diff options
| author | plepere <pierre-edouard.lepere@insa-rennes.fr> | 2014-05-15 16:31:11 +0200 |
|---|---|---|
| committer | Michael Niedermayer <michaelni@gmx.at> | 2014-05-16 21:11:03 +0200 |
| commit | de7b89fd43f850d77cf24ad6ae50185dfe391e91 (patch) | |
| tree | b09a832ce2711d357806cc3a2290c920f6254891 /libavcodec/x86/hevcdsp_init.c | |
| parent | 7ebf90ff472d05c7179f202df8686d84adfcb3d2 (diff) | |
| download | ffmpeg-de7b89fd43f850d77cf24ad6ae50185dfe391e91.tar.gz | |
avcodec/x86/hevc: added DBF assembly functions
Reviewed-by: James Almer <jamrial@gmail.com>
Reviewed-by: Ronald S. Bultje
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
Diffstat (limited to 'libavcodec/x86/hevcdsp_init.c')
| -rw-r--r-- | libavcodec/x86/hevcdsp_init.c | 37 |
1 files changed, 37 insertions, 0 deletions
diff --git a/libavcodec/x86/hevcdsp_init.c b/libavcodec/x86/hevcdsp_init.c index 61ac6a98dc..e5ef28759b 100644 --- a/libavcodec/x86/hevcdsp_init.c +++ b/libavcodec/x86/hevcdsp_init.c @@ -29,6 +29,27 @@ #include "libavcodec/x86/hevcdsp.h" +#define LFC_FUNC(DIR, DEPTH, OPT) \ +void ff_hevc_ ## DIR ## _loop_filter_chroma_ ## DEPTH ## _ ## OPT(uint8_t *_pix, ptrdiff_t _stride, int *_tc, uint8_t *_no_p, uint8_t *_no_q); + +#define LFL_FUNC(DIR, DEPTH, OPT) \ +void ff_hevc_ ## DIR ## _loop_filter_luma_ ## DEPTH ## _ ## OPT(uint8_t *_pix, ptrdiff_t stride, int *_beta, int *_tc, \ +uint8_t *_no_p, uint8_t *_no_q); + +#define LFC_FUNCS(type, depth) \ +LFC_FUNC(h, depth, sse2) \ +LFC_FUNC(v, depth, sse2) + +#define LFL_FUNCS(type, depth) \ +LFL_FUNC(h, depth, ssse3) \ +LFL_FUNC(v, depth, ssse3) + +LFC_FUNCS(uint8_t, 8) +LFC_FUNCS(uint8_t, 10) +LFL_FUNCS(uint8_t, 8) +LFL_FUNCS(uint8_t, 10) + + #define mc_rep_func(name, bitd, step, W, opt) \ void ff_hevc_put_hevc_##name##W##_##bitd##_##opt(int16_t *_dst, ptrdiff_t dststride, \ uint8_t *_src, ptrdiff_t _srcstride, int height, \ @@ -347,6 +368,14 @@ void ff_hevcdsp_init_x86(HEVCDSPContext *c, const int bit_depth) int mm_flags = av_get_cpu_flags(); if (bit_depth == 8) { + if (EXTERNAL_SSE2(mm_flags)) { + c->hevc_v_loop_filter_chroma = ff_hevc_v_loop_filter_chroma_8_sse2; + c->hevc_h_loop_filter_chroma = ff_hevc_h_loop_filter_chroma_8_sse2; + } + if (EXTERNAL_SSSE3(mm_flags) && ARCH_X86_64) { + c->hevc_v_loop_filter_luma = ff_hevc_v_loop_filter_luma_8_ssse3; + c->hevc_h_loop_filter_luma = ff_hevc_h_loop_filter_luma_8_ssse3; + } if (EXTERNAL_SSE4(mm_flags) && ARCH_X86_64) { EPEL_LINKS(c->put_hevc_epel, 0, 0, pel_pixels, 8, sse4); @@ -361,6 +390,14 @@ void ff_hevcdsp_init_x86(HEVCDSPContext *c, const int bit_depth) } } else if (bit_depth == 10) { + if (EXTERNAL_SSE2(mm_flags)) { + c->hevc_v_loop_filter_chroma = ff_hevc_v_loop_filter_chroma_10_sse2; + c->hevc_h_loop_filter_chroma = ff_hevc_h_loop_filter_chroma_10_sse2; + } + if (EXTERNAL_SSSE3(mm_flags) && ARCH_X86_64) { + c->hevc_v_loop_filter_luma = ff_hevc_v_loop_filter_luma_10_ssse3; + c->hevc_h_loop_filter_luma = ff_hevc_h_loop_filter_luma_10_ssse3; + } if (EXTERNAL_SSE4(mm_flags) && ARCH_X86_64) { EPEL_LINKS(c->put_hevc_epel, 0, 0, pel_pixels, 10, sse4); |