| Line | Branch | Exec | Source |
|---|---|---|---|
| 1 | /* | ||
| 2 | * SIMD-optimized halfpel functions | ||
| 3 | * Copyright (c) 2000, 2001 Fabrice Bellard | ||
| 4 | * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at> | ||
| 5 | * | ||
| 6 | * This file is part of FFmpeg. | ||
| 7 | * | ||
| 8 | * FFmpeg is free software; you can redistribute it and/or | ||
| 9 | * modify it under the terms of the GNU Lesser General Public | ||
| 10 | * License as published by the Free Software Foundation; either | ||
| 11 | * version 2.1 of the License, or (at your option) any later version. | ||
| 12 | * | ||
| 13 | * FFmpeg is distributed in the hope that it will be useful, | ||
| 14 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
| 15 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | ||
| 16 | * Lesser General Public License for more details. | ||
| 17 | * | ||
| 18 | * You should have received a copy of the GNU Lesser General Public | ||
| 19 | * License along with FFmpeg; if not, write to the Free Software | ||
| 20 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | ||
| 21 | * | ||
| 22 | * MMX optimization by Nick Kurshev <nickols_k@mail.ru> | ||
| 23 | */ | ||
| 24 | |||
| 25 | #include <stddef.h> | ||
| 26 | #include <stdint.h> | ||
| 27 | |||
| 28 | #include "libavutil/attributes.h" | ||
| 29 | #include "libavutil/cpu.h" | ||
| 30 | #include "libavutil/x86/cpu.h" | ||
| 31 | #include "libavcodec/avcodec.h" | ||
| 32 | #include "libavcodec/hpeldsp.h" | ||
| 33 | #include "libavcodec/pixels.h" | ||
| 34 | #include "fpel.h" | ||
| 35 | #include "hpeldsp.h" | ||
| 36 | #include "inline_asm.h" | ||
| 37 | |||
| 38 | void ff_put_pixels8_x2_mmxext(uint8_t *block, const uint8_t *pixels, | ||
| 39 | ptrdiff_t line_size, int h); | ||
| 40 | void ff_put_pixels16_x2_sse2(uint8_t *block, const uint8_t *pixels, | ||
| 41 | ptrdiff_t line_size, int h); | ||
| 42 | void ff_avg_pixels16_x2_sse2(uint8_t *block, const uint8_t *pixels, | ||
| 43 | ptrdiff_t line_size, int h); | ||
| 44 | void ff_put_pixels16_y2_sse2(uint8_t *block, const uint8_t *pixels, | ||
| 45 | ptrdiff_t line_size, int h); | ||
| 46 | void ff_avg_pixels16_y2_sse2(uint8_t *block, const uint8_t *pixels, | ||
| 47 | ptrdiff_t line_size, int h); | ||
