| 1 | /* |
| 2 | * Copyright (C) 2002-2012 Michael Niedermayer |
| 3 | * Copyright (C) 2012 Ronald S. Bultje |
| 4 | * |
| 5 | * This file is part of FFmpeg. |
| 6 | * |
| 7 | * FFmpeg is free software; you can redistribute it and/or |
| 8 | * modify it under the terms of the GNU Lesser General Public |
| 9 | * License as published by the Free Software Foundation; either |
| 10 | * version 2.1 of the License, or (at your option) any later version. |
| 11 | * |
| 12 | * FFmpeg is distributed in the hope that it will be useful, |
| 13 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 14 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 15 | * Lesser General Public License for more details. |
| 16 | * |
| 17 | * You should have received a copy of the GNU Lesser General Public |
| 18 | * License along with FFmpeg; if not, write to the Free Software |
| 19 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
| 20 | */ |
| 21 | |
| 22 | #include "config.h" |
| 23 | #include "libavutil/attributes.h" |
| 24 | #include "libavutil/avassert.h" |
| 25 | #include "libavutil/common.h" |
| 26 | #include "libavutil/cpu.h" |
| 27 | #include "libavutil/mem.h" |
| 28 | #include "libavutil/x86/asm.h" |
| 29 | #include "libavutil/x86/cpu.h" |
| 30 | #include "libavcodec/videodsp.h" |
| 31 | |
| 32 | #if HAVE_YASM |
| 33 | typedef void emu_edge_vfix_func(uint8_t *dst, x86_reg dst_stride, |
| 34 | const uint8_t *src, x86_reg src_stride, |
| 35 | x86_reg start_y, x86_reg end_y, x86_reg bh); |
| 36 | typedef void emu_edge_vvar_func(uint8_t *dst, x86_reg dst_stride, |
| 37 | const uint8_t *src, x86_reg src_stride, |
| 38 | x86_reg start_y, x86_reg end_y, x86_reg bh, |
| 39 | x86_reg w); |
| 40 | |
| 41 | extern emu_edge_vfix_func ff_emu_edge_vfix1_mmx; |
| 42 | extern emu_edge_vfix_func ff_emu_edge_vfix2_mmx; |
| 43 | extern emu_edge_vfix_func ff_emu_edge_vfix3_mmx; |
| 44 | extern emu_edge_vfix_func ff_emu_edge_vfix4_mmx; |
| 45 | extern emu_edge_vfix_func ff_emu_edge_vfix5_mmx; |
| 46 | extern emu_edge_vfix_func ff_emu_edge_vfix6_mmx; |
| 47 | extern emu_edge_vfix_func ff_emu_edge_vfix7_mmx; |
| 48 | extern emu_edge_vfix_func ff_emu_edge_vfix8_mmx; |
| 49 | extern emu_edge_vfix_func ff_emu_edge_vfix9_mmx; |
| 50 | extern emu_edge_vfix_func ff_emu_edge_vfix10_mmx; |
| 51 | extern emu_edge_vfix_func ff_emu_edge_vfix11_mmx; |
| 52 | extern emu_edge_vfix_func ff_emu_edge_vfix12_mmx; |
| 53 | extern emu_edge_vfix_func ff_emu_edge_vfix13_mmx; |
| 54 | extern emu_edge_vfix_func ff_emu_edge_vfix14_mmx; |
| 55 | extern emu_edge_vfix_func ff_emu_edge_vfix15_mmx; |
| 56 | extern emu_edge_vfix_func ff_emu_edge_vfix16_mmx; |
| 57 | extern emu_edge_vfix_func ff_emu_edge_vfix17_mmx; |
| 58 | extern emu_edge_vfix_func ff_emu_edge_vfix18_mmx; |
