generic-library/openh264
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Merge pull request #2436 from saamas/processing-add-avx2-vaa-routines

Browse Source 
[Processing] Add AVX2 VAA routines
This commit is contained in:
ruil2 2016-04-28 09:08:03 +08:00
commit 56618249d7
4 changed files with 1656 additions and 99 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
codec/processing/src/vaacalc/vaacalculation.cpp
View File

@ -64,6 +64,13 @@ void CVAACalculation::InitVaaFuncs (SVaaFuncs& sVaaFuncs, int32_t iCpuFlag) {
sVaaFuncs.pfVAACalcSadSsdBgd = VAACalcSadSsdBgd_sse2;
sVaaFuncs.pfVAACalcSadVar = VAACalcSadVar_sse2;
}
if (iCpuFlag & WELS_CPU_AVX2) {
sVaaFuncs.pfVAACalcSad = VAACalcSad_avx2;
sVaaFuncs.pfVAACalcSadBgd = VAACalcSadBgd_avx2;
sVaaFuncs.pfVAACalcSadSsd = VAACalcSadSsd_avx2;
sVaaFuncs.pfVAACalcSadSsdBgd = VAACalcSadSsdBgd_avx2;
sVaaFuncs.pfVAACalcSadVar = VAACalcSadVar_avx2;
}
#endif//X86_ASM
#ifdef HAVE_NEON
if ((iCpuFlag & WELS_CPU_NEON) == WELS_CPU_NEON) {

5
codec/processing/src/vaacalc/vaacalculation.h
View File

@ -104,6 +104,11 @@ VAACalcSadSsdBgdFunc VAACalcSadSsdBgd_sse2;
VAACalcSadFunc VAACalcSad_sse2;
VAACalcSadVarFunc VAACalcSadVar_sse2;
VAACalcSadSsdFunc VAACalcSadSsd_sse2;
VAACalcSadBgdFunc VAACalcSadBgd_avx2;
VAACalcSadSsdBgdFunc VAACalcSadSsdBgd_avx2;
VAACalcSadFunc VAACalcSad_avx2;
VAACalcSadVarFunc VAACalcSadVar_avx2;
VAACalcSadSsdFunc VAACalcSadSsd_avx2;
WELSVP_EXTERN_C_END
#endif

1529
codec/processing/src/x86/vaa.asm
View File

File diff suppressed because it is too large Load Diff

214
test/processing/ProcessUT_VaaCalc.cpp
View File

@ -590,20 +590,22 @@ TEST (VAACalcFuncTest, func) { \
int32_t pic_height_a; \
int32_t pic_stride_a; \
int32_t psadframe_a; \
pic_width_c = pic_width_a = 320-16; \
pic_height_c = pic_height_a = 320; \
pic_stride_c = pic_stride_a = 320; \
psadframe_c = psadframe_a = 0; \
for (int j=0; j<BUFFER_SIZE; j++) { \
cur_data_c[j] = cur_data_a[j] = (rand()%256); \
ref_data_c[j] = ref_data_a[j] = (rand()%256); \
psad8x8_c[j%(BUFFER_SIZE/64)] = psad8x8_a[j%(BUFFER_SIZE/64)] = (rand()%256); \
for (int i=0; i<4; i++) { \
pic_width_c = pic_width_a = 320-16*i; \
pic_height_c = pic_height_a = 320; \
pic_stride_c = pic_stride_a = 320; \
psadframe_c = psadframe_a = 0; \
for (int j=0; j<BUFFER_SIZE; j++) { \
cur_data_c[j] = cur_data_a[j] = (rand()%256); \
