmute(a)
}
@@ -23433,7 +23433,7 @@ pub unsafe fn vreinterpretq_s16_u8(a: uint8x16_t) -> int16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s32_p16(a: poly16x8_t) -> int32x4_t {
transmute(a)
}
@@ -23446,7 +23446,7 @@ pub unsafe fn vreinterpretq_s32_p16(a: poly16x8_t) -> int32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s32_s16(a: int16x8_t) -> int32x4_t {
transmute(a)
}
@@ -23459,7 +23459,7 @@ pub unsafe fn vreinterpretq_s32_s16(a: int16x8_t) -> int32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s32_u16(a: uint16x8_t) -> int32x4_t {
transmute(a)
}
@@ -23472,7 +23472,7 @@ pub unsafe fn vreinterpretq_s32_u16(a: uint16x8_t) -> int32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s64_s32(a: int32x4_t) -> int64x2_t {
transmute(a)
}
@@ -23485,7 +23485,7 @@ pub unsafe fn vreinterpretq_s64_s32(a: int32x4_t) -> int64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s64_u32(a: uint32x4_t) -> int64x2_t {
transmute(a)
}
@@ -23498,7 +23498,7 @@ pub unsafe fn vreinterpretq_s64_u32(a: uint32x4_t) -> int64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_u16_p8(a: poly8x8_t) -> uint16x4_t {
transmute(a)
}
@@ -23511,7 +23511,7 @@ pub unsafe fn vreinterpret_u16_p8(a: poly8x8_t) -> uint16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_u16_s8(a: int8x8_t) -> uint16x4_t {
transmute(a)
}
@@ -23524,7 +23524,7 @@ pub unsafe fn vreinterpret_u16_s8(a: int8x8_t) -> uint16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_u16_u8(a: uint8x8_t) -> uint16x4_t {
transmute(a)
}
@@ -23537,7 +23537,7 @@ pub unsafe fn vreinterpret_u16_u8(a: uint8x8_t) -> uint16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_u32_p16(a: poly16x4_t) -> uint32x2_t {
transmute(a)
}
@@ -23550,7 +23550,7 @@ pub unsafe fn vreinterpret_u32_p16(a: poly16x4_t) -> uint32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_u32_s16(a: int16x4_t) -> uint32x2_t {
transmute(a)
}
@@ -23563,7 +23563,7 @@ pub unsafe fn vreinterpret_u32_s16(a: int16x4_t) -> uint32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_u32_u16(a: uint16x4_t) -> uint32x2_t {
transmute(a)
}
@@ -23576,7 +23576,7 @@ pub unsafe fn vreinterpret_u32_u16(a: uint16x4_t) -> uint32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_u64_s32(a: int32x2_t) -> uint64x1_t {
transmute(a)
}
@@ -23589,7 +23589,7 @@ pub unsafe fn vreinterpret_u64_s32(a: int32x2_t) -> uint64x1_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_u64_u32(a: uint32x2_t) -> uint64x1_t {
transmute(a)
}
@@ -23602,7 +23602,7 @@ pub unsafe fn vreinterpret_u64_u32(a: uint32x2_t) -> uint64x1_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u16_p8(a: poly8x16_t) -> uint16x8_t {
transmute(a)
}
@@ -23615,7 +23615,7 @@ pub unsafe fn vreinterpretq_u16_p8(a: poly8x16_t) -> uint16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u16_s8(a: int8x16_t) -> uint16x8_t {
transmute(a)
}
@@ -23628,7 +23628,7 @@ pub unsafe fn vreinterpretq_u16_s8(a: int8x16_t) -> uint16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u16_u8(a: uint8x16_t) -> uint16x8_t {
transmute(a)
}
@@ -23641,7 +23641,7 @@ pub unsafe fn vreinterpretq_u16_u8(a: uint8x16_t) -> uint16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u32_p16(a: poly16x8_t) -> uint32x4_t {
transmute(a)
}
@@ -23654,7 +23654,7 @@ pub unsafe fn vreinterpretq_u32_p16(a: poly16x8_t) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u32_s16(a: int16x8_t) -> uint32x4_t {
transmute(a)
}
@@ -23667,7 +23667,7 @@ pub unsafe fn vreinterpretq_u32_s16(a: int16x8_t) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u32_u16(a: uint16x8_t) -> uint32x4_t {
transmute(a)
}
@@ -23680,7 +23680,7 @@ pub unsafe fn vreinterpretq_u32_u16(a: uint16x8_t) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u64_s32(a: int32x4_t) -> uint64x2_t {
transmute(a)
}
@@ -23693,7 +23693,7 @@ pub unsafe fn vreinterpretq_u64_s32(a: int32x4_t) -> uint64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u64_u32(a: uint32x4_t) -> uint64x2_t {
transmute(a)
}
@@ -23706,7 +23706,7 @@ pub unsafe fn vreinterpretq_u64_u32(a: uint32x4_t) -> uint64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_p16_p8(a: poly8x8_t) -> poly16x4_t {
transmute(a)
}
@@ -23719,7 +23719,7 @@ pub unsafe fn vreinterpret_p16_p8(a: poly8x8_t) -> poly16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_p16_s8(a: int8x8_t) -> poly16x4_t {
transmute(a)
}
@@ -23732,7 +23732,7 @@ pub unsafe fn vreinterpret_p16_s8(a: int8x8_t) -> poly16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_p16_u8(a: uint8x8_t) -> poly16x4_t {
transmute(a)
}
@@ -23745,7 +23745,7 @@ pub unsafe fn vreinterpret_p16_u8(a: uint8x8_t) -> poly16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p16_p8(a: poly8x16_t) -> poly16x8_t {
transmute(a)
}
@@ -23758,7 +23758,7 @@ pub unsafe fn vreinterpretq_p16_p8(a: poly8x16_t) -> poly16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p16_s8(a: int8x16_t) -> poly16x8_t {
transmute(a)
}
@@ -23771,7 +23771,7 @@ pub unsafe fn vreinterpretq_p16_s8(a: int8x16_t) -> poly16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p16_u8(a: uint8x16_t) -> poly16x8_t {
transmute(a)
}
@@ -23784,7 +23784,7 @@ pub unsafe fn vreinterpretq_p16_u8(a: uint8x16_t) -> poly16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_p64_s32(a: int32x2_t) -> poly64x1_t {
transmute(a)
}
@@ -23797,7 +23797,7 @@ pub unsafe fn vreinterpret_p64_s32(a: int32x2_t) -> poly64x1_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_p64_u32(a: uint32x2_t) -> poly64x1_t {
transmute(a)
}
@@ -23810,7 +23810,7 @@ pub unsafe fn vreinterpret_p64_u32(a: uint32x2_t) -> poly64x1_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p64_s32(a: int32x4_t) -> poly64x2_t {
transmute(a)
}
@@ -23823,7 +23823,7 @@ pub unsafe fn vreinterpretq_p64_s32(a: int32x4_t) -> poly64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p64_u32(a: uint32x4_t) -> poly64x2_t {
transmute(a)
}
@@ -23836,7 +23836,7 @@ pub unsafe fn vreinterpretq_p64_u32(a: uint32x4_t) -> poly64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p128_s64(a: int64x2_t) -> p128 {
transmute(a)
}
@@ -23849,7 +23849,7 @@ pub unsafe fn vreinterpretq_p128_s64(a: int64x2_t) -> p128 {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p128_u64(a: uint64x2_t) -> p128 {
transmute(a)
}
@@ -23862,7 +23862,7 @@ pub unsafe fn vreinterpretq_p128_u64(a: uint64x2_t) -> p128 {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p128_p64(a: poly64x2_t) -> p128 {
transmute(a)
}
@@ -23875,7 +23875,7 @@ pub unsafe fn vreinterpretq_p128_p64(a: poly64x2_t) -> p128 {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_s8_s32(a: int32x2_t) -> int8x8_t {
transmute(a)
}
@@ -23888,7 +23888,7 @@ pub unsafe fn vreinterpret_s8_s32(a: int32x2_t) -> int8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_s8_u32(a: uint32x2_t) -> int8x8_t {
transmute(a)
}
@@ -23901,7 +23901,7 @@ pub unsafe fn vreinterpret_s8_u32(a: uint32x2_t) -> int8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_s16_s64(a: int64x1_t) -> int16x4_t {
transmute(a)
}
@@ -23914,7 +23914,7 @@ pub unsafe fn vreinterpret_s16_s64(a: int64x1_t) -> int16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_s16_u64(a: uint64x1_t) -> int16x4_t {
transmute(a)
}
@@ -23927,7 +23927,7 @@ pub unsafe fn vreinterpret_s16_u64(a: uint64x1_t) -> int16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s8_s32(a: int32x4_t) -> int8x16_t {
transmute(a)
}
@@ -23940,7 +23940,7 @@ pub unsafe fn vreinterpretq_s8_s32(a: int32x4_t) -> int8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s8_u32(a: uint32x4_t) -> int8x16_t {
transmute(a)
}
@@ -23953,7 +23953,7 @@ pub unsafe fn vreinterpretq_s8_u32(a: uint32x4_t) -> int8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s16_s64(a: int64x2_t) -> int16x8_t {
transmute(a)
}
@@ -23966,7 +23966,7 @@ pub unsafe fn vreinterpretq_s16_s64(a: int64x2_t) -> int16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s16_u64(a: uint64x2_t) -> int16x8_t {
transmute(a)
}
@@ -23979,7 +23979,7 @@ pub unsafe fn vreinterpretq_s16_u64(a: uint64x2_t) -> int16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_u8_s32(a: int32x2_t) -> uint8x8_t {
transmute(a)
}
@@ -23992,7 +23992,7 @@ pub unsafe fn vreinterpret_u8_s32(a: int32x2_t) -> uint8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_u8_u32(a: uint32x2_t) -> uint8x8_t {
transmute(a)
}
@@ -24005,7 +24005,7 @@ pub unsafe fn vreinterpret_u8_u32(a: uint32x2_t) -> uint8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_u16_s64(a: int64x1_t) -> uint16x4_t {
transmute(a)
}
@@ -24018,7 +24018,7 @@ pub unsafe fn vreinterpret_u16_s64(a: int64x1_t) -> uint16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_u16_u64(a: uint64x1_t) -> uint16x4_t {
transmute(a)
}
@@ -24031,7 +24031,7 @@ pub unsafe fn vreinterpret_u16_u64(a: uint64x1_t) -> uint16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u8_s32(a: int32x4_t) -> uint8x16_t {
transmute(a)
}
@@ -24044,7 +24044,7 @@ pub unsafe fn vreinterpretq_u8_s32(a: int32x4_t) -> uint8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u8_u32(a: uint32x4_t) -> uint8x16_t {
transmute(a)
}
@@ -24057,7 +24057,7 @@ pub unsafe fn vreinterpretq_u8_u32(a: uint32x4_t) -> uint8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u16_s64(a: int64x2_t) -> uint16x8_t {
transmute(a)
}
@@ -24070,7 +24070,7 @@ pub unsafe fn vreinterpretq_u16_s64(a: int64x2_t) -> uint16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u16_u64(a: uint64x2_t) -> uint16x8_t {
transmute(a)
}
@@ -24083,7 +24083,7 @@ pub unsafe fn vreinterpretq_u16_u64(a: uint64x2_t) -> uint16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_p8_s32(a: int32x2_t) -> poly8x8_t {
transmute(a)
}
@@ -24096,7 +24096,7 @@ pub unsafe fn vreinterpret_p8_s32(a: int32x2_t) -> poly8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_p8_u32(a: uint32x2_t) -> poly8x8_t {
transmute(a)
}
@@ -24109,7 +24109,7 @@ pub unsafe fn vreinterpret_p8_u32(a: uint32x2_t) -> poly8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_p16_s64(a: int64x1_t) -> poly16x4_t {
transmute(a)
}
@@ -24122,7 +24122,7 @@ pub unsafe fn vreinterpret_p16_s64(a: int64x1_t) -> poly16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_p16_u64(a: uint64x1_t) -> poly16x4_t {
transmute(a)
}
@@ -24135,7 +24135,7 @@ pub unsafe fn vreinterpret_p16_u64(a: uint64x1_t) -> poly16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p8_s32(a: int32x4_t) -> poly8x16_t {
transmute(a)
}
@@ -24148,7 +24148,7 @@ pub unsafe fn vreinterpretq_p8_s32(a: int32x4_t) -> poly8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p8_u32(a: uint32x4_t) -> poly8x16_t {
transmute(a)
}
@@ -24161,7 +24161,7 @@ pub unsafe fn vreinterpretq_p8_u32(a: uint32x4_t) -> poly8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p16_s64(a: int64x2_t) -> poly16x8_t {
transmute(a)
}
@@ -24174,7 +24174,7 @@ pub unsafe fn vreinterpretq_p16_s64(a: int64x2_t) -> poly16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p16_u64(a: uint64x2_t) -> poly16x8_t {
transmute(a)
}
@@ -24187,7 +24187,7 @@ pub unsafe fn vreinterpretq_p16_u64(a: uint64x2_t) -> poly16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_s16_p64(a: poly64x1_t) -> int16x4_t {
transmute(a)
}
@@ -24200,7 +24200,7 @@ pub unsafe fn vreinterpret_s16_p64(a: poly64x1_t) -> int16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_u16_p64(a: poly64x1_t) -> uint16x4_t {
transmute(a)
}
@@ -24213,7 +24213,7 @@ pub unsafe fn vreinterpret_u16_p64(a: poly64x1_t) -> uint16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_p16_p64(a: poly64x1_t) -> poly16x4_t {
transmute(a)
}
@@ -24226,7 +24226,7 @@ pub unsafe fn vreinterpret_p16_p64(a: poly64x1_t) -> poly16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s16_p64(a: poly64x2_t) -> int16x8_t {
transmute(a)
}
@@ -24239,7 +24239,7 @@ pub unsafe fn vreinterpretq_s16_p64(a: poly64x2_t) -> int16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u16_p64(a: poly64x2_t) -> uint16x8_t {
transmute(a)
}
@@ -24252,7 +24252,7 @@ pub unsafe fn vreinterpretq_u16_p64(a: poly64x2_t) -> uint16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p16_p64(a: poly64x2_t) -> poly16x8_t {
transmute(a)
}
@@ -24265,7 +24265,7 @@ pub unsafe fn vreinterpretq_p16_p64(a: poly64x2_t) -> poly16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s32_p128(a: p128) -> int32x4_t {
transmute(a)
}
@@ -24278,7 +24278,7 @@ pub unsafe fn vreinterpretq_s32_p128(a: p128) -> int32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u32_p128(a: p128) -> uint32x4_t {
transmute(a)
}
@@ -24291,7 +24291,7 @@ pub unsafe fn vreinterpretq_u32_p128(a: p128) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_s32_p8(a: poly8x8_t) -> int32x2_t {
transmute(a)
}
@@ -24304,7 +24304,7 @@ pub unsafe fn vreinterpret_s32_p8(a: poly8x8_t) -> int32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_s32_s8(a: int8x8_t) -> int32x2_t {
transmute(a)
}
@@ -24317,7 +24317,7 @@ pub unsafe fn vreinterpret_s32_s8(a: int8x8_t) -> int32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_s32_u8(a: uint8x8_t) -> int32x2_t {
transmute(a)
}
@@ -24330,7 +24330,7 @@ pub unsafe fn vreinterpret_s32_u8(a: uint8x8_t) -> int32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_s64_p16(a: poly16x4_t) -> int64x1_t {
transmute(a)
}
@@ -24343,7 +24343,7 @@ pub unsafe fn vreinterpret_s64_p16(a: poly16x4_t) -> int64x1_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_s64_s16(a: int16x4_t) -> int64x1_t {
transmute(a)
}
@@ -24356,7 +24356,7 @@ pub unsafe fn vreinterpret_s64_s16(a: int16x4_t) -> int64x1_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_s64_u16(a: uint16x4_t) -> int64x1_t {
transmute(a)
}
@@ -24369,7 +24369,7 @@ pub unsafe fn vreinterpret_s64_u16(a: uint16x4_t) -> int64x1_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s32_p8(a: poly8x16_t) -> int32x4_t {
transmute(a)
}
@@ -24382,7 +24382,7 @@ pub unsafe fn vreinterpretq_s32_p8(a: poly8x16_t) -> int32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s32_s8(a: int8x16_t) -> int32x4_t {
transmute(a)
}
@@ -24395,7 +24395,7 @@ pub unsafe fn vreinterpretq_s32_s8(a: int8x16_t) -> int32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s32_u8(a: uint8x16_t) -> int32x4_t {
transmute(a)
}
@@ -24408,7 +24408,7 @@ pub unsafe fn vreinterpretq_s32_u8(a: uint8x16_t) -> int32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s64_p16(a: poly16x8_t) -> int64x2_t {
transmute(a)
}
@@ -24421,7 +24421,7 @@ pub unsafe fn vreinterpretq_s64_p16(a: poly16x8_t) -> int64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s64_s16(a: int16x8_t) -> int64x2_t {
transmute(a)
}
@@ -24434,7 +24434,7 @@ pub unsafe fn vreinterpretq_s64_s16(a: int16x8_t) -> int64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s64_u16(a: uint16x8_t) -> int64x2_t {
transmute(a)
}
@@ -24447,7 +24447,7 @@ pub unsafe fn vreinterpretq_s64_u16(a: uint16x8_t) -> int64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_u32_p8(a: poly8x8_t) -> uint32x2_t {
transmute(a)
}
@@ -24460,7 +24460,7 @@ pub unsafe fn vreinterpret_u32_p8(a: poly8x8_t) -> uint32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_u32_s8(a: int8x8_t) -> uint32x2_t {