| 59 | extern emu_edge_vfix_func ff_emu_edge_vfix19_mmx; |
| 60 | extern emu_edge_vfix_func ff_emu_edge_vfix20_mmx; |
| 61 | extern emu_edge_vfix_func ff_emu_edge_vfix21_mmx; |
| 62 | extern emu_edge_vfix_func ff_emu_edge_vfix22_mmx; |
| 63 | #if ARCH_X86_32 |
| 64 | static emu_edge_vfix_func * const vfixtbl_mmx[22] = { |
| 65 | &ff_emu_edge_vfix1_mmx, &ff_emu_edge_vfix2_mmx, &ff_emu_edge_vfix3_mmx, |
| 66 | &ff_emu_edge_vfix4_mmx, &ff_emu_edge_vfix5_mmx, &ff_emu_edge_vfix6_mmx, |
| 67 | &ff_emu_edge_vfix7_mmx, &ff_emu_edge_vfix8_mmx, &ff_emu_edge_vfix9_mmx, |
| 68 | &ff_emu_edge_vfix10_mmx, &ff_emu_edge_vfix11_mmx, &ff_emu_edge_vfix12_mmx, |
| 69 | &ff_emu_edge_vfix13_mmx, &ff_emu_edge_vfix14_mmx, &ff_emu_edge_vfix15_mmx, |
| 70 | &ff_emu_edge_vfix16_mmx, &ff_emu_edge_vfix17_mmx, &ff_emu_edge_vfix18_mmx, |
| 71 | &ff_emu_edge_vfix19_mmx, &ff_emu_edge_vfix20_mmx, &ff_emu_edge_vfix21_mmx, |
| 72 | &ff_emu_edge_vfix22_mmx |
| 73 | }; |
| 74 | #endif |
| 75 | extern emu_edge_vvar_func ff_emu_edge_vvar_mmx; |
| 76 | extern emu_edge_vfix_func ff_emu_edge_vfix16_sse; |
| 77 | extern emu_edge_vfix_func ff_emu_edge_vfix17_sse; |
| 78 | extern emu_edge_vfix_func ff_emu_edge_vfix18_sse; |
| 79 | extern emu_edge_vfix_func ff_emu_edge_vfix19_sse; |
| 80 | extern emu_edge_vfix_func ff_emu_edge_vfix20_sse; |
| 81 | extern emu_edge_vfix_func ff_emu_edge_vfix21_sse; |
| 82 | extern emu_edge_vfix_func ff_emu_edge_vfix22_sse; |
| 83 | static emu_edge_vfix_func * const vfixtbl_sse[22] = { |
| 84 | ff_emu_edge_vfix1_mmx, ff_emu_edge_vfix2_mmx, ff_emu_edge_vfix3_mmx, |
| 85 | ff_emu_edge_vfix4_mmx, ff_emu_edge_vfix5_mmx, ff_emu_edge_vfix6_mmx, |
| 86 | ff_emu_edge_vfix7_mmx, ff_emu_edge_vfix8_mmx, ff_emu_edge_vfix9_mmx, |
| 87 | ff_emu_edge_vfix10_mmx, ff_emu_edge_vfix11_mmx, ff_emu_edge_vfix12_mmx, |
| 88 | ff_emu_edge_vfix13_mmx, ff_emu_edge_vfix14_mmx, ff_emu_edge_vfix15_mmx, |
| 89 | ff_emu_edge_vfix16_sse, ff_emu_edge_vfix17_sse, ff_emu_edge_vfix18_sse, |
| 90 | ff_emu_edge_vfix19_sse, ff_emu_edge_vfix20_sse, ff_emu_edge_vfix21_sse, |
| 91 | ff_emu_edge_vfix22_sse |
| 92 | }; |
| 93 | extern emu_edge_vvar_func ff_emu_edge_vvar_sse; |
| 94 | |
| 95 | typedef void emu_edge_hfix_func(uint8_t *dst, x86_reg dst_stride, |
| 96 | x86_reg start_x, x86_reg bh); |
| 97 | typedef void emu_edge_hvar_func(uint8_t *dst, x86_reg dst_stride, |
| 98 | x86_reg start_x, x86_reg n_words, x86_reg bh); |
| 99 | |
| 100 | extern emu_edge_hfix_func ff_emu_edge_hfix2_mmx; |
| 101 | extern emu_edge_hfix_func ff_emu_edge_hfix4_mmx; |
| 102 | extern emu_edge_hfix_func ff_emu_edge_hfix6_mmx; |
| 103 | extern emu_edge_hfix_func ff_emu_edge_hfix8_mmx; |
| 104 | extern emu_edge_hfix_func ff_emu_edge_hfix10_mmx; |
