diff options
| author | Christophe Gisquet <Christophe.Gisquet@gmail.com> | 2008-01-11 08:29:58 +0000 |
|---|---|---|
| committer | Diego Biurrun <diego@biurrun.de> | 2008-01-11 08:29:58 +0000 |
| commit | 28748a912887bb1799f5fafb587a03e051f11201 (patch) | |
| tree | 6e33dcb10e7564dc2bb70f5254873b76d4e4b3d0 /libavcodec/i386/dsputil_mmx.h | |
| parent | c567c6945f23f5dc32bda2b4d30171e6d64586ca (diff) | |
| download | ffmpeg-28748a912887bb1799f5fafb587a03e051f11201.tar.gz | |
Factorize some duplicated code from CAVS and H.264 into a common file.
patch by Christophe Gisquet, christophe.gisquet free fr
Originally committed as revision 11504 to svn://svn.ffmpeg.org/ffmpeg/trunk
Diffstat (limited to 'libavcodec/i386/dsputil_mmx.h')
| -rw-r--r-- | libavcodec/i386/dsputil_mmx.h | 57 |
1 files changed, 57 insertions, 0 deletions
diff --git a/libavcodec/i386/dsputil_mmx.h b/libavcodec/i386/dsputil_mmx.h index fc5d01e70b..a1571fc1e3 100644 --- a/libavcodec/i386/dsputil_mmx.h +++ b/libavcodec/i386/dsputil_mmx.h @@ -52,4 +52,61 @@ extern const uint64_t ff_pb_FC; extern const double ff_pd_1[2]; extern const double ff_pd_2[2]; +/* in/out: mma=mma+mmb, mmb=mmb-mma */ +#define SUMSUB_BA( a, b ) \ + "paddw "#b", "#a" \n\t"\ + "paddw "#b", "#b" \n\t"\ + "psubw "#a", "#b" \n\t" + +#define SBUTTERFLY(a,b,t,n,m)\ + "mov" #m " " #a ", " #t " \n\t" /* abcd */\ + "punpckl" #n " " #b ", " #a " \n\t" /* aebf */\ + "punpckh" #n " " #b ", " #t " \n\t" /* cgdh */\ + +#define TRANSPOSE4(a,b,c,d,t)\ + SBUTTERFLY(a,b,t,wd,q) /* a=aebf t=cgdh */\ + SBUTTERFLY(c,d,b,wd,q) /* c=imjn b=kolp */\ + SBUTTERFLY(a,c,d,dq,q) /* a=aeim d=bfjn */\ + SBUTTERFLY(t,b,c,dq,q) /* t=cgko c=dhlp */ + +#ifdef ARCH_X86_64 +// permutes 01234567 -> 05736421 +#define TRANSPOSE8(a,b,c,d,e,f,g,h,t)\ + SBUTTERFLY(a,b,%%xmm8,wd,dqa)\ + SBUTTERFLY(c,d,b,wd,dqa)\ + SBUTTERFLY(e,f,d,wd,dqa)\ + SBUTTERFLY(g,h,f,wd,dqa)\ + SBUTTERFLY(a,c,h,dq,dqa)\ + SBUTTERFLY(%%xmm8,b,c,dq,dqa)\ + SBUTTERFLY(e,g,b,dq,dqa)\ + SBUTTERFLY(d,f,g,dq,dqa)\ + SBUTTERFLY(a,e,f,qdq,dqa)\ + SBUTTERFLY(%%xmm8,d,e,qdq,dqa)\ + SBUTTERFLY(h,b,d,qdq,dqa)\ + SBUTTERFLY(c,g,b,qdq,dqa)\ + "movdqa %%xmm8, "#g" \n\t" +#else +#define TRANSPOSE8(a,b,c,d,e,f,g,h,t)\ + "movdqa "#h", "#t" \n\t"\ + SBUTTERFLY(a,b,h,wd,dqa)\ + "movdqa "#h", 16"#t" \n\t"\ + "movdqa "#t", "#h" \n\t"\ + SBUTTERFLY(c,d,b,wd,dqa)\ + SBUTTERFLY(e,f,d,wd,dqa)\ + SBUTTERFLY(g,h,f,wd,dqa)\ + SBUTTERFLY(a,c,h,dq,dqa)\ + "movdqa "#h", "#t" \n\t"\ + "movdqa 16"#t", "#h" \n\t"\ + SBUTTERFLY(h,b,c,dq,dqa)\ + SBUTTERFLY(e,g,b,dq,dqa)\ + SBUTTERFLY(d,f,g,dq,dqa)\ + SBUTTERFLY(a,e,f,qdq,dqa)\ + SBUTTERFLY(h,d,e,qdq,dqa)\ + "movdqa "#h", 16"#t" \n\t"\ + "movdqa "#t", "#h" \n\t"\ + SBUTTERFLY(h,b,d,qdq,dqa)\ + SBUTTERFLY(c,g,b,qdq,dqa)\ + "movdqa 16"#t", "#g" \n\t" +#endif + #endif /* FFMPEG_DSPUTIL_MMX_H */ |