transmute(a)
}
@@ -24473,7 +24473,7 @@ pub unsafe fn vreinterpret_u32_s8(a: int8x8_t) -> uint32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_u32_u8(a: uint8x8_t) -> uint32x2_t {
transmute(a)
}
@@ -24486,7 +24486,7 @@ pub unsafe fn vreinterpret_u32_u8(a: uint8x8_t) -> uint32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_u64_p16(a: poly16x4_t) -> uint64x1_t {
transmute(a)
}
@@ -24499,7 +24499,7 @@ pub unsafe fn vreinterpret_u64_p16(a: poly16x4_t) -> uint64x1_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_u64_s16(a: int16x4_t) -> uint64x1_t {
transmute(a)
}
@@ -24512,7 +24512,7 @@ pub unsafe fn vreinterpret_u64_s16(a: int16x4_t) -> uint64x1_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_u64_u16(a: uint16x4_t) -> uint64x1_t {
transmute(a)
}
@@ -24525,7 +24525,7 @@ pub unsafe fn vreinterpret_u64_u16(a: uint16x4_t) -> uint64x1_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u32_p8(a: poly8x16_t) -> uint32x4_t {
transmute(a)
}
@@ -24538,7 +24538,7 @@ pub unsafe fn vreinterpretq_u32_p8(a: poly8x16_t) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u32_s8(a: int8x16_t) -> uint32x4_t {
transmute(a)
}
@@ -24551,7 +24551,7 @@ pub unsafe fn vreinterpretq_u32_s8(a: int8x16_t) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u32_u8(a: uint8x16_t) -> uint32x4_t {
transmute(a)
}
@@ -24564,7 +24564,7 @@ pub unsafe fn vreinterpretq_u32_u8(a: uint8x16_t) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u64_p16(a: poly16x8_t) -> uint64x2_t {
transmute(a)
}
@@ -24577,7 +24577,7 @@ pub unsafe fn vreinterpretq_u64_p16(a: poly16x8_t) -> uint64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u64_s16(a: int16x8_t) -> uint64x2_t {
transmute(a)
}
@@ -24590,7 +24590,7 @@ pub unsafe fn vreinterpretq_u64_s16(a: int16x8_t) -> uint64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u64_u16(a: uint16x8_t) -> uint64x2_t {
transmute(a)
}
@@ -24603,7 +24603,7 @@ pub unsafe fn vreinterpretq_u64_u16(a: uint16x8_t) -> uint64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_p64_p16(a: poly16x4_t) -> poly64x1_t {
transmute(a)
}
@@ -24616,7 +24616,7 @@ pub unsafe fn vreinterpret_p64_p16(a: poly16x4_t) -> poly64x1_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_p64_s16(a: int16x4_t) -> poly64x1_t {
transmute(a)
}
@@ -24629,7 +24629,7 @@ pub unsafe fn vreinterpret_p64_s16(a: int16x4_t) -> poly64x1_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_p64_u16(a: uint16x4_t) -> poly64x1_t {
transmute(a)
}
@@ -24642,7 +24642,7 @@ pub unsafe fn vreinterpret_p64_u16(a: uint16x4_t) -> poly64x1_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p64_p16(a: poly16x8_t) -> poly64x2_t {
transmute(a)
}
@@ -24655,7 +24655,7 @@ pub unsafe fn vreinterpretq_p64_p16(a: poly16x8_t) -> poly64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p64_s16(a: int16x8_t) -> poly64x2_t {
transmute(a)
}
@@ -24668,7 +24668,7 @@ pub unsafe fn vreinterpretq_p64_s16(a: int16x8_t) -> poly64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p64_u16(a: uint16x8_t) -> poly64x2_t {
transmute(a)
}
@@ -24681,7 +24681,7 @@ pub unsafe fn vreinterpretq_p64_u16(a: uint16x8_t) -> poly64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p128_s32(a: int32x4_t) -> p128 {
transmute(a)
}
@@ -24694,7 +24694,7 @@ pub unsafe fn vreinterpretq_p128_s32(a: int32x4_t) -> p128 {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p128_u32(a: uint32x4_t) -> p128 {
transmute(a)
}
@@ -24707,7 +24707,7 @@ pub unsafe fn vreinterpretq_p128_u32(a: uint32x4_t) -> p128 {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_s8_s64(a: int64x1_t) -> int8x8_t {
transmute(a)
}
@@ -24720,7 +24720,7 @@ pub unsafe fn vreinterpret_s8_s64(a: int64x1_t) -> int8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_s8_u64(a: uint64x1_t) -> int8x8_t {
transmute(a)
}
@@ -24733,7 +24733,7 @@ pub unsafe fn vreinterpret_s8_u64(a: uint64x1_t) -> int8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_u8_s64(a: int64x1_t) -> uint8x8_t {
transmute(a)
}
@@ -24746,7 +24746,7 @@ pub unsafe fn vreinterpret_u8_s64(a: int64x1_t) -> uint8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_u8_u64(a: uint64x1_t) -> uint8x8_t {
transmute(a)
}
@@ -24759,7 +24759,7 @@ pub unsafe fn vreinterpret_u8_u64(a: uint64x1_t) -> uint8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_p8_s64(a: int64x1_t) -> poly8x8_t {
transmute(a)
}
@@ -24772,7 +24772,7 @@ pub unsafe fn vreinterpret_p8_s64(a: int64x1_t) -> poly8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_p8_u64(a: uint64x1_t) -> poly8x8_t {
transmute(a)
}
@@ -24785,7 +24785,7 @@ pub unsafe fn vreinterpret_p8_u64(a: uint64x1_t) -> poly8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s8_s64(a: int64x2_t) -> int8x16_t {
transmute(a)
}
@@ -24798,7 +24798,7 @@ pub unsafe fn vreinterpretq_s8_s64(a: int64x2_t) -> int8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s8_u64(a: uint64x2_t) -> int8x16_t {
transmute(a)
}
@@ -24811,7 +24811,7 @@ pub unsafe fn vreinterpretq_s8_u64(a: uint64x2_t) -> int8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u8_s64(a: int64x2_t) -> uint8x16_t {
transmute(a)
}
@@ -24824,7 +24824,7 @@ pub unsafe fn vreinterpretq_u8_s64(a: int64x2_t) -> uint8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u8_u64(a: uint64x2_t) -> uint8x16_t {
transmute(a)
}
@@ -24837,7 +24837,7 @@ pub unsafe fn vreinterpretq_u8_u64(a: uint64x2_t) -> uint8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p8_s64(a: int64x2_t) -> poly8x16_t {
transmute(a)
}
@@ -24850,7 +24850,7 @@ pub unsafe fn vreinterpretq_p8_s64(a: int64x2_t) -> poly8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p8_u64(a: uint64x2_t) -> poly8x16_t {
transmute(a)
}
@@ -24863,7 +24863,7 @@ pub unsafe fn vreinterpretq_p8_u64(a: uint64x2_t) -> poly8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_s8_p64(a: poly64x1_t) -> int8x8_t {
transmute(a)
}
@@ -24876,7 +24876,7 @@ pub unsafe fn vreinterpret_s8_p64(a: poly64x1_t) -> int8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_u8_p64(a: poly64x1_t) -> uint8x8_t {
transmute(a)
}
@@ -24889,7 +24889,7 @@ pub unsafe fn vreinterpret_u8_p64(a: poly64x1_t) -> uint8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_p8_p64(a: poly64x1_t) -> poly8x8_t {
transmute(a)
}
@@ -24902,7 +24902,7 @@ pub unsafe fn vreinterpret_p8_p64(a: poly64x1_t) -> poly8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s8_p64(a: poly64x2_t) -> int8x16_t {
transmute(a)
}
@@ -24915,7 +24915,7 @@ pub unsafe fn vreinterpretq_s8_p64(a: poly64x2_t) -> int8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u8_p64(a: poly64x2_t) -> uint8x16_t {
transmute(a)
}
@@ -24928,7 +24928,7 @@ pub unsafe fn vreinterpretq_u8_p64(a: poly64x2_t) -> uint8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p8_p64(a: poly64x2_t) -> poly8x16_t {
transmute(a)
}
@@ -24941,7 +24941,7 @@ pub unsafe fn vreinterpretq_p8_p64(a: poly64x2_t) -> poly8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s16_p128(a: p128) -> int16x8_t {
transmute(a)
}
@@ -24954,7 +24954,7 @@ pub unsafe fn vreinterpretq_s16_p128(a: p128) -> int16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u16_p128(a: p128) -> uint16x8_t {
transmute(a)
}
@@ -24967,7 +24967,7 @@ pub unsafe fn vreinterpretq_u16_p128(a: p128) -> uint16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p16_p128(a: p128) -> poly16x8_t {
transmute(a)
}
@@ -24980,7 +24980,7 @@ pub unsafe fn vreinterpretq_p16_p128(a: p128) -> poly16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_s64_p8(a: poly8x8_t) -> int64x1_t {
transmute(a)
}
@@ -24993,7 +24993,7 @@ pub unsafe fn vreinterpret_s64_p8(a: poly8x8_t) -> int64x1_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_s64_s8(a: int8x8_t) -> int64x1_t {
transmute(a)
}
@@ -25006,7 +25006,7 @@ pub unsafe fn vreinterpret_s64_s8(a: int8x8_t) -> int64x1_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_s64_u8(a: uint8x8_t) -> int64x1_t {
transmute(a)
}
@@ -25019,7 +25019,7 @@ pub unsafe fn vreinterpret_s64_u8(a: uint8x8_t) -> int64x1_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_u64_p8(a: poly8x8_t) -> uint64x1_t {
transmute(a)
}
@@ -25032,7 +25032,7 @@ pub unsafe fn vreinterpret_u64_p8(a: poly8x8_t) -> uint64x1_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_u64_s8(a: int8x8_t) -> uint64x1_t {
transmute(a)
}
@@ -25045,7 +25045,7 @@ pub unsafe fn vreinterpret_u64_s8(a: int8x8_t) -> uint64x1_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_u64_u8(a: uint8x8_t) -> uint64x1_t {
transmute(a)
}
@@ -25058,7 +25058,7 @@ pub unsafe fn vreinterpret_u64_u8(a: uint8x8_t) -> uint64x1_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s64_p8(a: poly8x16_t) -> int64x2_t {
transmute(a)
}
@@ -25071,7 +25071,7 @@ pub unsafe fn vreinterpretq_s64_p8(a: poly8x16_t) -> int64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s64_s8(a: int8x16_t) -> int64x2_t {
transmute(a)
}
@@ -25084,7 +25084,7 @@ pub unsafe fn vreinterpretq_s64_s8(a: int8x16_t) -> int64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s64_u8(a: uint8x16_t) -> int64x2_t {
transmute(a)
}
@@ -25097,7 +25097,7 @@ pub unsafe fn vreinterpretq_s64_u8(a: uint8x16_t) -> int64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u64_p8(a: poly8x16_t) -> uint64x2_t {
transmute(a)
}
@@ -25110,7 +25110,7 @@ pub unsafe fn vreinterpretq_u64_p8(a: poly8x16_t) -> uint64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u64_s8(a: int8x16_t) -> uint64x2_t {
transmute(a)
}
@@ -25123,7 +25123,7 @@ pub unsafe fn vreinterpretq_u64_s8(a: int8x16_t) -> uint64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u64_u8(a: uint8x16_t) -> uint64x2_t {
transmute(a)
}
@@ -25136,7 +25136,7 @@ pub unsafe fn vreinterpretq_u64_u8(a: uint8x16_t) -> uint64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_p64_p8(a: poly8x8_t) -> poly64x1_t {
transmute(a)
}
@@ -25149,7 +25149,7 @@ pub unsafe fn vreinterpret_p64_p8(a: poly8x8_t) -> poly64x1_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_p64_s8(a: int8x8_t) -> poly64x1_t {
transmute(a)
}
@@ -25162,7 +25162,7 @@ pub unsafe fn vreinterpret_p64_s8(a: int8x8_t) -> poly64x1_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_p64_u8(a: uint8x8_t) -> poly64x1_t {
transmute(a)
}
@@ -25175,7 +25175,7 @@ pub unsafe fn vreinterpret_p64_u8(a: uint8x8_t) -> poly64x1_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p64_p8(a: poly8x16_t) -> poly64x2_t {
transmute(a)
}
@@ -25188,7 +25188,7 @@ pub unsafe fn vreinterpretq_p64_p8(a: poly8x16_t) -> poly64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p64_s8(a: int8x16_t) -> poly64x2_t {
transmute(a)
}
@@ -25201,7 +25201,7 @@ pub unsafe fn vreinterpretq_p64_s8(a: int8x16_t) -> poly64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p64_u8(a: uint8x16_t) -> poly64x2_t {
transmute(a)
}
@@ -25214,7 +25214,7 @@ pub unsafe fn vreinterpretq_p64_u8(a: uint8x16_t) -> poly64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p128_s16(a: int16x8_t) -> p128 {
transmute(a)
}
@@ -25227,7 +25227,7 @@ pub unsafe fn vreinterpretq_p128_s16(a: int16x8_t) -> p128 {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p128_u16(a: uint16x8_t) -> p128 {
transmute(a)
}
@@ -25240,7 +25240,7 @@ pub unsafe fn vreinterpretq_p128_u16(a: uint16x8_t) -> p128 {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p128_p16(a: poly16x8_t) -> p128 {
transmute(a)
}
@@ -25253,7 +25253,7 @@ pub unsafe fn vreinterpretq_p128_p16(a: poly16x8_t) -> p128 {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p128_s8(a: int8x16_t) -> p128 {
transmute(a)
}
@@ -25266,7 +25266,7 @@ pub unsafe fn vreinterpretq_p128_s8(a: int8x16_t) -> p128 {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p128_u8(a: uint8x16_t) -> p128 {
transmute(a)
}
@@ -25279,7 +25279,7 @@ pub unsafe fn vreinterpretq_p128_u8(a: uint8x16_t) -> p128 {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p128_p8(a: poly8x16_t) -> p128 {
transmute(a)
}
@@ -25292,7 +25292,7 @@ pub unsafe fn vreinterpretq_p128_p8(a: poly8x16_t) -> p128 {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s8_p128(a: p128) -> int8x16_t {
transmute(a)
}
@@ -25305,7 +25305,7 @@ pub unsafe fn vreinterpretq_s8_p128(a: p128) -> int8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u8_p128(a: p128) -> uint8x16_t {
transmute(a)
}
@@ -25318,7 +25318,7 @@ pub unsafe fn vreinterpretq_u8_p128(a: p128) -> uint8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "aes,v8"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p8_p128(a: p128) -> poly8x16_t {
transmute(a)
}
@@ -25331,7 +25331,7 @@ pub unsafe fn vreinterpretq_p8_p128(a: p128) -> poly8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_s8_f32(a: float32x2_t) -> int8x8_t {
transmute(a)
}
@@ -25344,7 +25344,7 @@ pub unsafe fn vreinterpret_s8_f32(a: float32x2_t) -> int8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_s16_f32(a: float32x2_t) -> int16x4_t {
transmute(a)
}
@@ -25357,7 +25357,7 @@ pub unsafe fn vreinterpret_s16_f32(a: float32x2_t) -> int16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_s32_f32(a: float32x2_t) -> int32x2_t {
transmute(a)
}
@@ -25370,7 +25370,7 @@ pub unsafe fn vreinterpret_s32_f32(a: float32x2_t) -> int32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_s64_f32(a: float32x2_t) -> int64x1_t {
transmute(a)
}
@@ -25383,7 +25383,7 @@ pub unsafe fn vreinterpret_s64_f32(a: float32x2_t) -> int64x1_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s8_f32(a: float32x4_t) -> int8x16_t {
transmute(a)
}
@@ -25396,7 +25396,7 @@ pub unsafe fn vreinterpretq_s8_f32(a: float32x4_t) -> int8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s16_f32(a: float32x4_t) -> int16x8_t {
transmute(a)
}
@@ -25409,7 +25409,7 @@ pub unsafe fn vreinterpretq_s16_f32(a: float32x4_t) -> int16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s32_f32(a: float32x4_t) -> int32x4_t {
transmute(a)
}
@@ -25422,7 +25422,7 @@ pub unsafe fn vreinterpretq_s32_f32(a: float32x4_t) -> int32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_s64_f32(a: float32x4_t) -> int64x2_t {
transmute(a)
}
@@ -25435,7 +25435,7 @@ pub unsafe fn vreinterpretq_s64_f32(a: float32x4_t) -> int64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_u8_f32(a: float32x2_t) -> uint8x8_t {
transmute(a)
}
@@ -25448,7 +25448,7 @@ pub unsafe fn vreinterpret_u8_f32(a: float32x2_t) -> uint8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_u16_f32(a: float32x2_t) -> uint16x4_t {
transmute(a)
}
@@ -25461,7 +25461,7 @@ pub unsafe fn vreinterpret_u16_f32(a: float32x2_t) -> uint16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_u32_f32(a: float32x2_t) -> uint32x2_t {
transmute(a)
}
@@ -25474,7 +25474,7 @@ pub unsafe fn vreinterpret_u32_f32(a: float32x2_t) -> uint32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_u64_f32(a: float32x2_t) -> uint64x1_t {
transmute(a)
}
@@ -25487,7 +25487,7 @@ pub unsafe fn vreinterpret_u64_f32(a: float32x2_t) -> uint64x1_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u8_f32(a: float32x4_t) -> uint8x16_t {
transmute(a)
}
