| 1 | /***************************************************************************** |
| 2 | * Copyright (C) 2013 x265 project |
| 3 | * |
| 4 | * Authors: Steve Borho <steve@borho.org> |
| 5 | * |
| 6 | * This program is free software; you can redistribute it and/or modify |
| 7 | * it under the terms of the GNU General Public License as published by |
| 8 | * the Free Software Foundation; either version 2 of the License, or |
| 9 | * (at your option) any later version. |
| 10 | * |
| 11 | * This program is distributed in the hope that it will be useful, |
| 12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 14 | * GNU General Public License for more details. |
| 15 | * |
| 16 | * You should have received a copy of the GNU General Public License |
| 17 | * along with this program; if not, write to the Free Software |
| 18 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA. |
| 19 | * |
| 20 | * This program is also available under a commercial proprietary license. |
| 21 | * For more information, contact us at license @ x265.com. |
| 22 | *****************************************************************************/ |
| 23 | |
| 24 | #ifndef X265_BLOCKCOPY8_H |
| 25 | #define X265_BLOCKCOPY8_H |
| 26 | |
| 27 | void x265_cpy2Dto1D_shl_4_sse2(int16_t* dst, const int16_t* src, intptr_t srcStride, int shift); |
| 28 | void x265_cpy2Dto1D_shl_8_sse2(int16_t* dst, const int16_t* src, intptr_t srcStride, int shift); |
| 29 | void x265_cpy2Dto1D_shl_16_sse2(int16_t* dst, const int16_t* src, intptr_t srcStride, int shift); |
| 30 | void x265_cpy2Dto1D_shl_32_sse2(int16_t* dst, const int16_t* src, intptr_t srcStride, int shift); |
| 31 | void x265_cpy2Dto1D_shr_4_sse2(int16_t* dst, const int16_t* src, intptr_t srcStride, int shift); |
| 32 | void x265_cpy2Dto1D_shr_8_sse2(int16_t* dst, const int16_t* src, intptr_t srcStride, int shift); |
| 33 | void x265_cpy2Dto1D_shr_16_sse2(int16_t* dst, const int16_t* src, intptr_t srcStride, int shift); |
| 34 | void x265_cpy2Dto1D_shr_32_sse2(int16_t* dst, const int16_t* src, intptr_t srcStride, int shift); |
| 35 | void x265_cpy1Dto2D_shl_4_avx2(int16_t* dst, const int16_t* src, intptr_t srcStride, int shift); |
| 36 | void x265_cpy1Dto2D_shl_8_avx2(int16_t* dst, const int16_t* src, intptr_t srcStride, int shift); |
| 37 | void x265_cpy1Dto2D_shl_16_avx2(int16_t* dst, const int16_t* src, intptr_t srcStride, int shift); |
| 38 | void x265_cpy1Dto2D_shl_32_avx2(int16_t* dst, const int16_t* src, intptr_t srcStride, int shift); |
| 39 | void x265_cpy1Dto2D_shl_4_sse2(int16_t* dst, const int16_t* src, intptr_t dstStride, int shift); |
