yading@10: /* yading@10: * Copyright (c) 2004-2005 Michael Niedermayer, Loren Merritt yading@10: * yading@10: * This file is part of FFmpeg. yading@10: * yading@10: * FFmpeg is free software; you can redistribute it and/or yading@10: * modify it under the terms of the GNU Lesser General Public yading@10: * License as published by the Free Software Foundation; either yading@10: * version 2.1 of the License, or (at your option) any later version. yading@10: * yading@10: * FFmpeg is distributed in the hope that it will be useful, yading@10: * but WITHOUT ANY WARRANTY; without even the implied warranty of yading@10: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU yading@10: * Lesser General Public License for more details. yading@10: * yading@10: * You should have received a copy of the GNU Lesser General Public yading@10: * License along with FFmpeg; if not, write to the Free Software yading@10: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA yading@10: */ yading@10: yading@10: #include "libavutil/attributes.h" yading@10: #include "libavutil/cpu.h" yading@10: #include "libavutil/x86/asm.h" yading@10: #include "libavutil/x86/cpu.h" yading@10: #include "libavcodec/h264dsp.h" yading@10: #include "dsputil_mmx.h" yading@10: yading@10: /***********************************/ yading@10: /* IDCT */ yading@10: #define IDCT_ADD_FUNC(NUM, DEPTH, OPT) \ yading@10: void ff_h264_idct ## NUM ## _add_ ## DEPTH ## _ ## OPT(uint8_t *dst, \ yading@10: int16_t *block, \ yading@10: int stride); yading@10: yading@10: IDCT_ADD_FUNC(, 8, mmx) yading@10: IDCT_ADD_FUNC(, 10, sse2) yading@10: IDCT_ADD_FUNC(_dc, 8, mmxext) yading@10: IDCT_ADD_FUNC(_dc, 10, mmxext) yading@10: IDCT_ADD_FUNC(8_dc, 8, mmxext) yading@10: IDCT_ADD_FUNC(8_dc, 10, sse2) yading@10: IDCT_ADD_FUNC(8, 8, mmx) yading@10: IDCT_ADD_FUNC(8, 8, sse2) yading@10: IDCT_ADD_FUNC(8, 10, sse2) yading@10: IDCT_ADD_FUNC(, 10, avx) yading@10: IDCT_ADD_FUNC(8_dc, 10, avx) yading@10: IDCT_ADD_FUNC(8, 10, avx) yading@10: yading@10: yading@10: #define IDCT_ADD_REP_FUNC(NUM, REP, DEPTH, OPT) \ yading@10: void ff_h264_idct ## NUM ## _add ## REP ## _ ## DEPTH ## _ ## OPT \ yading@10: (uint8_t *dst, const int *block_offset, \ yading@10: int16_t *block, int stride, const uint8_t nnzc[6 * 8]); yading@10: yading@10: IDCT_ADD_REP_FUNC(8, 4, 8, mmx) yading@10: IDCT_ADD_REP_FUNC(8, 4, 8, mmxext) yading@10: IDCT_ADD_REP_FUNC(8, 4, 8, sse2) yading@10: IDCT_ADD_REP_FUNC(8, 4, 10, sse2) yading@10: IDCT_ADD_REP_FUNC(8, 4, 10, avx) yading@10: IDCT_ADD_REP_FUNC(, 16, 8, mmx) yading@10: IDCT_ADD_REP_FUNC(, 16, 8, mmxext) yading@10: IDCT_ADD_REP_FUNC(, 16, 8, sse2) yading@10: IDCT_ADD_REP_FUNC(, 16, 10, sse2) yading@10: IDCT_ADD_REP_FUNC(, 16intra, 8, mmx) yading@10: IDCT_ADD_REP_FUNC(, 16intra, 8, mmxext) yading@10: IDCT_ADD_REP_FUNC(, 16intra, 8, sse2) yading@10: IDCT_ADD_REP_FUNC(, 16intra, 10, sse2) yading@10: IDCT_ADD_REP_FUNC(, 16, 10, avx) yading@10: IDCT_ADD_REP_FUNC(, 16intra, 10, avx) yading@10: yading@10: yading@10: #define IDCT_ADD_REP_FUNC2(NUM, REP, DEPTH, OPT) \ yading@10: void ff_h264_idct ## NUM ## _add ## REP ## _ ## DEPTH ## _ ## OPT \ yading@10: (uint8_t **dst, const int *block_offset, \ yading@10: int16_t *block, int stride, const uint8_t nnzc[6 * 8]); yading@10: yading@10: IDCT_ADD_REP_FUNC2(, 8, 8, mmx) yading@10: IDCT_ADD_REP_FUNC2(, 8, 8, mmxext) yading@10: IDCT_ADD_REP_FUNC2(, 8, 