@@ -25500,7 +25500,7 @@ pub unsafe fn vreinterpretq_u8_f32(a: float32x4_t) -> uint8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u16_f32(a: float32x4_t) -> uint16x8_t {
transmute(a)
}
@@ -25513,7 +25513,7 @@ pub unsafe fn vreinterpretq_u16_f32(a: float32x4_t) -> uint16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u32_f32(a: float32x4_t) -> uint32x4_t {
transmute(a)
}
@@ -25526,7 +25526,7 @@ pub unsafe fn vreinterpretq_u32_f32(a: float32x4_t) -> uint32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_u64_f32(a: float32x4_t) -> uint64x2_t {
transmute(a)
}
@@ -25539,7 +25539,7 @@ pub unsafe fn vreinterpretq_u64_f32(a: float32x4_t) -> uint64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_p8_f32(a: float32x2_t) -> poly8x8_t {
transmute(a)
}
@@ -25552,7 +25552,7 @@ pub unsafe fn vreinterpret_p8_f32(a: float32x2_t) -> poly8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_p16_f32(a: float32x2_t) -> poly16x4_t {
transmute(a)
}
@@ -25565,7 +25565,7 @@ pub unsafe fn vreinterpret_p16_f32(a: float32x2_t) -> poly16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p8_f32(a: float32x4_t) -> poly8x16_t {
transmute(a)
}
@@ -25578,7 +25578,7 @@ pub unsafe fn vreinterpretq_p8_f32(a: float32x4_t) -> poly8x16_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p16_f32(a: float32x4_t) -> poly16x8_t {
transmute(a)
}
@@ -25591,7 +25591,7 @@ pub unsafe fn vreinterpretq_p16_f32(a: float32x4_t) -> poly16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_p128_f32(a: float32x4_t) -> p128 {
transmute(a)
}
@@ -25604,7 +25604,7 @@ pub unsafe fn vreinterpretq_p128_f32(a: float32x4_t) -> p128 {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_f32_s8(a: int8x8_t) -> float32x2_t {
transmute(a)
}
@@ -25617,7 +25617,7 @@ pub unsafe fn vreinterpret_f32_s8(a: int8x8_t) -> float32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_f32_s16(a: int16x4_t) -> float32x2_t {
transmute(a)
}
@@ -25630,7 +25630,7 @@ pub unsafe fn vreinterpret_f32_s16(a: int16x4_t) -> float32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_f32_s32(a: int32x2_t) -> float32x2_t {
transmute(a)
}
@@ -25643,7 +25643,7 @@ pub unsafe fn vreinterpret_f32_s32(a: int32x2_t) -> float32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_f32_s64(a: int64x1_t) -> float32x2_t {
transmute(a)
}
@@ -25656,7 +25656,7 @@ pub unsafe fn vreinterpret_f32_s64(a: int64x1_t) -> float32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_f32_s8(a: int8x16_t) -> float32x4_t {
transmute(a)
}
@@ -25669,7 +25669,7 @@ pub unsafe fn vreinterpretq_f32_s8(a: int8x16_t) -> float32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_f32_s16(a: int16x8_t) -> float32x4_t {
transmute(a)
}
@@ -25682,7 +25682,7 @@ pub unsafe fn vreinterpretq_f32_s16(a: int16x8_t) -> float32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_f32_s32(a: int32x4_t) -> float32x4_t {
transmute(a)
}
@@ -25695,7 +25695,7 @@ pub unsafe fn vreinterpretq_f32_s32(a: int32x4_t) -> float32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_f32_s64(a: int64x2_t) -> float32x4_t {
transmute(a)
}
@@ -25708,7 +25708,7 @@ pub unsafe fn vreinterpretq_f32_s64(a: int64x2_t) -> float32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_f32_u8(a: uint8x8_t) -> float32x2_t {
transmute(a)
}
@@ -25721,7 +25721,7 @@ pub unsafe fn vreinterpret_f32_u8(a: uint8x8_t) -> float32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_f32_u16(a: uint16x4_t) -> float32x2_t {
transmute(a)
}
@@ -25734,7 +25734,7 @@ pub unsafe fn vreinterpret_f32_u16(a: uint16x4_t) -> float32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_f32_u32(a: uint32x2_t) -> float32x2_t {
transmute(a)
}
@@ -25747,7 +25747,7 @@ pub unsafe fn vreinterpret_f32_u32(a: uint32x2_t) -> float32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_f32_u64(a: uint64x1_t) -> float32x2_t {
transmute(a)
}
@@ -25760,7 +25760,7 @@ pub unsafe fn vreinterpret_f32_u64(a: uint64x1_t) -> float32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_f32_u8(a: uint8x16_t) -> float32x4_t {
transmute(a)
}
@@ -25773,7 +25773,7 @@ pub unsafe fn vreinterpretq_f32_u8(a: uint8x16_t) -> float32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_f32_u16(a: uint16x8_t) -> float32x4_t {
transmute(a)
}
@@ -25786,7 +25786,7 @@ pub unsafe fn vreinterpretq_f32_u16(a: uint16x8_t) -> float32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_f32_u32(a: uint32x4_t) -> float32x4_t {
transmute(a)
}
@@ -25799,7 +25799,7 @@ pub unsafe fn vreinterpretq_f32_u32(a: uint32x4_t) -> float32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_f32_u64(a: uint64x2_t) -> float32x4_t {
transmute(a)
}
@@ -25812,7 +25812,7 @@ pub unsafe fn vreinterpretq_f32_u64(a: uint64x2_t) -> float32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_f32_p8(a: poly8x8_t) -> float32x2_t {
transmute(a)
}
@@ -25825,7 +25825,7 @@ pub unsafe fn vreinterpret_f32_p8(a: poly8x8_t) -> float32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpret_f32_p16(a: poly16x4_t) -> float32x2_t {
transmute(a)
}
@@ -25838,7 +25838,7 @@ pub unsafe fn vreinterpret_f32_p16(a: poly16x4_t) -> float32x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_f32_p8(a: poly8x16_t) -> float32x4_t {
transmute(a)
}
@@ -25851,7 +25851,7 @@ pub unsafe fn vreinterpretq_f32_p8(a: poly8x16_t) -> float32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_f32_p16(a: poly16x8_t) -> float32x4_t {
transmute(a)
}
@@ -25864,7 +25864,7 @@ pub unsafe fn vreinterpretq_f32_p16(a: poly16x8_t) -> float32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vreinterpretq_f32_p128(a: p128) -> float32x4_t {
transmute(a)
}
@@ -25877,7 +25877,7 @@ pub unsafe fn vreinterpretq_f32_p128(a: p128) -> float32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(srshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrshl_s8(a: int8x8_t, b: int8x8_t) -> int8x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -25896,7 +25896,7 @@ vrshl_s8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(srshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrshlq_s8(a: int8x16_t, b: int8x16_t) -> int8x16_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -25915,7 +25915,7 @@ vrshlq_s8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(srshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrshl_s16(a: int16x4_t, b: int16x4_t) -> int16x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -25934,7 +25934,7 @@ vrshl_s16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(srshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrshlq_s16(a: int16x8_t, b: int16x8_t) -> int16x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -25953,7 +25953,7 @@ vrshlq_s16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(srshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrshl_s32(a: int32x2_t, b: int32x2_t) -> int32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -25972,7 +25972,7 @@ vrshl_s32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(srshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrshlq_s32(a: int32x4_t, b: int32x4_t) -> int32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -25991,7 +25991,7 @@ vrshlq_s32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(srshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrshl_s64(a: int64x1_t, b: int64x1_t) -> int64x1_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -26010,7 +26010,7 @@ vrshl_s64_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(srshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrshlq_s64(a: int64x2_t, b: int64x2_t) -> int64x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -26029,7 +26029,7 @@ vrshlq_s64_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(urshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrshl_u8(a: uint8x8_t, b: int8x8_t) -> uint8x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -26048,7 +26048,7 @@ vrshl_u8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(urshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrshlq_u8(a: uint8x16_t, b: int8x16_t) -> uint8x16_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -26067,7 +26067,7 @@ vrshlq_u8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(urshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrshl_u16(a: uint16x4_t, b: int16x4_t) -> uint16x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -26086,7 +26086,7 @@ vrshl_u16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(urshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrshlq_u16(a: uint16x8_t, b: int16x8_t) -> uint16x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -26105,7 +26105,7 @@ vrshlq_u16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(urshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrshl_u32(a: uint32x2_t, b: int32x2_t) -> uint32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -26124,7 +26124,7 @@ vrshl_u32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(urshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrshlq_u32(a: uint32x4_t, b: int32x4_t) -> uint32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -26143,7 +26143,7 @@ vrshlq_u32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(urshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrshl_u64(a: uint64x1_t, b: int64x1_t) -> uint64x1_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -26162,7 +26162,7 @@ vrshl_u64_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(urshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrshlq_u64(a: uint64x2_t, b: int64x2_t) -> uint64x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -26182,7 +26182,7 @@ vrshlq_u64_(a, b)
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrshr, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(srshr, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrshr_n_s8<const N: i32>(a: int8x8_t) -> int8x8_t {
static_assert!(N : i32 where N >= 1 && N <= 8);
vrshl_s8(a, vdup_n_s8((-N) as _))
@@ -26197,7 +26197,7 @@ pub unsafe fn vrshr_n_s8<const N: i32>(a: int8x8_t) -> int8x8_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrshr, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(srshr, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrshrq_n_s8<const N: i32>(a: int8x16_t) -> int8x16_t {
static_assert!(N : i32 where N >= 1 && N <= 8);
vrshlq_s8(a, vdupq_n_s8((-N) as _))
@@ -26212,7 +26212,7 @@ pub unsafe fn vrshrq_n_s8<const N: i32>(a: int8x16_t) -> int8x16_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrshr, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(srshr, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrshr_n_s16<const N: i32>(a: int16x4_t) -> int16x4_t {
static_assert!(N : i32 where N >= 1 && N <= 16);
vrshl_s16(a, vdup_n_s16((-N) as _))
@@ -26227,7 +26227,7 @@ pub unsafe fn vrshr_n_s16<const N: i32>(a: int16x4_t) -> int16x4_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrshr, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(srshr, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrshrq_n_s16<const N: i32>(a: int16x8_t) -> int16x8_t {
static_assert!(N : i32 where N >= 1 && N <= 16);
vrshlq_s16(a, vdupq_n_s16((-N) as _))
@@ -26242,7 +26242,7 @@ pub unsafe fn vrshrq_n_s16<const N: i32>(a: int16x8_t) -> int16x8_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrshr, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(srshr, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrshr_n_s32<const N: i32>(a: int32x2_t) -> int32x2_t {
static_assert!(N : i32 where N >= 1 && N <= 32);
vrshl_s32(a, vdup_n_s32((-N) as _))
@@ -26257,7 +26257,7 @@ pub unsafe fn vrshr_n_s32<const N: i32>(a: int32x2_t) -> int32x2_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrshr, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(srshr, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrshrq_n_s32<const N: i32>(a: int32x4_t) -> int32x4_t {
static_assert!(N : i32 where N >= 1 && N <= 32);
vrshlq_s32(a, vdupq_n_s32((-N) as _))
@@ -26272,7 +26272,7 @@ pub unsafe fn vrshrq_n_s32<const N: i32>(a: int32x4_t) -> int32x4_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrshr, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(srshr, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrshr_n_s64<const N: i32>(a: int64x1_t) -> int64x1_t {
static_assert!(N : i32 where N >= 1 && N <= 64);
vrshl_s64(a, vdup_n_s64((-N) as _))
@@ -26287,7 +26287,7 @@ pub unsafe fn vrshr_n_s64<const N: i32>(a: int64x1_t) -> int64x1_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrshr, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(srshr, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrshrq_n_s64<const N: i32>(a: int64x2_t) -> int64x2_t {
static_assert!(N : i32 where N >= 1 && N <= 64);
vrshlq_s64(a, vdupq_n_s64((-N) as _))
@@ -26302,7 +26302,7 @@ pub unsafe fn vrshrq_n_s64<const N: i32>(a: int64x2_t) -> int64x2_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrshr, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(urshr, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrshr_n_u8<const N: i32>(a: uint8x8_t) -> uint8x8_t {
static_assert!(N : i32 where N >= 1 && N <= 8);
vrshl_u8(a, vdup_n_s8((-N) as _))
@@ -26317,7 +26317,7 @@ pub unsafe fn vrshr_n_u8<const N: i32>(a: uint8x8_t) -> uint8x8_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrshr, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(urshr, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrshrq_n_u8<const N: i32>(a: uint8x16_t) -> uint8x16_t {
static_assert!(N : i32 where N >= 1 && N <= 8);
vrshlq_u8(a, vdupq_n_s8((-N) as _))
@@ -26332,7 +26332,7 @@ pub unsafe fn vrshrq_n_u8<const N: i32>(a: uint8x16_t) -> uint8x16_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrshr, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(urshr, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrshr_n_u16<const N: i32>(a: uint16x4_t) -> uint16x4_t {
static_assert!(N : i32 where N >= 1 && N <= 16);
vrshl_u16(a, vdup_n_s16((-N) as _))
@@ -26347,7 +26347,7 @@ pub unsafe fn vrshr_n_u16<const N: i32>(a: uint16x4_t) -> uint16x4_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrshr, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(urshr, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrshrq_n_u16<const N: i32>(a: uint16x8_t) -> uint16x8_t {
static_assert!(N : i32 where N >= 1 && N <= 16);
vrshlq_u16(a, vdupq_n_s16((-N) as _))
@@ -26362,7 +26362,7 @@ pub unsafe fn vrshrq_n_u16<const N: i32>(a: uint16x8_t) -> uint16x8_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrshr, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(urshr, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrshr_n_u32<const N: i32>(a: uint32x2_t) -> uint32x2_t {
static_assert!(N : i32 where N >= 1 && N <= 32);
vrshl_u32(a, vdup_n_s32((-N) as _))
@@ -26377,7 +26377,7 @@ pub unsafe fn vrshr_n_u32<const N: i32>(a: uint32x2_t) -> uint32x2_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrshr, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(urshr, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrshrq_n_u32<const N: i32>(a: uint32x4_t) -> uint32x4_t {
static_assert!(N : i32 where N >= 1 && N <= 32);