| 105 | extern emu_edge_hfix_func ff_emu_edge_hfix12_mmx; |
| 106 | extern emu_edge_hfix_func ff_emu_edge_hfix14_mmx; |
| 107 | extern emu_edge_hfix_func ff_emu_edge_hfix16_mmx; |
| 108 | extern emu_edge_hfix_func ff_emu_edge_hfix18_mmx; |
| 109 | extern emu_edge_hfix_func ff_emu_edge_hfix20_mmx; |
| 110 | extern emu_edge_hfix_func ff_emu_edge_hfix22_mmx; |
| 111 | #if ARCH_X86_32 |
| 112 | static emu_edge_hfix_func * const hfixtbl_mmx[11] = { |
| 113 | ff_emu_edge_hfix2_mmx, ff_emu_edge_hfix4_mmx, ff_emu_edge_hfix6_mmx, |
| 114 | ff_emu_edge_hfix8_mmx, ff_emu_edge_hfix10_mmx, ff_emu_edge_hfix12_mmx, |
| 115 | ff_emu_edge_hfix14_mmx, ff_emu_edge_hfix16_mmx, ff_emu_edge_hfix18_mmx, |
| 116 | ff_emu_edge_hfix20_mmx, ff_emu_edge_hfix22_mmx |
| 117 | }; |
| 118 | #endif |
| 119 | extern emu_edge_hvar_func ff_emu_edge_hvar_mmx; |
| 120 | extern emu_edge_hfix_func ff_emu_edge_hfix16_sse2; |
| 121 | extern emu_edge_hfix_func ff_emu_edge_hfix18_sse2; |
| 122 | extern emu_edge_hfix_func ff_emu_edge_hfix20_sse2; |
| 123 | extern emu_edge_hfix_func ff_emu_edge_hfix22_sse2; |
| 124 | static emu_edge_hfix_func * const hfixtbl_sse2[11] = { |
| 125 | ff_emu_edge_hfix2_mmx, ff_emu_edge_hfix4_mmx, ff_emu_edge_hfix6_mmx, |
| 126 | ff_emu_edge_hfix8_mmx, ff_emu_edge_hfix10_mmx, ff_emu_edge_hfix12_mmx, |
| 127 | ff_emu_edge_hfix14_mmx, ff_emu_edge_hfix16_sse2, ff_emu_edge_hfix18_sse2, |
| 128 | ff_emu_edge_hfix20_sse2, ff_emu_edge_hfix22_sse2 |
| 129 | }; |
| 130 | extern emu_edge_hvar_func ff_emu_edge_hvar_sse2; |
| 131 | #if HAVE_AVX2_EXTERNAL |
| 132 | extern emu_edge_hfix_func ff_emu_edge_hfix8_avx2; |
| 133 | extern emu_edge_hfix_func ff_emu_edge_hfix10_avx2; |
| 134 | extern emu_edge_hfix_func ff_emu_edge_hfix12_avx2; |
| 135 | extern emu_edge_hfix_func ff_emu_edge_hfix14_avx2; |
| 136 | extern emu_edge_hfix_func ff_emu_edge_hfix16_avx2; |
| 137 | extern emu_edge_hfix_func ff_emu_edge_hfix18_avx2; |
| 138 | extern emu_edge_hfix_func ff_emu_edge_hfix20_avx2; |
| 139 | extern emu_edge_hfix_func ff_emu_edge_hfix22_avx2; |
| 140 | static emu_edge_hfix_func * const hfixtbl_avx2[11] = { |
| 141 | ff_emu_edge_hfix2_mmx, ff_emu_edge_hfix4_mmx, ff_emu_edge_hfix6_mmx, |
| 142 | ff_emu_edge_hfix8_avx2, ff_emu_edge_hfix10_avx2, ff_emu_edge_hfix12_avx2, |
| 143 | ff_emu_edge_hfix14_avx2, ff_emu_edge_hfix16_avx2, ff_emu_edge_hfix18_avx2, |
| 144 | ff_emu_edge_hfix20_avx2, ff_emu_edge_hfix22_avx2 |
| 145 | }; |
| 146 | extern emu_edge_hvar_func ff_emu_edge_hvar_avx2; |
| 147 | #endif |
| 148 | |
| 149 | static av_always_inline void emulated_edge_mc(uint8_t *dst, const uint8_t *src, |
| 150 | ptrdiff_t dst_stride, |
| 151 | ptrdiff_t src_stride, |
| 152 | x86_reg block_w, x86_reg block_h, |
| 153 | x86_reg src_x, x86_reg src_y, |
