diff options
| author | Michael Niedermayer <michaelni@gmx.at> | 2006-10-10 01:08:39 +0000 |
|---|---|---|
| committer | Michael Niedermayer <michaelni@gmx.at> | 2006-10-10 01:08:39 +0000 |
| commit | ab0151d163287160038052778ebf21b5529c12fd (patch) | |
| tree | 0d3e7ef0e8bdd69f61d9828f375ee56475b7056d /libavcodec | |
| parent | c6a9271c564c35211498e9bef2309cd8f63814a4 (diff) | |
| download | ffmpeg-ab0151d163287160038052778ebf21b5529c12fd.tar.gz | |
replace a few and/sub/... by cmov
this is faster on P3, should be faster on AMD, and should be slower on P4
its disabled by default (benchmarks welcome so we know when to enable it)
Originally committed as revision 6615 to svn://svn.ffmpeg.org/ffmpeg/trunk
Diffstat (limited to 'libavcodec')
| -rw-r--r-- | libavcodec/cabac.h | 11 |
1 files changed, 10 insertions, 1 deletions
diff --git a/libavcodec/cabac.h b/libavcodec/cabac.h index 01163915a7..e35f6ea628 100644 --- a/libavcodec/cabac.h +++ b/libavcodec/cabac.h @@ -459,6 +459,14 @@ static int get_cabac(CABACContext *c, uint8_t * const state){ "movl "LOW "(%2), %%ebx \n\t" //eax:state ebx:low, edx:range, esi:RangeLPS "subl %%esi, %%edx \n\t" +#ifdef CMOV_IS_FAST //FIXME actually define this somewhere + "cmpl %%ebx, %%edx \n\t" + "cmova %%edx, %%esi \n\t" + "sbbl %%ecx, %%ecx \n\t" + "andl %%ecx, %%edx \n\t" + "subl %%edx, %%ebx \n\t" + "xorl %%ecx, %%eax \n\t" +#else "movl %%edx, %%ecx \n\t" "subl %%ebx, %%edx \n\t" "sarl $31, %%edx \n\t" //lps_mask @@ -467,9 +475,10 @@ static int get_cabac(CABACContext *c, uint8_t * const state){ "addl %%ecx, %%esi \n\t" //new range "andl %%edx, %%ecx \n\t" "subl %%ecx, %%ebx \n\t" + "xorl %%edx, %%eax \n\t" +#endif //eax:state ebx:low edx:mask esi:range - "xorl %%edx, %%eax \n\t" "movzbl "MPS_STATE"(%2, %%eax), %%ecx \n\t" "movb %%cl, (%1) \n\t" |