coffee/libyuv
1
0
Fork 0
mirror of https://chromium.googlesource.com/libyuv/libyuv synced 2026-10-28 18:07:12 +08:00
Code Issues Packages Projects Releases Wiki Activity

ARGBScale and UV use uint8_t for -fsanitize=alignment

Browse Source 
Test: libyuv_unittest built/run with -fsanitize=alignment

Bug: 553156658

TAG=agy
CONV=42e6e965-3e33-4fd8-8697-b46f2588e5b2

Change-Id: I1c9a3a520caa1b4681d499e1aa36b98a2bd40ba7
Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/8308575
Commit-Queue: Frank Barchard <fbarchard@google.com>
Reviewed-by: richard winterton <rrwinterton@gmail.com>
This commit is contained in:
Frank Barchard 2026-08-28 13:20:09 -07:00
parent 7ad044836d
commit 2fed9193c9
4 changed files with 113 additions and 198 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
README.chromium
View File

@ -1,6 +1,6 @@
Name: libyuv
URL: https://chromium.googlesource.com/libyuv/libyuv/
Version: 1965
Version: 1966
Revision: DEPS
License: BSD-3-Clause
License File: LICENSE

2
include/libyuv/version.h
View File

@ -11,6 +11,6 @@
#ifndef INCLUDE_LIBYUV_VERSION_H_
#define INCLUDE_LIBYUV_VERSION_H_
#define LIBYUV_VERSION 1966
#define LIBYUV_VERSION 1967
#endif // INCLUDE_LIBYUV_VERSION_H_