ref_data_c[j] = ref_data_a[j] = (rand()%256); \
psad8x8_c[j%(BUFFER_SIZE/64)] = psad8x8_a[j%(BUFFER_SIZE/64)] = (rand()%256); \
} \
VAACalcSad_ref (cur_data_c, ref_data_c, pic_width_c, pic_height_c, pic_stride_c, &psadframe_c, psad8x8_c); \
func (cur_data_a, ref_data_a, pic_width_a, pic_height_a, pic_stride_a, &psadframe_a, psad8x8_a); \
ASSERT_EQ (psadframe_a, psadframe_c); \
for (int j=0; j<(BUFFER_SIZE/64); j++) \
ASSERT_EQ (psad8x8_a[j], psad8x8_c[j]); \
} \
VAACalcSad_ref (cur_data_c, ref_data_c, pic_width_c, pic_height_c, pic_stride_c, &psadframe_c, psad8x8_c); \
func (cur_data_a, ref_data_a, pic_width_a, pic_height_a, pic_stride_a, &psadframe_a, psad8x8_a); \
ASSERT_EQ (psadframe_a, psadframe_c); \
for (int j=0; j<(BUFFER_SIZE/64); j++) \
ASSERT_EQ (psad8x8_a[j], psad8x8_c[j]); \
}
@ -633,24 +635,26 @@ TEST (VAACalcFuncTest, func) { \
int32_t pic_height_a; \
int32_t pic_stride_a; \
int32_t psadframe_a; \
pic_width_c = pic_width_a = 320-16; \
pic_height_c = pic_height_a = 320; \
pic_stride_c = pic_stride_a = 320; \
psadframe_c = psadframe_a = 0; \
for (int j=0; j<BUFFER_SIZE; j++) { \
cur_data_c[j] = cur_data_a[j] = (rand()%256); \
ref_data_c[j] = ref_data_a[j] = (rand()%256); \
psad8x8_c[j%(BUFFER_SIZE/64)] = psad8x8_a[j%(BUFFER_SIZE/64)] = (rand()%256); \
psd8x8_c[j%(BUFFER_SIZE/64)] = psd8x8_a[j%(BUFFER_SIZE/64)] = (rand()%256); \
pmad8x8_c[j%(BUFFER_SIZE/64)] = pmad8x8_a[j%(BUFFER_SIZE/64)] = (rand()%256); \
} \
VAACalcSadBgd_ref (cur_data_c, ref_data_c, pic_width_c, pic_height_c, pic_stride_c, &psadframe_c, psad8x8_c, psd8x8_c, pmad8x8_c); \
func (cur_data_a, ref_data_a, pic_width_a, pic_height_a, pic_stride_a, &psadframe_a, psad8x8_a, psd8x8_a, pmad8x8_a); \
ASSERT_EQ (psadframe_a, psadframe_c); \
for (int j=0; j<(BUFFER_SIZE/64); j++) {\
ASSERT_EQ (psad8x8_a[j], psad8x8_c[j]); \
ASSERT_EQ (psd8x8_a[j], psd8x8_c[j]); \
ASSERT_EQ (pmad8x8_a[j], pmad8x8_c[j]); \
for (int i=0; i<4; i++) { \
pic_width_c = pic_width_a = 320-16*i; \
pic_height_c = pic_height_a = 320; \
pic_stride_c = pic_stride_a = 320; \
psadframe_c = psadframe_a = 0; \
for (int j=0; j<BUFFER_SIZE; j++) { \
cur_data_c[j] = cur_data_a[j] = (rand()%256); \
ref_data_c[j] = ref_data_a[j] = (rand()%256); \
psad8x8_c[j%(BUFFER_SIZE/64)] = psad8x8_a[j%(BUFFER_SIZE/64)] = (rand()%256); \
psd8x8_c[j%(BUFFER_SIZE/64)] = psd8x8_a[j%(BUFFER_SIZE/64)] = (rand()%256); \