| 48 | void ff_put_no_rnd_pixels8_x2_mmxext(uint8_t *block, const uint8_t *pixels, | ||
| 49 | ptrdiff_t line_size, int h); | ||
| 50 | void ff_put_no_rnd_pixels8_x2_exact_mmxext(uint8_t *block, | ||
| 51 | const uint8_t *pixels, | ||
| 52 | ptrdiff_t line_size, int h); | ||
| 53 | void ff_put_no_rnd_pixels16_x2_sse2(uint8_t *block, const uint8_t *pixels, | ||
| 54 | ptrdiff_t line_size, int h); | ||
| 55 | void ff_avg_no_rnd_pixels16_x2_sse2(uint8_t *block, const uint8_t *pixels, | ||
| 56 | ptrdiff_t line_size, int h); | ||
| 57 | void ff_put_pixels8_y2_mmxext(uint8_t *block, const uint8_t *pixels, | ||
| 58 | ptrdiff_t line_size, int h); | ||
| 59 | void ff_put_no_rnd_pixels8_y2_mmxext(uint8_t *block, const uint8_t *pixels, | ||
| 60 | ptrdiff_t line_size, int h); | ||
| 61 | void ff_put_no_rnd_pixels8_y2_exact_mmxext(uint8_t *block, | ||
| 62 | const uint8_t *pixels, | ||
| 63 | ptrdiff_t line_size, int h); | ||
| 64 | void ff_put_no_rnd_pixels16_y2_sse2(uint8_t *block, const uint8_t *pixels, | ||
| 65 | ptrdiff_t line_size, int h); | ||
| 66 | void ff_avg_no_rnd_pixels16_y2_sse2(uint8_t *block, const uint8_t *pixels, | ||
| 67 | ptrdiff_t line_size, int h); | ||
| 68 | void ff_avg_pixels8_x2_mmxext(uint8_t *block, const uint8_t *pixels, | ||
| 69 | ptrdiff_t line_size, int h); | ||
| 70 | void ff_avg_pixels8_y2_mmxext(uint8_t *block, const uint8_t *pixels, | ||
| 71 | ptrdiff_t line_size, int h); | ||
| 72 | |||
| 73 | #if HAVE_INLINE_ASM | ||
| 74 | |||
| 75 | /***********************************/ | ||
| 76 | /* MMX no rounding */ | ||
| 77 | |||
| 78 | // put_pixels | ||
| 79 | 72222 | static void put_no_rnd_pixels8_xy2_mmx(uint8_t *block, const uint8_t *pixels, | |
| 80 | ptrdiff_t line_size, int h) | ||
| 81 | { | ||
| 82 | 72222 | MOVQ_ZERO(mm7); | |
| 83 | 72222 | MOVQ_WONE(mm6); // =1 for no_rnd version | |
| 84 | 72222 | __asm__ volatile( | |
| 85 | "movq (%1), %%mm0 \n\t" | ||
| 86 | "movq 1(%1), %%mm4 \n\t" | ||
| 87 | "movq %%mm0, %%mm1 \n\t" | ||
| 88 | "movq %%mm4, %%mm5 \n\t" | ||
| 89 | "punpcklbw %%mm7, %%mm0 \n\t" | ||
| 90 | "punpcklbw %%mm7, %%mm4 \n\t" | ||
| 91 | "punpckhbw %%mm7, %%mm1 \n\t" | ||
| 92 | "punpckhbw %%mm7, %%mm5 \n\t" | ||
| 93 | "paddusw %%mm0, %%mm4 \n\t" | ||
