diff options
Diffstat (limited to 'libavcodec/x86/dsputil.asm')
-rw-r--r-- | libavcodec/x86/dsputil.asm | 60 |
1 files changed, 60 insertions, 0 deletions
diff --git a/libavcodec/x86/dsputil.asm b/libavcodec/x86/dsputil.asm index c1ea9bf4e2..c91dd8eb69 100644 --- a/libavcodec/x86/dsputil.asm +++ b/libavcodec/x86/dsputil.asm @@ -513,3 +513,63 @@ BSWAP32_BUF INIT_XMM ssse3 BSWAP32_BUF + +;---------------------------------------- +; void ff_clear_block(int16_t *blocks); +;---------------------------------------- +; %1 = number of xmm registers used +; %2 = number of inline store loops +%macro CLEAR_BLOCK 2 +cglobal clear_block, 1, 1, %1, blocks + ZERO m0, m0 +%assign %%i 0 +%rep %2 + mova [blocksq+mmsize*(0+%%i)], m0 + mova [blocksq+mmsize*(1+%%i)], m0 + mova [blocksq+mmsize*(2+%%i)], m0 + mova [blocksq+mmsize*(3+%%i)], m0 + mova [blocksq+mmsize*(4+%%i)], m0 + mova [blocksq+mmsize*(5+%%i)], m0 + mova [blocksq+mmsize*(6+%%i)], m0 + mova [blocksq+mmsize*(7+%%i)], m0 +%assign %%i %%i+8 +%endrep + RET +%endmacro + +INIT_MMX mmx +%define ZERO pxor +CLEAR_BLOCK 0, 2 +INIT_XMM sse +%define ZERO xorps +CLEAR_BLOCK 1, 1 + +;----------------------------------------- +; void ff_clear_blocks(int16_t *blocks); +;----------------------------------------- +; %1 = number of xmm registers used +%macro CLEAR_BLOCKS 1 +cglobal clear_blocks, 1, 2, %1, blocks, len + add blocksq, 768 + mov lenq, -768 + ZERO m0, m0 +.loop + mova [blocksq+lenq+mmsize*0], m0 + mova [blocksq+lenq+mmsize*1], m0 + mova [blocksq+lenq+mmsize*2], m0 + mova [blocksq+lenq+mmsize*3], m0 + mova [blocksq+lenq+mmsize*4], m0 + mova [blocksq+lenq+mmsize*5], m0 + mova [blocksq+lenq+mmsize*6], m0 + mova [blocksq+lenq+mmsize*7], m0 + add lenq, mmsize*8 + js .loop + RET +%endmacro + +INIT_MMX mmx +%define ZERO pxor +CLEAR_BLOCKS 0 +INIT_XMM sse +%define ZERO xorps +CLEAR_BLOCKS 1 |