pmad8x8_c[j%(BUFFER_SIZE/64)] = pmad8x8_a[j%(BUFFER_SIZE/64)] = (rand()%256); \
} \
VAACalcSadBgd_ref (cur_data_c, ref_data_c, pic_width_c, pic_height_c, pic_stride_c, &psadframe_c, psad8x8_c, psd8x8_c, pmad8x8_c); \
func (cur_data_a, ref_data_a, pic_width_a, pic_height_a, pic_stride_a, &psadframe_a, psad8x8_a, psd8x8_a, pmad8x8_a); \
ASSERT_EQ (psadframe_a, psadframe_c); \
for (int j=0; j<(BUFFER_SIZE/64); j++) {\
ASSERT_EQ (psad8x8_a[j], psad8x8_c[j]); \
ASSERT_EQ (psd8x8_a[j], psd8x8_c[j]); \
ASSERT_EQ (pmad8x8_a[j], pmad8x8_c[j]); \
} \
} \
}
@ -682,28 +686,30 @@ TEST (VAACalcFuncTest, func) { \
int32_t pic_height_a; \
int32_t pic_stride_a; \
int32_t psadframe_a; \
pic_width_c = pic_width_a = 320-16; \
pic_height_c = pic_height_a = 320; \
pic_stride_c = pic_stride_a = 320; \
psadframe_c = psadframe_a = 0; \
for (int j=0; j<BUFFER_SIZE; j++) { \
cur_data_c[j] = cur_data_a[j] = (rand()%256); \
ref_data_c[j] = ref_data_a[j] = (rand()%256); \
psad8x8_c[j%(BUFFER_SIZE/64)] = psad8x8_a[j%(BUFFER_SIZE/64)] = (rand()%256); \
psum16x16_c[j%(BUFFER_SIZE/256)] = psum16x16_a[j%(BUFFER_SIZE/256)] = (rand()%256); \
psqsum16x16_c[j%(BUFFER_SIZE/256)] = psqsum16x16_a[j%(BUFFER_SIZE/256)] = (rand()%256); \
psqdiff16x16_c[j%(BUFFER_SIZE/256)] = psqdiff16x16_a[j%(BUFFER_SIZE/256)] = (rand()%256); \
} \
VAACalcSadSsd_ref (cur_data_c, ref_data_c, pic_width_c, pic_height_c, pic_stride_c, &psadframe_c, psad8x8_c, psum16x16_c, psqsum16x16_c, psqdiff16x16_c); \
func (cur_data_a, ref_data_a, pic_width_a, pic_height_a, pic_stride_a, &psadframe_a, psad8x8_a, psum16x16_a, psqsum16x16_a, psqdiff16x16_a); \
ASSERT_EQ (psadframe_a, psadframe_c); \
for (int j=0; j<(BUFFER_SIZE/64); j++) {\
ASSERT_EQ (psad8x8_a[j], psad8x8_c[j]); \
} \
for (int j=0; j<(BUFFER_SIZE/256); j++) {\
ASSERT_EQ (psum16x16_a[j], psum16x16_c[j]); \
ASSERT_EQ (psqsum16x16_a[j], psqsum16x16_c[j]); \
ASSERT_EQ (psqdiff16x16_a[j], psqdiff16x16_c[j]); \
for (int i=0; i<4; i++) { \
pic_width_c = pic_width_a = 320-16*i; \
pic_height_c = pic_height_a = 320; \
pic_stride_c = pic_stride_a = 320; \
psadframe_c = psadframe_a = 0; \
for (int j=0; j<BUFFER_SIZE; j++) { \
cur_data_c[j] = cur_data_a[j] = (rand()%256); \
ref_data_c[j] = ref_data_a[j] = (rand()%256); \
psad8x8_c[j%(BUFFER_SIZE/64)] = psad8x8_a[j%(BUFFER_SIZE/64)] = (rand()%256); \
