ref: 93db6511a84a56950390eec7d71f9e3255dd4c75
parent: 57fc3e991792ea277a309bcc9351bb800d46b380
author: Sindre Aamås <saamas@cisco.com>
date: Mon Apr 11 11:55:20 EDT 2016
[UT] Test VAA routines with a wider variety of resolutions Test even and odd multiples of 32 width because some AVX2 routines have conditional logic based on that.
--- a/test/processing/ProcessUT_VaaCalc.cpp
+++ b/test/processing/ProcessUT_VaaCalc.cpp
@@ -590,20 +590,22 @@
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]); \
}
@@ -633,25 +635,27 @@
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); \
+ 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]); \
+ } \
} \
- 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]); \
- } \
}
#define GENERATE_VAACalcSadSsd_UT(func, ASM, CPUFLAGS) \
@@ -682,29 +686,31 @@
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); \
+ 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]); \
+ } \
} \
- 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]); \
- } \
}
#define GENERATE_VAACalcSadVar_UT(func, ASM, CPUFLAGS) \
@@ -733,27 +739,29 @@
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); \
+ 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]); \
+ } \
} \
- 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]); \
- } \
}
#define GENERATE_VAACalcSadSsdBgd_UT(func, ASM, CPUFLAGS) \
@@ -788,32 +796,34 @@
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]); \
+ } \
} \
}