| 94 | "paddusw %%mm1, %%mm5 \n\t" | ||
| 95 | "xor %%"FF_REG_a", %%"FF_REG_a" \n\t" | ||
| 96 | "add %3, %1 \n\t" | ||
| 97 | ".p2align 3 \n\t" | ||
| 98 | "1: \n\t" | ||
| 99 | "movq (%1, %%"FF_REG_a"), %%mm0 \n\t" | ||
| 100 | "movq 1(%1, %%"FF_REG_a"), %%mm2 \n\t" | ||
| 101 | "movq %%mm0, %%mm1 \n\t" | ||
| 102 | "movq %%mm2, %%mm3 \n\t" | ||
| 103 | "punpcklbw %%mm7, %%mm0 \n\t" | ||
| 104 | "punpcklbw %%mm7, %%mm2 \n\t" | ||
| 105 | "punpckhbw %%mm7, %%mm1 \n\t" | ||
| 106 | "punpckhbw %%mm7, %%mm3 \n\t" | ||
| 107 | "paddusw %%mm2, %%mm0 \n\t" | ||
| 108 | "paddusw %%mm3, %%mm1 \n\t" | ||
| 109 | "paddusw %%mm6, %%mm4 \n\t" | ||
| 110 | "paddusw %%mm6, %%mm5 \n\t" | ||
| 111 | "paddusw %%mm0, %%mm4 \n\t" | ||
| 112 | "paddusw %%mm1, %%mm5 \n\t" | ||
| 113 | "psrlw $2, %%mm4 \n\t" | ||
| 114 | "psrlw $2, %%mm5 \n\t" | ||
| 115 | "packuswb %%mm5, %%mm4 \n\t" | ||
| 116 | "movq %%mm4, (%2, %%"FF_REG_a") \n\t" | ||
| 117 | "add %3, %%"FF_REG_a" \n\t" | ||
| 118 | |||
| 119 | "movq (%1, %%"FF_REG_a"), %%mm2 \n\t" // 0 <-> 2 1 <-> 3 | ||
| 120 | "movq 1(%1, %%"FF_REG_a"), %%mm4 \n\t" | ||
| 121 | "movq %%mm2, %%mm3 \n\t" | ||
| 122 | "movq %%mm4, %%mm5 \n\t" | ||
| 123 | "punpcklbw %%mm7, %%mm2 \n\t" | ||
| 124 | "punpcklbw %%mm7, %%mm4 \n\t" | ||
| 125 | "punpckhbw %%mm7, %%mm3 \n\t" | ||
| 126 | "punpckhbw %%mm7, %%mm5 \n\t" | ||
| 127 | "paddusw %%mm2, %%mm4 \n\t" | ||
| 128 | "paddusw %%mm3, %%mm5 \n\t" | ||
| 129 | "paddusw %%mm6, %%mm0 \n\t" | ||
| 130 | "paddusw %%mm6, %%mm1 \n\t" | ||
| 131 | "paddusw %%mm4, %%mm0 \n\t" | ||
| 132 | "paddusw %%mm5, %%mm1 \n\t" | ||
| 133 | "psrlw $2, %%mm0 \n\t" | ||
| 134 | "psrlw $2, %%mm1 \n\t" | ||
| 135 | "packuswb %%mm1, %%mm0 \n\t" | ||
| 136 | "movq %%mm0, (%2, %%"FF_REG_a") \n\t" | ||
| 137 | "add %3, %%"FF_REG_a" \n\t" | ||
| 138 | |||
| 139 | "subl $2, %0 \n\t" | ||
| 140 | "jnz 1b \n\t" | ||
| 141 | :"+g"(h), "+S"(pixels) | ||
| 142 | :"D"(block), "r"((x86_reg)line_size) | ||
| 143 | :FF_REG_a, "memory"); | ||
| 144 | 72222 | } | |
| 145 | |||
| 146 | // this routine is 'slightly' suboptimal but mostly unused | ||
| 147 | 2 | static void avg_no_rnd_pixels8_xy2_mmx(uint8_t *block, const uint8_t *pixels, | |
| 148 | ptrdiff_t line_size, int h) | ||
| 149 | { | ||
| 150 | 2 | MOVQ_ZERO(mm7); | |