psum16x16_c[j%(BUFFER_SIZE/256)] = psum16x16_a[j%(BUFFER_SIZE/256)] = (rand()%256); \
psqsum16x16_c[j%(BUFFER_SIZE/256)] = psqsum16x16_a[j%(BUFFER_SIZE/256)] = (rand()%256); \
psqdiff16x16_c[j%(BUFFER_SIZE/256)] = psqdiff16x16_a[j%(BUFFER_SIZE/256)] = (rand()%256); \
} \
VAACalcSadSsd_ref (cur_data_c, ref_data_c, pic_width_c, pic_height_c, pic_stride_c, &psadframe_c, psad8x8_c, psum16x16_c, psqsum16x16_c, psqdiff16x16_c); \
func (cur_data_a, ref_data_a, pic_width_a, pic_height_a, pic_stride_a, &psadframe_a, psad8x8_a, psum16x16_a, psqsum16x16_a, psqdiff16x16_a); \
ASSERT_EQ (psadframe_a, psadframe_c); \
for (int j=0; j<(BUFFER_SIZE/64); j++) {\
ASSERT_EQ (psad8x8_a[j], psad8x8_c[j]); \
} \
for (int j=0; j<(BUFFER_SIZE/256); j++) {\
ASSERT_EQ (psum16x16_a[j], psum16x16_c[j]); \
ASSERT_EQ (psqsum16x16_a[j], psqsum16x16_c[j]); \
ASSERT_EQ (psqdiff16x16_a[j], psqdiff16x16_c[j]); \
} \
} \
}
@ -733,26 +739,28 @@ TEST (VAACalcFuncTest, func) { \
int32_t pic_height_a; \
int32_t pic_stride_a; \
int32_t psadframe_a; \
pic_width_c = pic_width_a = 320-16; \
pic_height_c = pic_height_a = 320; \
pic_stride_c = pic_stride_a = 320; \
psadframe_c = psadframe_a = 0; \
for (int j=0; j<BUFFER_SIZE; j++) { \
cur_data_c[j] = cur_data_a[j] = (rand()%256); \
ref_data_c[j] = ref_data_a[j] = (rand()%256); \
psad8x8_c[j%(BUFFER_SIZE/64)] = psad8x8_a[j%(BUFFER_SIZE/64)] = (rand()%256); \
psum16x16_c[j%(BUFFER_SIZE/256)] = psum16x16_a[j%(BUFFER_SIZE/256)] = (rand()%256); \
psqsum16x16_c[j%(BUFFER_SIZE/256)] = psqsum16x16_a[j%(BUFFER_SIZE/256)] = (rand()%256); \
} \
VAACalcSadVar_ref (cur_data_c, ref_data_c, pic_width_c, pic_height_c, pic_stride_c, &psadframe_c, psad8x8_c, psum16x16_c, psqsum16x16_c); \
func (cur_data_a, ref_data_a, pic_width_a, pic_height_a, pic_stride_a, &psadframe_a, psad8x8_a, psum16x16_a, psqsum16x16_a); \
ASSERT_EQ (psadframe_a, psadframe_c); \
for (int j=0; j<(BUFFER_SIZE/64); j++) {\
ASSERT_EQ (psad8x8_a[j], psad8x8_c[j]); \
} \
for (int j=0; j<(BUFFER_SIZE/256); j++) {\
ASSERT_EQ (psum16x16_a[j], psum16x16_c[j]); \
ASSERT_EQ (psqsum16x16_a[j], psqsum16x16_c[j]); \
for (int i=0; i<4; i++) { \
pic_width_c = pic_width_a = 320-16*i; \
pic_height_c = pic_height_a = 320; \
pic_stride_c = pic_stride_a = 320; \
psadframe_c = psadframe_a = 0; \
for (int j=0; j<BUFFER_SIZE; j++) { \
cur_data_c[j] = cur_data_a[j] = (rand()%256); \
