Mercurial > libavcodec.hg
changeset 3213:57d31bdbebe8 libavcodec
added mmx implementation of h264_chroma_mc2
| author | lorenm |
|---|---|
| date | Wed, 22 Mar 2006 22:05:00 +0000 |
| parents | 5fb704618ec4 |
| children | 91f89a395b28 |
| files | i386/dsputil_h264_template_mmx.c i386/dsputil_mmx.c i386/h264dsp_mmx.c |
| diffstat | 3 files changed, 70 insertions(+), 1 deletions(-) [+] |
line wrap: on
line diff
--- a/i386/dsputil_h264_template_mmx.c Wed Mar 22 12:41:02 2006 +0000 +++ b/i386/dsputil_h264_template_mmx.c Wed Mar 22 22:05:00 2006 +0000 @@ -240,7 +240,7 @@ } } -static void H264_CHROMA_MC4_TMPL(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y) +static void H264_CHROMA_MC4_TMPL(uint8_t *dst/*align 4*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y) { DECLARE_ALIGNED_8(uint64_t, AA); DECLARE_ALIGNED_8(uint64_t, DD); @@ -319,3 +319,66 @@ dst += stride; } } + +#ifdef H264_CHROMA_MC2_TMPL +static void H264_CHROMA_MC2_TMPL(uint8_t *dst/*align 2*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y) +{ + int CD=((1<<16)-1)*x*y + 8*y; + int AB=((8<<16)-8)*x + 64 - CD; + int i; + + asm volatile( + /* mm5 = {A,B,A,B} */ + /* mm6 = {C,D,C,D} */ + "movd %0, %%mm5\n\t" + "movd %1, %%mm6\n\t" + "punpckldq %%mm5, %%mm5\n\t" + "punpckldq %%mm6, %%mm6\n\t" + "pxor %%mm7, %%mm7\n\t" + :: "r"(AB), "r"(CD)); + + asm volatile( + /* mm0 = src[0,1,1,2] */ + "movd %0, %%mm0\n\t" + "punpcklbw %%mm7, %%mm0\n\t" + "pshufw $0x94, %%mm0, %%mm0\n\t" + :: "m"(src[0])); + + for(i=0; i<h; i++) { + asm volatile( + /* mm1 = A * src[0,1] + B * src[1,2] */ + "movq %%mm0, %%mm1\n\t" + "pmaddwd %%mm5, %%mm1\n\t" + ::); + + src += stride; + asm volatile( + /* mm0 = src[0,1,1,2] */ + "movd %0, %%mm0\n\t" + "punpcklbw %%mm7, %%mm0\n\t" + "pshufw $0x94, %%mm0, %%mm0\n\t" + :: "m"(src[0])); + + asm volatile( + /* mm1 += C * src[0,1] + D * src[1,2] */ + "movq %%mm0, %%mm2\n\t" + "pmaddwd %%mm6, %%mm2\n\t" + "paddw %%mm2, %%mm1\n\t" + ::); + + asm volatile( + /* dst[0,1] = pack((mm1 + 32) >> 6) */ + "paddw %1, %%mm1\n\t" + "psrlw $6, %%mm1\n\t" + "packssdw %%mm7, %%mm1\n\t" + "packuswb %%mm7, %%mm1\n\t" + /* writes garbage to the right of dst. + * ok because partitions are processed from left to right. */ + H264_CHROMA_OP4(%0, %%mm1, %%mm3) + "movd %%mm1, %0\n\t" + : "=m" (dst[0]) : "m" (ff_pw_32) : "eax"); + dst += stride; + } +} +#endif +
--- a/i386/dsputil_mmx.c Wed Mar 22 12:41:02 2006 +0000 +++ b/i386/dsputil_mmx.c Wed Mar 22 22:05:00 2006 +0000 @@ -2848,6 +2848,8 @@ c->avg_h264_chroma_pixels_tab[0]= avg_h264_chroma_mc8_mmx2; c->avg_h264_chroma_pixels_tab[1]= avg_h264_chroma_mc4_mmx2; + c->avg_h264_chroma_pixels_tab[2]= avg_h264_chroma_mc2_mmx2; + c->put_h264_chroma_pixels_tab[2]= put_h264_chroma_mc2_mmx2; c->h264_v_loop_filter_luma= h264_v_loop_filter_luma_mmx2; c->h264_h_loop_filter_luma= h264_h_loop_filter_luma_mmx2; c->h264_v_loop_filter_chroma= h264_v_loop_filter_chroma_mmx2;
--- a/i386/h264dsp_mmx.c Wed Mar 22 12:41:02 2006 +0000 +++ b/i386/h264dsp_mmx.c Wed Mar 22 22:05:00 2006 +0000 @@ -1296,12 +1296,14 @@ #define H264_CHROMA_OP4(S,D,T) #define H264_CHROMA_MC8_TMPL put_h264_chroma_mc8_mmx #define H264_CHROMA_MC4_TMPL put_h264_chroma_mc4_mmx +#define H264_CHROMA_MC2_TMPL put_h264_chroma_mc2_mmx2 #define H264_CHROMA_MC8_MV0 put_pixels8_mmx #include "dsputil_h264_template_mmx.c" #undef H264_CHROMA_OP #undef H264_CHROMA_OP4 #undef H264_CHROMA_MC8_TMPL #undef H264_CHROMA_MC4_TMPL +#undef H264_CHROMA_MC2_TMPL #undef H264_CHROMA_MC8_MV0 #define H264_CHROMA_OP(S,D) "pavgb " #S ", " #D " \n\t" @@ -1309,12 +1311,14 @@ "pavgb " #T ", " #D " \n\t" #define H264_CHROMA_MC8_TMPL avg_h264_chroma_mc8_mmx2 #define H264_CHROMA_MC4_TMPL avg_h264_chroma_mc4_mmx2 +#define H264_CHROMA_MC2_TMPL avg_h264_chroma_mc2_mmx2 #define H264_CHROMA_MC8_MV0 avg_pixels8_mmx2 #include "dsputil_h264_template_mmx.c" #undef H264_CHROMA_OP #undef H264_CHROMA_OP4 #undef H264_CHROMA_MC8_TMPL #undef H264_CHROMA_MC4_TMPL +#undef H264_CHROMA_MC2_TMPL #undef H264_CHROMA_MC8_MV0 #define H264_CHROMA_OP(S,D) "pavgusb " #S ", " #D " \n\t"