| 151 | 2 | MOVQ_WONE(mm6); // =2 for rnd and =1 for no_rnd version | |
| 152 | 2 | __asm__ volatile( | |
| 153 | "movq (%1), %%mm0 \n\t" | ||
| 154 | "movq 1(%1), %%mm4 \n\t" | ||
| 155 | "movq %%mm0, %%mm1 \n\t" | ||
| 156 | "movq %%mm4, %%mm5 \n\t" | ||
| 157 | "punpcklbw %%mm7, %%mm0 \n\t" | ||
| 158 | "punpcklbw %%mm7, %%mm4 \n\t" | ||
| 159 | "punpckhbw %%mm7, %%mm1 \n\t" | ||
| 160 | "punpckhbw %%mm7, %%mm5 \n\t" | ||
| 161 | "paddusw %%mm0, %%mm4 \n\t" | ||
| 162 | "paddusw %%mm1, %%mm5 \n\t" | ||
| 163 | "xor %%"FF_REG_a", %%"FF_REG_a" \n\t" | ||
| 164 | "add %3, %1 \n\t" | ||
| 165 | ".p2align 3 \n\t" | ||
| 166 | "1: \n\t" | ||
| 167 | "movq (%1, %%"FF_REG_a"), %%mm0 \n\t" | ||
| 168 | "movq 1(%1, %%"FF_REG_a"), %%mm2 \n\t" | ||
| 169 | "movq %%mm0, %%mm1 \n\t" | ||
| 170 | "movq %%mm2, %%mm3 \n\t" | ||
| 171 | "punpcklbw %%mm7, %%mm0 \n\t" | ||
| 172 | "punpcklbw %%mm7, %%mm2 \n\t" | ||
| 173 | "punpckhbw %%mm7, %%mm1 \n\t" | ||
| 174 | "punpckhbw %%mm7, %%mm3 \n\t" | ||
| 175 | "paddusw %%mm2, %%mm0 \n\t" | ||
| 176 | "paddusw %%mm3, %%mm1 \n\t" | ||
| 177 | "paddusw %%mm6, %%mm4 \n\t" | ||
| 178 | "paddusw %%mm6, %%mm5 \n\t" | ||
| 179 | "paddusw %%mm0, %%mm4 \n\t" | ||
| 180 | "paddusw %%mm1, %%mm5 \n\t" | ||
| 181 | "psrlw $2, %%mm4 \n\t" | ||
| 182 | "psrlw $2, %%mm5 \n\t" | ||
| 183 | "movq (%2, %%"FF_REG_a"), %%mm3 \n\t" | ||
| 184 | "packuswb %%mm5, %%mm4 \n\t" | ||
| 185 | "pcmpeqd %%mm2, %%mm2 \n\t" | ||
| 186 | "paddb %%mm2, %%mm2 \n\t" | ||
| 187 | PAVGB_MMX(%%mm3, %%mm4, %%mm5, %%mm2) | ||
| 188 | "movq %%mm5, (%2, %%"FF_REG_a") \n\t" | ||
| 189 | "add %3, %%"FF_REG_a" \n\t" | ||
| 190 | |||
| 191 | "movq (%1, %%"FF_REG_a"), %%mm2 \n\t" // 0 <-> 2 1 <-> 3 | ||
| 192 | "movq 1(%1, %%"FF_REG_a"), %%mm4 \n\t" | ||
| 193 | "movq %%mm2, %%mm3 \n\t" | ||
| 194 | "movq %%mm4, %%mm5 \n\t" | ||
| 195 | "punpcklbw %%mm7, %%mm2 \n\t" | ||
| 196 | "punpcklbw %%mm7, %%mm4 \n\t" | ||
| 197 | "punpckhbw %%mm7, %%mm3 \n\t" | ||
| 198 | "punpckhbw %%mm7, %%mm5 \n\t" | ||
| 199 | "paddusw %%mm2, %%mm4 \n\t" | ||
| 200 | "paddusw %%mm3, %%mm5 \n\t" | ||
| 201 | "paddusw %%mm6, %%mm0 \n\t" | ||
| 202 | "paddusw %%mm6, %%mm1 \n\t" | ||
| 203 | "paddusw %%mm4, %%mm0 \n\t" | ||
| 204 | "paddusw %%mm5, %%mm1 \n\t" | ||
