1/* 2 * HEVC video decoder 3 * 4 * Copyright (C) 2012 - 2013 Guillaume Martres 5 * Copyright (C) 2013 - 2014 Pierre-Edouard Lepere 6 * 7 * 8 * This file is part of FFmpeg. 9 * 10 * FFmpeg is free software; you can redistribute it and/or 11 * modify it under the terms of the GNU Lesser General Public 12 * License as published by the Free Software Foundation; either 13 * version 2.1 of the License, or (at your option) any later version. 14 * 15 * FFmpeg is distributed in the hope that it will be useful, 16 * but WITHOUT ANY WARRANTY; without even the implied warranty of 17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 18 * Lesser General Public License for more details. 19 * 20 * You should have received a copy of the GNU Lesser General Public 21 * License along with FFmpeg; if not, write to the Free Software 22 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA 23 */ 24 25#ifndef AVCODEC_X86_HEVCDSP_H 26#define AVCODEC_X86_HEVCDSP_H 27 28#include <stddef.h> 29#include <stdint.h> 30 31 32#define PEL_LINK(dst, idx1, idx2, idx3, name, D, opt) \ 33dst[idx1][idx2][idx3] = ff_hevc_put_hevc_ ## name ## _ ## D ## _##opt; \ 34dst ## _bi[idx1][idx2][idx3] = ff_hevc_put_hevc_bi_ ## name ## _ ## D ## _##opt; \ 35dst ## _uni[idx1][idx2][idx3] = ff_hevc_put_hevc_uni_ ## name ## _ ## D ## _##opt; \ 36dst ## _uni_w[idx1][idx2][idx3] = ff_hevc_put_hevc_uni_w_ ## name ## _ ## D ## _##opt; \ 37dst ## _bi_w[idx1][idx2][idx3] = ff_hevc_put_hevc_bi_w_ ## name ## _ ## D ## _##opt 38 39 40#define PEL_PROTOTYPE(name, D, opt) \ 41void ff_hevc_put_hevc_ ## name ## _ ## D ## _##opt(int16_t *dst, uint8_t *_src, ptrdiff_t _srcstride, int height, intptr_t mx, intptr_t my,int width); \ 42void ff_hevc_put_hevc_bi_ ## name ## _ ## D ## _##opt(uint8_t *_dst, ptrdiff_t _dststride, uint8_t *_src, ptrdiff_t _srcstride, int16_t *src2, int height, intptr_t mx, intptr_t my, int width); \ 43void ff_hevc_put_hevc_uni_ ## name ## _ ## D ## _##opt(uint8_t *_dst, ptrdiff_t _dststride, uint8_t *_src, ptrdiff_t _srcstride, int height, intptr_t mx, intptr_t my, int width); \ 44void ff_hevc_put_hevc_uni_w_ ## name ## _ ## D ## _##opt(uint8_t *_dst, ptrdiff_t _dststride, uint8_t *_src, ptrdiff_t _srcstride, int height, int denom, int wx, int ox, intptr_t mx, intptr_t my, int width); \ 45void ff_hevc_put_hevc_bi_w_ ## name ## _ ## D ## _##opt(uint8_t *_dst, ptrdiff_t _dststride, uint8_t *_src, ptrdiff_t _srcstride, int16_t *src2, int height, int denom, int wx0, int wx1, int ox0, int ox1, intptr_t mx, intptr_t my, int width) 46 47 48/////////////////////////////////////////////////////////////////////////////// 49// MC functions 50/////////////////////////////////////////////////////////////////////////////// 51 52#define EPEL_PROTOTYPES(fname, bitd, opt) \ 53 PEL_PROTOTYPE(fname##4, bitd, opt); \ 54 PEL_PROTOTYPE(fname##6, bitd, opt); \ 55 PEL_PROTOTYPE(fname##8, bitd, opt); \ 56 PEL_PROTOTYPE(fname##12, bitd, opt); \ 57 