| 40 | void x265_cpy1Dto2D_shl_8_sse2(int16_t* dst, const int16_t* src, intptr_t dstStride, int shift); |
| 41 | void x265_cpy1Dto2D_shl_16_sse2(int16_t* dst, const int16_t* src, intptr_t dstStride, int shift); |
| 42 | void x265_cpy1Dto2D_shl_32_sse2(int16_t* dst, const int16_t* src, intptr_t dstStride, int shift); |
| 43 | void x265_cpy1Dto2D_shr_4_avx2(int16_t* dst, const int16_t* src, intptr_t srcStride, int shift); |
| 44 | void x265_cpy1Dto2D_shr_8_avx2(int16_t* dst, const int16_t* src, intptr_t dstStride, int shift); |
| 45 | void x265_cpy1Dto2D_shr_16_avx2(int16_t* dst, const int16_t* src, intptr_t dstStride, int shift); |
| 46 | void x265_cpy1Dto2D_shr_32_avx2(int16_t* dst, const int16_t* src, intptr_t dstStride, int shift); |
| 47 | void x265_cpy1Dto2D_shr_4_sse2(int16_t* dst, const int16_t* src, intptr_t dstStride, int shift); |
| 48 | void x265_cpy1Dto2D_shr_8_sse2(int16_t* dst, const int16_t* src, intptr_t dstStride, int shift); |
| 49 | void x265_cpy1Dto2D_shr_16_sse2(int16_t* dst, const int16_t* src, intptr_t dstStride, int shift); |
| 50 | void x265_cpy1Dto2D_shr_32_sse2(int16_t* dst, const int16_t* src, intptr_t dstStride, int shift); |
| 51 | uint32_t x265_copy_cnt_4_sse4(int16_t* dst, const int16_t* src, intptr_t srcStride); |
| 52 | uint32_t x265_copy_cnt_8_sse4(int16_t* dst, const int16_t* src, intptr_t srcStride); |
| 53 | uint32_t x265_copy_cnt_16_sse4(int16_t* dst, const int16_t* src, intptr_t srcStride); |
| 54 | uint32_t x265_copy_cnt_32_sse4(int16_t* dst, const int16_t* src, intptr_t srcStride); |
| 55 | uint32_t x265_copy_cnt_4_avx2(int16_t* dst, const int16_t* src, intptr_t srcStride); |
| 56 | uint32_t x265_copy_cnt_8_avx2(int16_t* dst, const int16_t* src, intptr_t srcStride); |
| 57 | uint32_t x265_copy_cnt_16_avx2(int16_t* dst, const int16_t* src, intptr_t srcStride); |
| 58 | uint32_t x265_copy_cnt_32_avx2(int16_t* dst, const int16_t* src, intptr_t srcStride); |
| 59 | |
| 60 | #define SETUP_BLOCKCOPY_FUNC(W, H, cpu) \ |
| 61 | void x265_blockcopy_pp_ ## W ## x ## H ## cpu(pixel* a, intptr_t stridea, const pixel* b, intptr_t strideb); \ |
| 62 | void x265_blockcopy_sp_ ## W ## x ## H ## cpu(pixel* a, intptr_t stridea, const int16_t* b, intptr_t strideb); \ |
| 63 | void x265_blockcopy_ss_ ## W ## x ## H ## cpu(int16_t* a, intptr_t stridea, const int16_t* b, intptr_t strideb); |
| 64 | |
| 65 | #define SETUP_BLOCKCOPY_PS(W, H, cpu) \ |
| 66 | void x265_blockcopy_ps_ ## W ## x ## H ## cpu(int16_t* dst, intptr_t dstStride, const pixel* src, intptr_t srcStride); |
| 67 | |
| 68 | #define SETUP_BLOCKCOPY_SP(W, H, cpu) \ |
