diff options
| author | michael <michael@b3059339-0415-0410-9bf9-f77b7e298cf2> | 2001-12-23 17:16:07 +0000 |
|---|---|---|
| committer | michael <michael@b3059339-0415-0410-9bf9-f77b7e298cf2> | 2001-12-23 17:16:07 +0000 |
| commit | c2aab979095885ccade4813dda3eeacfb51c4693 (patch) | |
| tree | 11cd9a75e05335fa3e0e7cb8748b97c512595289 /liba52 | |
| parent | 1ce2b453224d971d8f0bf0bb32db3255d5fe6860 (diff) | |
| download | mpv-c2aab979095885ccade4813dda3eeacfb51c4693.tar.bz2 mpv-c2aab979095885ccade4813dda3eeacfb51c4693.tar.xz | |
mix5to1 & mix4to1 in SSE
git-svn-id: svn://svn.mplayerhq.hu/mplayer/trunk@3678 b3059339-0415-0410-9bf9-f77b7e298cf2
Diffstat (limited to 'liba52')
| -rw-r--r-- | liba52/downmix.c | 41 |
1 files changed, 41 insertions, 0 deletions
diff --git a/liba52/downmix.c b/liba52/downmix.c index ac8a4b33fa..4f1ba231e8 100644 --- a/liba52/downmix.c +++ b/liba52/downmix.c @@ -342,18 +342,59 @@ static void mix4to1 (sample_t * samples, sample_t bias) { int i; +#ifdef HAVE_SSE // FIXME untested + asm volatile( + "movlps %1, %%xmm7 \n\t" + "shufps $0x00, %%xmm7, %%xmm7 \n\t" + "movl $-1024, %%esi \n\t" + "1: \n\t" + "movaps (%0, %%esi), %%xmm0 \n\t" + "movaps 1024(%0, %%esi), %%xmm1 \n\t" + "addps 2048(%0, %%esi), %%xmm0 \n\t" + "addps 3072(%0, %%esi), %%xmm1 \n\t" + "addps %%xmm7, %%xmm0 \n\t" + "addps %%xmm1, %%xmm0 \n\t" + "movaps %%xmm0, (%0, %%esi) \n\t" + "addl $16, %%esi \n\t" + " jnz 1b \n\t" + :: "r" (samples+256), "m" (bias) + : "%esi" + ); +#else for (i = 0; i < 256; i++) samples[i] += (samples[i + 256] + samples[i + 512] + samples[i + 768] + bias); +#endif } static void mix5to1 (sample_t * samples, sample_t bias) { int i; +#ifdef HAVE_SSE + asm volatile( + "movlps %1, %%xmm7 \n\t" + "shufps $0x00, %%xmm7, %%xmm7 \n\t" + "movl $-1024, %%esi \n\t" + "1: \n\t" + "movaps (%0, %%esi), %%xmm0 \n\t" + "movaps 1024(%0, %%esi), %%xmm1 \n\t" + "addps 2048(%0, %%esi), %%xmm0 \n\t" + "addps 3072(%0, %%esi), %%xmm1 \n\t" + "addps %%xmm7, %%xmm0 \n\t" + "addps 4096(%0, %%esi), %%xmm1 \n\t" + "addps %%xmm1, %%xmm0 \n\t" + "movaps %%xmm0, (%0, %%esi) \n\t" + "addl $16, %%esi \n\t" + " jnz 1b \n\t" + :: "r" (samples+256), "m" (bias) + : "%esi" + ); +#else for (i = 0; i < 256; i++) samples[i] += (samples[i + 256] + samples[i + 512] + samples[i + 768] + samples[i + 1024] + bias); +#endif } static void mix3to2 (sample_t * samples, sample_t bias) |