| 154 | x86_reg w, x86_reg h, |
| 155 | emu_edge_vfix_func * const *vfix_tbl, |
| 156 | emu_edge_vvar_func *v_extend_var, |
| 157 | emu_edge_hfix_func * const *hfix_tbl, |
| 158 | emu_edge_hvar_func *h_extend_var) |
| 159 | { |
| 160 | x86_reg start_y, start_x, end_y, end_x, src_y_add = 0, p; |
| 161 | |
| 162 | if (!w || !h) |
| 163 | return; |
| 164 | |
| 165 | if (src_y >= h) { |
| 166 | src -= src_y*src_stride; |
| 167 | src_y_add = h - 1; |
| 168 | src_y = h - 1; |
| 169 | } else if (src_y <= -block_h) { |
| 170 | src -= src_y*src_stride; |
| 171 | src_y_add = 1 - block_h; |
| 172 | src_y = 1 - block_h; |
| 173 | } |
| 174 | if (src_x >= w) { |
| 175 | src += w - 1 - src_x; |
| 176 | src_x = w - 1; |
| 177 | } else if (src_x <= -block_w) { |
| 178 | src += 1 - block_w - src_x; |
| 179 | src_x = 1 - block_w; |
| 180 | } |
| 181 | |
| 182 | start_y = FFMAX(0, -src_y); |
| 183 | start_x = FFMAX(0, -src_x); |
| 184 | end_y = FFMIN(block_h, h-src_y); |
| 185 | end_x = FFMIN(block_w, w-src_x); |
| 186 | av_assert2(start_x < end_x && block_w > 0); |
| 187 | av_assert2(start_y < end_y && block_h > 0); |
| 188 | |
| 189 | // fill in the to-be-copied part plus all above/below |
| 190 | src += (src_y_add + start_y) * src_stride + start_x; |
| 191 | w = end_x - start_x; |
| 192 | if (w <= 22) { |
| 193 | vfix_tbl[w - 1](dst + start_x, dst_stride, src, src_stride, |
| 194 | start_y, end_y, block_h); |
| 195 | } else { |
| 196 | v_extend_var(dst + start_x, dst_stride, src, src_stride, |
| 197 | start_y, end_y, block_h, w); |
| 198 | } |
| 199 | |
| 200 | // fill left |
| 201 | if (start_x) { |
| 202 | if (start_x <= 22) { |
| 203 | hfix_tbl[(start_x - 1) >> 1](dst, dst_stride, start_x, block_h); |
| 204 | } else { |
| 205 | h_extend_var(dst, dst_stride, |
| 206 | start_x, (start_x + 1) >> 1, block_h); |
| 207 | } |
| 208 | } |
| 209 | |
| 210 | // fill right |
| 211 | p = block_w - end_x; |
| 212 | if (p) { |
| 213 | if (p <= 22) { |
| 214 | hfix_tbl[(p - 1) >> 1](dst + end_x - (p & 1), dst_stride, |
| 215 | -!(p & 1), block_h); |
| 216 | } else { |
| 217 | h_extend_var(dst + end_x - (p & 1), dst_stride, |
| 218 | -!(p & 1), (p + 1) >> 1, block_h); |
| 219 | } |
| 220 | } |
| 221 | } |
| 222 | |
| 223 | #if ARCH_X86_32 |
| 224 | static av_noinline void emulated_edge_mc_mmx(uint8_t *buf, const uint8_t *src, |
| 225 | ptrdiff_t buf_stride, |
| 226 | ptrdiff_t src_stride, |
| 227 | int block_w, int block_h, |
| 228 | int src_x, int src_y, int w, int h) |
| 229 | { |
| 230 | emulated_edge_mc(buf, src, buf_stride, src_stride, block_w, block_h, |
| 231 | src_x, src_y, w, h, vfixtbl_mmx, &ff_emu_edge_vvar_mmx, |
| 232 | hfixtbl_mmx, &ff_emu_edge_hvar_mmx); |
| 233 | } |
| 234 | |
