From 1878f685c0f69d1bf0acc78c5fc09dae03ac48d5 Mon Sep 17 00:00:00 2001 From: "Ronald S. Bultje" Date: Mon, 19 Jul 2010 21:53:28 +0000 Subject: Implement chroma (width=8) inner loopfilter MMX/MMX2/SSE2 functions. Originally committed as revision 24339 to svn://svn.ffmpeg.org/ffmpeg/trunk --- libavcodec/x86/vp8dsp-init.c | 19 +++++++++++++++++++ 1 file changed, 19 insertions(+) (limited to 'libavcodec/x86/vp8dsp-init.c') diff --git a/libavcodec/x86/vp8dsp-init.c b/libavcodec/x86/vp8dsp-init.c index dbaa0dc0a5..c7b02d1541 100644 --- a/libavcodec/x86/vp8dsp-init.c +++ b/libavcodec/x86/vp8dsp-init.c @@ -242,6 +242,19 @@ extern void ff_vp8_h_loop_filter16y_inner_mmxext(uint8_t *dst, int stride, int e, int i, int hvt); extern void ff_vp8_h_loop_filter16y_inner_sse2 (uint8_t *dst, int stride, int e, int i, int hvt); + +extern void ff_vp8_v_loop_filter8uv_inner_mmx (uint8_t *dstU, uint8_t *dstV, + int s, int e, int i, int hvt); +extern void ff_vp8_v_loop_filter8uv_inner_mmxext(uint8_t *dstU, uint8_t *dstV, + int s, int e, int i, int hvt); +extern void ff_vp8_v_loop_filter8uv_inner_sse2 (uint8_t *dstU, uint8_t *dstV, + int s, int e, int i, int hvt); +extern void ff_vp8_h_loop_filter8uv_inner_mmx (uint8_t *dstU, uint8_t *dstV, + int s, int e, int i, int hvt); +extern void ff_vp8_h_loop_filter8uv_inner_mmxext(uint8_t *dstU, uint8_t *dstV, + int s, int e, int i, int hvt); +extern void ff_vp8_h_loop_filter8uv_inner_sse2 (uint8_t *dstU, uint8_t *dstV, + int s, int e, int i, int hvt); #endif #define VP8_LUMA_MC_FUNC(IDX, SIZE, OPT) \ @@ -286,6 +299,8 @@ av_cold void ff_vp8dsp_init_x86(VP8DSPContext* c) c->vp8_v_loop_filter16y_inner = ff_vp8_v_loop_filter16y_inner_mmx; c->vp8_h_loop_filter16y_inner = ff_vp8_h_loop_filter16y_inner_mmx; + c->vp8_v_loop_filter8uv_inner = ff_vp8_v_loop_filter8uv_inner_mmx; + c->vp8_h_loop_filter8uv_inner = ff_vp8_h_loop_filter8uv_inner_mmx; } /* note that 4-tap width=16 functions are missing because w=16 @@ -304,6 +319,8 @@ av_cold void ff_vp8dsp_init_x86(VP8DSPContext* c) c->vp8_v_loop_filter16y_inner = ff_vp8_v_loop_filter16y_inner_mmxext; c->vp8_h_loop_filter16y_inner = ff_vp8_h_loop_filter16y_inner_mmxext; + c->vp8_v_loop_filter8uv_inner = ff_vp8_v_loop_filter8uv_inner_mmxext; + c->vp8_h_loop_filter8uv_inner = ff_vp8_h_loop_filter8uv_inner_mmxext; } if (mm_flags & FF_MM_SSE) { @@ -322,6 +339,8 @@ av_cold void ff_vp8dsp_init_x86(VP8DSPContext* c) c->vp8_v_loop_filter16y_inner = ff_vp8_v_loop_filter16y_inner_sse2; c->vp8_h_loop_filter16y_inner = ff_vp8_h_loop_filter16y_inner_sse2; + c->vp8_v_loop_filter8uv_inner = ff_vp8_v_loop_filter8uv_inner_sse2; + c->vp8_h_loop_filter8uv_inner = ff_vp8_h_loop_filter8uv_inner_sse2; } if (mm_flags & FF_MM_SSSE3) { -- cgit v1.2.3