275
source/scale_common.cc
View File

@ -878,18 +878,15 @@ void ScaleARGBRowDown2_C(const uint8_t* src_argb,
ptrdiff_t src_stride,
uint8_t* dst_argb,
int dst_width) {
const uint32_t* src = (const uint32_t*)(src_argb);
uint32_t* dst = (uint32_t*)(dst_argb);
int x;
(void)src_stride;
for (x = 0; x < dst_width - 1; x += 2) {
dst[0] = src[1];
dst[1] = src[3];
src += 4;
dst += 2;
}
if (dst_width & 1) {
dst[0] = src[1];
for (x = 0; x < dst_width; ++x) {
dst_argb[0] = src_argb[4];
dst_argb[1] = src_argb[5];
dst_argb[2] = src_argb[6];
dst_argb[3] = src_argb[7];
src_argb += 8;
dst_argb += 4;
}
}
@ -937,18 +934,15 @@ void ScaleARGBRowDownEven_C(const uint8_t* src_argb,
int src_stepx,
uint8_t* dst_argb,
int dst_width) {
const uint32_t* src = (const uint32_t*)(src_argb);
uint32_t* dst = (uint32_t*)(dst_argb);
(void)src_stride;
int x;
for (x = 0; x < dst_width - 1; x += 2) {
dst[0] = src[0];
dst[1] = src[src_stepx];
src += src_stepx * 2;
dst += 2;
}
if (dst_width & 1) {
dst[0] = src[0];
(void)src_stride;
for (x = 0; x < dst_width; ++x) {
dst_argb[0] = src_argb[0];
dst_argb[1] = src_argb[1];
dst_argb[2] = src_argb[2];
dst_argb[3] = src_argb[3];
src_argb += src_stepx * 4;
dst_argb += 4;
}
}
@ -982,18 +976,15 @@ void ScaleARGBCols_C(uint8_t* dst_argb,
int dst_width,
int x,
int dx) {
const uint32_t* src = (const uint32_t*)(src_argb);
uint32_t* dst = (uint32_t*)(dst_argb);
int j;
for (j = 0; j < dst_width - 1; j += 2) {
dst[0] = src[x >> 16];
for (j = 0; j < dst_width; ++j) {
const uint8_t* src = src_argb + (x >> 16) * 4;
dst_argb[0] = src[0];
dst_argb[1] = src[1];
dst_argb[2] = src[2];
dst_argb[3] = src[3];
dst_argb += 4;
x += dx;
dst[1] = src[x >> 16];
x += dx;
dst += 2;
}
if (dst_width & 1) {
dst[0] = src[x >> 16];
}
}
@ -1003,18 +994,15 @@ void ScaleARGBCols64_C(uint8_t* dst_argb,
int x32,
int dx) {
int64_t x = (int64_t)(x32);
const uint32_t* src = (const uint32_t*)(src_argb);
uint32_t* dst = (uint32_t*)(dst_argb);
int j;
for (j = 0; j < dst_width - 1; j += 2) {
dst[0] = src[x >> 16];
for (j = 0; j < dst_width; ++j) {
const uint8_t* src = src_argb + (x >> 16) * 4;
dst_argb[0] = src[0];
dst_argb[1] = src[1];
dst_argb[2] = src[2];
dst_argb[3] = src[3];
dst_argb += 4;
x += dx;
dst[1] = src[x >> 16];
x += dx;
dst += 2;
}
if (dst_width & 1) {
dst[0] = src[x >> 16];
}
}
@ -1024,59 +1012,46 @@ void ScaleARGBColsUp2_C(uint8_t* dst_argb,
int dst_width,
int x,
int dx) {
const uint32_t* src = (const uint32_t*)(src_argb);
uint32_t* dst = (uint32_t*)(dst_argb);
int j;
(void)x;
(void)dx;
for (j = 0; j < dst_width - 1; j += 2) {
dst[1] = dst[0] = src[0];
src += 1;
dst += 2;
dst_argb[0] = dst_argb[4] = src_argb[0];
dst_argb[1] = dst_argb[5] = src_argb[1];
dst_argb[2] = dst_argb[6] = src_argb[2];
dst_argb[3] = dst_argb[7] = src_argb[3];
src_argb += 4;
dst_argb += 8;
}
if (dst_width & 1) {
dst[0] = src[0];
dst_argb[0] = src_argb[0];
dst_argb[1] = src_argb[1];
dst_argb[2] = src_argb[2];
dst_argb[3] = src_argb[3];
}
}
// TODO(fbarchard): Replace 0x7f ^ f with 128-f. bug=607.
// Mimics SSSE3 blender
#define BLENDER1(a, b, f) ((a) * (0x7f ^ f) + (b) * f) >> 7
#define BLENDERC(a, b, f, s) \
(uint32_t)(BLENDER1(((a) >> s) & 255, ((b) >> s) & 255, f) << s)
#define BLENDER(a, b, f) \