| 205 | "psrlw $2, %%mm0 \n\t" | ||
| 206 | "psrlw $2, %%mm1 \n\t" | ||
| 207 | "movq (%2, %%"FF_REG_a"), %%mm3 \n\t" | ||
| 208 | "packuswb %%mm1, %%mm0 \n\t" | ||
| 209 | "pcmpeqd %%mm2, %%mm2 \n\t" | ||
| 210 | "paddb %%mm2, %%mm2 \n\t" | ||
| 211 | PAVGB_MMX(%%mm3, %%mm0, %%mm1, %%mm2) | ||
| 212 | "movq %%mm1, (%2, %%"FF_REG_a") \n\t" | ||
| 213 | "add %3, %%"FF_REG_a" \n\t" | ||
| 214 | |||
| 215 | "subl $2, %0 \n\t" | ||
| 216 | "jnz 1b \n\t" | ||
| 217 | :"+g"(h), "+S"(pixels) | ||
| 218 | :"D"(block), "r"((x86_reg)line_size) | ||
| 219 | :FF_REG_a, "memory"); | ||
| 220 | 2 | } | |
| 221 | |||
| 222 | #if HAVE_MMX | ||
| 223 | 1 | CALL_2X_PIXELS(avg_no_rnd_pixels16_xy2_mmx, avg_no_rnd_pixels8_xy2_mmx, 8) | |
| 224 | 10119 | CALL_2X_PIXELS(put_no_rnd_pixels16_xy2_mmx, put_no_rnd_pixels8_xy2_mmx, 8) | |
| 225 | #endif | ||
| 226 | #endif /* HAVE_INLINE_ASM */ | ||
| 227 | |||
| 228 | 171 | static void hpeldsp_init_mmx(HpelDSPContext *c, int flags) | |
| 229 | { | ||
| 230 | #if HAVE_MMX_INLINE | ||
| 231 | 171 | c->put_no_rnd_pixels_tab[0][3] = put_no_rnd_pixels16_xy2_mmx; | |
| 232 | 171 | c->put_no_rnd_pixels_tab[1][3] = put_no_rnd_pixels8_xy2_mmx; | |
| 233 | 171 | c->avg_no_rnd_pixels_tab[3] = avg_no_rnd_pixels16_xy2_mmx; | |
| 234 | #endif | ||
| 235 | #if HAVE_MMX_EXTERNAL | ||
| 236 | 171 | c->put_no_rnd_pixels_tab[1][0] = | |
| 237 | 171 | c->put_pixels_tab[1][0] = ff_put_pixels8_mmx; | |
| 238 | #endif | ||
| 239 | 171 | } | |
| 240 | |||
| 241 | 170 | static void hpeldsp_init_mmxext(HpelDSPContext *c, int flags) | |
| 242 | { | ||
| 243 | #if HAVE_MMXEXT_EXTERNAL | ||
| 244 | 170 | c->put_pixels_tab[1][1] = ff_put_pixels8_x2_mmxext; | |
| 245 | 170 | c->put_pixels_tab[1][2] = ff_put_pixels8_y2_mmxext; | |
| 246 | |||
| 247 | 170 | c->avg_pixels_tab[1][0] = ff_avg_pixels8_mmxext; | |
| 248 | 170 | c->avg_pixels_tab[1][1] = ff_avg_pixels8_x2_mmxext; | |
| 249 | 170 | c->avg_pixels_tab[1][2] = ff_avg_pixels8_y2_mmxext; | |
| 250 | |||
| 251 | 170 | c->put_no_rnd_pixels_tab[1][1] = ff_put_no_rnd_pixels8_x2_exact_mmxext; | |
| 252 | 170 | c->put_no_rnd_pixels_tab[1][2] = ff_put_no_rnd_pixels8_y2_exact_mmxext; | |
| 253 | |||
| 254 |
2/2✓ Branch 0 taken 142 times.
✓ Branch 1 taken 28 times.