8, sse2) yading@10: IDCT_ADD_REP_FUNC2(, 8, 10, sse2) yading@10: IDCT_ADD_REP_FUNC2(, 8, 10, avx) yading@10: yading@10: void ff_h264_luma_dc_dequant_idct_mmx(int16_t *output, int16_t *input, int qmul); yading@10: void ff_h264_luma_dc_dequant_idct_sse2(int16_t *output, int16_t *input, int qmul); yading@10: yading@10: /***********************************/ yading@10: /* deblocking */ yading@10: yading@10: void ff_h264_loop_filter_strength_mmxext(int16_t bS[2][4][4], uint8_t nnz[40], yading@10: int8_t ref[2][40], yading@10: int16_t mv[2][40][2], yading@10: int bidir, int edges, int step, yading@10: int mask_mv0, int mask_mv1, int field); yading@10: yading@10: #define LF_FUNC(DIR, TYPE, DEPTH, OPT) \ yading@10: void ff_deblock_ ## DIR ## _ ## TYPE ## _ ## DEPTH ## _ ## OPT(uint8_t *pix, \ yading@10: int stride, \ yading@10: int alpha, \ yading@10: int beta, \ yading@10: int8_t *tc0); yading@10: #define LF_IFUNC(DIR, TYPE, DEPTH, OPT) \ yading@10: void ff_deblock_ ## DIR ## _ ## TYPE ## _ ## DEPTH ## _ ## OPT(uint8_t *pix, \ yading@10: int stride, \ yading@10: int alpha, \ yading@10: int beta); yading@10: yading@10: #define LF_FUNCS(type, depth) \ yading@10: LF_FUNC(h, chroma, depth, mmxext) \ yading@10: LF_IFUNC(h, chroma_intra, depth, mmxext) \ yading@10: LF_FUNC(v, chroma, depth, mmxext) \ yading@10: LF_IFUNC(v, chroma_intra, depth, mmxext) \ yading@10: LF_FUNC(h, luma, depth, mmxext) \ yading@10: LF_IFUNC(h, luma_intra, depth, mmxext) \ yading@10: LF_FUNC(h, luma, depth, sse2) \ yading@10: LF_IFUNC(h, luma_intra, depth, sse2) \ yading@10: LF_FUNC(v, luma, depth, sse2) \ yading@10: LF_IFUNC(v, luma_intra, depth, sse2) \ yading@10: LF_FUNC(h, chroma, depth, sse2) \ yading@10: LF_IFUNC(h, chroma_intra, depth, sse2) \ yading@10: LF_FUNC(v, chroma, depth, sse2) \ yading@10: LF_IFUNC(v, chroma_intra, depth, sse2) \ yading@10: LF_FUNC(h, luma, depth, avx) \ yading@10: LF_IFUNC(h, luma_intra, depth, avx) \ yading@10: LF_FUNC(v, luma, depth, avx) \ yading@10: LF_IFUNC(v, luma_intra, depth, avx) \ yading@10: LF_FUNC(h, chroma, depth, avx) \ yading@10: LF_IFUNC(h, chroma_intra, depth, avx) \ yading@10: LF_FUNC(v, chroma, depth, avx) \ yading@10: LF_IFUNC(v, chroma_intra, depth, avx) yading@10: yading@10: LF_FUNCS(uint8_t, 8) yading@10: LF_FUNCS(uint16_t, 10) yading@10: yading@10: #if ARCH_X86_32 && HAVE_MMXEXT_EXTERNAL yading@10: LF_FUNC(v8, luma, 8, mmxext) yading@10: static void ff_deblock_v_luma_8_mmxext(uint8_t *pix, int stride, int alpha, yading@10: int beta, int8_t *tc0) yading@10: { yading@10: if ((tc0[0] & tc0[1]) >= 0) yading@10: ff_deblock_v8_luma_8_mmxext(pix + 0, stride, alpha, beta, tc0); yading@10: if ((tc0[2] & tc0[3]) >= 0) yading@10: ff_deblock_v8_luma_8_mmxext(pix + 8, stride, alpha, beta, tc0 + 2); yading@10: } yading@10: LF_IFUNC(v8, luma_intra, 8, mmxext) yading@10: static void ff_deblock_v_luma_intra_8_mmxext(uint8_t *pix, int stride, yading@10: int alpha, int beta) yading@10: { yading@10: ff_deblock_v8_luma_intra_8_mmxext(pix + 0, stride, alpha, beta); yading@10: ff_deblock_v8_luma_intra_8_mmxext(pix + 8, stride, alpha, beta); yading@10: } yading@10: #endif /* ARCH_X86_32 && HAVE_MMXEXT_EXTERNAL */ yading@10: yading@10: LF_FUNC(v, luma, 10, mmxext) yading@10: LF_IFUNC(v, luma_intra, 10, mmxext) yading@10: yading@10: /***********************************/ yading@10: /* weighted prediction */ yading@10: yading@10: #define