coffee/libyuv
1
0
Fork 0
mirror of https://chromium.googlesource.com/libyuv/libyuv synced 2025-12-06 16:56:55 +08:00
Code Issues Packages Projects Releases Wiki Activity

Change ARGBMultiplyRow_C to match Neon

Browse Source 
The existing behaviour does not round correctly in all cases, so adjust
it to match the existing Neon implementation.

Update the tests to require bit-exactness and disable other
implementations that do not round correctly.

Change-Id: Ie790fb4b4805b555d74d689d83802e1dd4f33df5
Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/5869115
Reviewed-by: Frank Barchard <fbarchard@chromium.org>
This commit is contained in:
George Steed 2024-09-17 11:39:37 +01:00 committed by Frank Barchard
parent a37e6bc81b
commit 02c6e8baca
3 changed files with 22 additions and 22 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
include/libyuv/row.h
View File

@ -106,7 +106,8 @@ extern "C" {
#define HAS_ARGBGRAYROW_SSSE3 #define HAS_ARGBGRAYROW_SSSE3
#define HAS_ARGBLUMACOLORTABLEROW_SSSE3 #define HAS_ARGBLUMACOLORTABLEROW_SSSE3
#define HAS_ARGBMIRRORROW_SSE2 #define HAS_ARGBMIRRORROW_SSE2
#define HAS_ARGBMULTIPLYROW_SSE2 // TODO: Re-enable once rounding behaviour is fixed.
// #define HAS_ARGBMULTIPLYROW_SSE2
#define HAS_ARGBPOLYNOMIALROW_SSE2 #define HAS_ARGBPOLYNOMIALROW_SSE2
#define HAS_ARGBQUANTIZEROW_SSE2 #define HAS_ARGBQUANTIZEROW_SSE2
#define HAS_ARGBSEPIAROW_SSSE3 #define HAS_ARGBSEPIAROW_SSSE3
@ -186,7 +187,8 @@ extern "C" {
// Effects: // Effects:
#define HAS_ARGBADDROW_AVX2 #define HAS_ARGBADDROW_AVX2
#define HAS_ARGBMULTIPLYROW_AVX2 // TODO: Re-enable once rounding behaviour is fixed.
// #define HAS_ARGBMULTIPLYROW_AVX2
#define HAS_ARGBSUBTRACTROW_AVX2 #define HAS_ARGBSUBTRACTROW_AVX2
#define HAS_BLENDPLANEROW_AVX2 #define HAS_BLENDPLANEROW_AVX2
@ -585,7 +587,8 @@ extern "C" {
#define HAS_ARGBEXTRACTALPHAROW_MSA #define HAS_ARGBEXTRACTALPHAROW_MSA
#define HAS_ARGBGRAYROW_MSA #define HAS_ARGBGRAYROW_MSA
#define HAS_ARGBMIRRORROW_MSA #define HAS_ARGBMIRRORROW_MSA
#define HAS_ARGBMULTIPLYROW_MSA // TODO: Re-enable once rounding behaviour is fixed.
// #define HAS_ARGBMULTIPLYROW_MSA
#define HAS_ARGBQUANTIZEROW_MSA #define HAS_ARGBQUANTIZEROW_MSA
#define HAS_ARGBSEPIAROW_MSA #define HAS_ARGBSEPIAROW_MSA
#define HAS_ARGBSETROW_MSA #define HAS_ARGBSETROW_MSA
@ -684,7 +687,8 @@ extern "C" {
#define HAS_ARGBTOUVROW_LSX #define HAS_ARGBTOUVROW_LSX
#define HAS_ARGBTOYJROW_LSX #define HAS_ARGBTOYJROW_LSX
#define HAS_ARGBMIRRORROW_LSX #define HAS_ARGBMIRRORROW_LSX
#define HAS_ARGBMULTIPLYROW_LSX // TODO: Re-enable once rounding behaviour is fixed.
// #define HAS_ARGBMULTIPLYROW_LSX
#define HAS_BGRATOUVROW_LSX #define HAS_BGRATOUVROW_LSX
#define HAS_BGRATOYROW_LSX #define HAS_BGRATOYROW_LSX
#define HAS_I400TOARGBROW_LSX #define HAS_I400TOARGBROW_LSX
@ -751,7 +755,8 @@ extern "C" {
#define HAS_ARGBATTENUATEROW_LASX #define HAS_ARGBATTENUATEROW_LASX
#define HAS_ARGBGRAYROW_LASX #define HAS_ARGBGRAYROW_LASX
#define HAS_ARGBMIRRORROW_LASX #define HAS_ARGBMIRRORROW_LASX
#define HAS_ARGBMULTIPLYROW_LASX // TODO: Re-enable once rounding behaviour is fixed.
// #define HAS_ARGBMULTIPLYROW_LASX
#define HAS_ARGBSEPIAROW_LASX #define HAS_ARGBSEPIAROW_LASX
#define HAS_ARGBSHADEROW_LASX #define HAS_ARGBSHADEROW_LASX
#define HAS_ARGBSHUFFLEROW_LASX #define HAS_ARGBSHUFFLEROW_LASX