BLENDERC(a, b, f, 24) | BLENDERC(a, b, f, 16) | BLENDERC(a, b, f, 8) | \
BLENDERC(a, b, f, 0)
#define BLENDER1(a, b, f) (((a) * (0x7f ^ (f)) + (b) * (f)) >> 7)
void ScaleARGBFilterCols_C(uint8_t* dst_argb,
const uint8_t* src_argb,
int dst_width,
int x,
int dx) {
const uint32_t* src = (const uint32_t*)(src_argb);
uint32_t* dst = (uint32_t*)(dst_argb);
int j;
for (j = 0; j < dst_width - 1; j += 2) {
int xi = x >> 16;
for (j = 0; j < dst_width; ++j) {
int xi = (x >> 16) * 4;
int xf = (x >> 9) & 0x7f;
uint32_t a = src[xi];
uint32_t b = src[xi + 1];
dst[0] = BLENDER(a, b, xf);
const uint8_t* a = src_argb + xi;
const uint8_t* b = a + 4;
dst_argb[0] = (uint8_t)BLENDER1(a[0], b[0], xf);
dst_argb[1] = (uint8_t)BLENDER1(a[1], b[1], xf);
dst_argb[2] = (uint8_t)BLENDER1(a[2], b[2], xf);
dst_argb[3] = (uint8_t)BLENDER1(a[3], b[3], xf);
dst_argb += 4;
x += dx;
xi = x >> 16;
xf = (x >> 9) & 0x7f;
a = src[xi];
b = src[xi + 1];
dst[1] = BLENDER(a, b, xf);
x += dx;
dst += 2;
}
if (dst_width & 1) {
int xi = x >> 16;
int xf = (x >> 9) & 0x7f;
uint32_t a = src[xi];
uint32_t b = src[xi + 1];
dst[0] = BLENDER(a, b, xf);
}
}
@ -1086,35 +1061,21 @@ void ScaleARGBFilterCols64_C(uint8_t* dst_argb,
int x32,
int dx) {
int64_t x = (int64_t)(x32);
const uint32_t* src = (const uint32_t*)(src_argb);
uint32_t* dst = (uint32_t*)(dst_argb);
int j;
for (j = 0; j < dst_width - 1; j += 2) {
int64_t xi = x >> 16;
for (j = 0; j < dst_width; ++j) {
int64_t xi = (x >> 16) * 4;
int xf = (x >> 9) & 0x7f;
uint32_t a = src[xi];
uint32_t b = src[xi + 1];
dst[0] = BLENDER(a, b, xf);
const uint8_t* a = src_argb + xi;
const uint8_t* b = a + 4;
dst_argb[0] = (uint8_t)BLENDER1(a[0], b[0], xf);
dst_argb[1] = (uint8_t)BLENDER1(a[1], b[1], xf);
dst_argb[2] = (uint8_t)BLENDER1(a[2], b[2], xf);
dst_argb[3] = (uint8_t)BLENDER1(a[3], b[3], xf);
dst_argb += 4;
x += dx;
xi = x >> 16;
xf = (x >> 9) & 0x7f;
a = src[xi];
b = src[xi + 1];
dst[1] = BLENDER(a, b, xf);
x += dx;
dst += 2;
}
if (dst_width & 1) {
int64_t xi = x >> 16;
int xf = (x >> 9) & 0x7f;
uint32_t a = src[xi];
uint32_t b = src[xi + 1];
dst[0] = BLENDER(a, b, xf);
}
}
#undef BLENDER1
#undef BLENDERC
#undef BLENDER
// UV scale row functions
// same as ARGB but 2 channels
@ -1169,18 +1130,13 @@ void ScaleUVRowDownEven_C(const uint8_t* src_uv,
int src_stepx,
uint8_t* dst_uv,
int dst_width) {
const uint16_t* src = (const uint16_t*)(src_uv);
uint16_t* dst = (uint16_t*)(dst_uv);
(void)src_stride;
int x;
for (x = 0; x < dst_width - 1; x += 2) {
dst[0] = src[0];
dst[1] = src[src_stepx];
src += src_stepx * 2;
dst += 2;
}
if (dst_width & 1) {
dst[0] = src[0];
(void)src_stride;
for (x = 0; x < dst_width; ++x) {
dst_uv[0] = src_uv[0];
dst_uv[1] = src_uv[1];
src_uv += src_stepx * 2;
dst_uv += 2;
}
}
@ -1324,18 +1280,13 @@ void ScaleUVCols_C(uint8_t* dst_uv,
int dst_width,
int x,
int dx) {
const uint16_t* src = (const uint16_t*)(src_uv);
uint16_t* dst = (uint16_t*)(dst_uv);
int j;
for (j = 0; j < dst_width - 1; j += 2) {
dst[0] = src[x >> 16];
for (j = 0; j < dst_width; ++j) {
const uint8_t* src = src_uv + (x >> 16) * 2;
dst_uv[0] = src[0];
dst_uv[1] = src[1];
dst_uv += 2;
x += dx;
dst[1] = src[x >> 16];
x += dx;
dst += 2;
}
if (dst_width & 1) {
