1 files changed, 230 insertions, 0 deletions
diff --git a/gfx/qcms/src/transform_avx.rs b/gfx/qcms/src/transform_avx.rs
new file mode 100644
index 0000000000..b34fc869d5
--- /dev/null
+++ b/gfx/qcms/src/transform_avx.rs
@@ -0,0 +1,230 @@
+use crate::transform::{qcms_transform, Format, BGRA, CLAMPMAXVAL, FLOATSCALE, RGB, RGBA};
+#[cfg(target_arch = "x86")]
+pub use std::arch::x86::{
+    __m128, __m128i, __m256, __m256i, _mm256_add_ps, _mm256_broadcast_ps, _mm256_castps128_ps256,
+    _mm256_castps256_ps128, _mm256_cvtps_epi32, _mm256_insertf128_ps, _mm256_max_ps, _mm256_min_ps,
+    _mm256_mul_ps, _mm256_set1_ps, _mm256_set_ps, _mm256_setzero_ps, _mm256_store_si256,
+    _mm_add_ps, _mm_broadcast_ss, _mm_cvtps_epi32, _mm_loadu_ps, _mm_max_ps, _mm_min_ps,
+    _mm_mul_ps, _mm_setzero_ps, _mm_store_si128,
+};
+#[cfg(target_arch = "x86_64")]
+pub use std::arch::x86_64::{
+    __m128, __m128i, __m256, __m256i, _mm256_add_ps, _mm256_broadcast_ps, _mm256_castps128_ps256,
+    _mm256_castps256_ps128, _mm256_cvtps_epi32, _mm256_insertf128_ps, _mm256_max_ps, _mm256_min_ps,
+    _mm256_mul_ps, _mm256_set1_ps, _mm256_set_ps, _mm256_setzero_ps, _mm256_store_si256,
+    _mm_add_ps, _mm_broadcast_ss, _mm_cvtps_epi32, _mm_loadu_ps, _mm_max_ps, _mm_min_ps,
+    _mm_mul_ps, _mm_setzero_ps, _mm_store_si128,
+};
+
+#[repr(align(32))]
+struct Output([u32; 8]);
+
+#[target_feature(enable = "avx")]
+unsafe extern "C" fn qcms_transform_data_template_lut_avx<F: Format>(
+    transform: &qcms_transform,
+    mut src: *const u8,
+    mut dest: *mut u8,
+    mut length: usize,
+) {
+    let mat: *const [f32; 4] = (*transform).matrix.as_ptr();
+    let mut input: Output = std::mem::zeroed();
+    /* share input and output locations to save having to keep the
+     * locations in separate registers */
+    let output: *const u32 = &mut input as *mut Output as *mut u32;
+    /* deref *transform now to avoid it in loop */
+    let igtbl_r: *const f32 = (*transform).input_gamma_table_r.as_ref().unwrap().as_ptr();
+    let igtbl_g: *const f32 = (*transform).input_gamma_table_g.as_ref().unwrap().as_ptr();
+    let igtbl_b: *const f32 = (*transform).input_gamma_table_b.as_ref().unwrap().as_ptr();
+    /* deref *transform now to avoid it in loop */
+    let otdata_r: *const u8 = (*transform)
+        .output_table_r
+        .as_deref()
+        .unwrap()
+        .data
+        .as_ptr();
+    let otdata_g: *const u8 = (*transform)
+        .output_table_g
+        .as_deref()
+        .unwrap()
+        .data
+        .as_ptr();
+    let otdata_b: *const u8 = (*transform)
+        .output_table_b
+        .as_deref()
+        .unwrap()
+        .data
+        .as_ptr();
+    /* input matrix values never change */
+    let mat0: __m256 = _mm256_broadcast_ps(&*((*mat.offset(0isize)).as_ptr() as *const __m128));
+    let mat1: __m256 = _mm256_broadcast_ps(&*((*mat.offset(1isize)).as_ptr() as *const __m128));
+    let mat2: __m256 = _mm256_broadcast_ps(&*((*mat.offset(2isize)).as_ptr() as *const __m128));
+    /* these values don't change, either */
+    let max: __m256 = _mm256_set1_ps(CLAMPMAXVAL);
+    let min: __m256 = _mm256_setzero_ps();
+    let scale: __m256 = _mm256_set1_ps(FLOATSCALE);
+    let components: u32 = if F::kAIndex == 0xff { 3 } else { 4 } as u32;
+    /* working variables */
+    let mut vec_r: __m256 = _mm256_setzero_ps();
+    let mut vec_g: __m256 = _mm256_setzero_ps();
+    let mut vec_b: __m256 = _mm256_setzero_ps();
+    let mut result: __m256;
+    let mut vec_r0: __m128;
+    let mut vec_g0: __m128;
+    let mut vec_b0: __m128;
+    let mut vec_r1: __m128;
+    let mut vec_g1: __m128;
+    let mut vec_b1: __m128;
+    let mut alpha1: u8 = 0;
+    let mut alpha2: u8 = 0;
+    /* CYA */
+    if length == 0 {
+        return;
+    }