vrshlq_u32(a, vdupq_n_s32((-N) as _))
@@ -26392,7 +26392,7 @@ pub unsafe fn vrshrq_n_u32<const N: i32>(a: uint32x4_t) -> uint32x4_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrshr, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(urshr, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrshr_n_u64<const N: i32>(a: uint64x1_t) -> uint64x1_t {
static_assert!(N : i32 where N >= 1 && N <= 64);
vrshl_u64(a, vdup_n_s64((-N) as _))
@@ -26407,7 +26407,7 @@ pub unsafe fn vrshr_n_u64<const N: i32>(a: uint64x1_t) -> uint64x1_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrshr, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(urshr, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrshrq_n_u64<const N: i32>(a: uint64x2_t) -> uint64x2_t {
static_assert!(N : i32 where N >= 1 && N <= 64);
vrshlq_u64(a, vdupq_n_s64((-N) as _))
@@ -26533,7 +26533,7 @@ vrshrn_n_s64_(a, N)
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrshrn, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(rshrn, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrshrn_n_u16<const N: i32>(a: uint16x8_t) -> uint8x8_t {
static_assert!(N : i32 where N >= 1 && N <= 8);
transmute(vrshrn_n_s16::<N>(transmute(a)))
@@ -26548,7 +26548,7 @@ pub unsafe fn vrshrn_n_u16<const N: i32>(a: uint16x8_t) -> uint8x8_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrshrn, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(rshrn, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrshrn_n_u32<const N: i32>(a: uint32x4_t) -> uint16x4_t {
static_assert!(N : i32 where N >= 1 && N <= 16);
transmute(vrshrn_n_s32::<N>(transmute(a)))
@@ -26563,7 +26563,7 @@ pub unsafe fn vrshrn_n_u32<const N: i32>(a: uint32x4_t) -> uint16x4_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrshrn, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(rshrn, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrshrn_n_u64<const N: i32>(a: uint64x2_t) -> uint32x2_t {
static_assert!(N : i32 where N >= 1 && N <= 32);
transmute(vrshrn_n_s64::<N>(transmute(a)))
@@ -26578,7 +26578,7 @@ pub unsafe fn vrshrn_n_u64<const N: i32>(a: uint64x2_t) -> uint32x2_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrsra, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(srsra, N = 2))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrsra_n_s8<const N: i32>(a: int8x8_t, b: int8x8_t) -> int8x8_t {
static_assert!(N : i32 where N >= 1 && N <= 8);
simd_add(a, vrshr_n_s8::<N>(b))
@@ -26593,7 +26593,7 @@ pub unsafe fn vrsra_n_s8<const N: i32>(a: int8x8_t, b: int8x8_t) -> int8x8_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrsra, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(srsra, N = 2))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrsraq_n_s8<const N: i32>(a: int8x16_t, b: int8x16_t) -> int8x16_t {
static_assert!(N : i32 where N >= 1 && N <= 8);
simd_add(a, vrshrq_n_s8::<N>(b))
@@ -26608,7 +26608,7 @@ pub unsafe fn vrsraq_n_s8<const N: i32>(a: int8x16_t, b: int8x16_t) -> int8x16_t
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrsra, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(srsra, N = 2))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrsra_n_s16<const N: i32>(a: int16x4_t, b: int16x4_t) -> int16x4_t {
static_assert!(N : i32 where N >= 1 && N <= 16);
simd_add(a, vrshr_n_s16::<N>(b))
@@ -26623,7 +26623,7 @@ pub unsafe fn vrsra_n_s16<const N: i32>(a: int16x4_t, b: int16x4_t) -> int16x4_t
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrsra, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(srsra, N = 2))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrsraq_n_s16<const N: i32>(a: int16x8_t, b: int16x8_t) -> int16x8_t {
static_assert!(N : i32 where N >= 1 && N <= 16);
simd_add(a, vrshrq_n_s16::<N>(b))
@@ -26638,7 +26638,7 @@ pub unsafe fn vrsraq_n_s16<const N: i32>(a: int16x8_t, b: int16x8_t) -> int16x8_
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrsra, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(srsra, N = 2))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrsra_n_s32<const N: i32>(a: int32x2_t, b: int32x2_t) -> int32x2_t {
static_assert!(N : i32 where N >= 1 && N <= 32);
simd_add(a, vrshr_n_s32::<N>(b))
@@ -26653,7 +26653,7 @@ pub unsafe fn vrsra_n_s32<const N: i32>(a: int32x2_t, b: int32x2_t) -> int32x2_t
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrsra, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(srsra, N = 2))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrsraq_n_s32<const N: i32>(a: int32x4_t, b: int32x4_t) -> int32x4_t {
static_assert!(N : i32 where N >= 1 && N <= 32);
simd_add(a, vrshrq_n_s32::<N>(b))
@@ -26668,7 +26668,7 @@ pub unsafe fn vrsraq_n_s32<const N: i32>(a: int32x4_t, b: int32x4_t) -> int32x4_
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrsra, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(srsra, N = 2))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrsra_n_s64<const N: i32>(a: int64x1_t, b: int64x1_t) -> int64x1_t {
static_assert!(N : i32 where N >= 1 && N <= 64);
simd_add(a, vrshr_n_s64::<N>(b))
@@ -26683,7 +26683,7 @@ pub unsafe fn vrsra_n_s64<const N: i32>(a: int64x1_t, b: int64x1_t) -> int64x1_t
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrsra, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(srsra, N = 2))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrsraq_n_s64<const N: i32>(a: int64x2_t, b: int64x2_t) -> int64x2_t {
static_assert!(N : i32 where N >= 1 && N <= 64);
simd_add(a, vrshrq_n_s64::<N>(b))
@@ -26698,7 +26698,7 @@ pub unsafe fn vrsraq_n_s64<const N: i32>(a: int64x2_t, b: int64x2_t) -> int64x2_
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrsra, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ursra, N = 2))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrsra_n_u8<const N: i32>(a: uint8x8_t, b: uint8x8_t) -> uint8x8_t {
static_assert!(N : i32 where N >= 1 && N <= 8);
simd_add(a, vrshr_n_u8::<N>(b))
@@ -26713,7 +26713,7 @@ pub unsafe fn vrsra_n_u8<const N: i32>(a: uint8x8_t, b: uint8x8_t) -> uint8x8_t
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrsra, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ursra, N = 2))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrsraq_n_u8<const N: i32>(a: uint8x16_t, b: uint8x16_t) -> uint8x16_t {
static_assert!(N : i32 where N >= 1 && N <= 8);
simd_add(a, vrshrq_n_u8::<N>(b))
@@ -26728,7 +26728,7 @@ pub unsafe fn vrsraq_n_u8<const N: i32>(a: uint8x16_t, b: uint8x16_t) -> uint8x1
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrsra, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ursra, N = 2))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrsra_n_u16<const N: i32>(a: uint16x4_t, b: uint16x4_t) -> uint16x4_t {
static_assert!(N : i32 where N >= 1 && N <= 16);
simd_add(a, vrshr_n_u16::<N>(b))
@@ -26743,7 +26743,7 @@ pub unsafe fn vrsra_n_u16<const N: i32>(a: uint16x4_t, b: uint16x4_t) -> uint16x
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrsra, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ursra, N = 2))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrsraq_n_u16<const N: i32>(a: uint16x8_t, b: uint16x8_t) -> uint16x8_t {
static_assert!(N : i32 where N >= 1 && N <= 16);
simd_add(a, vrshrq_n_u16::<N>(b))
@@ -26758,7 +26758,7 @@ pub unsafe fn vrsraq_n_u16<const N: i32>(a: uint16x8_t, b: uint16x8_t) -> uint16
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrsra, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ursra, N = 2))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrsra_n_u32<const N: i32>(a: uint32x2_t, b: uint32x2_t) -> uint32x2_t {
static_assert!(N : i32 where N >= 1 && N <= 32);
simd_add(a, vrshr_n_u32::<N>(b))
@@ -26773,7 +26773,7 @@ pub unsafe fn vrsra_n_u32<const N: i32>(a: uint32x2_t, b: uint32x2_t) -> uint32x
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrsra, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ursra, N = 2))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrsraq_n_u32<const N: i32>(a: uint32x4_t, b: uint32x4_t) -> uint32x4_t {
static_assert!(N : i32 where N >= 1 && N <= 32);
simd_add(a, vrshrq_n_u32::<N>(b))
@@ -26788,7 +26788,7 @@ pub unsafe fn vrsraq_n_u32<const N: i32>(a: uint32x4_t, b: uint32x4_t) -> uint32
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrsra, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ursra, N = 2))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrsra_n_u64<const N: i32>(a: uint64x1_t, b: uint64x1_t) -> uint64x1_t {
static_assert!(N : i32 where N >= 1 && N <= 64);
simd_add(a, vrshr_n_u64::<N>(b))
@@ -26803,7 +26803,7 @@ pub unsafe fn vrsra_n_u64<const N: i32>(a: uint64x1_t, b: uint64x1_t) -> uint64x
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrsra, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ursra, N = 2))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrsraq_n_u64<const N: i32>(a: uint64x2_t, b: uint64x2_t) -> uint64x2_t {
static_assert!(N : i32 where N >= 1 && N <= 64);
simd_add(a, vrshrq_n_u64::<N>(b))
@@ -26817,7 +26817,7 @@ pub unsafe fn vrsraq_n_u64<const N: i32>(a: uint64x2_t, b: uint64x2_t) -> uint64
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrsubhn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(rsubhn))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrsubhn_s16(a: int16x8_t, b: int16x8_t) -> int8x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -26836,7 +26836,7 @@ vrsubhn_s16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrsubhn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(rsubhn))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrsubhn_s32(a: int32x4_t, b: int32x4_t) -> int16x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -26855,7 +26855,7 @@ vrsubhn_s32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrsubhn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(rsubhn))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrsubhn_s64(a: int64x2_t, b: int64x2_t) -> int32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -26874,7 +26874,7 @@ vrsubhn_s64_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrsubhn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(rsubhn))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrsubhn_u16(a: uint16x8_t, b: uint16x8_t) -> uint8x8_t {
transmute(vrsubhn_s16(transmute(a), transmute(b)))
}
@@ -26887,7 +26887,7 @@ pub unsafe fn vrsubhn_u16(a: uint16x8_t, b: uint16x8_t) -> uint8x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrsubhn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(rsubhn))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrsubhn_u32(a: uint32x4_t, b: uint32x4_t) -> uint16x4_t {
transmute(vrsubhn_s32(transmute(a), transmute(b)))
}
@@ -26900,7 +26900,7 @@ pub unsafe fn vrsubhn_u32(a: uint32x4_t, b: uint32x4_t) -> uint16x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vrsubhn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(rsubhn))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vrsubhn_u64(a: uint64x2_t, b: uint64x2_t) -> uint32x2_t {
transmute(vrsubhn_s64(transmute(a), transmute(b)))
}
@@ -26914,7 +26914,7 @@ pub unsafe fn vrsubhn_u64(a: uint64x2_t, b: uint64x2_t) -> uint32x2_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vset_lane_s8<const LANE: i32>(a: i8, b: int8x8_t) -> int8x8_t {
static_assert_imm3!(LANE);
simd_insert(b, LANE as u32, a)
@@ -26929,7 +26929,7 @@ pub unsafe fn vset_lane_s8<const LANE: i32>(a: i8, b: int8x8_t) -> int8x8_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vset_lane_s16<const LANE: i32>(a: i16, b: int16x4_t) -> int16x4_t {
static_assert_imm2!(LANE);
simd_insert(b, LANE as u32, a)
@@ -26944,7 +26944,7 @@ pub unsafe fn vset_lane_s16<const LANE: i32>(a: i16, b: int16x4_t) -> int16x4_t
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vset_lane_s32<const LANE: i32>(a: i32, b: int32x2_t) -> int32x2_t {
static_assert_imm1!(LANE);
simd_insert(b, LANE as u32, a)
@@ -26959,7 +26959,7 @@ pub unsafe fn vset_lane_s32<const LANE: i32>(a: i32, b: int32x2_t) -> int32x2_t
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vset_lane_s64<const LANE: i32>(a: i64, b: int64x1_t) -> int64x1_t {
static_assert!(LANE : i32 where LANE == 0);
simd_insert(b, LANE as u32, a)
@@ -26974,7 +26974,7 @@ pub unsafe fn vset_lane_s64<const LANE: i32>(a: i64, b: int64x1_t) -> int64x1_t
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vset_lane_u8<const LANE: i32>(a: u8, b: uint8x8_t) -> uint8x8_t {
static_assert_imm3!(LANE);
simd_insert(b, LANE as u32, a)
@@ -26989,7 +26989,7 @@ pub unsafe fn vset_lane_u8<const LANE: i32>(a: u8, b: uint8x8_t) -> uint8x8_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vset_lane_u16<const LANE: i32>(a: u16, b: uint16x4_t) -> uint16x4_t {
static_assert_imm2!(LANE);
simd_insert(b, LANE as u32, a)
@@ -27004,7 +27004,7 @@ pub unsafe fn vset_lane_u16<const LANE: i32>(a: u16, b: uint16x4_t) -> uint16x4_
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vset_lane_u32<const LANE: i32>(a: u32, b: uint32x2_t) -> uint32x2_t {
static_assert_imm1!(LANE);
simd_insert(b, LANE as u32, a)
@@ -27019,7 +27019,7 @@ pub unsafe fn vset_lane_u32<const LANE: i32>(a: u32, b: uint32x2_t) -> uint32x2_
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vset_lane_u64<const LANE: i32>(a: u64, b: uint64x1_t) -> uint64x1_t {
static_assert!(LANE : i32 where LANE == 0);
simd_insert(b, LANE as u32, a)
@@ -27034,7 +27034,7 @@ pub unsafe fn vset_lane_u64<const LANE: i32>(a: u64, b: uint64x1_t) -> uint64x1_
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vset_lane_p8<const LANE: i32>(a: p8, b: poly8x8_t) -> poly8x8_t {
static_assert_imm3!(LANE);
simd_insert(b, LANE as u32, a)
@@ -27049,7 +27049,7 @@ pub unsafe fn vset_lane_p8<const LANE: i32>(a: p8, b: poly8x8_t) -> poly8x8_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vset_lane_p16<const LANE: i32>(a: p16, b: poly16x4_t) -> poly16x4_t {
static_assert_imm2!(LANE);
simd_insert(b, LANE as u32, a)
@@ -27064,7 +27064,7 @@ pub unsafe fn vset_lane_p16<const LANE: i32>(a: p16, b: poly16x4_t) -> poly16x4_
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vset_lane_p64<const LANE: i32>(a: p64, b: poly64x1_t) -> poly64x1_t {
static_assert!(LANE : i32 where LANE == 0);
simd_insert(b, LANE as u32, a)
@@ -27079,7 +27079,7 @@ pub unsafe fn vset_lane_p64<const LANE: i32>(a: p64, b: poly64x1_t) -> poly64x1_
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsetq_lane_s8<const LANE: i32>(a: i8, b: int8x16_t) -> int8x16_t {
static_assert_imm4!(LANE);
simd_insert(b, LANE as u32, a)
@@ -27094,7 +27094,7 @@ pub unsafe fn vsetq_lane_s8<const LANE: i32>(a: i8, b: int8x16_t) -> int8x16_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsetq_lane_s16<const LANE: i32>(a: i16, b: int16x8_t) -> int16x8_t {
static_assert_imm3!(LANE);
simd_insert(b, LANE as u32, a)
@@ -27109,7 +27109,7 @@ pub unsafe fn vsetq_lane_s16<const LANE: i32>(a: i16, b: int16x8_t) -> int16x8_t
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsetq_lane_s32<const LANE: i32>(a: i32, b: int32x4_t) -> int32x4_t {
static_assert_imm2!(LANE);