dst[0] = src[x >> 16];
}
}
@ -1345,18 +1296,13 @@ void ScaleUVCols64_C(uint8_t* dst_uv,
int x32,
int dx) {
int64_t x = (int64_t)(x32);
const uint16_t* src = (const uint16_t*)(src_uv);
uint16_t* dst = (uint16_t*)(dst_uv);
int j;
for (j = 0; j < dst_width - 1; j += 2) {
dst[0] = src[x >> 16];
for (j = 0; j < dst_width; ++j) {
const uint8_t* src = src_uv + (x >> 16) * 2;
dst_uv[0] = src[0];
dst_uv[1] = src[1];
dst_uv += 2;
x += dx;
dst[1] = src[x >> 16];
x += dx;
dst += 2;
}
if (dst_width & 1) {
dst[0] = src[x >> 16];
}
}
@ -1366,57 +1312,40 @@ void ScaleUVColsUp2_C(uint8_t* dst_uv,
int dst_width,
int x,
int dx) {
const uint16_t* src = (const uint16_t*)(src_uv);
uint16_t* dst = (uint16_t*)(dst_uv);
int j;
(void)x;
(void)dx;
for (j = 0; j < dst_width - 1; j += 2) {
dst[1] = dst[0] = src[0];
src += 1;
dst += 2;
dst_uv[0] = dst_uv[2] = src_uv[0];
dst_uv[1] = dst_uv[3] = src_uv[1];
src_uv += 2;
dst_uv += 4;
}
if (dst_width & 1) {
dst[0] = src[0];
dst_uv[0] = src_uv[0];
dst_uv[1] = src_uv[1];
}
}
// Performs (a + ((f * (b - a) + 64) >> 7)) which is equivalent of
// ((a * (128 - f) + b * f + 64) >> 7).
#define BLENDER1(a, b, f) ((a) + (((f) * ((b) - (a)) + 64) >> 7))
#define BLENDERC(a, b, f, s) \
(uint16_t)(BLENDER1(((a) >> s) & 255, ((b) >> s) & 255, f) << s)
#define BLENDER(a, b, f) BLENDERC(a, b, f, 8) | BLENDERC(a, b, f, 0)
void ScaleUVFilterCols_C(uint8_t* dst_uv,
const uint8_t* src_uv,
int dst_width,
int x,
int dx) {
const uint16_t* src = (const uint16_t*)(src_uv);
uint16_t* dst = (uint16_t*)(dst_uv);
int j;
for (j = 0; j < dst_width - 1; j += 2) {
int xi = x >> 16;
for (j = 0; j < dst_width; ++j) {
int xi = (x >> 16) * 2;
int xf = (x >> 9) & 0x7f;
uint16_t a = src[xi];
uint16_t b = src[xi + 1];
dst[0] = BLENDER(a, b, xf);
const uint8_t* a = src_uv + xi;
const uint8_t* b = a + 2;
dst_uv[0] = (uint8_t)BLENDER1(a[0], b[0], xf);
dst_uv[1] = (uint8_t)BLENDER1(a[1], b[1], xf);
dst_uv += 2;
x += dx;
xi = x >> 16;
xf = (x >> 9) & 0x7f;
a = src[xi];
b = src[xi + 1];
dst[1] = BLENDER(a, b, xf);
x += dx;
dst += 2;
}
if (dst_width & 1) {
int xi = x >> 16;
int xf = (x >> 9) & 0x7f;
uint16_t a = src[xi];
uint16_t b = src[xi + 1];
dst[0] = BLENDER(a, b, xf);
}
}
@ -1426,35 +1355,19 @@ void ScaleUVFilterCols64_C(uint8_t* dst_uv,
int x32,
int dx) {
int64_t x = (int64_t)(x32);
const uint16_t* src = (const uint16_t*)(src_uv);
uint16_t* dst = (uint16_t*)(dst_uv);
int j;
for (j = 0; j < dst_width - 1; j += 2) {
int64_t xi = x >> 16;
for (j = 0; j < dst_width; ++j) {
int64_t xi = (x >> 16) * 2;
int xf = (x >> 9) & 0x7f;
uint16_t a = src[xi];
uint16_t b = src[xi + 1];
dst[0] = BLENDER(a, b, xf);
const uint8_t* a = src_uv + xi;
const uint8_t* b = a + 2;
dst_uv[0] = (uint8_t)BLENDER1(a[0], b[0], xf);
dst_uv[1] = (uint8_t)BLENDER1(a[1], b[1], xf);
dst_uv += 2;
x += dx;
xi = x >> 16;
xf = (x >> 9) & 0x7f;
a = src[xi];
b = src[xi + 1];
dst[1] = BLENDER(a, b, xf);
x += dx;
dst += 2;
}
if (dst_width & 1) {
int64_t xi = x >> 16;
int xf = (x >> 9) & 0x7f;
uint16_t a = src[xi];
uint16_t b = src[xi + 1];
dst[0] = BLENDER(a, b, xf);
}
}
#undef BLENDER1
#undef BLENDERC
#undef BLENDER
// Scale plane vertically with bilinear interpolation.
void ScalePlaneVertical(int src_height,