PEL_PROTOTYPE(fname##16, bitd, opt); \ 58 PEL_PROTOTYPE(fname##24, bitd, opt); \ 59 PEL_PROTOTYPE(fname##32, bitd, opt); \ 60 PEL_PROTOTYPE(fname##48, bitd, opt); \ 61 PEL_PROTOTYPE(fname##64, bitd, opt) 62 63#define QPEL_PROTOTYPES(fname, bitd, opt) \ 64 PEL_PROTOTYPE(fname##4, bitd, opt); \ 65 PEL_PROTOTYPE(fname##8, bitd, opt); \ 66 PEL_PROTOTYPE(fname##12, bitd, opt); \ 67 PEL_PROTOTYPE(fname##16, bitd, opt); \ 68 PEL_PROTOTYPE(fname##24, bitd, opt); \ 69 PEL_PROTOTYPE(fname##32, bitd, opt); \ 70 PEL_PROTOTYPE(fname##48, bitd, opt); \ 71 PEL_PROTOTYPE(fname##64, bitd, opt) 72 73#define WEIGHTING_PROTOTYPE(width, bitd, opt) \ 74void ff_hevc_put_hevc_uni_w##width##_##bitd##_##opt(uint8_t *dst, ptrdiff_t dststride, int16_t *_src, int height, int denom, int _wx, int _ox); \ 75void ff_hevc_put_hevc_bi_w##width##_##bitd##_##opt(uint8_t *dst, ptrdiff_t dststride, int16_t *_src, int16_t *_src2, int height, int denom, int _wx0, int _wx1, int _ox0, int _ox1) 76 77#define WEIGHTING_PROTOTYPES(bitd, opt) \ 78 WEIGHTING_PROTOTYPE(2, bitd, opt); \ 79 WEIGHTING_PROTOTYPE(4, bitd, opt); \ 80 WEIGHTING_PROTOTYPE(6, bitd, opt); \ 81 WEIGHTING_PROTOTYPE(8, bitd, opt); \ 82 WEIGHTING_PROTOTYPE(12, bitd, opt); \ 83 WEIGHTING_PROTOTYPE(16, bitd, opt); \ 84 WEIGHTING_PROTOTYPE(24, bitd, opt); \ 85 WEIGHTING_PROTOTYPE(32, bitd, opt); \ 86 WEIGHTING_PROTOTYPE(48, bitd, opt); \ 87 WEIGHTING_PROTOTYPE(64, bitd, opt) 88 89 90/////////////////////////////////////////////////////////////////////////////// 91// QPEL_PIXELS EPEL_PIXELS 92/////////////////////////////////////////////////////////////////////////////// 93EPEL_PROTOTYPES(pel_pixels , 8, sse4); 94EPEL_PROTOTYPES(pel_pixels , 10, sse4); 95EPEL_PROTOTYPES(pel_pixels , 12, sse4); 96 97void ff_hevc_put_hevc_pel_pixels16_8_avx2(int16_t *dst, uint8_t *_src, ptrdiff_t _srcstride, int height, intptr_t mx, intptr_t my,int width); 98void ff_hevc_put_hevc_pel_pixels24_8_avx2(int16_t *dst, uint8_t *_src, ptrdiff_t _srcstride, int height, intptr_t mx, intptr_t my,int width); 99void ff_hevc_put_hevc_pel_pixels32_8_avx2(int16_t *dst, uint8_t *_src, ptrdiff_t _srcstride, int height, intptr_t mx, intptr_t my,int width); 100void ff_hevc_put_hevc_pel_pixels48_8_avx2(int16_t *dst, uint8_t *_src, ptrdiff_t _srcstride, int height, intptr_t mx, intptr_t my,int width); 101void ff_hevc_put_hevc_pel_pixels64_8_avx2(int16_t *dst, uint8_t *_src, ptrdiff_t _srcstride, int height, intptr_t mx, intptr_t my,int width); 102 103void ff_hevc_put_hevc_pel_pixels16_10_avx2(int16_t *dst, uint8_t *_src, ptrdiff_t _srcstride, int height, intptr_t mx, intptr_t my,int width); 104void ff_hevc_put_hevc_pel_pixels24_10_avx2(int16_t *dst, uint8_t *_src, ptrdiff_t _srcstride, int height, intptr_t mx, intptr_t my,int width); 105void ff_hevc_put_hevc_pel_pixels32_10_avx2(int16_t *dst, uint8_t *_src, ptrdiff_t _srcstride, int height, intptr_t mx, intptr_t my,int width); 106void ff_hevc_put_hevc_pel_pixels48_10_avx2(int16_t *dst, uint8_t *_src, ptrdiff_t _srcstride, int height, intptr_t mx, intptr_t my,int width); 107void ff_hevc_put_hevc_pel_pixels64_10_avx2(int16_t *dst, uint8_t *_src, ptrdiff_t _srcstride, int height, intptr_t mx, intptr_t my,int width); 108 109 110 111void ff_hevc_put_hevc_uni_pel_pixels32_8_avx2(uint8_t *dst, ptrdiff_t dststride,uint8_t *_src, ptrdiff_t _srcstride, int height, intptr_t mx, intptr_t my,int width); 112void ff_hevc_put_hevc_uni_pel_pixels48_8_avx2(uint8_t *dst, ptrdiff_t dststride,uint8_t *_src, ptrdiff_t _srcstride, int height, intptr_t mx, intptr_t my,int width); 113void ff_hevc_put_hevc_uni_pel_pixels64_8_avx2(uint8_t *dst, ptrdiff_t dststride,uint8_t *_src, ptrdiff_t _srcstride, int height, intptr_t mx, intptr_t my,int width); 114void ff_hevc_put_hevc_uni_pel_pixels96_8_avx2(uint8_t *dst, ptrdiff_t dststride,uint8_t *_src, ptrdiff_t _srcstride, int height, intptr_t mx, intptr_t my,int width); //used for 10bit 115void ff_hevc_put_hevc_uni_pel_pixels128_8_avx2(uint8_t *dst, ptrdiff_t dststride,uint8_t *_src, ptrdiff_t _srcstride, int height, intptr_t mx, intptr_t my,int width);//used for 10bit 116 117 118void ff_hevc_put_hevc_bi_pel_pixels16_8_avx2(uint8_t *_dst, ptrdiff_t _dststride, uint8_t *_src, ptrdiff_t _srcstride, int16_t *src2, int height, intptr_t mx, intptr_t my, int width); 119void ff_hevc_put_hevc_bi_pel_pixels24_8_avx2(uint8_t *_dst, ptrdiff_t _dststride, uint8_t *_src, ptrdiff_t _srcstride, int16_t *src2, int height, intptr_t mx, intptr_t my, int width); 120void ff_hevc_put_hevc_bi_pel_pixels32_8_avx2(uint8_t *_dst, ptrdiff_t _dststride, uint8_t *_src, ptrdiff_t _srcstride, int16_t *src2, int height, intptr_t mx, intptr_t my, int width); 121void ff_hevc_put_hevc_bi_pel_pixels48_8_avx2(uint8_t *_dst, ptrdiff_t _dststride, uint8_t *_src, ptrdiff_t _srcstride, int16_t *src2, int height, intptr_t mx, intptr_t my, int width); 122void ff_hevc_put_hevc_bi_pel_pixels64_8_avx2(uint8_t *_dst, ptrdiff_t _dststride, uint8_t *_src, ptrdiff_t _srcstride, int16_t *src2, int height, intptr_t mx, intptr_t my, int width); 123 124void ff_hevc_put_hevc_bi_pel_pixels16_10_avx2(uint8_t *_dst, ptrdiff_t _dststride, uint8_t *_src, ptrdiff_t _srcstride, int16_t *src2, int height, intptr_t mx, intptr_t my, int width); 125void ff_hevc_put_hevc_bi_pel_pixels24_10_avx2(uint8_t *_dst, ptrdiff_t _dststride, uint8_t *_src, ptrdiff_t _srcstride, int16_t *src2, int height, intptr_t mx, intptr_t my, int width); 126void ff_hevc_put_hevc_bi_pel_pixels32_10_avx2(uint8_t *_dst, ptrdiff_t _dststride, uint8_t *_src, ptrdiff_t _srcstride, int16_t *src2, int height, intptr_t mx, intptr_t my, int width); 127void ff_hevc_put_hevc_bi_pel_pixels48_10_avx2(uint8_t *_dst, ptrdiff_t _dststride, uint8_t *_src, ptrdiff_t _srcstride, int16_t *src2, int height, intptr_t mx, intptr_t my, int width); 128void ff_hevc_put_hevc_bi_pel_pixels64_10_avx2(uint8_t *_dst, ptrdiff_t _dststride, uint8_t *_src, ptrdiff_t _srcstride, int16_t *src2, int height, intptr_t mx, intptr_t my, int width); 129 130/////////////////////////////////////////////////////////////////////////////// 131// EPEL 132/////////////////////////////////////////////////////////////////////////////// 133EPEL_PROTOTYPES(epel_h , 8, sse4); 134EPEL_PROTOTYPES(epel_h , 10, sse4); 135EPEL_PROTOTYPES(epel_h , 12, sse4); 136 137EPEL_PROTOTYPES(epel_v , 8, sse4); 138EPEL_PROTOTYPES(epel_v , 10, sse4); 139EPEL_PROTOTYPES(epel_v , 12, sse4); 140 141EPEL_PROTOTYPES(epel_hv , 8, sse4); 142EPEL_PROTOTYPES(epel_hv , 10, sse4); 143EPEL_PROTOTYPES(epel_hv , 12, sse4); 144 145PEL_PROTOTYPE(epel_h16, 8, avx2); 146PEL_PROTOTYPE(epel_h24, 8, avx2); 147PEL_PROTOTYPE(epel_h32, 8, avx2); 148PEL_PROTOTYPE(epel_h48, 8, avx2); 149PEL_PROTOTYPE(epel_h64, 8, avx2); 150 151PEL_PROTOTYPE(epel_h16,10, avx2); 152PEL_PROTOTYPE(epel_h24,10, avx2); 153PEL_PROTOTYPE(epel_h32,10, avx2); 154PEL_PROTOTYPE(epel_h48,10, avx2); 155PEL_PROTOTYPE(epel_h64,10, avx2); 156 157PEL_PROTOTYPE(epel_v16, 8, avx2); 158PEL_PROTOTYPE(epel_v24, 8, avx2); 159PEL_PROTOTYPE(epel_v32, 8, avx2); 160PEL_PROTOTYPE(epel_v48, 8, avx2); 161PEL_PROTOTYPE(epel_v64, 8, avx2); 162 163PEL_PROTOTYPE(epel_v16,10, avx2); 164PEL_PROTOTYPE(epel_v24,10, avx2); 165PEL_PROTOTYPE(epel_v32,10, avx2); 166PEL_PROTOTYPE(epel_v48,10, avx2); 167PEL_PROTOTYPE(epel_v64,10, avx2); 168 169PEL_PROTOTYPE(epel_hv16, 8, avx2); 170PEL_PROTOTYPE(epel_hv24, 8, avx2); 171PEL_PROTOTYPE(epel_hv32, 8, avx2); 172PEL_PROTOTYPE(epel_hv48, 8, avx2); 173PEL_PROTOTYPE(epel_hv64, 8, avx2); 174 175PEL_PROTOTYPE(epel_hv16,10, avx2); 176PEL_PROTOTYPE(epel_hv24,10, avx2); 177PEL_PROTOTYPE(epel_hv32,10, avx2); 178PEL_PROTOTYPE(epel_hv48,10, avx2); 179PEL_PROTOTYPE(epel_hv64,10, avx2); 180 181/////////////////////////////////////////////////////////////////////////////// 182// QPEL 183/////////////////////////////////////////////////////////////////////////////// 184QPEL_PROTOTYPES(qpel_h , 8, sse4); 185QPEL_PROTOTYPES(qpel_h , 10, sse4); 186QPEL_PROTOTYPES(qpel_h , 12, sse4); 187 188QPEL_PROTOTYPES(qpel_v, 8, sse4); 189QPEL_PROTOTYPES(qpel_v, 10, sse4); 190QPEL_PROTOTYPES(qpel_v, 12, sse4); 191 192QPEL_PROTOTYPES(qpel_hv, 8, sse4); 193QPEL_PROTOTYPES(qpel_hv, 10, sse4); 194QPEL_PROTOTYPES(qpel_hv, 12, sse4); 195 196PEL_PROTOTYPE(qpel_h16, 8, avx2); 197PEL_PROTOTYPE(qpel_h24, 8, avx2); 198PEL_PROTOTYPE(qpel_h32, 8, avx2); 199PEL_PROTOTYPE(qpel_h48, 8, avx2); 200PEL_PROTOTYPE(qpel_h64, 8, avx2); 201 202PEL_PROTOTYPE(qpel_h16,10, avx2); 203PEL_PROTOTYPE(qpel_h24,10, avx2); 204PEL_PROTOTYPE(qpel_h32,10, avx2); 205PEL_PROTOTYPE(qpel_h48,10, avx2); 206PEL_PROTOTYPE(qpel_h64,10, avx2); 207 208PEL_PROTOTYPE(qpel_v16, 8, avx2); 209PEL_PROTOTYPE(qpel_v24, 8, avx2); 210PEL_PROTOTYPE(qpel_v32, 8, avx2); 211PEL_PROTOTYPE(qpel_v48, 8, avx2); 212PEL_PROTOTYPE(qpel_v64, 8, avx2); 213 214PEL_PROTOTYPE(qpel_v16,10, avx2); 215PEL_PROTOTYPE(qpel_v24,10, avx2); 