| 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 idct_dc_proto(size, bitd, opt) \ |
| 33 | void ff_hevc_idct##size##_dc_add_##bitd##_##opt(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride) |
| 34 | |
| 35 | #define PEL_LINK(dst, idx1, idx2, idx3, name, D, opt) \ |
| 36 | dst[idx1][idx2][idx3] = ff_hevc_put_hevc_ ## name ## _ ## D ## _##opt; \ |
| 37 | dst ## _bi[idx1][idx2][idx3] = ff_hevc_put_hevc_bi_ ## name ## _ ## D ## _##opt; \ |
| 38 | dst ## _uni[idx1][idx2][idx3] = ff_hevc_put_hevc_uni_ ## name ## _ ## D ## _##opt; \ |
| 39 | dst ## _uni_w[idx1][idx2][idx3] = ff_hevc_put_hevc_uni_w_ ## name ## _ ## D ## _##opt; \ |
| 40 | dst ## _bi_w[idx1][idx2][idx3] = ff_hevc_put_hevc_bi_w_ ## name ## _ ## D ## _##opt |
| 41 | |
| 42 | |
| 43 | #define PEL_PROTOTYPE(name, D, opt) \ |
| 44 | void 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); \ |
| 45 | void 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); \ |
| 46 | void 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); \ |
| 47 | void 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); \ |
| 48 | void 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) |
| 49 | |
| 50 | |
| 51 | /////////////////////////////////////////////////////////////////////////////// |
| 52 | // MC functions |
| 53 | /////////////////////////////////////////////////////////////////////////////// |
| 54 | |
| 55 | #define EPEL_PROTOTYPES(fname, bitd, opt) \ |
| 56 | PEL_PROTOTYPE(fname##4, bitd, opt); \ |
| 57 | PEL_PROTOTYPE(fname##6, bitd, opt); \ |
| 58 | PEL_PROTOTYPE(fname##8, bitd, opt); \ |
| 59 | PEL_PROTOTYPE(fname##12, bitd, opt); \ |
| 60 | PEL_PROTOTYPE(fname##16, bitd, opt); \ |
| 61 | PEL_PROTOTYPE(fname##24, bitd, opt); \ |
| 62 | PEL_PROTOTYPE(fname##32, bitd, opt); \ |
| 63 | PEL_PROTOTYPE(fname##48, bitd, opt); \ |
| 64 | PEL_PROTOTYPE(fname##64, bitd, opt) |
| 65 | |
| 66 | #define QPEL_PROTOTYPES(fname, bitd, opt) \ |
| 67 | PEL_PROTOTYPE(fname##4, bitd, opt); \ |
| 68 | PEL_PROTOTYPE(fname##8, bitd, opt); \ |
| 69 | PEL_PROTOTYPE(fname##12, bitd, opt); \ |
| 70 | PEL_PROTOTYPE(fname##16, bitd, opt); \ |
| 71 | PEL_PROTOTYPE(fname##24, bitd, opt); \ |
| 72 | PEL_PROTOTYPE(fname##32, bitd, opt); \ |
| 73 | PEL_PROTOTYPE(fname##48, bitd, opt); \ |
| 74 | PEL_PROTOTYPE(fname##64, bitd, opt) |
| 75 | |
| 76 | #define WEIGHTING_PROTOTYPE(width, bitd, opt) \ |
| 77 | void ff_hevc_put_hevc_uni_w##width##_##bitd##_##opt(uint8_t *dst, ptrdiff_t dststride, int16_t *_src, ptrdiff_t _srcstride, int height, int denom, int _wx, int _ox); \ |
| 78 | void ff_hevc_put_hevc_bi_w##width##_##bitd##_##opt(uint8_t *dst, ptrdiff_t dststride, int16_t *_src, ptrdiff_t _srcstride, int16_t *_src2, int height, int denom, int _wx0, int _wx1, int _ox0, int _ox1) |
| 79 | |
| 80 | #define WEIGHTING_PROTOTYPES(bitd, opt) \ |
| 81 | WEIGHTING_PROTOTYPE(2, bitd, opt); \ |
| 82 | WEIGHTING_PROTOTYPE(4, bitd, opt); \ |
| 83 | WEIGHTING_PROTOTYPE(6, bitd, opt); \ |
| 84 | WEIGHTING_PROTOTYPE(8, bitd, opt); \ |
| 85 | WEIGHTING_PROTOTYPE(12, bitd, opt); \ |
| 86 | WEIGHTING_PROTOTYPE(16, bitd, opt); \ |
| 87 | WEIGHTING_PROTOTYPE(24, bitd, opt); \ |
| 88 | WEIGHTING_PROTOTYPE(32, bitd, opt); \ |
| 89 | WEIGHTING_PROTOTYPE(48, bitd, opt); \ |
| 90 | WEIGHTING_PROTOTYPE(64, bitd, opt) |
| 91 | |
| 92 | |
| 93 | /////////////////////////////////////////////////////////////////////////////// |
| 94 | // QPEL_PIXELS EPEL_PIXELS |
