diff options
| author | James Almer <jamrial@gmail.com> | 2017-01-31 15:36:49 -0300 |
|---|---|---|
| committer | James Almer <jamrial@gmail.com> | 2017-01-31 15:36:49 -0300 |
| commit | ba5d08938130b2545f8b3e73a701989563316e49 (patch) | |
| tree | 1a445dcfab3b46302f48e631dd6208095c182f5c /libavcodec/x86 | |
| parent | 8d5df204d002bef25aadc4d7667e32c4fecaa2b7 (diff) | |
| parent | d06dfaa5cbdd20acfd2364b16c0f4ae4ddb30a65 (diff) | |
Merge commit 'd06dfaa5cbdd20acfd2364b16c0f4ae4ddb30a65'
* commit 'd06dfaa5cbdd20acfd2364b16c0f4ae4ddb30a65':
x86: huffyuv: Use EXTERNAL_SSSE3_FAST convenience macro where appropriate
Merged-by: James Almer <jamrial@gmail.com>
Diffstat (limited to 'libavcodec/x86')
| -rw-r--r-- | libavcodec/x86/lossless_videodsp.asm | 4 | ||||
| -rw-r--r-- | libavcodec/x86/lossless_videodsp_init.c | 11 |
2 files changed, 8 insertions, 7 deletions
diff --git a/libavcodec/x86/lossless_videodsp.asm b/libavcodec/x86/lossless_videodsp.asm index 7fcae898b5..443fe02951 100644 --- a/libavcodec/x86/lossless_videodsp.asm +++ b/libavcodec/x86/lossless_videodsp.asm @@ -160,8 +160,8 @@ cglobal add_left_pred, 3,3,7, dst, src, w, left psllq m0, 56 ADD_LEFT_LOOP 1, 1 -INIT_XMM sse4 -cglobal add_left_pred, 3,3,7, dst, src, w, left +INIT_XMM ssse3 +cglobal add_left_pred_unaligned, 3,3,7, dst, src, w, left mova m5, [pb_15] mova m6, [pb_zzzzzzzz77777777] mova m4, [pb_zzzz3333zzzzbbbb] diff --git a/libavcodec/x86/lossless_videodsp_init.c b/libavcodec/x86/lossless_videodsp_init.c index 58715e290a..8d583447dc 100644 --- a/libavcodec/x86/lossless_videodsp_init.c +++ b/libavcodec/x86/lossless_videodsp_init.c @@ -35,8 +35,8 @@ void ff_add_median_pred_sse2(uint8_t *dst, const uint8_t *top, int ff_add_left_pred_ssse3(uint8_t *dst, const uint8_t *src, intptr_t w, int left); -int ff_add_left_pred_sse4(uint8_t *dst, const uint8_t *src, - intptr_t w, int left); +int ff_add_left_pred_unaligned_ssse3(uint8_t *dst, const uint8_t *src, + intptr_t w, int left); int ff_add_left_pred_int16_ssse3(uint16_t *dst, const uint16_t *src, unsigned mask, int w, unsigned acc); int ff_add_left_pred_int16_sse4(uint16_t *dst, const uint16_t *src, unsigned mask, int w, unsigned acc); @@ -105,12 +105,13 @@ void ff_llviddsp_init_x86(LLVidDSPContext *c) if (EXTERNAL_SSSE3(cpu_flags)) { c->add_left_pred = ff_add_left_pred_ssse3; - if (cpu_flags & AV_CPU_FLAG_SSE4) // not really SSE4, just slow on Conroe - c->add_left_pred = ff_add_left_pred_sse4; - c->add_left_pred_int16 = ff_add_left_pred_int16_ssse3; } + if (EXTERNAL_SSSE3_FAST(cpu_flags)) { + c->add_left_pred = ff_add_left_pred_unaligned_ssse3; + } + if (EXTERNAL_SSE4(cpu_flags)) { c->add_left_pred_int16 = ff_add_left_pred_int16_sse4; } |