|
170 | if (!(flags & AV_CODEC_FLAG_BITEXACT)) { |
| 255 | 142 | c->put_no_rnd_pixels_tab[1][1] = ff_put_no_rnd_pixels8_x2_mmxext; | |
| 256 | 142 | c->put_no_rnd_pixels_tab[1][2] = ff_put_no_rnd_pixels8_y2_mmxext; | |
| 257 | } | ||
| 258 | #endif /* HAVE_MMXEXT_EXTERNAL */ | ||
| 259 | 170 | } | |
| 260 | |||
| 261 | 168 | static void hpeldsp_init_sse2(HpelDSPContext *c, int flags) | |
| 262 | { | ||
| 263 | #if HAVE_SSE2_EXTERNAL | ||
| 264 | 168 | c->put_pixels_tab[0][0] = ff_put_pixels16_sse2; | |
| 265 | 168 | c->put_pixels_tab[0][1] = ff_put_pixels16_x2_sse2; | |
| 266 | 168 | c->put_pixels_tab[0][2] = ff_put_pixels16_y2_sse2; | |
| 267 | 168 | c->put_pixels_tab[0][3] = ff_put_pixels16_xy2_sse2; | |
| 268 | |||
| 269 | 168 | c->put_no_rnd_pixels_tab[0][0] = ff_put_pixels16_sse2; | |
| 270 | 168 | c->put_no_rnd_pixels_tab[0][1] = ff_put_no_rnd_pixels16_x2_sse2; | |
| 271 | 168 | c->put_no_rnd_pixels_tab[0][2] = ff_put_no_rnd_pixels16_y2_sse2; | |
| 272 | |||
| 273 | 168 | c->avg_pixels_tab[0][0] = ff_avg_pixels16_sse2; | |
| 274 | 168 | c->avg_pixels_tab[0][1] = ff_avg_pixels16_x2_sse2; | |
| 275 | 168 | c->avg_pixels_tab[0][2] = ff_avg_pixels16_y2_sse2; | |
| 276 | 168 | c->avg_pixels_tab[0][3] = ff_avg_pixels16_xy2_sse2; | |
| 277 | |||
| 278 | 168 | c->avg_no_rnd_pixels_tab[0] = ff_avg_pixels16_sse2; | |
| 279 | 168 | c->avg_no_rnd_pixels_tab[1] = ff_avg_no_rnd_pixels16_x2_sse2; | |
| 280 | 168 | c->avg_no_rnd_pixels_tab[2] = ff_avg_no_rnd_pixels16_y2_sse2; | |
| 281 | #endif /* HAVE_SSE2_EXTERNAL */ | ||
| 282 | 168 | } | |
| 283 | |||
| 284 | 166 | static void hpeldsp_init_ssse3(HpelDSPContext *c, int flags) | |
| 285 | { | ||
| 286 | #if HAVE_SSSE3_EXTERNAL | ||
| 287 | 166 | c->put_pixels_tab[0][3] = ff_put_pixels16_xy2_ssse3; | |
| 288 | 166 | c->avg_pixels_tab[0][3] = ff_avg_pixels16_xy2_ssse3; | |
| 289 | 166 | c->put_pixels_tab[1][3] = ff_put_pixels8_xy2_ssse3; | |
| 290 | 166 | c->avg_pixels_tab[1][3] = ff_avg_pixels8_xy2_ssse3; | |
| 291 | #endif | ||
| 292 | 166 | } | |
| 293 | |||
| 294 | 941 | av_cold void ff_hpeldsp_init_x86(HpelDSPContext *c, int flags) | |
| 295 | { | ||
| 296 | 941 | int cpu_flags = av_get_cpu_flags(); | |
| 297 | |||
| 298 |
2/2✓ Branch 0 taken 171 times.
✓ Branch 1 taken 770 times.
|
941 | if (INLINE_MMX(cpu_flags)) |
| 299 | 171 | hpeldsp_init_mmx(c, flags); | |
| 300 | |||
| 301 |
2/2✓ Branch 0 taken 170 times.
✓ Branch 1 taken 771 times.
|
941 | if (EXTERNAL_MMXEXT(cpu_flags)) |
| 302 | 170 | hpeldsp_init_mmxext(c, flags); | |
| 303 | |||
| 304 |
2/2✓ Branch 0 taken 168 times.
✓ Branch 1 taken 773 times.
|
941 | if (EXTERNAL_SSE2(cpu_flags)) |
| 305 | 168 | hpeldsp_init_sse2(c, flags); | |
| 306 | |||
| 307 |
2/2✓ Branch 0 taken 166 times.
✓ Branch 1 taken 775 times.
|
941 | if (EXTERNAL_SSSE3(cpu_flags)) |
| 308 | 166 | hpeldsp_init_ssse3(c, flags); | |
| 309 | 941 | } | |
| 310 |