| 69 | void x265_blockcopy_sp_ ## W ## x ## H ## cpu(pixel* a, intptr_t stridea, const int16_t* b, intptr_t strideb); |
| 70 | |
| 71 | #define SETUP_BLOCKCOPY_SS_PP(W, H, cpu) \ |
| 72 | void x265_blockcopy_pp_ ## W ## x ## H ## cpu(pixel* a, intptr_t stridea, const pixel* b, intptr_t strideb); \ |
| 73 | void x265_blockcopy_ss_ ## W ## x ## H ## cpu(int16_t* a, intptr_t stridea, const int16_t* b, intptr_t strideb); |
| 74 | |
| 75 | #define BLOCKCOPY_COMMON(cpu) \ |
| 76 | SETUP_BLOCKCOPY_FUNC(4, 4, cpu); \ |
| 77 | SETUP_BLOCKCOPY_FUNC(4, 2, cpu); \ |
| 78 | SETUP_BLOCKCOPY_FUNC(8, 8, cpu); \ |
| 79 | SETUP_BLOCKCOPY_FUNC(8, 4, cpu); \ |
| 80 | SETUP_BLOCKCOPY_FUNC(4, 8, cpu); \ |
| 81 | SETUP_BLOCKCOPY_FUNC(8, 6, cpu); \ |
| 82 | SETUP_BLOCKCOPY_FUNC(8, 2, cpu); \ |
| 83 | SETUP_BLOCKCOPY_FUNC(16, 16, cpu); \ |
| 84 | SETUP_BLOCKCOPY_FUNC(16, 8, cpu); \ |
| 85 | SETUP_BLOCKCOPY_FUNC(8, 16, cpu); \ |
| 86 | SETUP_BLOCKCOPY_FUNC(16, 12, cpu); \ |
| 87 | SETUP_BLOCKCOPY_FUNC(12, 16, cpu); \ |
| 88 | SETUP_BLOCKCOPY_FUNC(16, 4, cpu); \ |
| 89 | SETUP_BLOCKCOPY_FUNC(4, 16, cpu); \ |
| 90 | SETUP_BLOCKCOPY_FUNC(32, 32, cpu); \ |
| 91 | SETUP_BLOCKCOPY_FUNC(32, 16, cpu); \ |
| 92 | SETUP_BLOCKCOPY_FUNC(16, 32, cpu); \ |
| 93 | SETUP_BLOCKCOPY_FUNC(32, 24, cpu); \ |
| 94 | SETUP_BLOCKCOPY_FUNC(24, 32, cpu); \ |
| 95 | SETUP_BLOCKCOPY_FUNC(32, 8, cpu); \ |
| 96 | SETUP_BLOCKCOPY_FUNC(8, 32, cpu); \ |
| 97 | SETUP_BLOCKCOPY_FUNC(64, 64, cpu); \ |
| 98 | SETUP_BLOCKCOPY_FUNC(64, 32, cpu); \ |
| 99 | SETUP_BLOCKCOPY_FUNC(32, 64, cpu); \ |
| 100 | SETUP_BLOCKCOPY_FUNC(64, 48, cpu); \ |
| 101 | SETUP_BLOCKCOPY_FUNC(48, 64, cpu); \ |
| 102 | SETUP_BLOCKCOPY_FUNC(64, 16, cpu); \ |
| 103 | SETUP_BLOCKCOPY_FUNC(16, 64, cpu); |
| 104 | |
| 105 | #define BLOCKCOPY_SP(cpu) \ |
| 106 | SETUP_BLOCKCOPY_SP(2, 4, cpu); \ |
| 107 | SETUP_BLOCKCOPY_SP(2, 8, cpu); \ |
| 108 | SETUP_BLOCKCOPY_SP(6, 8, cpu); \ |
| 109 | \ |
| 110 | SETUP_BLOCKCOPY_SP(2, 16, cpu); \ |
| 111 | SETUP_BLOCKCOPY_SP(4, 32, cpu); \ |
| 112 | SETUP_BLOCKCOPY_SP(6, 16, cpu); \ |
| 113 | SETUP_BLOCKCOPY_SP(8, 12, cpu); \ |
| 114 | SETUP_BLOCKCOPY_SP(8, 64, cpu); \ |
| 115 | SETUP_BLOCKCOPY_SP(12, 32, cpu); \ |
| 116 | SETUP_BLOCKCOPY_SP(16, 24, cpu); \ |
| 117 | SETUP_BLOCKCOPY_SP(24, 64, cpu); \ |
| 118 | SETUP_BLOCKCOPY_SP(32, 48, cpu); |
| 119 | |
| 120 | #define BLOCKCOPY_SS_PP(cpu) \ |
| 121 | SETUP_BLOCKCOPY_SS_PP(2, 4, cpu); \ |
| 122 | SETUP_BLOCKCOPY_SS_PP(2, 8, cpu); \ |
| 123 | SETUP_BLOCKCOPY_SS_PP(6, 8, cpu); \ |
| 124 | \ |
| 125 | SETUP_BLOCKCOPY_SS_PP(2, 16, cpu); \ |
