From d8bbc7858622b6d9c278469aab701ca0b609cddf Mon Sep 17 00:00:00 2001 From: Daniel Baumann Date: Wed, 15 May 2024 05:35:49 +0200 Subject: Merging upstream version 126.0. Signed-off-by: Daniel Baumann --- third_party/dav1d/src/arm/32/itx16.S | 19 +++++++++++++------ 1 file changed, 13 insertions(+), 6 deletions(-) (limited to 'third_party/dav1d/src/arm/32/itx16.S') diff --git a/third_party/dav1d/src/arm/32/itx16.S b/third_party/dav1d/src/arm/32/itx16.S index aa6c272e71..7691272517 100644 --- a/third_party/dav1d/src/arm/32/itx16.S +++ b/third_party/dav1d/src/arm/32/itx16.S @@ -547,11 +547,11 @@ function inv_txfm_add_wht_wht_4x4_16bpc_neon, export=1 vmov.i16 q15, #0 vld1.32 {q8, q9}, [r2, :128] vst1.32 {q14, q15}, [r2, :128]! - vshr.s16 q8, q8, #2 + vshr.s32 q8, q8, #2 vld1.32 {q10, q11}, [r2, :128] - vshr.s16 q9, q9, #2 - vshr.s16 q10, q10, #2 - vshr.s16 q11, q11, #2 + vshr.s32 q9, q9, #2 + vshr.s32 q10, q10, #2 + vshr.s32 q11, q11, #2 iwht4 @@ -598,7 +598,9 @@ function inv_txfm_add_4x4_neon vld1.16 {d3}, [r0, :64], r1 L(itx_4x4_end): - vmvn.i16 q15, #0xfc00 // 0x3ff + // read bitdepth_max from the callers stack + ldr r4, [sp, #44] + vdup.i16 q15, r4 sub r0, r0, r1, lsl #2 vqadd.s16 q8, q8, q0 vqadd.s16 q9, q9, q1 @@ -1487,6 +1489,10 @@ function inv_txfm_horz\suffix\()_16x2_neon vqrshrn.s32 d21, q13, #\shift vqrshrn.s32 d22, q14, #\shift vqrshrn.s32 d23, q15, #\shift +.if \scale + b L(horz_16x2_epilog) +.else +L(horz_16x2_epilog): vuzp.16 q8, q9 vuzp.16 q10, q11 @@ -1495,11 +1501,12 @@ function inv_txfm_horz\suffix\()_16x2_neon .endr pop {pc} +.endif endfunc .endm -def_horz_16 scale=0, shift=2 def_horz_16 scale=1, shift=1, suffix=_scale +def_horz_16 scale=0, shift=2 function inv_txfm_add_vert_4x16_neon push {lr} -- cgit v1.2.3