21
source/row_common.cc
View File

@ -1375,34 +1375,29 @@ void ARGBShadeRow_C(const uint8_t* src_argb,
#undef REPEAT8 #undef REPEAT8
#undef SHADE #undef SHADE
#define REPEAT8(v) (v) | ((v) << 8)
#define SHADE(f, v) v* f >> 16
void ARGBMultiplyRow_C(const uint8_t* src_argb, void ARGBMultiplyRow_C(const uint8_t* src_argb,
const uint8_t* src_argb1, const uint8_t* src_argb1,
uint8_t* dst_argb, uint8_t* dst_argb,
int width) { int width) {
int i; int i;
for (i = 0; i < width; ++i) { for (i = 0; i < width; ++i) {
const uint32_t b = REPEAT8(src_argb[0]); const uint32_t b = src_argb[0];
const uint32_t g = REPEAT8(src_argb[1]); const uint32_t g = src_argb[1];
const uint32_t r = REPEAT8(src_argb[2]); const uint32_t r = src_argb[2];
const uint32_t a = REPEAT8(src_argb[3]); const uint32_t a = src_argb[3];
const uint32_t b_scale = src_argb1[0]; const uint32_t b_scale = src_argb1[0];
const uint32_t g_scale = src_argb1[1]; const uint32_t g_scale = src_argb1[1];
const uint32_t r_scale = src_argb1[2]; const uint32_t r_scale = src_argb1[2];
const uint32_t a_scale = src_argb1[3]; const uint32_t a_scale = src_argb1[3];
dst_argb[0] = STATIC_CAST(uint8_t, SHADE(b, b_scale)); dst_argb[0] = STATIC_CAST(uint8_t, (b * b_scale + 128) >> 8);
dst_argb[1] = STATIC_CAST(uint8_t, SHADE(g, g_scale)); dst_argb[1] = STATIC_CAST(uint8_t, (g * g_scale + 128) >> 8);
dst_argb[2] = STATIC_CAST(uint8_t, SHADE(r, r_scale)); dst_argb[2] = STATIC_CAST(uint8_t, (r * r_scale + 128) >> 8);
dst_argb[3] = STATIC_CAST(uint8_t, SHADE(a, a_scale)); dst_argb[3] = STATIC_CAST(uint8_t, (a * a_scale + 128) >> 8);
src_argb += 4; src_argb += 4;
src_argb1 += 4; src_argb1 += 4;
dst_argb += 4; dst_argb += 4;
} }
} }
#undef REPEAT8
#undef SHADE
#define SHADE(f, v) clamp255(v + f) #define SHADE(f, v) clamp255(v + f)

8
unit_test/planar_test.cc
View File

@ -1906,28 +1906,28 @@ TEST_F(LibYUVPlanarTest, ARGBMultiply_Any) {
int max_diff = TestMultiply(benchmark_width_ + 1, benchmark_height_, int max_diff = TestMultiply(benchmark_width_ + 1, benchmark_height_,
benchmark_iterations_, disable_cpu_flags_, benchmark_iterations_, disable_cpu_flags_,
benchmark_cpu_info_, +1, 0); benchmark_cpu_info_, +1, 0);
EXPECT_LE(max_diff, 1); EXPECT_LE(max_diff, 0);
} }
TEST_F(LibYUVPlanarTest, ARGBMultiply_Unaligned) { TEST_F(LibYUVPlanarTest, ARGBMultiply_Unaligned) {
int max_diff = int max_diff =
TestMultiply(benchmark_width_, benchmark_height_, benchmark_iterations_, TestMultiply(benchmark_width_, benchmark_height_, benchmark_iterations_,
disable_cpu_flags_, benchmark_cpu_info_, +1, 1); disable_cpu_flags_, benchmark_cpu_info_, +1, 1);
EXPECT_LE(max_diff, 1); EXPECT_LE(max_diff, 0);
} }
TEST_F(LibYUVPlanarTest, ARGBMultiply_Invert) { TEST_F(LibYUVPlanarTest, ARGBMultiply_Invert) {
int max_diff = int max_diff =
TestMultiply(benchmark_width_, benchmark_height_, benchmark_iterations_, TestMultiply(benchmark_width_, benchmark_height_, benchmark_iterations_,
disable_cpu_flags_, benchmark_cpu_info_, -1, 0); disable_cpu_flags_, benchmark_cpu_info_, -1, 0);
EXPECT_LE(max_diff, 1); EXPECT_LE(max_diff, 0);
} }
TEST_F(LibYUVPlanarTest, ARGBMultiply_Opt) { TEST_F(LibYUVPlanarTest, ARGBMultiply_Opt) {
int max_diff = int max_diff =
TestMultiply(benchmark_width_, benchmark_height_, benchmark_iterations_, TestMultiply(benchmark_width_, benchmark_height_, benchmark_iterations_,
disable_cpu_flags_, benchmark_cpu_info_, +1, 0); disable_cpu_flags_, benchmark_cpu_info_, +1, 0);
EXPECT_LE(max_diff, 1); EXPECT_LE(max_diff, 0);
} }
static int TestAdd(int width, static int TestAdd(int width,