| 126 | SETUP_BLOCKCOPY_SS_PP(4, 32, cpu); \ |
| 127 | SETUP_BLOCKCOPY_SS_PP(6, 16, cpu); \ |
| 128 | SETUP_BLOCKCOPY_SS_PP(8, 12, cpu); \ |
| 129 | SETUP_BLOCKCOPY_SS_PP(8, 64, cpu); \ |
| 130 | SETUP_BLOCKCOPY_SS_PP(12, 32, cpu); \ |
| 131 | SETUP_BLOCKCOPY_SS_PP(16, 24, cpu); \ |
| 132 | SETUP_BLOCKCOPY_SS_PP(24, 64, cpu); \ |
| 133 | SETUP_BLOCKCOPY_SS_PP(32, 48, cpu); |
| 134 | |
| 135 | |
| 136 | #define BLOCKCOPY_PS(cpu) \ |
| 137 | SETUP_BLOCKCOPY_PS(2, 4, cpu); \ |
| 138 | SETUP_BLOCKCOPY_PS(2, 8, cpu); \ |
| 139 | SETUP_BLOCKCOPY_PS(4, 2, cpu); \ |
| 140 | SETUP_BLOCKCOPY_PS(4, 4, cpu); \ |
| 141 | SETUP_BLOCKCOPY_PS(4, 8, cpu); \ |
| 142 | SETUP_BLOCKCOPY_PS(4, 16, cpu); \ |
| 143 | SETUP_BLOCKCOPY_PS(6, 8, cpu); \ |
| 144 | SETUP_BLOCKCOPY_PS(8, 2, cpu); \ |
| 145 | SETUP_BLOCKCOPY_PS(8, 4, cpu); \ |
| 146 | SETUP_BLOCKCOPY_PS(8, 6, cpu); \ |
| 147 | SETUP_BLOCKCOPY_PS(8, 8, cpu); \ |
| 148 | SETUP_BLOCKCOPY_PS(8, 16, cpu); \ |
| 149 | SETUP_BLOCKCOPY_PS(8, 32, cpu); \ |
| 150 | SETUP_BLOCKCOPY_PS(12, 16, cpu); \ |
| 151 | SETUP_BLOCKCOPY_PS(16, 4, cpu); \ |
| 152 | SETUP_BLOCKCOPY_PS(16, 8, cpu); \ |
| 153 | SETUP_BLOCKCOPY_PS(16, 12, cpu); \ |
| 154 | SETUP_BLOCKCOPY_PS(16, 16, cpu); \ |
| 155 | SETUP_BLOCKCOPY_PS(16, 32, cpu); \ |
| 156 | SETUP_BLOCKCOPY_PS(24, 32, cpu); \ |
| 157 | SETUP_BLOCKCOPY_PS(32, 8, cpu); \ |
| 158 | SETUP_BLOCKCOPY_PS(32, 16, cpu); \ |
| 159 | SETUP_BLOCKCOPY_PS(32, 24, cpu); \ |
| 160 | SETUP_BLOCKCOPY_PS(32, 32, cpu); \ |
| 161 | SETUP_BLOCKCOPY_PS(16, 64, cpu); \ |
| 162 | SETUP_BLOCKCOPY_PS(32, 64, cpu); \ |
| 163 | SETUP_BLOCKCOPY_PS(48, 64, cpu); \ |
| 164 | SETUP_BLOCKCOPY_PS(64, 16, cpu); \ |
| 165 | SETUP_BLOCKCOPY_PS(64, 32, cpu); \ |
| 166 | SETUP_BLOCKCOPY_PS(64, 48, cpu); \ |
| 167 | SETUP_BLOCKCOPY_PS(64, 64, cpu); \ |
| 168 | \ |
| 169 | SETUP_BLOCKCOPY_PS(2, 16, cpu); \ |
| 170 | SETUP_BLOCKCOPY_PS(4, 32, cpu); \ |
| 171 | SETUP_BLOCKCOPY_PS(6, 16, cpu); \ |
| 172 | SETUP_BLOCKCOPY_PS(8, 12, cpu); \ |
| 173 | SETUP_BLOCKCOPY_PS(8, 64, cpu); \ |
| 174 | SETUP_BLOCKCOPY_PS(12, 32, cpu); \ |
| 175 | SETUP_BLOCKCOPY_PS(16, 24, cpu); \ |
| 176 | SETUP_BLOCKCOPY_PS(24, 64, cpu); \ |
| 177 | SETUP_BLOCKCOPY_PS(32, 48, cpu); |
| 178 | |
| 179 | BLOCKCOPY_COMMON(_sse2); |
| 180 | BLOCKCOPY_SS_PP(_sse2); |
| 181 | BLOCKCOPY_SP(_sse4); |
| 182 | BLOCKCOPY_PS(_sse4); |
| 183 | |
| 184 | BLOCKCOPY_SP(_sse2); |
| 185 | |
| 186 | void x265_blockfill_s_4x4_sse2(int16_t* dst, intptr_t dstride, int16_t val); |
| 187 | void x265_blockfill_s_8x8_sse2(int16_t* dst, intptr_t dstride, int16_t val); |