H264_WEIGHT(W, OPT) \ yading@10: void ff_h264_weight_ ## W ## _ ## OPT(uint8_t *dst, int stride, \ yading@10: int height, int log2_denom, \ yading@10: int weight, int offset); yading@10: yading@10: #define H264_BIWEIGHT(W, OPT) \ yading@10: void ff_h264_biweight_ ## W ## _ ## OPT(uint8_t *dst, uint8_t *src, \ yading@10: int stride, int height, \ yading@10: int log2_denom, int weightd, \ yading@10: int weights, int offset); yading@10: yading@10: #define H264_BIWEIGHT_MMX(W) \ yading@10: H264_WEIGHT(W, mmxext) \ yading@10: H264_BIWEIGHT(W, mmxext) yading@10: yading@10: #define H264_BIWEIGHT_MMX_SSE(W) \ yading@10: H264_BIWEIGHT_MMX(W) \ yading@10: H264_WEIGHT(W, sse2) \ yading@10: H264_BIWEIGHT(W, sse2) \ yading@10: H264_BIWEIGHT(W, ssse3) yading@10: yading@10: H264_BIWEIGHT_MMX_SSE(16) yading@10: H264_BIWEIGHT_MMX_SSE(8) yading@10: H264_BIWEIGHT_MMX(4) yading@10: yading@10: #define H264_WEIGHT_10(W, DEPTH, OPT) \ yading@10: void ff_h264_weight_ ## W ## _ ## DEPTH ## _ ## OPT(uint8_t *dst, \ yading@10: int stride, \ yading@10: int height, \ yading@10: int log2_denom, \ yading@10: int weight, \ yading@10: int offset); yading@10: yading@10: #define H264_BIWEIGHT_10(W, DEPTH, OPT) \ yading@10: void ff_h264_biweight_ ## W ## _ ## DEPTH ## _ ## OPT(uint8_t *dst, \ yading@10: uint8_t *src, \ yading@10: int stride, \ yading@10: int height, \ yading@10: int log2_denom, \ yading@10: int weightd, \ yading@10: int weights, \ yading@10: int offset); yading@10: yading@10: #define H264_BIWEIGHT_10_SSE(W, DEPTH) \ yading@10: H264_WEIGHT_10(W, DEPTH, sse2) \ yading@10: H264_WEIGHT_10(W, DEPTH, sse4) \ yading@10: H264_BIWEIGHT_10(W, DEPTH, sse2) \ yading@10: H264_BIWEIGHT_10(W, DEPTH, sse4) yading@10: yading@10: H264_BIWEIGHT_10_SSE(16, 10) yading@10: H264_BIWEIGHT_10_SSE(8, 10) yading@10: H264_BIWEIGHT_10_SSE(4, 10) yading@10: yading@10: av_cold void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth, yading@10: const int chroma_format_idc) yading@10: { yading@10: #if HAVE_YASM yading@10: int mm_flags = av_get_cpu_flags(); yading@10: yading@10: if (chroma_format_idc == 1 && EXTERNAL_MMXEXT(mm_flags)) yading@10: c->h264_loop_filter_strength = ff_h264_loop_filter_strength_mmxext; yading@10: yading@10: if (bit_depth == 8) { yading@10: if (EXTERNAL_MMX(mm_flags)) { yading@10: c->h264_idct_dc_add = yading@10: c->h264_idct_add = ff_h264_idct_add_8_mmx; yading@10: c->h264_idct8_dc_add = yading@10: c->h264_idct8_add = ff_h264_idct8_add_8_mmx; yading@10: yading@10: c->h264_idct_add16 = ff_h264_idct_add16_8_mmx; yading@10: c->h264_idct8_add4 = ff_h264_idct8_add4_8_mmx; yading@10: if (chroma_format_idc == 1) yading@10: c->h264_idct_add8 = ff_h264_idct_add8_8_mmx; yading@10: c->h264_idct_add16intra = ff_h264_idct_add16intra_8_mmx; yading@10: if (mm_flags & AV_CPU_FLAG_CMOV) yading@10: c->h264_luma_dc_dequant_idct = ff_h264_luma_dc_dequant_idct_mmx; yading@10: yading@10: if (EXTERNAL_MMXEXT(mm_flags)) { yading@10: c->h264_idct_dc_add = ff_h264_idct_dc_add_8_mmxext; yading@10: c->h264_idct8_dc_add = ff_h264_idct8_dc_add_8_mmxext; yading@10: c->h264_idct_add16 = ff_h264_idct_add16_8_mmxext; yading@10: c->h264_idct8_add4 = ff_h264_idct8_add4_8_mmxext; yading@10: if (chroma_format_idc == 1) yading@10: c->h264_idct_add8 = ff_h264_idct_add8_8_mmxext; yading@10: c->h264_idct_add16intra = ff_h264_idct_add16intra_8_mmxext; yading@10: yading@10: c->h264_v_loop_filter_chroma = ff_deblock_v_chroma_8_mmxext; yading@10: c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_8_mmxext; yading@10: if (chroma_format_idc == 1) { yading@10: c->h264_h_loop_filter_chroma = ff_deblock_h_chroma_8_mmxext; yading@10: c->h264_h_loop_filter_chroma_intra = ff_deblock_h_chroma_intra_8_mmxext; yading@10: } yading@10: #if ARCH_X86_32 && HAVE_MMXEXT_EXTERNAL yading@10: c->h264_v_loop_filter_luma = ff_deblock_v_luma_8_mmxext; yading@10: c->h264_h_loop_filter_luma = ff_deblock_h_luma_8_mmxext; yading@10: c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_mmxext; yading@10: c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_mmxext; yading@10: #endif /* ARCH_X86_32 && HAVE_MMXEXT_EXTERNAL */ yading@10: c->weight_h264_pixels_tab[0] = ff_h264_weight_16_mmxext; yading@10: c->weight_h264_pixels_tab[1] = ff_h264_weight_8_mmxext; yading@10: c->weight_h264_pixels_tab[2] = ff_h264_weight_4_mmxext; yading@10: yading@10: c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_mmxext; yading@10: c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_mmxext; yading@10: c->biweight_h264_pixels_tab[2] = ff_h264_biweight_4_mmxext; yading@10: yading@10: if (EXTERNAL_SSE2(mm_flags)) { yading@10: c->h264_idct8_add = ff_h264_idct8_add_8_sse2; yading@10: yading@10: c->h264_idct_add16 = ff_h264_idct_add16_8_sse2; yading@10: c->h264_idct8_add4 = ff_h264_idct8_add4_8_sse2; yading@10: if (chroma_format_idc == 1) yading@10: c->h264_idct_add8 = ff_h264_idct_add8_8_sse2; yading@10: c->h264_idct_add16intra = ff_h264_idct_add16intra_8_sse2; yading@10: c->h264_luma_dc_dequant_idct = ff_h264_luma_dc_dequant_idct_sse2; yading@10: yading@10: c->weight_h264_pixels_tab[0] = ff_h264_weight_16_sse2; yading@10: c->weight_h264_pixels_tab[1] = ff_h264_weight_8_sse2; yading@10: yading@10: c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_sse2; yading@10: c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_sse2; yading@10: yading@10: c->h264_v_loop_filter_luma = ff_deblock_v_luma_8_sse2; yading@10: c->h264_h_loop_filter_luma = ff_deblock_h_luma_8_sse2; yading@10: c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_sse2; yading@10: c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_sse2; yading@10: } yading@10: if (EXTERNAL_SSSE3(mm_flags)) { yading@10: c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_ssse3; yading@10: c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_ssse3; yading@10: } yading@10: if (EXTERNAL_AVX(mm_flags)) { yading@10: c->h264_v_loop_filter_luma = ff_deblock_v_luma_8_avx; yading@10: c->h264_h_loop_filter_luma = ff_deblock_h_luma_8_avx; yading@10: c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_avx; yading@10: c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_avx; yading@10: } yading@10: } yading@10: } yading@10: } else if (bit_depth == 10) { yading@10: if (EXTERNAL_MMX(mm_flags)) { yading@10: if (EXTERNAL_MMXEXT(mm_flags)) { yading@10: #if ARCH_X86_32 yading@10: c->h264_v_loop_filter_chroma = ff_deblock_v_chroma_10_mmxext; yading@10: c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_10_mmxext; yading@10: c->h264_v_loop_filter_luma = ff_deblock_v_luma_10_mmxext; yading@10: c->h264_h_loop_filter_luma = ff_deblock_h_luma_10_mmxext; yading@10: c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_10_mmxext; yading@10: c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_10_mmxext; yading@10: #endif /* ARCH_X86_32 */ yading@10: c->h264_idct_dc_add = ff_h264_idct_dc_add_10_mmxext; yading@10: if (EXTERNAL_SSE2(mm_flags)) { yading@10: c->h264_idct_add = ff_h264_idct_add_10_sse2; yading@10: c->h264_idct8_dc_add = ff_h264_idct8_dc_add_10_sse2; yading@10: yading@10: c->h264_idct_add16 = ff_h264_idct_add16_10_sse2; yading@10: if (chroma_format_idc == 1) yading@10: c->h264_idct_add8 = ff_h264_idct_add8_10_sse2; yading@10: c->h264_idct_add16intra = ff_h264_idct_add16intra_10_sse2; yading@10: #if HAVE_ALIGNED_STACK yading@10: c->h264_idct8_add = ff_h264_idct8_add_10_sse2; yading@10: c->h264_idct8_add4 = ff_h264_idct8_add4_10_sse2; yading@10: #endif /* HAVE_ALIGNED_STACK */ yading@10: yading@10: c->weight_h264_pixels_tab[0] = ff_h264_weight_16_10_sse2; yading@10: c->weight_h264_pixels_tab[1] = ff_h264_weight_8_10_sse2; yading@10: c->weight_h264_pixels_tab[2] = ff_h264_weight_4_10_sse2; yading@10: yading@10: c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_10_sse2; yading@10: c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_10_sse2; yading@10: c->biweight_h264_pixels_tab[2] = ff_h264_biweight_4_10_sse2; yading@10: yading@10: c->h264_v_loop_filter_chroma = ff_deblock_v_chroma_10_sse2; yading@10: c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_10_sse2; yading@10: #if HAVE_ALIGNED_STACK yading@10: c->h264_v_loop_filter_luma = ff_deblock_v_luma_10_sse2; yading@10: c->h264_h_loop_filter_luma = ff_deblock_h_luma_10_sse2; yading@10: c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_10_sse2; yading@10: c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_10_sse2; yading@10: #endif /* HAVE_ALIGNED_STACK */ yading@10: } yading@10: if (EXTERNAL_SSE4(mm_flags)) { yading@10: c->weight_h264_pixels_tab[0] = ff_h264_weight_16_10_sse4; yading@10: c->weight_h264_pixels_tab[1] = ff_h264_weight_8_10_sse4; yading@10: c->weight_h264_pixels_tab[2] = ff_h264_weight_4_10_sse4; yading@10: yading@10: c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_10_sse4; yading@10: c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_10_sse4; yading@10: c->biweight_h264_pixels_tab[2] = ff_h264_biweight_4_10_sse4; yading@10: } yading@10: if (EXTERNAL_AVX(mm_flags)) { yading@10: c->h264_idct_dc_add = yading@10: c->h264_idct_add = ff_h264_idct_add_10_avx; yading@10: c->h264_idct8_dc_add = ff_h264_idct8_dc_add_10_avx; yading@10: yading@10: c->h264_idct_add16 = ff_h264_idct_add16_10_avx; yading@10: if (chroma_format_idc == 1) yading@10: c->h264_idct_add8 = ff_h264_idct_add8_10_avx; yading@10: c->h264_idct_add16intra = ff_h264_idct_add16intra_10_avx; yading@10: #if HAVE_ALIGNED_STACK yading@10: c->h264_idct8_add = ff_h264_idct8_add_10_avx; yading@10: c->h264_idct8_add4 = ff_h264_idct8_add4_10_avx; yading@10: #endif /* HAVE_ALIGNED_STACK */ yading@10: yading@10: c->h264_v_loop_filter_chroma = ff_deblock_v_chroma_10_avx; yading@10: c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_10_avx; yading@10: #if HAVE_ALIGNED_STACK yading@10: c->h264_v_loop_filter_luma = ff_deblock_v_luma_10_avx; yading@10: c->h264_h_loop_filter_luma = ff_deblock_h_luma_10_avx; yading@10: c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_10_avx; yading@10: c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_10_avx; yading@10: #endif /* HAVE_ALIGNED_STACK */ yading@10: } yading@10: } yading@10: } yading@10: } yading@10: #endif yading@10: }