ref_data_c[j] = ref_data_a[j] = (rand()%256); \
psad8x8_c[j%(BUFFER_SIZE/64)] = psad8x8_a[j%(BUFFER_SIZE/64)] = (rand()%256); \
psum16x16_c[j%(BUFFER_SIZE/256)] = psum16x16_a[j%(BUFFER_SIZE/256)] = (rand()%256); \
psqsum16x16_c[j%(BUFFER_SIZE/256)] = psqsum16x16_a[j%(BUFFER_SIZE/256)] = (rand()%256); \
} \
VAACalcSadVar_ref (cur_data_c, ref_data_c, pic_width_c, pic_height_c, pic_stride_c, &psadframe_c, psad8x8_c, psum16x16_c, psqsum16x16_c); \
func (cur_data_a, ref_data_a, pic_width_a, pic_height_a, pic_stride_a, &psadframe_a, psad8x8_a, psum16x16_a, psqsum16x16_a); \
ASSERT_EQ (psadframe_a, psadframe_c); \
for (int j=0; j<(BUFFER_SIZE/64); j++) {\
ASSERT_EQ (psad8x8_a[j], psad8x8_c[j]); \
} \
for (int j=0; j<(BUFFER_SIZE/256); j++) {\
ASSERT_EQ (psum16x16_a[j], psum16x16_c[j]); \
ASSERT_EQ (psqsum16x16_a[j], psqsum16x16_c[j]); \
} \
} \
}
@ -788,32 +796,34 @@ TEST (VAACalcFuncTest, func) { \
int32_t pic_height_a; \
int32_t pic_stride_a; \
int32_t psadframe_a; \
pic_width_c = pic_width_a = 320-16; \
pic_height_c = pic_height_a = 320; \
pic_stride_c = pic_stride_a = 320; \
psadframe_c = psadframe_a = 0; \
for (int j=0; j<BUFFER_SIZE; j++) { \
cur_data_c[j] = cur_data_a[j] = (rand()%256); \
ref_data_c[j] = ref_data_a[j] = (rand()%256); \
psad8x8_c[j%(BUFFER_SIZE/64)] = psad8x8_a[j%(BUFFER_SIZE/64)] = (rand()%256); \
psd8x8_c[j%(BUFFER_SIZE/64)] = psd8x8_a[j%(BUFFER_SIZE/64)] = (rand()%256); \
pmad8x8_c[j%(BUFFER_SIZE/64)] = pmad8x8_a[j%(BUFFER_SIZE/64)] = (rand()%256); \
psum16x16_c[j%(BUFFER_SIZE/256)] = psum16x16_a[j%(BUFFER_SIZE/256)] = (rand()%256); \
psqsum16x16_c[j%(BUFFER_SIZE/256)] = psqsum16x16_a[j%(BUFFER_SIZE/256)] = (rand()%256); \
psqdiff16x16_c[j%(BUFFER_SIZE/256)] = psqdiff16x16_a[j%(BUFFER_SIZE/256)] = (rand()%256); \
} \
VAACalcSadSsdBgd_ref (cur_data_c, ref_data_c, pic_width_c, pic_height_c, pic_stride_c, &psadframe_c, psad8x8_c, psum16x16_c, psqsum16x16_c, psqdiff16x16_c, psd8x8_c, pmad8x8_c); \
func (cur_data_a, ref_data_a, pic_width_a, pic_height_a, pic_stride_a, &psadframe_a, psad8x8_a, psum16x16_a, psqsum16x16_a, psqdiff16x16_a, psd8x8_a, pmad8x8_a); \
ASSERT_EQ (psadframe_a, psadframe_c); \
for (int j=0; j<(BUFFER_SIZE/64); j++) {\
ASSERT_EQ (psad8x8_a[j], psad8x8_c[j]); \
ASSERT_EQ (psd8x8_a[j], psd8x8_c[j]); \
ASSERT_EQ (pmad8x8_a[j], pmad8x8_c[j]); \
} \
for (int j=0; j<(BUFFER_SIZE/256); j++) {\
