Mercurial > libavcodec.hg
diff x86/vp8dsp-init.c @ 12054:b8f80fe02861 libavcodec
SSSE3 versions of width4 VP8 6-tap MC functions
Also make some small changes to saturation order of 4-tap SSSE3 MC to fix a
non-bitexactness bug.
Patch mostly by Eli Friedman <eli.friedman AT gmail DOT com>.
| author | darkshikari |
|---|---|
| date | Fri, 02 Jul 2010 05:27:41 +0000 |
| parents | dc4feabd4dab |
| children | 8527154f6e81 |
line wrap: on
line diff
--- a/x86/vp8dsp-init.c Fri Jul 02 00:30:27 2010 +0000 +++ b/x86/vp8dsp-init.c Fri Jul 02 05:27:41 2010 +0000 @@ -54,6 +54,18 @@ uint8_t *src, int srcstride, int height, int mx, int my); +extern void ff_put_vp8_epel4_h4_ssse3 (uint8_t *dst, int dststride, + uint8_t *src, int srcstride, + int height, int mx, int my); +extern void ff_put_vp8_epel4_h6_ssse3 (uint8_t *dst, int dststride, + uint8_t *src, int srcstride, + int height, int mx, int my); +extern void ff_put_vp8_epel4_v4_ssse3 (uint8_t *dst, int dststride, + uint8_t *src, int srcstride, + int height, int mx, int my); +extern void ff_put_vp8_epel4_v6_ssse3 (uint8_t *dst, int dststride, + uint8_t *src, int srcstride, + int height, int mx, int my); extern void ff_put_vp8_epel8_h4_ssse3 (uint8_t *dst, int dststride, uint8_t *src, int srcstride, int height, int mx, int my); @@ -173,6 +185,11 @@ HVTAPSSE2(6, 6, 8) HVTAPSSE2(6, 6, 16) +HVTAP(ssse3, 16, 4, 4, 4, 8) +HVTAP(ssse3, 16, 4, 6, 4, 8) +HVTAP(ssse3, 16, 6, 4, 4, 8) +HVTAP(ssse3, 16, 6, 6, 4, 8) + #define HVBILIN(OPT, ALIGN, SIZE, MAXHEIGHT) \ static void ff_put_vp8_bilinear ## SIZE ## _hv_ ## OPT( \ uint8_t *dst, int dststride, uint8_t *src, \ @@ -264,6 +281,7 @@ if (mm_flags & FF_MM_SSSE3) { VP8_LUMA_MC_FUNC(0, 16, ssse3); VP8_MC_FUNC(1, 8, ssse3); + VP8_MC_FUNC(2, 4, ssse3); VP8_BILINEAR_MC_FUNC(0, 16, ssse3); VP8_BILINEAR_MC_FUNC(1, 8, ssse3); }