diff options
| author | Alex Converse <aconverse@google.com> | 2016-01-13 00:25:54 +0000 |
|---|---|---|
| committer | Gerrit Code Review <noreply-gerritcodereview@google.com> | 2016-01-13 00:25:54 +0000 |
| commit | ea48370a500537906d62544ca4ed75301d79e772 (patch) | |
| tree | 782752236c957a8286831a08652f90c341219069 /vpx_dsp/x86 | |
| parent | 15939cb2d76c773950cda40988ede89e111872ea (diff) | |
| parent | 022c848b4d04a2c1f56a54748eee503767269023 (diff) | |
| download | libvpx-ea48370a500537906d62544ca4ed75301d79e772.tar libvpx-ea48370a500537906d62544ca4ed75301d79e772.tar.gz libvpx-ea48370a500537906d62544ca4ed75301d79e772.tar.bz2 libvpx-ea48370a500537906d62544ca4ed75301d79e772.zip | |
Merge "Change highbd variance rounding to prevent negative variance."
Diffstat (limited to 'vpx_dsp/x86')
| -rw-r--r-- | vpx_dsp/x86/highbd_variance_sse2.c | 21 |
1 files changed, 11 insertions, 10 deletions
diff --git a/vpx_dsp/x86/highbd_variance_sse2.c b/vpx_dsp/x86/highbd_variance_sse2.c index 81ec5dbdb..ade5eacad 100644 --- a/vpx_dsp/x86/highbd_variance_sse2.c +++ b/vpx_dsp/x86/highbd_variance_sse2.c @@ -9,6 +9,7 @@ */ #include "./vpx_config.h" +#include "vpx_dsp/variance.h" #include "vpx_ports/mem.h" typedef uint32_t (*high_variance_fn_t) (const uint16_t *src, int src_stride, @@ -62,8 +63,8 @@ static void highbd_10_variance_sse2(const uint16_t *src, int src_stride, sum_long += sum0; } } - *sum = ROUND_POWER_OF_TWO(sum_long, 2); - *sse = (uint32_t)ROUND_POWER_OF_TWO(sse_long, 4); + *sum = ROUND_ZERO_POWER_OF_TWO(sum_long, 2); + *sse = (uint32_t)ROUND_ZERO_POWER_OF_TWO(sse_long, 4); } static void highbd_12_variance_sse2(const uint16_t *src, int src_stride, @@ -84,8 +85,8 @@ static void highbd_12_variance_sse2(const uint16_t *src, int src_stride, sum_long += sum0; } } - *sum = ROUND_POWER_OF_TWO(sum_long, 4); - *sse = (uint32_t)ROUND_POWER_OF_TWO(sse_long, 8); + *sum = ROUND_ZERO_POWER_OF_TWO(sum_long, 4); + *sse = (uint32_t)ROUND_ZERO_POWER_OF_TWO(sse_long, 8); } @@ -106,7 +107,7 @@ void vpx_highbd_10_get##S##x##S##var_sse2(const uint8_t *src8, int src_stride, \ uint16_t *ref = CONVERT_TO_SHORTPTR(ref8); \ vpx_highbd_calc##S##x##S##var_sse2(src, src_stride, ref, ref_stride, \ sse, sum); \ - *sum = ROUND_POWER_OF_TWO(*sum, 2); \ + *sum = ROUND_ZERO_POWER_OF_TWO(*sum, 2); \ *sse = ROUND_POWER_OF_TWO(*sse, 4); \ } \ \ @@ -117,7 +118,7 @@ void vpx_highbd_12_get##S##x##S##var_sse2(const uint8_t *src8, int src_stride, \ uint16_t *ref = CONVERT_TO_SHORTPTR(ref8); \ vpx_highbd_calc##S##x##S##var_sse2(src, src_stride, ref, ref_stride, \ sse, sum); \ - *sum = ROUND_POWER_OF_TWO(*sum, 4); \ + *sum = ROUND_ZERO_POWER_OF_TWO(*sum, 4); \ *sse = ROUND_POWER_OF_TWO(*sse, 8); \ } @@ -345,7 +346,7 @@ uint32_t vpx_highbd_10_sub_pixel_variance##w##x##h##_##opt( \ sse += sse2; \ } \ } \ - se = ROUND_POWER_OF_TWO(se, 2); \ + se = ROUND_ZERO_POWER_OF_TWO(se, 2); \ sse = ROUND_POWER_OF_TWO(sse, 4); \ *sse_ptr = sse; \ return sse - ((cast se * se) >> (wlog2 + hlog2)); \ @@ -392,7 +393,7 @@ uint32_t vpx_highbd_12_sub_pixel_variance##w##x##h##_##opt( \ }\ } \ } \ - se = ROUND_POWER_OF_TWO(se, 4); \ + se = ROUND_ZERO_POWER_OF_TWO(se, 4); \ sse = (uint32_t)ROUND_POWER_OF_TWO(long_sse, 8); \ *sse_ptr = sse; \ return sse - ((cast se * se) >> (wlog2 + hlog2)); \ @@ -514,7 +515,7 @@ uint32_t vpx_highbd_10_sub_pixel_avg_variance##w##x##h##_##opt( \ sse += sse2; \ } \ } \ - se = ROUND_POWER_OF_TWO(se, 2); \ + se = ROUND_ZERO_POWER_OF_TWO(se, 2); \ sse = ROUND_POWER_OF_TWO(sse, 4); \ *sse_ptr = sse; \ return sse - ((cast se * se) >> (wlog2 + hlog2)); \ @@ -566,7 +567,7 @@ uint32_t vpx_highbd_12_sub_pixel_avg_variance##w##x##h##_##opt( \ } \ } \ } \ - se = ROUND_POWER_OF_TWO(se, 4); \ + se = ROUND_ZERO_POWER_OF_TWO(se, 4); \ sse = (uint32_t)ROUND_POWER_OF_TWO(long_sse, 8); \ *sse_ptr = sse; \ return sse - ((cast se * se) >> (wlog2 + hlog2)); \ |