simd_insert(b, LANE as u32, a)
@@ -27124,7 +27124,7 @@ pub unsafe fn vsetq_lane_s32<const LANE: i32>(a: i32, b: int32x4_t) -> int32x4_t
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsetq_lane_s64<const LANE: i32>(a: i64, b: int64x2_t) -> int64x2_t {
static_assert_imm1!(LANE);
simd_insert(b, LANE as u32, a)
@@ -27139,7 +27139,7 @@ pub unsafe fn vsetq_lane_s64<const LANE: i32>(a: i64, b: int64x2_t) -> int64x2_t
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsetq_lane_u8<const LANE: i32>(a: u8, b: uint8x16_t) -> uint8x16_t {
static_assert_imm4!(LANE);
simd_insert(b, LANE as u32, a)
@@ -27154,7 +27154,7 @@ pub unsafe fn vsetq_lane_u8<const LANE: i32>(a: u8, b: uint8x16_t) -> uint8x16_t
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsetq_lane_u16<const LANE: i32>(a: u16, b: uint16x8_t) -> uint16x8_t {
static_assert_imm3!(LANE);
simd_insert(b, LANE as u32, a)
@@ -27169,7 +27169,7 @@ pub unsafe fn vsetq_lane_u16<const LANE: i32>(a: u16, b: uint16x8_t) -> uint16x8
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsetq_lane_u32<const LANE: i32>(a: u32, b: uint32x4_t) -> uint32x4_t {
static_assert_imm2!(LANE);
simd_insert(b, LANE as u32, a)
@@ -27184,7 +27184,7 @@ pub unsafe fn vsetq_lane_u32<const LANE: i32>(a: u32, b: uint32x4_t) -> uint32x4
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsetq_lane_u64<const LANE: i32>(a: u64, b: uint64x2_t) -> uint64x2_t {
static_assert_imm1!(LANE);
simd_insert(b, LANE as u32, a)
@@ -27199,7 +27199,7 @@ pub unsafe fn vsetq_lane_u64<const LANE: i32>(a: u64, b: uint64x2_t) -> uint64x2
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsetq_lane_p8<const LANE: i32>(a: p8, b: poly8x16_t) -> poly8x16_t {
static_assert_imm4!(LANE);
simd_insert(b, LANE as u32, a)
@@ -27214,7 +27214,7 @@ pub unsafe fn vsetq_lane_p8<const LANE: i32>(a: p8, b: poly8x16_t) -> poly8x16_t
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsetq_lane_p16<const LANE: i32>(a: p16, b: poly16x8_t) -> poly16x8_t {
static_assert_imm3!(LANE);
simd_insert(b, LANE as u32, a)
@@ -27229,7 +27229,7 @@ pub unsafe fn vsetq_lane_p16<const LANE: i32>(a: p16, b: poly16x8_t) -> poly16x8
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsetq_lane_p64<const LANE: i32>(a: p64, b: poly64x2_t) -> poly64x2_t {
static_assert_imm1!(LANE);
simd_insert(b, LANE as u32, a)
@@ -27244,7 +27244,7 @@ pub unsafe fn vsetq_lane_p64<const LANE: i32>(a: p64, b: poly64x2_t) -> poly64x2
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vset_lane_f32<const LANE: i32>(a: f32, b: float32x2_t) -> float32x2_t {
static_assert_imm1!(LANE);
simd_insert(b, LANE as u32, a)
@@ -27259,7 +27259,7 @@ pub unsafe fn vset_lane_f32<const LANE: i32>(a: f32, b: float32x2_t) -> float32x
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop, LANE = 0))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(nop, LANE = 0))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsetq_lane_f32<const LANE: i32>(a: f32, b: float32x4_t) -> float32x4_t {
static_assert_imm2!(LANE);
simd_insert(b, LANE as u32, a)
@@ -27273,7 +27273,7 @@ pub unsafe fn vsetq_lane_f32<const LANE: i32>(a: f32, b: float32x4_t) -> float32
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshl_s8(a: int8x8_t, b: int8x8_t) -> int8x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -27292,7 +27292,7 @@ vshl_s8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshlq_s8(a: int8x16_t, b: int8x16_t) -> int8x16_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -27311,7 +27311,7 @@ vshlq_s8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshl_s16(a: int16x4_t, b: int16x4_t) -> int16x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -27330,7 +27330,7 @@ vshl_s16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshlq_s16(a: int16x8_t, b: int16x8_t) -> int16x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -27349,7 +27349,7 @@ vshlq_s16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshl_s32(a: int32x2_t, b: int32x2_t) -> int32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -27368,7 +27368,7 @@ vshl_s32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshlq_s32(a: int32x4_t, b: int32x4_t) -> int32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -27387,7 +27387,7 @@ vshlq_s32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshl_s64(a: int64x1_t, b: int64x1_t) -> int64x1_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -27406,7 +27406,7 @@ vshl_s64_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sshl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshlq_s64(a: int64x2_t, b: int64x2_t) -> int64x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -27425,7 +27425,7 @@ vshlq_s64_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ushl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshl_u8(a: uint8x8_t, b: int8x8_t) -> uint8x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -27444,7 +27444,7 @@ vshl_u8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ushl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshlq_u8(a: uint8x16_t, b: int8x16_t) -> uint8x16_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -27463,7 +27463,7 @@ vshlq_u8_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ushl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshl_u16(a: uint16x4_t, b: int16x4_t) -> uint16x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -27482,7 +27482,7 @@ vshl_u16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ushl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshlq_u16(a: uint16x8_t, b: int16x8_t) -> uint16x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -27501,7 +27501,7 @@ vshlq_u16_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ushl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshl_u32(a: uint32x2_t, b: int32x2_t) -> uint32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -27520,7 +27520,7 @@ vshl_u32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ushl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshlq_u32(a: uint32x4_t, b: int32x4_t) -> uint32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -27539,7 +27539,7 @@ vshlq_u32_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ushl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshl_u64(a: uint64x1_t, b: int64x1_t) -> uint64x1_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -27558,7 +27558,7 @@ vshl_u64_(a, b)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vshl))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ushl))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshlq_u64(a: uint64x2_t, b: int64x2_t) -> uint64x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -27578,7 +27578,7 @@ vshlq_u64_(a, b)
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vshl, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(shl, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshl_n_s8<const N: i32>(a: int8x8_t) -> int8x8_t {
static_assert_imm3!(N);
simd_shl(a, vdup_n_s8(N as _))
@@ -27593,7 +27593,7 @@ pub unsafe fn vshl_n_s8<const N: i32>(a: int8x8_t) -> int8x8_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vshl, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(shl, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshlq_n_s8<const N: i32>(a: int8x16_t) -> int8x16_t {
static_assert_imm3!(N);
simd_shl(a, vdupq_n_s8(N as _))
@@ -27608,7 +27608,7 @@ pub unsafe fn vshlq_n_s8<const N: i32>(a: int8x16_t) -> int8x16_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vshl, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(shl, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshl_n_s16<const N: i32>(a: int16x4_t) -> int16x4_t {
static_assert_imm4!(N);
simd_shl(a, vdup_n_s16(N as _))
@@ -27623,7 +27623,7 @@ pub unsafe fn vshl_n_s16<const N: i32>(a: int16x4_t) -> int16x4_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vshl, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(shl, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshlq_n_s16<const N: i32>(a: int16x8_t) -> int16x8_t {
static_assert_imm4!(N);
simd_shl(a, vdupq_n_s16(N as _))
@@ -27638,7 +27638,7 @@ pub unsafe fn vshlq_n_s16<const N: i32>(a: int16x8_t) -> int16x8_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vshl, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(shl, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshl_n_s32<const N: i32>(a: int32x2_t) -> int32x2_t {
static_assert_imm5!(N);
simd_shl(a, vdup_n_s32(N as _))
@@ -27653,7 +27653,7 @@ pub unsafe fn vshl_n_s32<const N: i32>(a: int32x2_t) -> int32x2_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vshl, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(shl, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshlq_n_s32<const N: i32>(a: int32x4_t) -> int32x4_t {
static_assert_imm5!(N);
simd_shl(a, vdupq_n_s32(N as _))
@@ -27668,7 +27668,7 @@ pub unsafe fn vshlq_n_s32<const N: i32>(a: int32x4_t) -> int32x4_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vshl, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(shl, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshl_n_u8<const N: i32>(a: uint8x8_t) -> uint8x8_t {
static_assert_imm3!(N);
simd_shl(a, vdup_n_u8(N as _))
@@ -27683,7 +27683,7 @@ pub unsafe fn vshl_n_u8<const N: i32>(a: uint8x8_t) -> uint8x8_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vshl, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(shl, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshlq_n_u8<const N: i32>(a: uint8x16_t) -> uint8x16_t {
static_assert_imm3!(N);
simd_shl(a, vdupq_n_u8(N as _))
@@ -27698,7 +27698,7 @@ pub unsafe fn vshlq_n_u8<const N: i32>(a: uint8x16_t) -> uint8x16_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vshl, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(shl, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshl_n_u16<const N: i32>(a: uint16x4_t) -> uint16x4_t {
static_assert_imm4!(N);
simd_shl(a, vdup_n_u16(N as _))
@@ -27713,7 +27713,7 @@ pub unsafe fn vshl_n_u16<const N: i32>(a: uint16x4_t) -> uint16x4_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vshl, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(shl, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshlq_n_u16<const N: i32>(a: uint16x8_t) -> uint16x8_t {
static_assert_imm4!(N);
simd_shl(a, vdupq_n_u16(N as _))
@@ -27728,7 +27728,7 @@ pub unsafe fn vshlq_n_u16<const N: i32>(a: uint16x8_t) -> uint16x8_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vshl, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(shl, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshl_n_u32<const N: i32>(a: uint32x2_t) -> uint32x2_t {
static_assert_imm5!(N);
simd_shl(a, vdup_n_u32(N as _))
@@ -27743,7 +27743,7 @@ pub unsafe fn vshl_n_u32<const N: i32>(a: uint32x2_t) -> uint32x2_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vshl, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(shl, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshlq_n_u32<const N: i32>(a: uint32x4_t) -> uint32x4_t {
static_assert_imm5!(N);
simd_shl(a, vdupq_n_u32(N as _))
@@ -27758,7 +27758,7 @@ pub unsafe fn vshlq_n_u32<const N: i32>(a: uint32x4_t) -> uint32x4_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vshl, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(shl, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshl_n_s64<const N: i32>(a: int64x1_t) -> int64x1_t {
static_assert_imm6!(N);
simd_shl(a, vdup_n_s64(N as _))
@@ -27773,7 +27773,7 @@ pub unsafe fn vshl_n_s64<const N: i32>(a: int64x1_t) -> int64x1_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vshl, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(shl, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshlq_n_s64<const N: i32>(a: int64x2_t) -> int64x2_t {
static_assert_imm6!(N);
simd_shl(a, vdupq_n_s64(N as _))
@@ -27788,7 +27788,7 @@ pub unsafe fn vshlq_n_s64<const N: i32>(a: int64x2_t) -> int64x2_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vshl, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(shl, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshl_n_u64<const N: i32>(a: uint64x1_t) -> uint64x1_t {
static_assert_imm6!(N);
simd_shl(a, vdup_n_u64(N as _))
@@ -27803,7 +27803,7 @@ pub unsafe fn vshl_n_u64<const N: i32>(a: uint64x1_t) -> uint64x1_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vshl, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(shl, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshlq_n_u64<const N: i32>(a: uint64x2_t) -> uint64x2_t {
static_assert_imm6!(N);
simd_shl(a, vdupq_n_u64(N as _))
@@ -27818,7 +27818,7 @@ pub unsafe fn vshlq_n_u64<const N: i32>(a: uint64x2_t) -> uint64x2_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vshll.s8", N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sshll, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshll_n_s8<const N: i32>(a: int8x8_t) -> int16x8_t {
static_assert!(N : i32 where N >= 0 && N <= 8);
simd_shl(simd_cast(a), vdupq_n_s16(N as _))
@@ -27833,7 +27833,7 @@ pub unsafe fn vshll_n_s8<const N: i32>(a: int8x8_t) -> int16x8_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vshll.s16", N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sshll, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshll_n_s16<const N: i32>(a: int16x4_t) -> int32x4_t {
static_assert!(N : i32 where N >= 0 && N <= 16);
simd_shl(simd_cast(a), vdupq_n_s32(N as _))
@@ -27848,7 +27848,7 @@ pub unsafe fn vshll_n_s16<const N: i32>(a: int16x4_t) -> int32x4_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vshll.s32", N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sshll, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshll_n_s32<const N: i32>(a: int32x2_t) -> int64x2_t {
static_assert!(N : i32 where N >= 0 && N <= 32);
simd_shl(simd_cast(a), vdupq_n_s64(N as _))
@@ -27863,7 +27863,7 @@ pub unsafe fn vshll_n_s32<const N: i32>(a: int32x2_t) -> int64x2_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vshll.u8", N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ushll, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshll_n_u8<const N: i32>(a: uint8x8_t) -> uint16x8_t {
static_assert!(N : i32 where N >= 0 && N <= 8);
simd_shl(simd_cast(a), vdupq_n_u16(N as _))
@@ -27878,7 +27878,7 @@ pub unsafe fn vshll_n_u8<const N: i32>(a: uint8x8_t) -> uint16x8_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vshll.u16", N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ushll, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshll_n_u16<const N: i32>(a: uint16x4_t) -> uint32x4_t {
static_assert!(N : i32 where N >= 0 && N <= 16);
simd_shl(simd_cast(a), vdupq_n_u32(N as _))
@@ -27893,7 +27893,7 @@ pub unsafe fn vshll_n_u16<const N: i32>(a: uint16x4_t) -> uint32x4_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vshll.u32", N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ushll, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshll_n_u32<const N: i32>(a: uint32x2_t) -> uint64x2_t {
static_assert!(N : i32 where N >= 0 && N <= 32);
simd_shl(simd_cast(a), vdupq_n_u64(N as _))