+    /* If there are at least 2 pixels, then we can load their components into
+    a single 256-bit register for processing. */
+    if length > 1 {
+        vec_r0 = _mm_broadcast_ss(&*igtbl_r.offset(*src.add(F::kRIndex) as isize));
+        vec_g0 = _mm_broadcast_ss(&*igtbl_g.offset(*src.add(F::kGIndex) as isize));
+        vec_b0 = _mm_broadcast_ss(&*igtbl_b.offset(*src.add(F::kBIndex) as isize));
+        vec_r1 =
+            _mm_broadcast_ss(&*igtbl_r.offset(*src.add(F::kRIndex + components as usize) as isize));
+        vec_g1 =
+            _mm_broadcast_ss(&*igtbl_g.offset(*src.add(F::kGIndex + components as usize) as isize));
+        vec_b1 =
+            _mm_broadcast_ss(&*igtbl_b.offset(*src.add(F::kBIndex + components as usize) as isize));
+        vec_r = _mm256_insertf128_ps(_mm256_castps128_ps256(vec_r0), vec_r1, 1);
+        vec_g = _mm256_insertf128_ps(_mm256_castps128_ps256(vec_g0), vec_g1, 1);
+        vec_b = _mm256_insertf128_ps(_mm256_castps128_ps256(vec_b0), vec_b1, 1);
+        if F::kAIndex != 0xff {
+            alpha1 = *src.add(F::kAIndex);
+            alpha2 = *src.add(F::kAIndex + components as usize)
+        }
+    }
+    /* If there are at least 4 pixels, then we can iterate and preload the
+    next 2 while we store the result of the current 2. */
+    while length > 3 {
+        /* Ensure we are pointing at the next 2 pixels for the next load. */
+        src = src.offset((2 * components) as isize);
+        /* gamma * matrix */
+        vec_r = _mm256_mul_ps(vec_r, mat0);
+        vec_g = _mm256_mul_ps(vec_g, mat1);
+        vec_b = _mm256_mul_ps(vec_b, mat2);
+        /* store alpha for these pixels; load alpha for next two */
+        if F::kAIndex != 0xff {
+            *dest.add(F::kAIndex) = alpha1;
+            *dest.add(F::kAIndex + components as usize) = alpha2;
+            alpha1 = *src.add(F::kAIndex);
+            alpha2 = *src.add(F::kAIndex + components as usize)
+        }
+        /* crunch, crunch, crunch */
+        vec_r = _mm256_add_ps(vec_r, _mm256_add_ps(vec_g, vec_b));
+        vec_r = _mm256_max_ps(min, vec_r);
+        vec_r = _mm256_min_ps(max, vec_r);
+        result = _mm256_mul_ps(vec_r, scale);
+        /* store calc'd output tables indices */
+        _mm256_store_si256(output as *mut __m256i, _mm256_cvtps_epi32(result));
+        /* load gamma values for next loop while store completes */
+        vec_r0 = _mm_broadcast_ss(&*igtbl_r.offset(*src.add(F::kRIndex) as isize));
+        vec_g0 = _mm_broadcast_ss(&*igtbl_g.offset(*src.add(F::kGIndex) as isize));
+        vec_b0 = _mm_broadcast_ss(&*igtbl_b.offset(*src.add(F::kBIndex) as isize));
+        vec_r1 =
+            _mm_broadcast_ss(&*igtbl_r.offset(*src.add(F::kRIndex + components as usize) as isize));
+        vec_g1 =
+            _mm_broadcast_ss(&*igtbl_g.offset(*src.add(F::kGIndex + components as usize) as isize));
+        vec_b1 =
+            _mm_broadcast_ss(&*igtbl_b.offset(*src.add(F::kBIndex + components as usize) as isize));
+        vec_r = _mm256_insertf128_ps(_mm256_castps128_ps256(vec_r0), vec_r1, 1);
+        vec_g = _mm256_insertf128_ps(_mm256_castps128_ps256(vec_g0), vec_g1, 1);
+        vec_b = _mm256_insertf128_ps(_mm256_castps128_ps256(vec_b0), vec_b1, 1);
+        /* use calc'd indices to output RGB values */
+        *dest.add(F::kRIndex) = *otdata_r.offset(*output.offset(0isize) as isize);
+        *dest.add(F::kGIndex) = *otdata_g.offset(*output.offset(1isize) as isize);
+        *dest.add(F::kBIndex) = *otdata_b.offset(*output.offset(2isize) as isize);
+        *dest.add(F::kRIndex + components as usize) =
+            *otdata_r.offset(*output.offset(4isize) as isize);
+        *dest.add(F::kGIndex + components as usize) =
+            *otdata_g.offset(*output.offset(5isize) as isize);