| 95 | /////////////////////////////////////////////////////////////////////////////// |
| 96 | EPEL_PROTOTYPES(pel_pixels , 8, sse4); |
| 97 | EPEL_PROTOTYPES(pel_pixels , 10, sse4); |
| 98 | EPEL_PROTOTYPES(pel_pixels , 12, sse4); |
| 99 | /////////////////////////////////////////////////////////////////////////////// |
| 100 | // EPEL |
| 101 | /////////////////////////////////////////////////////////////////////////////// |
| 102 | EPEL_PROTOTYPES(epel_h , 8, sse4); |
| 103 | EPEL_PROTOTYPES(epel_h , 10, sse4); |
| 104 | EPEL_PROTOTYPES(epel_h , 12, sse4); |
| 105 | |
| 106 | EPEL_PROTOTYPES(epel_v , 8, sse4); |
| 107 | EPEL_PROTOTYPES(epel_v , 10, sse4); |
| 108 | EPEL_PROTOTYPES(epel_v , 12, sse4); |
| 109 | |
| 110 | EPEL_PROTOTYPES(epel_hv , 8, sse4); |
| 111 | EPEL_PROTOTYPES(epel_hv , 10, sse4); |
| 112 | EPEL_PROTOTYPES(epel_hv , 12, sse4); |
| 113 | |
| 114 | /////////////////////////////////////////////////////////////////////////////// |
| 115 | // QPEL |
| 116 | /////////////////////////////////////////////////////////////////////////////// |
| 117 | QPEL_PROTOTYPES(qpel_h , 8, sse4); |
| 118 | QPEL_PROTOTYPES(qpel_h , 10, sse4); |
| 119 | QPEL_PROTOTYPES(qpel_h , 12, sse4); |
| 120 | |
| 121 | QPEL_PROTOTYPES(qpel_v, 8, sse4); |
| 122 | QPEL_PROTOTYPES(qpel_v, 10, sse4); |
| 123 | QPEL_PROTOTYPES(qpel_v, 12, sse4); |
| 124 | |
| 125 | QPEL_PROTOTYPES(qpel_hv, 8, sse4); |
| 126 | QPEL_PROTOTYPES(qpel_hv, 10, sse4); |
| 127 | QPEL_PROTOTYPES(qpel_hv, 12, sse4); |
| 128 | |
| 129 | |
| 130 | WEIGHTING_PROTOTYPES(8, sse4); |
| 131 | WEIGHTING_PROTOTYPES(10, sse4); |
| 132 | WEIGHTING_PROTOTYPES(12, sse4); |
| 133 | |
| 134 | /////////////////////////////////////////////////////////////////////////////// |
| 135 | // TRANSFORM_ADD |
| 136 | /////////////////////////////////////////////////////////////////////////////// |
| 137 | void ff_hevc_transform_add4_8_mmxext(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride); |
| 138 | void ff_hevc_transform_add8_8_sse2(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride); |
| 139 | void ff_hevc_transform_add16_8_sse2(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride); |
| 140 | void ff_hevc_transform_add32_8_sse2(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride); |
| 141 | |
| 142 | void ff_hevc_transform_add8_8_avx(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride); |
| 143 | void ff_hevc_transform_add16_8_avx(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride); |
| 144 | void ff_hevc_transform_add32_8_avx(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride); |
| 145 | |
| 146 | void ff_hevc_transform_add32_8_avx2(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride); |
| 147 | |
| 148 | void ff_hevc_transform_add4_10_mmxext(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride); |
| 149 | void ff_hevc_transform_add8_10_sse2(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride); |
| 150 | void ff_hevc_transform_add16_10_sse2(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride); |
| 151 | void ff_hevc_transform_add32_10_sse2(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride); |
| 152 | |
| 153 | void ff_hevc_transform_add16_10_avx2(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride); |
| 154 | void ff_hevc_transform_add32_10_avx2(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride); |
| 155 | |
| 156 | #endif // AVCODEC_X86_HEVCDSP_H |