216PEL_PROTOTYPE(qpel_v32,10, avx2); 217PEL_PROTOTYPE(qpel_v48,10, avx2); 218PEL_PROTOTYPE(qpel_v64,10, avx2); 219 220PEL_PROTOTYPE(qpel_hv16, 8, avx2); 221PEL_PROTOTYPE(qpel_hv24, 8, avx2); 222PEL_PROTOTYPE(qpel_hv32, 8, avx2); 223PEL_PROTOTYPE(qpel_hv48, 8, avx2); 224PEL_PROTOTYPE(qpel_hv64, 8, avx2); 225 226PEL_PROTOTYPE(qpel_hv16,10, avx2); 227PEL_PROTOTYPE(qpel_hv24,10, avx2); 228PEL_PROTOTYPE(qpel_hv32,10, avx2); 229PEL_PROTOTYPE(qpel_hv48,10, avx2); 230PEL_PROTOTYPE(qpel_hv64,10, avx2); 231 232WEIGHTING_PROTOTYPES(8, sse4); 233WEIGHTING_PROTOTYPES(10, sse4); 234WEIGHTING_PROTOTYPES(12, sse4); 235 236void ff_hevc_put_hevc_qpel_h4_8_avx512icl(int16_t *dst, uint8_t *_src, ptrdiff_t _srcstride, int height, intptr_t mx, intptr_t my, int width); 237void ff_hevc_put_hevc_qpel_h8_8_avx512icl(int16_t *dst, uint8_t *_src, ptrdiff_t _srcstride, int height, intptr_t mx, intptr_t my, int width); 238void ff_hevc_put_hevc_qpel_h16_8_avx512icl(int16_t *dst, uint8_t *_src, ptrdiff_t _srcstride, int height, intptr_t mx, intptr_t my, int width); 239void ff_hevc_put_hevc_qpel_h32_8_avx512icl(int16_t *dst, uint8_t *_src, ptrdiff_t _srcstride, int height, intptr_t mx, intptr_t my, int width); 240void ff_hevc_put_hevc_qpel_h64_8_avx512icl(int16_t *dst, uint8_t *_src, ptrdiff_t _srcstride, int height, intptr_t mx, intptr_t my, int width); 241void ff_hevc_put_hevc_qpel_hv8_8_avx512icl(int16_t *dst, uint8_t *_src, ptrdiff_t _srcstride, int height, intptr_t mx, intptr_t my, int width); 242 243/////////////////////////////////////////////////////////////////////////////// 244// TRANSFORM_ADD 245/////////////////////////////////////////////////////////////////////////////// 246 247void ff_hevc_add_residual_4_8_mmxext(uint8_t *dst, int16_t *res, ptrdiff_t stride); 248void ff_hevc_add_residual_8_8_sse2(uint8_t *dst, int16_t *res, ptrdiff_t stride); 249void ff_hevc_add_residual_16_8_sse2(uint8_t *dst, int16_t *res, ptrdiff_t stride); 250void ff_hevc_add_residual_32_8_sse2(uint8_t *dst, int16_t *res, ptrdiff_t stride); 251 252void ff_hevc_add_residual_8_8_avx(uint8_t *dst, int16_t *res, ptrdiff_t stride); 253void ff_hevc_add_residual_16_8_avx(uint8_t *dst, int16_t *res, ptrdiff_t stride); 254void ff_hevc_add_residual_32_8_avx(uint8_t *dst, int16_t *res, ptrdiff_t stride); 255 256void ff_hevc_add_residual_32_8_avx2(uint8_t *dst, int16_t *res, ptrdiff_t stride); 257 258void ff_hevc_add_residual_4_10_mmxext(uint8_t *dst, int16_t *res, ptrdiff_t stride); 259void ff_hevc_add_residual_8_10_sse2(uint8_t *dst, int16_t *res, ptrdiff_t stride); 260void ff_hevc_add_residual_16_10_sse2(uint8_t *dst, int16_t *res, ptrdiff_t stride); 261void ff_hevc_add_residual_32_10_sse2(uint8_t *dst, int16_t *res, ptrdiff_t stride); 262 263void ff_hevc_add_residual_16_10_avx2(uint8_t *dst, int16_t *res, ptrdiff_t stride); 264void ff_hevc_add_residual_32_10_avx2(uint8_t *dst, int16_t *res, ptrdiff_t stride); 265 266#endif // AVCODEC_X86_HEVCDSP_H 267