@@ -27908,7 +27908,7 @@ pub unsafe fn vshll_n_u32<const N: i32>(a: uint32x2_t) -> uint64x2_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vshr.s8", N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sshr, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshr_n_s8<const N: i32>(a: int8x8_t) -> int8x8_t {
static_assert!(N : i32 where N >= 1 && N <= 8);
let n: i32 = if N == 8 { 7 } else { N };
@@ -27924,7 +27924,7 @@ pub unsafe fn vshr_n_s8<const N: i32>(a: int8x8_t) -> int8x8_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vshr.s8", N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sshr, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshrq_n_s8<const N: i32>(a: int8x16_t) -> int8x16_t {
static_assert!(N : i32 where N >= 1 && N <= 8);
let n: i32 = if N == 8 { 7 } else { N };
@@ -27940,7 +27940,7 @@ pub unsafe fn vshrq_n_s8<const N: i32>(a: int8x16_t) -> int8x16_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vshr.s16", N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sshr, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshr_n_s16<const N: i32>(a: int16x4_t) -> int16x4_t {
static_assert!(N : i32 where N >= 1 && N <= 16);
let n: i32 = if N == 16 { 15 } else { N };
@@ -27956,7 +27956,7 @@ pub unsafe fn vshr_n_s16<const N: i32>(a: int16x4_t) -> int16x4_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vshr.s16", N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sshr, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshrq_n_s16<const N: i32>(a: int16x8_t) -> int16x8_t {
static_assert!(N : i32 where N >= 1 && N <= 16);
let n: i32 = if N == 16 { 15 } else { N };
@@ -27972,7 +27972,7 @@ pub unsafe fn vshrq_n_s16<const N: i32>(a: int16x8_t) -> int16x8_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vshr.s32", N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sshr, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshr_n_s32<const N: i32>(a: int32x2_t) -> int32x2_t {
static_assert!(N : i32 where N >= 1 && N <= 32);
let n: i32 = if N == 32 { 31 } else { N };
@@ -27988,7 +27988,7 @@ pub unsafe fn vshr_n_s32<const N: i32>(a: int32x2_t) -> int32x2_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vshr.s32", N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sshr, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshrq_n_s32<const N: i32>(a: int32x4_t) -> int32x4_t {
static_assert!(N : i32 where N >= 1 && N <= 32);
let n: i32 = if N == 32 { 31 } else { N };
@@ -28004,7 +28004,7 @@ pub unsafe fn vshrq_n_s32<const N: i32>(a: int32x4_t) -> int32x4_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vshr.s64", N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sshr, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshr_n_s64<const N: i32>(a: int64x1_t) -> int64x1_t {
static_assert!(N : i32 where N >= 1 && N <= 64);
let n: i32 = if N == 64 { 63 } else { N };
@@ -28020,7 +28020,7 @@ pub unsafe fn vshr_n_s64<const N: i32>(a: int64x1_t) -> int64x1_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vshr.s64", N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sshr, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshrq_n_s64<const N: i32>(a: int64x2_t) -> int64x2_t {
static_assert!(N : i32 where N >= 1 && N <= 64);
let n: i32 = if N == 64 { 63 } else { N };
@@ -28036,7 +28036,7 @@ pub unsafe fn vshrq_n_s64<const N: i32>(a: int64x2_t) -> int64x2_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vshr.u8", N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ushr, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshr_n_u8<const N: i32>(a: uint8x8_t) -> uint8x8_t {
static_assert!(N : i32 where N >= 1 && N <= 8);
let n: i32 = if N == 8 { return vdup_n_u8(0); } else { N };
@@ -28052,7 +28052,7 @@ pub unsafe fn vshr_n_u8<const N: i32>(a: uint8x8_t) -> uint8x8_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vshr.u8", N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ushr, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshrq_n_u8<const N: i32>(a: uint8x16_t) -> uint8x16_t {
static_assert!(N : i32 where N >= 1 && N <= 8);
let n: i32 = if N == 8 { return vdupq_n_u8(0); } else { N };
@@ -28068,7 +28068,7 @@ pub unsafe fn vshrq_n_u8<const N: i32>(a: uint8x16_t) -> uint8x16_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vshr.u16", N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ushr, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshr_n_u16<const N: i32>(a: uint16x4_t) -> uint16x4_t {
static_assert!(N : i32 where N >= 1 && N <= 16);
let n: i32 = if N == 16 { return vdup_n_u16(0); } else { N };
@@ -28084,7 +28084,7 @@ pub unsafe fn vshr_n_u16<const N: i32>(a: uint16x4_t) -> uint16x4_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vshr.u16", N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ushr, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshrq_n_u16<const N: i32>(a: uint16x8_t) -> uint16x8_t {
static_assert!(N : i32 where N >= 1 && N <= 16);
let n: i32 = if N == 16 { return vdupq_n_u16(0); } else { N };
@@ -28100,7 +28100,7 @@ pub unsafe fn vshrq_n_u16<const N: i32>(a: uint16x8_t) -> uint16x8_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vshr.u32", N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ushr, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshr_n_u32<const N: i32>(a: uint32x2_t) -> uint32x2_t {
static_assert!(N : i32 where N >= 1 && N <= 32);
let n: i32 = if N == 32 { return vdup_n_u32(0); } else { N };
@@ -28116,7 +28116,7 @@ pub unsafe fn vshr_n_u32<const N: i32>(a: uint32x2_t) -> uint32x2_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vshr.u32", N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ushr, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshrq_n_u32<const N: i32>(a: uint32x4_t) -> uint32x4_t {
static_assert!(N : i32 where N >= 1 && N <= 32);
let n: i32 = if N == 32 { return vdupq_n_u32(0); } else { N };
@@ -28132,7 +28132,7 @@ pub unsafe fn vshrq_n_u32<const N: i32>(a: uint32x4_t) -> uint32x4_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vshr.u64", N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ushr, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshr_n_u64<const N: i32>(a: uint64x1_t) -> uint64x1_t {
static_assert!(N : i32 where N >= 1 && N <= 64);
let n: i32 = if N == 64 { return vdup_n_u64(0); } else { N };
@@ -28148,7 +28148,7 @@ pub unsafe fn vshr_n_u64<const N: i32>(a: uint64x1_t) -> uint64x1_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vshr.u64", N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ushr, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshrq_n_u64<const N: i32>(a: uint64x2_t) -> uint64x2_t {
static_assert!(N : i32 where N >= 1 && N <= 64);
let n: i32 = if N == 64 { return vdupq_n_u64(0); } else { N };
@@ -28164,7 +28164,7 @@ pub unsafe fn vshrq_n_u64<const N: i32>(a: uint64x2_t) -> uint64x2_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vshrn.i16", N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(shrn, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshrn_n_s16<const N: i32>(a: int16x8_t) -> int8x8_t {
static_assert!(N : i32 where N >= 1 && N <= 8);
simd_cast(simd_shr(a, vdupq_n_s16(N as _)))
@@ -28179,7 +28179,7 @@ pub unsafe fn vshrn_n_s16<const N: i32>(a: int16x8_t) -> int8x8_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vshrn.i32", N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(shrn, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshrn_n_s32<const N: i32>(a: int32x4_t) -> int16x4_t {
static_assert!(N : i32 where N >= 1 && N <= 16);
simd_cast(simd_shr(a, vdupq_n_s32(N as _)))
@@ -28194,7 +28194,7 @@ pub unsafe fn vshrn_n_s32<const N: i32>(a: int32x4_t) -> int16x4_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vshrn.i64", N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(shrn, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshrn_n_s64<const N: i32>(a: int64x2_t) -> int32x2_t {
static_assert!(N : i32 where N >= 1 && N <= 32);
simd_cast(simd_shr(a, vdupq_n_s64(N as _)))
@@ -28209,7 +28209,7 @@ pub unsafe fn vshrn_n_s64<const N: i32>(a: int64x2_t) -> int32x2_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vshrn.i16", N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(shrn, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshrn_n_u16<const N: i32>(a: uint16x8_t) -> uint8x8_t {
static_assert!(N : i32 where N >= 1 && N <= 8);
simd_cast(simd_shr(a, vdupq_n_u16(N as _)))
@@ -28224,7 +28224,7 @@ pub unsafe fn vshrn_n_u16<const N: i32>(a: uint16x8_t) -> uint8x8_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vshrn.i32", N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(shrn, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshrn_n_u32<const N: i32>(a: uint32x4_t) -> uint16x4_t {
static_assert!(N : i32 where N >= 1 && N <= 16);
simd_cast(simd_shr(a, vdupq_n_u32(N as _)))
@@ -28239,7 +28239,7 @@ pub unsafe fn vshrn_n_u32<const N: i32>(a: uint32x4_t) -> uint16x4_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vshrn.i64", N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(shrn, N = 2))]
#[rustc_legacy_const_generics(1)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vshrn_n_u64<const N: i32>(a: uint64x2_t) -> uint32x2_t {
static_assert!(N : i32 where N >= 1 && N <= 32);
simd_cast(simd_shr(a, vdupq_n_u64(N as _)))
@@ -28254,7 +28254,7 @@ pub unsafe fn vshrn_n_u64<const N: i32>(a: uint64x2_t) -> uint32x2_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vsra, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ssra, N = 2))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsra_n_s8<const N: i32>(a: int8x8_t, b: int8x8_t) -> int8x8_t {
static_assert!(N : i32 where N >= 1 && N <= 8);
simd_add(a, vshr_n_s8::<N>(b))
@@ -28269,7 +28269,7 @@ pub unsafe fn vsra_n_s8<const N: i32>(a: int8x8_t, b: int8x8_t) -> int8x8_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vsra, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ssra, N = 2))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsraq_n_s8<const N: i32>(a: int8x16_t, b: int8x16_t) -> int8x16_t {
static_assert!(N : i32 where N >= 1 && N <= 8);
simd_add(a, vshrq_n_s8::<N>(b))
@@ -28284,7 +28284,7 @@ pub unsafe fn vsraq_n_s8<const N: i32>(a: int8x16_t, b: int8x16_t) -> int8x16_t
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vsra, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ssra, N = 2))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsra_n_s16<const N: i32>(a: int16x4_t, b: int16x4_t) -> int16x4_t {
static_assert!(N : i32 where N >= 1 && N <= 16);
simd_add(a, vshr_n_s16::<N>(b))
@@ -28299,7 +28299,7 @@ pub unsafe fn vsra_n_s16<const N: i32>(a: int16x4_t, b: int16x4_t) -> int16x4_t
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vsra, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ssra, N = 2))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsraq_n_s16<const N: i32>(a: int16x8_t, b: int16x8_t) -> int16x8_t {
static_assert!(N : i32 where N >= 1 && N <= 16);
simd_add(a, vshrq_n_s16::<N>(b))
@@ -28314,7 +28314,7 @@ pub unsafe fn vsraq_n_s16<const N: i32>(a: int16x8_t, b: int16x8_t) -> int16x8_t
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vsra, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ssra, N = 2))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsra_n_s32<const N: i32>(a: int32x2_t, b: int32x2_t) -> int32x2_t {
static_assert!(N : i32 where N >= 1 && N <= 32);
simd_add(a, vshr_n_s32::<N>(b))
@@ -28329,7 +28329,7 @@ pub unsafe fn vsra_n_s32<const N: i32>(a: int32x2_t, b: int32x2_t) -> int32x2_t
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vsra, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ssra, N = 2))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsraq_n_s32<const N: i32>(a: int32x4_t, b: int32x4_t) -> int32x4_t {
static_assert!(N : i32 where N >= 1 && N <= 32);
simd_add(a, vshrq_n_s32::<N>(b))
@@ -28344,7 +28344,7 @@ pub unsafe fn vsraq_n_s32<const N: i32>(a: int32x4_t, b: int32x4_t) -> int32x4_t
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vsra, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ssra, N = 2))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsra_n_s64<const N: i32>(a: int64x1_t, b: int64x1_t) -> int64x1_t {
static_assert!(N : i32 where N >= 1 && N <= 64);
simd_add(a, vshr_n_s64::<N>(b))
@@ -28359,7 +28359,7 @@ pub unsafe fn vsra_n_s64<const N: i32>(a: int64x1_t, b: int64x1_t) -> int64x1_t
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vsra, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ssra, N = 2))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsraq_n_s64<const N: i32>(a: int64x2_t, b: int64x2_t) -> int64x2_t {
static_assert!(N : i32 where N >= 1 && N <= 64);
simd_add(a, vshrq_n_s64::<N>(b))
@@ -28374,7 +28374,7 @@ pub unsafe fn vsraq_n_s64<const N: i32>(a: int64x2_t, b: int64x2_t) -> int64x2_t
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vsra, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(usra, N = 2))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsra_n_u8<const N: i32>(a: uint8x8_t, b: uint8x8_t) -> uint8x8_t {
static_assert!(N : i32 where N >= 1 && N <= 8);
simd_add(a, vshr_n_u8::<N>(b))
@@ -28389,7 +28389,7 @@ pub unsafe fn vsra_n_u8<const N: i32>(a: uint8x8_t, b: uint8x8_t) -> uint8x8_t {
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vsra, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(usra, N = 2))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsraq_n_u8<const N: i32>(a: uint8x16_t, b: uint8x16_t) -> uint8x16_t {
static_assert!(N : i32 where N >= 1 && N <= 8);
simd_add(a, vshrq_n_u8::<N>(b))
@@ -28404,7 +28404,7 @@ pub unsafe fn vsraq_n_u8<const N: i32>(a: uint8x16_t, b: uint8x16_t) -> uint8x16
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vsra, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(usra, N = 2))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsra_n_u16<const N: i32>(a: uint16x4_t, b: uint16x4_t) -> uint16x4_t {
static_assert!(N : i32 where N >= 1 && N <= 16);
simd_add(a, vshr_n_u16::<N>(b))
@@ -28419,7 +28419,7 @@ pub unsafe fn vsra_n_u16<const N: i32>(a: uint16x4_t, b: uint16x4_t) -> uint16x4
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vsra, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(usra, N = 2))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsraq_n_u16<const N: i32>(a: uint16x8_t, b: uint16x8_t) -> uint16x8_t {
static_assert!(N : i32 where N >= 1 && N <= 16);
simd_add(a, vshrq_n_u16::<N>(b))
@@ -28434,7 +28434,7 @@ pub unsafe fn vsraq_n_u16<const N: i32>(a: uint16x8_t, b: uint16x8_t) -> uint16x
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vsra, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(usra, N = 2))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsra_n_u32<const N: i32>(a: uint32x2_t, b: uint32x2_t) -> uint32x2_t {
static_assert!(N : i32 where N >= 1 && N <= 32);
simd_add(a, vshr_n_u32::<N>(b))
@@ -28449,7 +28449,7 @@ pub unsafe fn vsra_n_u32<const N: i32>(a: uint32x2_t, b: uint32x2_t) -> uint32x2
