X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libavcodec%2Fx86%2Fh264dsp_mmx.c;h=68e543681f6b6271979419358cc571944bfb2ece;hb=bb258fb995a42112d1fe14f53ec599b2cd19b707;hp=01791604d6498399bc7feca94dfda63e7b6d67c3;hpb=80d156d7fdc44b09783ba242fe2681a6d4cc8df5;p=ffmpeg diff --git a/libavcodec/x86/h264dsp_mmx.c b/libavcodec/x86/h264dsp_mmx.c index 01791604d64..68e543681f6 100644 --- a/libavcodec/x86/h264dsp_mmx.c +++ b/libavcodec/x86/h264dsp_mmx.c @@ -27,38 +27,61 @@ DECLARE_ALIGNED(8, static const uint64_t, ff_pb_3_1 ) = 0x0103010301030103ULL; /***********************************/ /* IDCT */ +#define IDCT_ADD_FUNC(NUM, DEPTH, OPT) \ +void ff_h264_idct ## NUM ## _add_ ## DEPTH ## _ ## OPT (uint8_t *dst, int16_t *block, int stride); + +IDCT_ADD_FUNC(, 8, mmx) +IDCT_ADD_FUNC(, 10, sse2) +IDCT_ADD_FUNC(_dc, 8, mmx2) +IDCT_ADD_FUNC(_dc, 10, mmx2) +IDCT_ADD_FUNC(8_dc, 8, mmx2) +IDCT_ADD_FUNC(8_dc, 10, sse2) +IDCT_ADD_FUNC(8, 8, mmx) +IDCT_ADD_FUNC(8, 8, sse2) +IDCT_ADD_FUNC(8, 10, sse2) +#if HAVE_AVX +IDCT_ADD_FUNC(, 10, avx) +IDCT_ADD_FUNC(8_dc, 10, avx) +IDCT_ADD_FUNC(8, 10, avx) +#endif + + +#define IDCT_ADD_REP_FUNC(NUM, REP, DEPTH, OPT) \ +void ff_h264_idct ## NUM ## _add ## REP ## _ ## DEPTH ## _ ## OPT \ + (uint8_t *dst, const int *block_offset, \ + DCTELEM *block, int stride, const uint8_t nnzc[6*8]); + +IDCT_ADD_REP_FUNC(8, 4, 8, mmx) +IDCT_ADD_REP_FUNC(8, 4, 8, mmx2) +IDCT_ADD_REP_FUNC(8, 4, 8, sse2) +IDCT_ADD_REP_FUNC(8, 4, 10, sse2) +IDCT_ADD_REP_FUNC(8, 4, 10, avx) +IDCT_ADD_REP_FUNC(, 16, 8, mmx) +IDCT_ADD_REP_FUNC(, 16, 8, mmx2) +IDCT_ADD_REP_FUNC(, 16, 8, sse2) +IDCT_ADD_REP_FUNC(, 16, 10, sse2) +IDCT_ADD_REP_FUNC(, 16intra, 8, mmx) +IDCT_ADD_REP_FUNC(, 16intra, 8, mmx2) +IDCT_ADD_REP_FUNC(, 16intra, 8, sse2) +IDCT_ADD_REP_FUNC(, 16intra, 10, sse2) +#if HAVE_AVX +IDCT_ADD_REP_FUNC(, 16, 10, avx) +IDCT_ADD_REP_FUNC(, 16intra, 10, avx) +#endif + + +#define IDCT_ADD_REP_FUNC2(NUM, REP, DEPTH, OPT) \ +void ff_h264_idct ## NUM ## _add ## REP ## _ ## DEPTH ## _ ## OPT \ + (uint8_t **dst, const int *block_offset, \ + DCTELEM *block, int stride, const uint8_t nnzc[6*8]); +IDCT_ADD_REP_FUNC2(, 8, 8, mmx) +IDCT_ADD_REP_FUNC2(, 8, 8, mmx2) +IDCT_ADD_REP_FUNC2(, 8, 8, sse2) +IDCT_ADD_REP_FUNC2(, 8, 10, sse2) +#if HAVE_AVX +IDCT_ADD_REP_FUNC2(, 8, 10, avx) +#endif -void ff_h264_idct_add_mmx (uint8_t *dst, int16_t *block, int stride); -void ff_h264_idct8_add_mmx (uint8_t *dst, int16_t *block, int stride); -void ff_h264_idct8_add_sse2 (uint8_t *dst, int16_t *block, int stride); -void ff_h264_idct_dc_add_mmx2 (uint8_t *dst, int16_t *block, int stride); -void ff_h264_idct8_dc_add_mmx2(uint8_t *dst, int16_t *block, int stride); - -void ff_h264_idct_add16_mmx (uint8_t *dst, const int *block_offset, - DCTELEM *block, int stride, const uint8_t nnzc[6*8]); -void ff_h264_idct8_add4_mmx (uint8_t *dst, const int *block_offset, - DCTELEM *block, int stride, const uint8_t nnzc[6*8]); -void ff_h264_idct_add16_mmx2 (uint8_t *dst, const int *block_offset, - DCTELEM *block, int stride, const uint8_t nnzc[6*8]); -void ff_h264_idct_add16intra_mmx (uint8_t *dst, const int *block_offset, - DCTELEM *block, int stride, const uint8_t nnzc[6*8]); -void ff_h264_idct_add16intra_mmx2(uint8_t *dst, const int *block_offset, - DCTELEM *block, int stride, const uint8_t nnzc[6*8]); -void ff_h264_idct8_add4_mmx2 (uint8_t *dst, const int *block_offset, - DCTELEM *block, int stride, const uint8_t nnzc[6*8]); -void ff_h264_idct8_add4_sse2 (uint8_t *dst, const int *block_offset, - DCTELEM *block, int stride, const uint8_t nnzc[6*8]); -void ff_h264_idct_add8_mmx (uint8_t **dest, const int *block_offset, - DCTELEM *block, int stride, const uint8_t nnzc[6*8]); -void ff_h264_idct_add8_mmx2 (uint8_t **dest, const int *block_offset, - DCTELEM *block, int stride, const uint8_t nnzc[6*8]); - -void ff_h264_idct_add16_sse2 (uint8_t *dst, const int *block_offset, DCTELEM *block, - int stride, const uint8_t nnzc[6*8]); -void ff_h264_idct_add16intra_sse2(uint8_t *dst, const int *block_offset, DCTELEM *block, - int stride, const uint8_t nnzc[6*8]); -void ff_h264_idct_add8_sse2 (uint8_t **dest, const int *block_offset, DCTELEM *block, - int stride, const uint8_t nnzc[6*8]); void ff_h264_luma_dc_dequant_idct_mmx (DCTELEM *output, DCTELEM *input, int qmul); void ff_h264_luma_dc_dequant_idct_sse2(DCTELEM *output, DCTELEM *input, int qmul); @@ -303,34 +326,59 @@ H264_BIWEIGHT_MMX ( 4, 8) H264_BIWEIGHT_MMX ( 4, 4) H264_BIWEIGHT_MMX ( 4, 2) +#define H264_WEIGHT_10(W, H, DEPTH, OPT) \ +void ff_h264_weight_ ## W ## x ## H ## _ ## DEPTH ## _ ## OPT(uint8_t *dst, \ + int stride, int log2_denom, int weight, int offset); + +#define H264_BIWEIGHT_10(W, H, DEPTH, OPT) \ +void ff_h264_biweight_ ## W ## x ## H ## _ ## DEPTH ## _ ## OPT \ + (uint8_t *dst, uint8_t *src, int stride, int log2_denom, \ + int weightd, int weights, int offset); + +#define H264_BIWEIGHT_10_SSE(W, H, DEPTH) \ +H264_WEIGHT_10 (W, H, DEPTH, sse2) \ +H264_WEIGHT_10 (W, H, DEPTH, sse4) \ +H264_BIWEIGHT_10(W, H, DEPTH, sse2) \ +H264_BIWEIGHT_10(W, H, DEPTH, sse4) + +H264_BIWEIGHT_10_SSE(16, 16, 10) +H264_BIWEIGHT_10_SSE(16, 8, 10) +H264_BIWEIGHT_10_SSE( 8, 16, 10) +H264_BIWEIGHT_10_SSE( 8, 8, 10) +H264_BIWEIGHT_10_SSE( 8, 4, 10) +H264_BIWEIGHT_10_SSE( 4, 8, 10) +H264_BIWEIGHT_10_SSE( 4, 4, 10) +H264_BIWEIGHT_10_SSE( 4, 2, 10) + void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth) { int mm_flags = av_get_cpu_flags(); - if (bit_depth == 8) { if (mm_flags & AV_CPU_FLAG_MMX2) { c->h264_loop_filter_strength= h264_loop_filter_strength_mmx2; } + + if (bit_depth == 8) { #if HAVE_YASM if (mm_flags & AV_CPU_FLAG_MMX) { - c->h264_idct_dc_add= - c->h264_idct_add= ff_h264_idct_add_mmx; - c->h264_idct8_dc_add= - c->h264_idct8_add= ff_h264_idct8_add_mmx; - - c->h264_idct_add16 = ff_h264_idct_add16_mmx; - c->h264_idct8_add4 = ff_h264_idct8_add4_mmx; - c->h264_idct_add8 = ff_h264_idct_add8_mmx; - c->h264_idct_add16intra= ff_h264_idct_add16intra_mmx; + c->h264_idct_dc_add = + c->h264_idct_add = ff_h264_idct_add_8_mmx; + c->h264_idct8_dc_add = + c->h264_idct8_add = ff_h264_idct8_add_8_mmx; + + c->h264_idct_add16 = ff_h264_idct_add16_8_mmx; + c->h264_idct8_add4 = ff_h264_idct8_add4_8_mmx; + c->h264_idct_add8 = ff_h264_idct_add8_8_mmx; + c->h264_idct_add16intra = ff_h264_idct_add16intra_8_mmx; c->h264_luma_dc_dequant_idct= ff_h264_luma_dc_dequant_idct_mmx; if (mm_flags & AV_CPU_FLAG_MMX2) { - c->h264_idct_dc_add= ff_h264_idct_dc_add_mmx2; - c->h264_idct8_dc_add= ff_h264_idct8_dc_add_mmx2; - c->h264_idct_add16 = ff_h264_idct_add16_mmx2; - c->h264_idct8_add4 = ff_h264_idct8_add4_mmx2; - c->h264_idct_add8 = ff_h264_idct_add8_mmx2; - c->h264_idct_add16intra= ff_h264_idct_add16intra_mmx2; + c->h264_idct_dc_add = ff_h264_idct_dc_add_8_mmx2; + c->h264_idct8_dc_add = ff_h264_idct8_dc_add_8_mmx2; + c->h264_idct_add16 = ff_h264_idct_add16_8_mmx2; + c->h264_idct8_add4 = ff_h264_idct8_add4_8_mmx2; + c->h264_idct_add8 = ff_h264_idct_add8_8_mmx2; + c->h264_idct_add16intra= ff_h264_idct_add16intra_8_mmx2; c->h264_v_loop_filter_chroma= ff_deblock_v_chroma_8_mmxext; c->h264_h_loop_filter_chroma= ff_deblock_h_chroma_8_mmxext; @@ -361,8 +409,12 @@ void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth) c->biweight_h264_pixels_tab[7]= ff_h264_biweight_4x2_mmx2; if (mm_flags&AV_CPU_FLAG_SSE2) { - c->h264_idct8_add = ff_h264_idct8_add_sse2; - c->h264_idct8_add4= ff_h264_idct8_add4_sse2; + c->h264_idct8_add = ff_h264_idct8_add_8_sse2; + + c->h264_idct_add16 = ff_h264_idct_add16_8_sse2; + c->h264_idct8_add4 = ff_h264_idct8_add4_8_sse2; + c->h264_idct_add8 = ff_h264_idct_add8_8_sse2; + c->h264_idct_add16intra = ff_h264_idct_add16intra_8_sse2; c->h264_luma_dc_dequant_idct= ff_h264_luma_dc_dequant_idct_sse2; c->weight_h264_pixels_tab[0]= ff_h264_weight_16x16_sse2; @@ -383,10 +435,6 @@ void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth) c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_sse2; c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_sse2; #endif - - c->h264_idct_add16 = ff_h264_idct_add16_sse2; - c->h264_idct_add8 = ff_h264_idct_add8_sse2; - c->h264_idct_add16intra = ff_h264_idct_add16intra_sse2; } if (mm_flags&AV_CPU_FLAG_SSSE3) { c->biweight_h264_pixels_tab[0]= ff_h264_biweight_16x16_ssse3; @@ -395,7 +443,7 @@ void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth) c->biweight_h264_pixels_tab[3]= ff_h264_biweight_8x8_ssse3; c->biweight_h264_pixels_tab[4]= ff_h264_biweight_8x4_ssse3; } - if (mm_flags&AV_CPU_FLAG_AVX) { + if (HAVE_AVX && mm_flags&AV_CPU_FLAG_AVX) { #if HAVE_ALIGNED_STACK c->h264_v_loop_filter_luma = ff_deblock_v_luma_8_avx; c->h264_h_loop_filter_luma = ff_deblock_h_luma_8_avx; @@ -418,7 +466,37 @@ void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth) c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_10_mmxext; c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_10_mmxext; #endif + c->h264_idct_dc_add= ff_h264_idct_dc_add_10_mmx2; if (mm_flags&AV_CPU_FLAG_SSE2) { + c->h264_idct_add = ff_h264_idct_add_10_sse2; + c->h264_idct8_dc_add = ff_h264_idct8_dc_add_10_sse2; + + c->h264_idct_add16 = ff_h264_idct_add16_10_sse2; + c->h264_idct_add8 = ff_h264_idct_add8_10_sse2; + c->h264_idct_add16intra= ff_h264_idct_add16intra_10_sse2; +#if HAVE_ALIGNED_STACK + c->h264_idct8_add = ff_h264_idct8_add_10_sse2; + c->h264_idct8_add4 = ff_h264_idct8_add4_10_sse2; +#endif + + c->weight_h264_pixels_tab[0] = ff_h264_weight_16x16_10_sse2; + c->weight_h264_pixels_tab[1] = ff_h264_weight_16x8_10_sse2; + c->weight_h264_pixels_tab[2] = ff_h264_weight_8x16_10_sse2; + c->weight_h264_pixels_tab[3] = ff_h264_weight_8x8_10_sse2; + c->weight_h264_pixels_tab[4] = ff_h264_weight_8x4_10_sse2; + c->weight_h264_pixels_tab[5] = ff_h264_weight_4x8_10_sse2; + c->weight_h264_pixels_tab[6] = ff_h264_weight_4x4_10_sse2; + c->weight_h264_pixels_tab[7] = ff_h264_weight_4x2_10_sse2; + + c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16x16_10_sse2; + c->biweight_h264_pixels_tab[1] = ff_h264_biweight_16x8_10_sse2; + c->biweight_h264_pixels_tab[2] = ff_h264_biweight_8x16_10_sse2; + c->biweight_h264_pixels_tab[3] = ff_h264_biweight_8x8_10_sse2; + c->biweight_h264_pixels_tab[4] = ff_h264_biweight_8x4_10_sse2; + c->biweight_h264_pixels_tab[5] = ff_h264_biweight_4x8_10_sse2; + c->biweight_h264_pixels_tab[6] = ff_h264_biweight_4x4_10_sse2; + c->biweight_h264_pixels_tab[7] = ff_h264_biweight_4x2_10_sse2; + c->h264_v_loop_filter_chroma= ff_deblock_v_chroma_10_sse2; c->h264_v_loop_filter_chroma_intra= ff_deblock_v_chroma_intra_10_sse2; #if HAVE_ALIGNED_STACK @@ -428,7 +506,39 @@ void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth) c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_10_sse2; #endif } + if (mm_flags&AV_CPU_FLAG_SSE4) { + c->weight_h264_pixels_tab[0] = ff_h264_weight_16x16_10_sse4; + c->weight_h264_pixels_tab[1] = ff_h264_weight_16x8_10_sse4; + c->weight_h264_pixels_tab[2] = ff_h264_weight_8x16_10_sse4; + c->weight_h264_pixels_tab[3] = ff_h264_weight_8x8_10_sse4; + c->weight_h264_pixels_tab[4] = ff_h264_weight_8x4_10_sse4; + c->weight_h264_pixels_tab[5] = ff_h264_weight_4x8_10_sse4; + c->weight_h264_pixels_tab[6] = ff_h264_weight_4x4_10_sse4; + c->weight_h264_pixels_tab[7] = ff_h264_weight_4x2_10_sse4; + + c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16x16_10_sse4; + c->biweight_h264_pixels_tab[1] = ff_h264_biweight_16x8_10_sse4; + c->biweight_h264_pixels_tab[2] = ff_h264_biweight_8x16_10_sse4; + c->biweight_h264_pixels_tab[3] = ff_h264_biweight_8x8_10_sse4; + c->biweight_h264_pixels_tab[4] = ff_h264_biweight_8x4_10_sse4; + c->biweight_h264_pixels_tab[5] = ff_h264_biweight_4x8_10_sse4; + c->biweight_h264_pixels_tab[6] = ff_h264_biweight_4x4_10_sse4; + c->biweight_h264_pixels_tab[7] = ff_h264_biweight_4x2_10_sse4; + } +#if HAVE_AVX if (mm_flags&AV_CPU_FLAG_AVX) { + c->h264_idct_dc_add = + c->h264_idct_add = ff_h264_idct_add_10_avx; + c->h264_idct8_dc_add = ff_h264_idct8_dc_add_10_avx; + + c->h264_idct_add16 = ff_h264_idct_add16_10_avx; + c->h264_idct_add8 = ff_h264_idct_add8_10_avx; + c->h264_idct_add16intra= ff_h264_idct_add16intra_10_avx; +#if HAVE_ALIGNED_STACK + c->h264_idct8_add = ff_h264_idct8_add_10_avx; + c->h264_idct8_add4 = ff_h264_idct8_add4_10_avx; +#endif + c->h264_v_loop_filter_chroma= ff_deblock_v_chroma_10_avx; c->h264_v_loop_filter_chroma_intra= ff_deblock_v_chroma_intra_10_avx; #if HAVE_ALIGNED_STACK @@ -438,6 +548,7 @@ void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth) c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_10_avx; #endif } +#endif /* HAVE_AVX */ } } #endif