+        *dest.add(F::kBIndex + components as usize) =
+            *otdata_b.offset(*output.offset(6isize) as isize);
+        dest = dest.offset((2 * components) as isize);
+        length -= 2
+    }
+    /* There are 0-3 pixels remaining. If there are 2-3 remaining, then we know
+    we have already populated the necessary registers to start the transform. */
+    if length > 1 {
+        vec_r = _mm256_mul_ps(vec_r, mat0);
+        vec_g = _mm256_mul_ps(vec_g, mat1);
+        vec_b = _mm256_mul_ps(vec_b, mat2);
+        if F::kAIndex != 0xff {
+            *dest.add(F::kAIndex) = alpha1;
+            *dest.add(F::kAIndex + components as usize) = alpha2
+        }
+        vec_r = _mm256_add_ps(vec_r, _mm256_add_ps(vec_g, vec_b));
+        vec_r = _mm256_max_ps(min, vec_r);
+        vec_r = _mm256_min_ps(max, vec_r);
+        result = _mm256_mul_ps(vec_r, scale);
+        _mm256_store_si256(output as *mut __m256i, _mm256_cvtps_epi32(result));
+        *dest.add(F::kRIndex) = *otdata_r.offset(*output.offset(0isize) as isize);
+        *dest.add(F::kGIndex) = *otdata_g.offset(*output.offset(1isize) as isize);
+        *dest.add(F::kBIndex) = *otdata_b.offset(*output.offset(2isize) as isize);
+        *dest.add(F::kRIndex + components as usize) =
+            *otdata_r.offset(*output.offset(4isize) as isize);
+        *dest.add(F::kGIndex + components as usize) =
+            *otdata_g.offset(*output.offset(5isize) as isize);
+        *dest.add(F::kBIndex + components as usize) =
+            *otdata_b.offset(*output.offset(6isize) as isize);
+        src = src.offset((2 * components) as isize);
+        dest = dest.offset((2 * components) as isize);
+        length -= 2
+    }
+    /* There may be 0-1 pixels remaining. */
+    if length == 1 {
+        vec_r0 = _mm_broadcast_ss(&*igtbl_r.offset(*src.add(F::kRIndex) as isize));
+        vec_g0 = _mm_broadcast_ss(&*igtbl_g.offset(*src.add(F::kGIndex) as isize));
+        vec_b0 = _mm_broadcast_ss(&*igtbl_b.offset(*src.add(F::kBIndex) as isize));
+        vec_r0 = _mm_mul_ps(vec_r0, _mm256_castps256_ps128(mat0));
+        vec_g0 = _mm_mul_ps(vec_g0, _mm256_castps256_ps128(mat1));
+        vec_b0 = _mm_mul_ps(vec_b0, _mm256_castps256_ps128(mat2));
+        if F::kAIndex != 0xff {
+            *dest.add(F::kAIndex) = *src.add(F::kAIndex)
+        }
+        vec_r0 = _mm_add_ps(vec_r0, _mm_add_ps(vec_g0, vec_b0));
+        vec_r0 = _mm_max_ps(_mm256_castps256_ps128(min), vec_r0);
+        vec_r0 = _mm_min_ps(_mm256_castps256_ps128(max), vec_r0);
+        vec_r0 = _mm_mul_ps(vec_r0, _mm256_castps256_ps128(scale));
+        _mm_store_si128(output as *mut __m128i, _mm_cvtps_epi32(vec_r0));
+        *dest.add(F::kRIndex) = *otdata_r.offset(*output.offset(0isize) as isize);
+        *dest.add(F::kGIndex) = *otdata_g.offset(*output.offset(1isize) as isize);
+        *dest.add(F::kBIndex) = *otdata_b.offset(*output.offset(2isize) as isize)
+    };
+}
+#[no_mangle]
+#[target_feature(enable = "avx")]
+pub unsafe fn qcms_transform_data_rgb_out_lut_avx(
+    transform: &qcms_transform,
+    src: *const u8,
+    dest: *mut u8,
+    length: usize,
+) {
+    qcms_transform_data_template_lut_avx::<RGB>(transform, src, dest, length);
+}
+#[no_mangle]
+#[target_feature(enable = "avx")]
+pub unsafe fn qcms_transform_data_rgba_out_lut_avx(
+    transform: &qcms_transform,
+    src: *const u8,
+    dest: *mut u8,
+    length: usize,
+) {
+    qcms_transform_data_template_lut_avx::<RGBA>(transform, src, dest, length);
+}
+#[no_mangle]
+#[target_feature(enable = "avx")]
+pub unsafe fn qcms_transform_data_bgra_out_lut_avx(
+    transform: &qcms_transform,
+    src: *const u8,
+    dest: *mut u8,
+    length: usize,
+) {
+    qcms_transform_data_template_lut_avx::<BGRA>(transform, src, dest, length);
+}