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vsra, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(usra, N = 2))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsraq_n_u32<const N: i32>(a: uint32x4_t, b: uint32x4_t) -> uint32x4_t {
static_assert!(N : i32 where N >= 1 && N <= 32);
simd_add(a, vshrq_n_u32::<N>(b))
@@ -28464,7 +28464,7 @@ pub unsafe fn vsraq_n_u32<const N: i32>(a: uint32x4_t, b: uint32x4_t) -> uint32x
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vsra, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(usra, N = 2))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsra_n_u64<const N: i32>(a: uint64x1_t, b: uint64x1_t) -> uint64x1_t {
static_assert!(N : i32 where N >= 1 && N <= 64);
simd_add(a, vshr_n_u64::<N>(b))
@@ -28479,7 +28479,7 @@ pub unsafe fn vsra_n_u64<const N: i32>(a: uint64x1_t, b: uint64x1_t) -> uint64x1
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vsra, N = 2))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(usra, N = 2))]
#[rustc_legacy_const_generics(2)]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vsraq_n_u64<const N: i32>(a: uint64x2_t, b: uint64x2_t) -> uint64x2_t {
static_assert!(N : i32 where N >= 1 && N <= 64);
simd_add(a, vshrq_n_u64::<N>(b))
@@ -28493,7 +28493,7 @@ pub unsafe fn vsraq_n_u64<const N: i32>(a: uint64x2_t, b: uint64x2_t) -> uint64x
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vtrn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(trn))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vtrn_s8(a: int8x8_t, b: int8x8_t) -> int8x8x2_t {
let a1: int8x8_t = simd_shuffle8!(a, b, [0, 8, 2, 10, 4, 12, 6, 14]);
let b1: int8x8_t = simd_shuffle8!(a, b, [1, 9, 3, 11, 5, 13, 7, 15]);
@@ -28508,7 +28508,7 @@ pub unsafe fn vtrn_s8(a: int8x8_t, b: int8x8_t) -> int8x8x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vtrn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(trn))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vtrn_s16(a: int16x4_t, b: int16x4_t) -> int16x4x2_t {
let a1: int16x4_t = simd_shuffle4!(a, b, [0, 4, 2, 6]);
let b1: int16x4_t = simd_shuffle4!(a, b, [1, 5, 3, 7]);
@@ -28523,7 +28523,7 @@ pub unsafe fn vtrn_s16(a: int16x4_t, b: int16x4_t) -> int16x4x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vtrn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(trn))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vtrnq_s8(a: int8x16_t, b: int8x16_t) -> int8x16x2_t {
let a1: int8x16_t = simd_shuffle16!(a, b, [0, 16, 2, 18, 4, 20, 6, 22, 8, 24, 10, 26, 12, 28, 14, 30]);
let b1: int8x16_t = simd_shuffle16!(a, b, [1, 17, 3, 19, 5, 21, 7, 23, 9, 25, 11, 27, 13, 29, 15, 31]);
@@ -28538,7 +28538,7 @@ pub unsafe fn vtrnq_s8(a: int8x16_t, b: int8x16_t) -> int8x16x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vtrn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(trn))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vtrnq_s16(a: int16x8_t, b: int16x8_t) -> int16x8x2_t {
let a1: int16x8_t = simd_shuffle8!(a, b, [0, 8, 2, 10, 4, 12, 6, 14]);
let b1: int16x8_t = simd_shuffle8!(a, b, [1, 9, 3, 11, 5, 13, 7, 15]);
@@ -28553,7 +28553,7 @@ pub unsafe fn vtrnq_s16(a: int16x8_t, b: int16x8_t) -> int16x8x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vtrn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(trn))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vtrnq_s32(a: int32x4_t, b: int32x4_t) -> int32x4x2_t {
let a1: int32x4_t = simd_shuffle4!(a, b, [0, 4, 2, 6]);
let b1: int32x4_t = simd_shuffle4!(a, b, [1, 5, 3, 7]);
@@ -28568,7 +28568,7 @@ pub unsafe fn vtrnq_s32(a: int32x4_t, b: int32x4_t) -> int32x4x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vtrn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(trn))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vtrn_u8(a: uint8x8_t, b: uint8x8_t) -> uint8x8x2_t {
let a1: uint8x8_t = simd_shuffle8!(a, b, [0, 8, 2, 10, 4, 12, 6, 14]);
let b1: uint8x8_t = simd_shuffle8!(a, b, [1, 9, 3, 11, 5, 13, 7, 15]);
@@ -28583,7 +28583,7 @@ pub unsafe fn vtrn_u8(a: uint8x8_t, b: uint8x8_t) -> uint8x8x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vtrn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(trn))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vtrn_u16(a: uint16x4_t, b: uint16x4_t) -> uint16x4x2_t {
let a1: uint16x4_t = simd_shuffle4!(a, b, [0, 4, 2, 6]);
let b1: uint16x4_t = simd_shuffle4!(a, b, [1, 5, 3, 7]);
@@ -28598,7 +28598,7 @@ pub unsafe fn vtrn_u16(a: uint16x4_t, b: uint16x4_t) -> uint16x4x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vtrn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(trn))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vtrnq_u8(a: uint8x16_t, b: uint8x16_t) -> uint8x16x2_t {
let a1: uint8x16_t = simd_shuffle16!(a, b, [0, 16, 2, 18, 4, 20, 6, 22, 8, 24, 10, 26, 12, 28, 14, 30]);
let b1: uint8x16_t = simd_shuffle16!(a, b, [1, 17, 3, 19, 5, 21, 7, 23, 9, 25, 11, 27, 13, 29, 15, 31]);
@@ -28613,7 +28613,7 @@ pub unsafe fn vtrnq_u8(a: uint8x16_t, b: uint8x16_t) -> uint8x16x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vtrn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(trn))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vtrnq_u16(a: uint16x8_t, b: uint16x8_t) -> uint16x8x2_t {
let a1: uint16x8_t = simd_shuffle8!(a, b, [0, 8, 2, 10, 4, 12, 6, 14]);
let b1: uint16x8_t = simd_shuffle8!(a, b, [1, 9, 3, 11, 5, 13, 7, 15]);
@@ -28628,7 +28628,7 @@ pub unsafe fn vtrnq_u16(a: uint16x8_t, b: uint16x8_t) -> uint16x8x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vtrn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(trn))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vtrnq_u32(a: uint32x4_t, b: uint32x4_t) -> uint32x4x2_t {
let a1: uint32x4_t = simd_shuffle4!(a, b, [0, 4, 2, 6]);
let b1: uint32x4_t = simd_shuffle4!(a, b, [1, 5, 3, 7]);
@@ -28643,7 +28643,7 @@ pub unsafe fn vtrnq_u32(a: uint32x4_t, b: uint32x4_t) -> uint32x4x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vtrn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(trn))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vtrn_p8(a: poly8x8_t, b: poly8x8_t) -> poly8x8x2_t {
let a1: poly8x8_t = simd_shuffle8!(a, b, [0, 8, 2, 10, 4, 12, 6, 14]);
let b1: poly8x8_t = simd_shuffle8!(a, b, [1, 9, 3, 11, 5, 13, 7, 15]);
@@ -28658,7 +28658,7 @@ pub unsafe fn vtrn_p8(a: poly8x8_t, b: poly8x8_t) -> poly8x8x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vtrn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(trn))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vtrn_p16(a: poly16x4_t, b: poly16x4_t) -> poly16x4x2_t {
let a1: poly16x4_t = simd_shuffle4!(a, b, [0, 4, 2, 6]);
let b1: poly16x4_t = simd_shuffle4!(a, b, [1, 5, 3, 7]);
@@ -28673,7 +28673,7 @@ pub unsafe fn vtrn_p16(a: poly16x4_t, b: poly16x4_t) -> poly16x4x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vtrn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(trn))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vtrnq_p8(a: poly8x16_t, b: poly8x16_t) -> poly8x16x2_t {
let a1: poly8x16_t = simd_shuffle16!(a, b, [0, 16, 2, 18, 4, 20, 6, 22, 8, 24, 10, 26, 12, 28, 14, 30]);
let b1: poly8x16_t = simd_shuffle16!(a, b, [1, 17, 3, 19, 5, 21, 7, 23, 9, 25, 11, 27, 13, 29, 15, 31]);
@@ -28688,7 +28688,7 @@ pub unsafe fn vtrnq_p8(a: poly8x16_t, b: poly8x16_t) -> poly8x16x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vtrn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(trn))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vtrnq_p16(a: poly16x8_t, b: poly16x8_t) -> poly16x8x2_t {
let a1: poly16x8_t = simd_shuffle8!(a, b, [0, 8, 2, 10, 4, 12, 6, 14]);
let b1: poly16x8_t = simd_shuffle8!(a, b, [1, 9, 3, 11, 5, 13, 7, 15]);
@@ -28703,7 +28703,7 @@ pub unsafe fn vtrnq_p16(a: poly16x8_t, b: poly16x8_t) -> poly16x8x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vtrn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(zip))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vtrn_s32(a: int32x2_t, b: int32x2_t) -> int32x2x2_t {
let a1: int32x2_t = simd_shuffle2!(a, b, [0, 2]);
let b1: int32x2_t = simd_shuffle2!(a, b, [1, 3]);
@@ -28718,7 +28718,7 @@ pub unsafe fn vtrn_s32(a: int32x2_t, b: int32x2_t) -> int32x2x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vtrn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(zip))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vtrn_u32(a: uint32x2_t, b: uint32x2_t) -> uint32x2x2_t {
let a1: uint32x2_t = simd_shuffle2!(a, b, [0, 2]);
let b1: uint32x2_t = simd_shuffle2!(a, b, [1, 3]);
@@ -28733,7 +28733,7 @@ pub unsafe fn vtrn_u32(a: uint32x2_t, b: uint32x2_t) -> uint32x2x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vtrn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(zip))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vtrn_f32(a: float32x2_t, b: float32x2_t) -> float32x2x2_t {
let a1: float32x2_t = simd_shuffle2!(a, b, [0, 2]);
let b1: float32x2_t = simd_shuffle2!(a, b, [1, 3]);
@@ -28748,7 +28748,7 @@ pub unsafe fn vtrn_f32(a: float32x2_t, b: float32x2_t) -> float32x2x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vtrn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(trn))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vtrnq_f32(a: float32x4_t, b: float32x4_t) -> float32x4x2_t {
let a1: float32x4_t = simd_shuffle4!(a, b, [0, 4, 2, 6]);
let b1: float32x4_t = simd_shuffle4!(a, b, [1, 5, 3, 7]);
@@ -28763,7 +28763,7 @@ pub unsafe fn vtrnq_f32(a: float32x4_t, b: float32x4_t) -> float32x4x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vzip))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(zip))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vzip_s8(a: int8x8_t, b: int8x8_t) -> int8x8x2_t {
let a0: int8x8_t = simd_shuffle8!(a, b, [0, 8, 1, 9, 2, 10, 3, 11]);
let b0: int8x8_t = simd_shuffle8!(a, b, [4, 12, 5, 13, 6, 14, 7, 15]);
@@ -28778,7 +28778,7 @@ pub unsafe fn vzip_s8(a: int8x8_t, b: int8x8_t) -> int8x8x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vzip))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(zip))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vzip_s16(a: int16x4_t, b: int16x4_t) -> int16x4x2_t {
let a0: int16x4_t = simd_shuffle4!(a, b, [0, 4, 1, 5]);
let b0: int16x4_t = simd_shuffle4!(a, b, [2, 6, 3, 7]);
@@ -28793,7 +28793,7 @@ pub unsafe fn vzip_s16(a: int16x4_t, b: int16x4_t) -> int16x4x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vzip))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(zip))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vzip_u8(a: uint8x8_t, b: uint8x8_t) -> uint8x8x2_t {
let a0: uint8x8_t = simd_shuffle8!(a, b, [0, 8, 1, 9, 2, 10, 3, 11]);
let b0: uint8x8_t = simd_shuffle8!(a, b, [4, 12, 5, 13, 6, 14, 7, 15]);
@@ -28808,7 +28808,7 @@ pub unsafe fn vzip_u8(a: uint8x8_t, b: uint8x8_t) -> uint8x8x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vzip))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(zip))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vzip_u16(a: uint16x4_t, b: uint16x4_t) -> uint16x4x2_t {
let a0: uint16x4_t = simd_shuffle4!(a, b, [0, 4, 1, 5]);
let b0: uint16x4_t = simd_shuffle4!(a, b, [2, 6, 3, 7]);
@@ -28823,7 +28823,7 @@ pub unsafe fn vzip_u16(a: uint16x4_t, b: uint16x4_t) -> uint16x4x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vzip))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(zip))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vzip_p8(a: poly8x8_t, b: poly8x8_t) -> poly8x8x2_t {
let a0: poly8x8_t = simd_shuffle8!(a, b, [0, 8, 1, 9, 2, 10, 3, 11]);
let b0: poly8x8_t = simd_shuffle8!(a, b, [4, 12, 5, 13, 6, 14, 7, 15]);
@@ -28838,7 +28838,7 @@ pub unsafe fn vzip_p8(a: poly8x8_t, b: poly8x8_t) -> poly8x8x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vzip))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(zip))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vzip_p16(a: poly16x4_t, b: poly16x4_t) -> poly16x4x2_t {
let a0: poly16x4_t = simd_shuffle4!(a, b, [0, 4, 1, 5]);
let b0: poly16x4_t = simd_shuffle4!(a, b, [2, 6, 3, 7]);
@@ -28853,7 +28853,7 @@ pub unsafe fn vzip_p16(a: poly16x4_t, b: poly16x4_t) -> poly16x4x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vtrn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(zip))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vzip_s32(a: int32x2_t, b: int32x2_t) -> int32x2x2_t {
let a0: int32x2_t = simd_shuffle2!(a, b, [0, 2]);
let b0: int32x2_t = simd_shuffle2!(a, b, [1, 3]);
@@ -28868,7 +28868,7 @@ pub unsafe fn vzip_s32(a: int32x2_t, b: int32x2_t) -> int32x2x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vtrn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(zip))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vzip_u32(a: uint32x2_t, b: uint32x2_t) -> uint32x2x2_t {
let a0: uint32x2_t = simd_shuffle2!(a, b, [0, 2]);
let b0: uint32x2_t = simd_shuffle2!(a, b, [1, 3]);
@@ -28883,7 +28883,7 @@ pub unsafe fn vzip_u32(a: uint32x2_t, b: uint32x2_t) -> uint32x2x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vorr))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ext))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vzipq_s8(a: int8x16_t, b: int8x16_t) -> int8x16x2_t {
let a0: int8x16_t = simd_shuffle16!(a, b, [0, 16, 1, 17, 2, 18, 3, 19, 4, 20, 5, 21, 6, 22, 7, 23]);
let b0: int8x16_t = simd_shuffle16!(a, b, [8, 24, 9, 25, 10, 26, 11, 27, 12, 28, 13, 29, 14, 30, 15, 31]);
@@ -28898,7 +28898,7 @@ pub unsafe fn vzipq_s8(a: int8x16_t, b: int8x16_t) -> int8x16x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vorr))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ext))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vzipq_s16(a: int16x8_t, b: int16x8_t) -> int16x8x2_t {
let a0: int16x8_t = simd_shuffle8!(a, b, [0, 8, 1, 9, 2, 10, 3, 11]);
let b0: int16x8_t = simd_shuffle8!(a, b, [4, 12, 5, 13, 6, 14, 7, 15]);
@@ -28913,7 +28913,7 @@ pub unsafe fn vzipq_s16(a: int16x8_t, b: int16x8_t) -> int16x8x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vorr))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ext))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vzipq_s32(a: int32x4_t, b: int32x4_t) -> int32x4x2_t {
let a0: int32x4_t = simd_shuffle4!(a, b, [0, 4, 1, 5]);
let b0: int32x4_t = simd_shuffle4!(a, b, [2, 6, 3, 7]);
@@ -28928,7 +28928,7 @@ pub unsafe fn vzipq_s32(a: int32x4_t, b: int32x4_t) -> int32x4x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vorr))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ext))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vzipq_u8(a: uint8x16_t, b: uint8x16_t) -> uint8x16x2_t {
let a0: uint8x16_t = simd_shuffle16!(a, b, [0, 16, 1, 17, 2, 18, 3, 19, 4, 20, 5, 21, 6, 22, 7, 23]);
let b0: uint8x16_t = simd_shuffle16!(a, b, [8, 24, 9, 25, 10, 26, 11, 27, 12, 28, 13, 29, 14, 30, 15, 31]);
@@ -28943,7 +28943,7 @@ pub unsafe fn vzipq_u8(a: uint8x16_t, b: uint8x16_t) -> uint8x16x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vorr))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ext))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vzipq_u16(a: uint16x8_t, b: uint16x8_t) -> uint16x8x2_t {