| 188 | void x265_blockfill_s_16x16_sse2(int16_t* dst, intptr_t dstride, int16_t val); |
| 189 | void x265_blockfill_s_32x32_sse2(int16_t* dst, intptr_t dstride, int16_t val); |
| 190 | void x265_blockcopy_ss_16x4_avx(int16_t* dst, intptr_t dstStride, const int16_t* src, intptr_t srcStride); |
| 191 | void x265_blockcopy_ss_16x8_avx(int16_t* dst, intptr_t dstStride, const int16_t* src, intptr_t srcStride); |
| 192 | void x265_blockcopy_ss_16x12_avx(int16_t* dst, intptr_t dstStride, const int16_t* src, intptr_t srcStride); |
| 193 | void x265_blockcopy_ss_16x16_avx(int16_t* dst, intptr_t dstStride, const int16_t* src, intptr_t srcStride); |
| 194 | void x265_blockcopy_ss_16x24_avx(int16_t* dst, intptr_t dstStride, const int16_t* src, intptr_t srcStride); |
| 195 | void x265_blockcopy_ss_16x32_avx(int16_t* dst, intptr_t dstStride, const int16_t* src, intptr_t srcStride); |
| 196 | void x265_blockcopy_ss_16x64_avx(int16_t* dst, intptr_t dstStride, const int16_t* src, intptr_t srcStride); |
| 197 | void x265_blockcopy_ss_64x16_avx(int16_t* dst, intptr_t dstStride, const int16_t* src, intptr_t srcStride); |
| 198 | void x265_blockcopy_ss_64x32_avx(int16_t* dst, intptr_t dstStride, const int16_t* src, intptr_t srcStride); |
| 199 | void x265_blockcopy_ss_64x48_avx(int16_t* dst, intptr_t dstStride, const int16_t* src, intptr_t srcStride); |
| 200 | void x265_blockcopy_ss_64x64_avx(int16_t* dst, intptr_t dstStride, const int16_t* src, intptr_t srcStride); |
| 201 | |
| 202 | void x265_blockcopy_pp_32x8_avx(pixel* a, intptr_t stridea, const pixel* b, intptr_t strideb); |
| 203 | void x265_blockcopy_pp_32x16_avx(pixel* a, intptr_t stridea, const pixel* b, intptr_t strideb); |
| 204 | void x265_blockcopy_pp_32x24_avx(pixel* a, intptr_t stridea, const pixel* b, intptr_t strideb); |
| 205 | void x265_blockcopy_pp_32x32_avx(pixel* a, intptr_t stridea, const pixel* b, intptr_t strideb); |
| 206 | void x265_blockcopy_pp_32x48_avx(pixel* a, intptr_t stridea, const pixel* b, intptr_t strideb); |
| 207 | void x265_blockcopy_pp_32x64_avx(pixel* a, intptr_t stridea, const pixel* b, intptr_t strideb); |
| 208 | |
| 209 | void x265_blockfill_s_16x16_avx2(int16_t* dst, intptr_t dstride, int16_t val); |
| 210 | void x265_blockfill_s_32x32_avx2(int16_t* dst, intptr_t dstride, int16_t val); |
| 211 | |
| 212 | #undef BLOCKCOPY_COMMON |
| 213 | #undef BLOCKCOPY_SS_PP |
| 214 | #undef BLOCKCOPY_SP |
| 215 | #undef BLOCKCOPY_PS |
| 216 | #undef SETUP_BLOCKCOPY_PS |
| 217 | #undef SETUP_BLOCKCOPY_SP |
| 218 | #undef SETUP_BLOCKCOPY_SS_PP |
| 219 | #undef SETUP_BLOCKCOPY_FUNC |
| 220 | |
| 221 | #endif // ifndef X265_I386_PIXEL_H |