32
unit_test/planar_test.cc
View File

@ -1741,10 +1741,11 @@ TEST_F(LibYUVPlanarTest, TestDetileSplitUVPlane_Correctness) {
int i, j;
// orig is tiled. Allocate enough memory for tiles.
int tile_width = (benchmark_width_ + 15) & ~15;
int uv_width = (benchmark_width_ + 1) & ~1;
int tile_width = (uv_width + 15) & ~15;
int tile_height = (benchmark_height_ + 15) & ~15;
int tile_plane_size = tile_width * tile_height;
int uv_plane_size = ((benchmark_width_ + 1) / 2) * benchmark_height_;
int uv_plane_size = (uv_width / 2) * benchmark_height_;
align_buffer_page_end(tile_uv, tile_plane_size);
align_buffer_page_end(detiled_uv, tile_plane_size);
align_buffer_page_end(dst_u_two_stage, uv_plane_size);
@ -1759,17 +1760,17 @@ TEST_F(LibYUVPlanarTest, TestDetileSplitUVPlane_Correctness) {
memset(dst_v_two_stage, 0, uv_plane_size);
memset(dst_v_opt, 0, uv_plane_size);
DetileSplitUVPlane(tile_uv, tile_width, dst_u_opt, (benchmark_width_ + 1) / 2,
dst_v_opt, (benchmark_width_ + 1) / 2, benchmark_width_,
DetileSplitUVPlane(tile_uv, tile_width, dst_u_opt, uv_width / 2,
dst_v_opt, uv_width / 2, uv_width,
benchmark_height_, 16);
// Benchmark 2 step conversion for comparison.
for (j = 0; j < benchmark_iterations_; j++) {
DetilePlane(tile_uv, tile_width, detiled_uv, benchmark_width_,
benchmark_width_, benchmark_height_, 16);
SplitUVPlane(detiled_uv, tile_width, dst_u_two_stage,
(benchmark_width_ + 1) / 2, dst_v_two_stage,
(benchmark_width_ + 1) / 2, (benchmark_width_ + 1) / 2,
DetilePlane(tile_uv, tile_width, detiled_uv, uv_width,
uv_width, benchmark_height_, 16);
SplitUVPlane(detiled_uv, uv_width, dst_u_two_stage,
uv_width / 2, dst_v_two_stage,
uv_width / 2, uv_width / 2,
benchmark_height_);
}
@ -1790,10 +1791,11 @@ TEST_F(LibYUVPlanarTest, TestDetileSplitUVPlane_Benchmark) {
int i, j;
// orig is tiled. Allocate enough memory for tiles.
int tile_width = (benchmark_width_ + 15) & ~15;
int uv_width = (benchmark_width_ + 1) & ~1;
int tile_width = (uv_width + 15) & ~15;
int tile_height = (benchmark_height_ + 15) & ~15;
int tile_plane_size = tile_width * tile_height;
int uv_plane_size = ((benchmark_width_ + 1) / 2) * benchmark_height_;
int uv_plane_size = (uv_width / 2) * benchmark_height_;
align_buffer_page_end(tile_uv, tile_plane_size);
align_buffer_page_end(dst_u_c, uv_plane_size);
align_buffer_page_end(dst_u_opt, uv_plane_size);
@ -1809,8 +1811,8 @@ TEST_F(LibYUVPlanarTest, TestDetileSplitUVPlane_Benchmark) {
// Disable all optimizations.
MaskCpuFlags(disable_cpu_flags_);
DetileSplitUVPlane(tile_uv, tile_width, dst_u_c, (benchmark_width_ + 1) / 2,
dst_v_c, (benchmark_width_ + 1) / 2, benchmark_width_,
DetileSplitUVPlane(tile_uv, tile_width, dst_u_c, uv_width / 2,
dst_v_c, uv_width / 2, uv_width,
benchmark_height_, 16);
// Enable optimizations.
@ -1818,8 +1820,8 @@ TEST_F(LibYUVPlanarTest, TestDetileSplitUVPlane_Benchmark) {
for (j = 0; j < benchmark_iterations_; j++) {
DetileSplitUVPlane(
tile_uv, tile_width, dst_u_opt, (benchmark_width_ + 1) / 2, dst_v_opt,
(benchmark_width_ + 1) / 2, benchmark_width_, benchmark_height_, 16);
tile_uv, tile_width, dst_u_opt, uv_width / 2, dst_v_opt,
uv_width / 2, uv_width, benchmark_height_, 16);
}
for (i = 0; i < uv_plane_size; ++i) {