let a0: uint16x8_t = simd_shuffle8!(a, b, [0, 8, 1, 9, 2, 10, 3, 11]);
let b0: uint16x8_t = simd_shuffle8!(a, b, [4, 12, 5, 13, 6, 14, 7, 15]);
@@ -28958,7 +28958,7 @@ pub unsafe fn vzipq_u16(a: uint16x8_t, b: uint16x8_t) -> uint16x8x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vorr))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ext))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vzipq_u32(a: uint32x4_t, b: uint32x4_t) -> uint32x4x2_t {
let a0: uint32x4_t = simd_shuffle4!(a, b, [0, 4, 1, 5]);
let b0: uint32x4_t = simd_shuffle4!(a, b, [2, 6, 3, 7]);
@@ -28973,7 +28973,7 @@ pub unsafe fn vzipq_u32(a: uint32x4_t, b: uint32x4_t) -> uint32x4x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vorr))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ext))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vzipq_p8(a: poly8x16_t, b: poly8x16_t) -> poly8x16x2_t {
let a0: poly8x16_t = simd_shuffle16!(a, b, [0, 16, 1, 17, 2, 18, 3, 19, 4, 20, 5, 21, 6, 22, 7, 23]);
let b0: poly8x16_t = simd_shuffle16!(a, b, [8, 24, 9, 25, 10, 26, 11, 27, 12, 28, 13, 29, 14, 30, 15, 31]);
@@ -28988,7 +28988,7 @@ pub unsafe fn vzipq_p8(a: poly8x16_t, b: poly8x16_t) -> poly8x16x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vorr))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ext))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vzipq_p16(a: poly16x8_t, b: poly16x8_t) -> poly16x8x2_t {
let a0: poly16x8_t = simd_shuffle8!(a, b, [0, 8, 1, 9, 2, 10, 3, 11]);
let b0: poly16x8_t = simd_shuffle8!(a, b, [4, 12, 5, 13, 6, 14, 7, 15]);
@@ -29003,7 +29003,7 @@ pub unsafe fn vzipq_p16(a: poly16x8_t, b: poly16x8_t) -> poly16x8x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vtrn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(zip))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vzip_f32(a: float32x2_t, b: float32x2_t) -> float32x2x2_t {
let a0: float32x2_t = simd_shuffle2!(a, b, [0, 2]);
let b0: float32x2_t = simd_shuffle2!(a, b, [1, 3]);
@@ -29018,7 +29018,7 @@ pub unsafe fn vzip_f32(a: float32x2_t, b: float32x2_t) -> float32x2x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vorr))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ext))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vzipq_f32(a: float32x4_t, b: float32x4_t) -> float32x4x2_t {
let a0: float32x4_t = simd_shuffle4!(a, b, [0, 4, 1, 5]);
let b0: float32x4_t = simd_shuffle4!(a, b, [2, 6, 3, 7]);
@@ -29033,7 +29033,7 @@ pub unsafe fn vzipq_f32(a: float32x4_t, b: float32x4_t) -> float32x4x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vuzp))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uzp))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vuzp_s8(a: int8x8_t, b: int8x8_t) -> int8x8x2_t {
let a0: int8x8_t = simd_shuffle8!(a, b, [0, 2, 4, 6, 8, 10, 12, 14]);
let b0: int8x8_t = simd_shuffle8!(a, b, [1, 3, 5, 7, 9, 11, 13, 15]);
@@ -29048,7 +29048,7 @@ pub unsafe fn vuzp_s8(a: int8x8_t, b: int8x8_t) -> int8x8x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vuzp))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uzp))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vuzp_s16(a: int16x4_t, b: int16x4_t) -> int16x4x2_t {
let a0: int16x4_t = simd_shuffle4!(a, b, [0, 2, 4, 6]);
let b0: int16x4_t = simd_shuffle4!(a, b, [1, 3, 5, 7]);
@@ -29063,7 +29063,7 @@ pub unsafe fn vuzp_s16(a: int16x4_t, b: int16x4_t) -> int16x4x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vuzp))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uzp))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vuzpq_s8(a: int8x16_t, b: int8x16_t) -> int8x16x2_t {
let a0: int8x16_t = simd_shuffle16!(a, b, [0, 2, 4, 6, 8, 10, 12, 14, 16, 18, 20, 22, 24, 26, 28, 30]);
let b0: int8x16_t = simd_shuffle16!(a, b, [1, 3, 5, 7, 9, 11, 13, 15, 17, 19, 21, 23, 25, 27, 29, 31]);
@@ -29078,7 +29078,7 @@ pub unsafe fn vuzpq_s8(a: int8x16_t, b: int8x16_t) -> int8x16x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vuzp))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uzp))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vuzpq_s16(a: int16x8_t, b: int16x8_t) -> int16x8x2_t {
let a0: int16x8_t = simd_shuffle8!(a, b, [0, 2, 4, 6, 8, 10, 12, 14]);
let b0: int16x8_t = simd_shuffle8!(a, b, [1, 3, 5, 7, 9, 11, 13, 15]);
@@ -29093,7 +29093,7 @@ pub unsafe fn vuzpq_s16(a: int16x8_t, b: int16x8_t) -> int16x8x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vuzp))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uzp))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vuzpq_s32(a: int32x4_t, b: int32x4_t) -> int32x4x2_t {
let a0: int32x4_t = simd_shuffle4!(a, b, [0, 2, 4, 6]);
let b0: int32x4_t = simd_shuffle4!(a, b, [1, 3, 5, 7]);
@@ -29108,7 +29108,7 @@ pub unsafe fn vuzpq_s32(a: int32x4_t, b: int32x4_t) -> int32x4x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vuzp))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uzp))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vuzp_u8(a: uint8x8_t, b: uint8x8_t) -> uint8x8x2_t {
let a0: uint8x8_t = simd_shuffle8!(a, b, [0, 2, 4, 6, 8, 10, 12, 14]);
let b0: uint8x8_t = simd_shuffle8!(a, b, [1, 3, 5, 7, 9, 11, 13, 15]);
@@ -29123,7 +29123,7 @@ pub unsafe fn vuzp_u8(a: uint8x8_t, b: uint8x8_t) -> uint8x8x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vuzp))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uzp))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vuzp_u16(a: uint16x4_t, b: uint16x4_t) -> uint16x4x2_t {
let a0: uint16x4_t = simd_shuffle4!(a, b, [0, 2, 4, 6]);
let b0: uint16x4_t = simd_shuffle4!(a, b, [1, 3, 5, 7]);
@@ -29138,7 +29138,7 @@ pub unsafe fn vuzp_u16(a: uint16x4_t, b: uint16x4_t) -> uint16x4x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vuzp))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uzp))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vuzpq_u8(a: uint8x16_t, b: uint8x16_t) -> uint8x16x2_t {
let a0: uint8x16_t = simd_shuffle16!(a, b, [0, 2, 4, 6, 8, 10, 12, 14, 16, 18, 20, 22, 24, 26, 28, 30]);
let b0: uint8x16_t = simd_shuffle16!(a, b, [1, 3, 5, 7, 9, 11, 13, 15, 17, 19, 21, 23, 25, 27, 29, 31]);
@@ -29153,7 +29153,7 @@ pub unsafe fn vuzpq_u8(a: uint8x16_t, b: uint8x16_t) -> uint8x16x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vuzp))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uzp))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vuzpq_u16(a: uint16x8_t, b: uint16x8_t) -> uint16x8x2_t {
let a0: uint16x8_t = simd_shuffle8!(a, b, [0, 2, 4, 6, 8, 10, 12, 14]);
let b0: uint16x8_t = simd_shuffle8!(a, b, [1, 3, 5, 7, 9, 11, 13, 15]);
@@ -29168,7 +29168,7 @@ pub unsafe fn vuzpq_u16(a: uint16x8_t, b: uint16x8_t) -> uint16x8x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vuzp))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uzp))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vuzpq_u32(a: uint32x4_t, b: uint32x4_t) -> uint32x4x2_t {
let a0: uint32x4_t = simd_shuffle4!(a, b, [0, 2, 4, 6]);
let b0: uint32x4_t = simd_shuffle4!(a, b, [1, 3, 5, 7]);
@@ -29183,7 +29183,7 @@ pub unsafe fn vuzpq_u32(a: uint32x4_t, b: uint32x4_t) -> uint32x4x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vuzp))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uzp))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vuzp_p8(a: poly8x8_t, b: poly8x8_t) -> poly8x8x2_t {
let a0: poly8x8_t = simd_shuffle8!(a, b, [0, 2, 4, 6, 8, 10, 12, 14]);
let b0: poly8x8_t = simd_shuffle8!(a, b, [1, 3, 5, 7, 9, 11, 13, 15]);
@@ -29198,7 +29198,7 @@ pub unsafe fn vuzp_p8(a: poly8x8_t, b: poly8x8_t) -> poly8x8x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vuzp))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uzp))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vuzp_p16(a: poly16x4_t, b: poly16x4_t) -> poly16x4x2_t {
let a0: poly16x4_t = simd_shuffle4!(a, b, [0, 2, 4, 6]);
let b0: poly16x4_t = simd_shuffle4!(a, b, [1, 3, 5, 7]);
@@ -29213,7 +29213,7 @@ pub unsafe fn vuzp_p16(a: poly16x4_t, b: poly16x4_t) -> poly16x4x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vuzp))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uzp))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vuzpq_p8(a: poly8x16_t, b: poly8x16_t) -> poly8x16x2_t {
let a0: poly8x16_t = simd_shuffle16!(a, b, [0, 2, 4, 6, 8, 10, 12, 14, 16, 18, 20, 22, 24, 26, 28, 30]);
let b0: poly8x16_t = simd_shuffle16!(a, b, [1, 3, 5, 7, 9, 11, 13, 15, 17, 19, 21, 23, 25, 27, 29, 31]);
@@ -29228,7 +29228,7 @@ pub unsafe fn vuzpq_p8(a: poly8x16_t, b: poly8x16_t) -> poly8x16x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vuzp))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uzp))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vuzpq_p16(a: poly16x8_t, b: poly16x8_t) -> poly16x8x2_t {
let a0: poly16x8_t = simd_shuffle8!(a, b, [0, 2, 4, 6, 8, 10, 12, 14]);
let b0: poly16x8_t = simd_shuffle8!(a, b, [1, 3, 5, 7, 9, 11, 13, 15]);
@@ -29243,7 +29243,7 @@ pub unsafe fn vuzpq_p16(a: poly16x8_t, b: poly16x8_t) -> poly16x8x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vtrn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(zip))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vuzp_s32(a: int32x2_t, b: int32x2_t) -> int32x2x2_t {
let a0: int32x2_t = simd_shuffle2!(a, b, [0, 2]);
let b0: int32x2_t = simd_shuffle2!(a, b, [1, 3]);
@@ -29258,7 +29258,7 @@ pub unsafe fn vuzp_s32(a: int32x2_t, b: int32x2_t) -> int32x2x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vtrn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(zip))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vuzp_u32(a: uint32x2_t, b: uint32x2_t) -> uint32x2x2_t {
let a0: uint32x2_t = simd_shuffle2!(a, b, [0, 2]);
let b0: uint32x2_t = simd_shuffle2!(a, b, [1, 3]);
@@ -29273,7 +29273,7 @@ pub unsafe fn vuzp_u32(a: uint32x2_t, b: uint32x2_t) -> uint32x2x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vtrn))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(zip))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vuzp_f32(a: float32x2_t, b: float32x2_t) -> float32x2x2_t {
let a0: float32x2_t = simd_shuffle2!(a, b, [0, 2]);
let b0: float32x2_t = simd_shuffle2!(a, b, [1, 3]);
@@ -29288,7 +29288,7 @@ pub unsafe fn vuzp_f32(a: float32x2_t, b: float32x2_t) -> float32x2x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vuzp))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uzp))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vuzpq_f32(a: float32x4_t, b: float32x4_t) -> float32x4x2_t {
let a0: float32x4_t = simd_shuffle4!(a, b, [0, 2, 4, 6]);
let b0: float32x4_t = simd_shuffle4!(a, b, [1, 3, 5, 7]);
@@ -29303,7 +29303,7 @@ pub unsafe fn vuzpq_f32(a: float32x4_t, b: float32x4_t) -> float32x4x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vabal.u8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uabal))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vabal_u8(a: uint16x8_t, b: uint8x8_t, c: uint8x8_t) -> uint16x8_t {
let d: uint8x8_t = vabd_u8(b, c);
simd_add(a, simd_cast(d))
@@ -29317,7 +29317,7 @@ pub unsafe fn vabal_u8(a: uint16x8_t, b: uint8x8_t, c: uint8x8_t) -> uint16x8_t
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vabal.u16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uabal))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vabal_u16(a: uint32x4_t, b: uint16x4_t, c: uint16x4_t) -> uint32x4_t {
let d: uint16x4_t = vabd_u16(b, c);
simd_add(a, simd_cast(d))
@@ -29331,7 +29331,7 @@ pub unsafe fn vabal_u16(a: uint32x4_t, b: uint16x4_t, c: uint16x4_t) -> uint32x4
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vabal.u32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(uabal))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vabal_u32(a: uint64x2_t, b: uint32x2_t, c: uint32x2_t) -> uint64x2_t {
let d: uint32x2_t = vabd_u32(b, c);
simd_add(a, simd_cast(d))
@@ -29345,7 +29345,7 @@ pub unsafe fn vabal_u32(a: uint64x2_t, b: uint32x2_t, c: uint32x2_t) -> uint64x2
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vabal.s8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sabal))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vabal_s8(a: int16x8_t, b: int8x8_t, c: int8x8_t) -> int16x8_t {
let d: int8x8_t = vabd_s8(b, c);
let e: uint8x8_t = simd_cast(d);
@@ -29360,7 +29360,7 @@ pub unsafe fn vabal_s8(a: int16x8_t, b: int8x8_t, c: int8x8_t) -> int16x8_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vabal.s16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sabal))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vabal_s16(a: int32x4_t, b: int16x4_t, c: int16x4_t) -> int32x4_t {
let d: int16x4_t = vabd_s16(b, c);
let e: uint16x4_t = simd_cast(d);
@@ -29375,7 +29375,7 @@ pub unsafe fn vabal_s16(a: int32x4_t, b: int16x4_t, c: int16x4_t) -> int32x4_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vabal.s32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sabal))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vabal_s32(a: int64x2_t, b: int32x2_t, c: int32x2_t) -> int64x2_t {
let d: int32x2_t = vabd_s32(b, c);
let e: uint32x2_t = simd_cast(d);
@@ -29390,7 +29390,7 @@ pub unsafe fn vabal_s32(a: int64x2_t, b: int32x2_t, c: int32x2_t) -> int64x2_t {
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vqabs.s8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqabs))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqabs_s8(a: int8x8_t) -> int8x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -29409,7 +29409,7 @@ vqabs_s8_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vqabs.s8"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqabs))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqabsq_s8(a: int8x16_t) -> int8x16_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -29428,7 +29428,7 @@ vqabsq_s8_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vqabs.s16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqabs))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqabs_s16(a: int16x4_t) -> int16x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -29447,7 +29447,7 @@ vqabs_s16_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vqabs.s16"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqabs))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqabsq_s16(a: int16x8_t) -> int16x8_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -29466,7 +29466,7 @@ vqabsq_s16_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vqabs.s32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqabs))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqabs_s32(a: int32x2_t) -> int32x2_t {
#[allow(improper_ctypes)]
extern "unadjusted" {
@@ -29485,7 +29485,7 @@ vqabs_s32_(a)
#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vqabs.s32"))]
#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(sqabs))]
-#[cfg_attr(target_arch = "aarch64", stable(feature = "neon_intrinsics", since = "1.59.0"))]
+#[cfg_attr(not(target_arch = "arm"), stable(feature = "neon_intrinsics", since = "1.59.0"))]
pub unsafe fn vqabsq_s32(a: int32x4_t) -> int32x4_t {
#[allow(improper_ctypes)]
extern "unadjusted" {