ASSERT_EQ (psum16x16_a[j], psum16x16_c[j]); \
ASSERT_EQ (psqsum16x16_a[j], psqsum16x16_c[j]); \
ASSERT_EQ (psqdiff16x16_a[j], psqdiff16x16_c[j]); \
for (int i=0; i<4; i++) { \
pic_width_c = pic_width_a = 320-16*i; \
pic_height_c = pic_height_a = 320; \
pic_stride_c = pic_stride_a = 320; \
psadframe_c = psadframe_a = 0; \
for (int j=0; j<BUFFER_SIZE; j++) { \
cur_data_c[j] = cur_data_a[j] = (rand()%256); \
ref_data_c[j] = ref_data_a[j] = (rand()%256); \
psad8x8_c[j%(BUFFER_SIZE/64)] = psad8x8_a[j%(BUFFER_SIZE/64)] = (rand()%256); \
psd8x8_c[j%(BUFFER_SIZE/64)] = psd8x8_a[j%(BUFFER_SIZE/64)] = (rand()%256); \
pmad8x8_c[j%(BUFFER_SIZE/64)] = pmad8x8_a[j%(BUFFER_SIZE/64)] = (rand()%256); \
psum16x16_c[j%(BUFFER_SIZE/256)] = psum16x16_a[j%(BUFFER_SIZE/256)] = (rand()%256); \
psqsum16x16_c[j%(BUFFER_SIZE/256)] = psqsum16x16_a[j%(BUFFER_SIZE/256)] = (rand()%256); \
psqdiff16x16_c[j%(BUFFER_SIZE/256)] = psqdiff16x16_a[j%(BUFFER_SIZE/256)] = (rand()%256); \
} \
VAACalcSadSsdBgd_ref (cur_data_c, ref_data_c, pic_width_c, pic_height_c, pic_stride_c, &psadframe_c, psad8x8_c, psum16x16_c, psqsum16x16_c, psqdiff16x16_c, psd8x8_c, pmad8x8_c); \
func (cur_data_a, ref_data_a, pic_width_a, pic_height_a, pic_stride_a, &psadframe_a, psad8x8_a, psum16x16_a, psqsum16x16_a, psqdiff16x16_a, psd8x8_a, pmad8x8_a); \
ASSERT_EQ (psadframe_a, psadframe_c); \
for (int j=0; j<(BUFFER_SIZE/64); j++) {\
ASSERT_EQ (psad8x8_a[j], psad8x8_c[j]); \
ASSERT_EQ (psd8x8_a[j], psd8x8_c[j]); \
ASSERT_EQ (pmad8x8_a[j], pmad8x8_c[j]); \
} \
for (int j=0; j<(BUFFER_SIZE/256); j++) {\
ASSERT_EQ (psum16x16_a[j], psum16x16_c[j]); \
ASSERT_EQ (psqsum16x16_a[j], psqsum16x16_c[j]); \
ASSERT_EQ (psqdiff16x16_a[j], psqdiff16x16_c[j]); \
} \
} \
}
@ -828,6 +838,12 @@ GENERATE_VAACalcSadBgd_UT (VAACalcSadBgd_sse2, 1, WELS_CPU_SSE2)
GENERATE_VAACalcSadSsdBgd_UT (VAACalcSadSsdBgd_sse2, 1, WELS_CPU_SSE2)
GENERATE_VAACalcSadSsd_UT (VAACalcSadSsd_sse2, 1, WELS_CPU_SSE2)
GENERATE_VAACalcSadVar_UT (VAACalcSadVar_sse2, 1, WELS_CPU_SSE2)
GENERATE_VAACalcSad_UT (VAACalcSad_avx2, 1, WELS_CPU_AVX2)
GENERATE_VAACalcSadBgd_UT (VAACalcSadBgd_avx2, 1, WELS_CPU_AVX2)
GENERATE_VAACalcSadSsdBgd_UT (VAACalcSadSsdBgd_avx2, 1, WELS_CPU_AVX2)
GENERATE_VAACalcSadSsd_UT (VAACalcSadSsd_avx2, 1, WELS_CPU_AVX2)
GENERATE_VAACalcSadVar_UT (VAACalcSadVar_avx2, 1, WELS_CPU_AVX2)
#endif
#if defined(HAVE_NEON)