diff options
| author | James Zern <jzern@google.com> | 2015-04-25 15:59:58 -0700 |
|---|---|---|
| committer | James Zern <jzern@google.com> | 2015-04-30 19:14:17 -0700 |
| commit | c77b1f5acd09852aff1ba09d7f371728a60634d7 (patch) | |
| tree | 8bba66034836498a3aa559c8ec8e3cac7b8ba628 | |
| parent | 778845da05b2d92e9802839bd5b438ea6112de16 (diff) | |
| download | libvpx-c77b1f5acd09852aff1ba09d7f371728a60634d7.tar.gz | |
vp9: RECON_AND_STORE4X4: remove dest offset
offsetting by a variable stride prevents instruction reordering,
resulting in poor assembly
Change-Id: Id62d6b3299cdd23f8c44f97b630abf4fea241446
| -rw-r--r-- | vp9/common/x86/vp9_idct_intrin_sse2.c | 11 |
1 files changed, 5 insertions, 6 deletions
diff --git a/vp9/common/x86/vp9_idct_intrin_sse2.c b/vp9/common/x86/vp9_idct_intrin_sse2.c index 1a641446b..824f2d043 100644 --- a/vp9/common/x86/vp9_idct_intrin_sse2.c +++ b/vp9/common/x86/vp9_idct_intrin_sse2.c @@ -17,8 +17,7 @@ d0 = _mm_unpacklo_epi8(d0, zero); \ d0 = _mm_add_epi16(in_x, d0); \ d0 = _mm_packus_epi16(d0, d0); \ - *(int *)dest = _mm_cvtsi128_si32(d0); \ - dest += stride; \ + *(int *)(dest) = _mm_cvtsi128_si32(d0); \ } void vp9_idct4x4_16_add_sse2(const int16_t *input, uint8_t *dest, int stride) { @@ -162,10 +161,10 @@ void vp9_idct4x4_1_add_sse2(const int16_t *input, uint8_t *dest, int stride) { dc_value = _mm_set1_epi16(a); - RECON_AND_STORE4X4(dest, dc_value); - RECON_AND_STORE4X4(dest, dc_value); - RECON_AND_STORE4X4(dest, dc_value); - RECON_AND_STORE4X4(dest, dc_value); + RECON_AND_STORE4X4(dest + 0 * stride, dc_value); + RECON_AND_STORE4X4(dest + 1 * stride, dc_value); + RECON_AND_STORE4X4(dest + 2 * stride, dc_value); + RECON_AND_STORE4X4(dest + 3 * stride, dc_value); } static INLINE void transpose_4x4(__m128i *res) { |