| 235 | static av_noinline void emulated_edge_mc_sse(uint8_t *buf, const uint8_t *src, |
| 236 | ptrdiff_t buf_stride, |
| 237 | ptrdiff_t src_stride, |
| 238 | int block_w, int block_h, |
| 239 | int src_x, int src_y, int w, int h) |
| 240 | { |
| 241 | emulated_edge_mc(buf, src, buf_stride, src_stride, block_w, block_h, |
| 242 | src_x, src_y, w, h, vfixtbl_sse, &ff_emu_edge_vvar_sse, |
| 243 | hfixtbl_mmx, &ff_emu_edge_hvar_mmx); |
| 244 | } |
| 245 | #endif |
| 246 | |
| 247 | static av_noinline void emulated_edge_mc_sse2(uint8_t *buf, const uint8_t *src, |
| 248 | ptrdiff_t buf_stride, |
| 249 | ptrdiff_t src_stride, |
| 250 | int block_w, int block_h, |
| 251 | int src_x, int src_y, int w, |
| 252 | int h) |
| 253 | { |
| 254 | emulated_edge_mc(buf, src, buf_stride, src_stride, block_w, block_h, |
| 255 | src_x, src_y, w, h, vfixtbl_sse, &ff_emu_edge_vvar_sse, |
| 256 | hfixtbl_sse2, &ff_emu_edge_hvar_sse2); |
| 257 | } |
| 258 | |
| 259 | #if HAVE_AVX2_EXTERNAL |
| 260 | static av_noinline void emulated_edge_mc_avx2(uint8_t *buf, const uint8_t *src, |
| 261 | ptrdiff_t buf_stride, |
| 262 | ptrdiff_t src_stride, |
| 263 | int block_w, int block_h, |
| 264 | int src_x, int src_y, int w, |
| 265 | int h) |
| 266 | { |
| 267 | emulated_edge_mc(buf, src, buf_stride, src_stride, block_w, block_h, |
| 268 | src_x, src_y, w, h, vfixtbl_sse, &ff_emu_edge_vvar_sse, |
| 269 | hfixtbl_avx2, &ff_emu_edge_hvar_avx2); |
| 270 | } |
| 271 | #endif /* HAVE_AVX2_EXTERNAL */ |
| 272 | #endif /* HAVE_YASM */ |
| 273 | |
| 274 | void ff_prefetch_mmxext(uint8_t *buf, ptrdiff_t stride, int h); |
| 275 | void ff_prefetch_3dnow(uint8_t *buf, ptrdiff_t stride, int h); |
| 276 | |
| 277 | av_cold void ff_videodsp_init_x86(VideoDSPContext *ctx, int bpc) |
| 278 | { |
| 279 | #if HAVE_YASM |
| 280 | int cpu_flags = av_get_cpu_flags(); |
| 281 | |
| 282 | #if ARCH_X86_32 |
| 283 | if (EXTERNAL_MMX(cpu_flags) && bpc <= 8) { |
| 284 | ctx->emulated_edge_mc = emulated_edge_mc_mmx; |
| 285 | } |
| 286 | if (EXTERNAL_AMD3DNOW(cpu_flags)) { |
| 287 | ctx->prefetch = ff_prefetch_3dnow; |
| 288 | } |
| 289 | #endif /* ARCH_X86_32 */ |
| 290 | if (EXTERNAL_MMXEXT(cpu_flags)) { |
| 291 | ctx->prefetch = ff_prefetch_mmxext; |
| 292 | } |
| 293 | #if ARCH_X86_32 |
| 294 | if (EXTERNAL_SSE(cpu_flags) && bpc <= 8) { |
| 295 | ctx->emulated_edge_mc = emulated_edge_mc_sse; |
| 296 | } |
| 297 | #endif /* ARCH_X86_32 */ |
| 298 | if (EXTERNAL_SSE2(cpu_flags) && bpc <= 8) { |
| 299 | ctx->emulated_edge_mc = emulated_edge_mc_sse2; |
| 300 | } |
| 301 | #if HAVE_AVX2_EXTERNAL |
| 302 | if (EXTERNAL_AVX2(cpu_flags) && bpc <= 8) { |
| 303 | ctx->emulated_edge_mc = emulated_edge_mc_avx2; |
| 304 | } |
| 305 | #endif |
| 306 | #endif /* HAVE_YASM */ |
| 307 | } |