Since a51279bbdea0d6db920d71980262bccd0ce78226, hpeldsp_rnd_template.c was only included once and one of the two functions in rnd_template.c was only enabled once.
Unified · Split
| @@ -22,6 +22,9 @@ · 21 unmodified lines | |||
| 22 | * MMX optimization by Nick Kurshev <nickols_k@mail.ru> | 22 | * MMX optimization by Nick Kurshev <nickols_k@mail.ru> |
| 23 | */ | 23 | */ |
| 24 | 24 | ||
| 25 | #include <stddef.h> | ||
| 26 | #include <stdint.h> | ||
| 27 | |||
| 25 | #include "libavutil/attributes.h" | 28 | #include "libavutil/attributes.h" |
| 26 | #include "libavutil/cpu.h" | 29 | #include "libavutil/cpu.h" |
| 27 | #include "libavutil/x86/cpu.h" | 30 | #include "libavutil/x86/cpu.h" |
| @@ -74,19 +77,263 @@ void ff_avg_approx_pixels8_xy2_mmxext(uint8_t *block, const uint8_t *pixels, · 46 unmodified lines | |||
| 74 | /* MMX no rounding */ | 77 | /* MMX no rounding */ |
| 75 | #define DEF(x, y) x ## _no_rnd_ ## y ## _mmx | 78 | #define DEF(x, y) x ## _no_rnd_ ## y ## _mmx |
| 76 | #define SET_RND MOVQ_WONE | 79 | #define SET_RND MOVQ_WONE |
| 77 | #define PAVGBP(a, b, c, d, e, f) PAVGBP_MMX_NO_RND(a, b, c, d, e, f) | ||
| 78 | #define PAVGB(a, b, c, e) PAVGB_MMX_NO_RND(a, b, c, e) | ||
| 79 | #define STATIC static | 80 | #define STATIC static |
| 80 | 81 | ||
| 81 | #include "rnd_template.c" | 82 | #include "rnd_template.c" |
| 82 | #include "hpeldsp_rnd_template.c" | ||
| 83 | 83 | ||
| 84 | #undef DEF | 84 | #undef DEF |
| 85 | #undef SET_RND | 85 | #undef SET_RND |
| 86 | #undef PAVGBP | ||
| 87 | #undef PAVGB | ||
| 88 | #undef STATIC | 86 | #undef STATIC |
| 89 | 87 | ||
| 88 | // this routine is 'slightly' suboptimal but mostly unused | ||
| 89 | static void avg_no_rnd_pixels8_xy2_mmx(uint8_t *block, const uint8_t *pixels, | ||
| 90 | ptrdiff_t line_size, int h) | ||
| 91 | { | ||
| 92 | MOVQ_ZERO(mm7); | ||
| 93 | MOVQ_WONE(mm6); // =2 for rnd and =1 for no_rnd version | ||
| 94 | __asm__ volatile( | ||
| 95 | "movq (%1), %%mm0 \n\t" | ||
| 96 | "movq 1(%1), %%mm4 \n\t" | ||
| 97 | "movq %%mm0, %%mm1 \n\t" | ||
| 98 | "movq %%mm4, %%mm5 \n\t" | ||
| 99 | "punpcklbw %%mm7, %%mm0 \n\t" | ||
| 100 | "punpcklbw %%mm7, %%mm4 \n\t" | ||
| 101 | "punpckhbw %%mm7, %%mm1 \n\t" | ||
| 102 | "punpckhbw %%mm7, %%mm5 \n\t" | ||
| 103 | "paddusw %%mm0, %%mm4 \n\t" | ||
| 104 | "paddusw %%mm1, %%mm5 \n\t" | ||
| 105 | "xor %%"FF_REG_a", %%"FF_REG_a" \n\t" | ||
| 106 | "add %3, %1 \n\t" | ||
| 107 | ".p2align 3 \n\t" | ||
| 108 | "1: \n\t" | ||
| 109 | "movq (%1, %%"FF_REG_a"), %%mm0 \n\t" | ||
| 110 | "movq 1(%1, %%"FF_REG_a"), %%mm2 \n\t" | ||
| 111 | "movq %%mm0, %%mm1 \n\t" | ||
| 112 | "movq %%mm2, %%mm3 \n\t" | ||
| 113 | "punpcklbw %%mm7, %%mm0 \n\t" | ||
| 114 | "punpcklbw %%mm7, %%mm2 \n\t" | ||
| 115 | "punpckhbw %%mm7, %%mm1 \n\t" | ||
| 116 | "punpckhbw %%mm7, %%mm3 \n\t" | ||
| 117 | "paddusw %%mm2, %%mm0 \n\t" | ||
| 118 | "paddusw %%mm3, %%mm1 \n\t" | ||
| 119 | "paddusw %%mm6, %%mm4 \n\t" | ||
| 120 | "paddusw %%mm6, %%mm5 \n\t" | ||
| 121 | "paddusw %%mm0, %%mm4 \n\t" | ||
| 122 | "paddusw %%mm1, %%mm5 \n\t" | ||
| 123 | "psrlw $2, %%mm4 \n\t" | ||
| 124 | "psrlw $2, %%mm5 \n\t" | ||
| 125 | "movq (%2, %%"FF_REG_a"), %%mm3 \n\t" | ||
| 126 | "packuswb %%mm5, %%mm4 \n\t" | ||
| 127 | "pcmpeqd %%mm2, %%mm2 \n\t" | ||
| 128 | "paddb %%mm2, %%mm2 \n\t" | ||
| 129 | PAVGB_MMX(%%mm3, %%mm4, %%mm5, %%mm2) | ||
| 130 | "movq %%mm5, (%2, %%"FF_REG_a") \n\t" | ||
| 131 | "add %3, %%"FF_REG_a" \n\t" | ||
| 132 | |||
| 133 | "movq (%1, %%"FF_REG_a"), %%mm2 \n\t" // 0 <-> 2 1 <-> 3 | ||
| 134 | "movq 1(%1, %%"FF_REG_a"), %%mm4 \n\t" | ||
| 135 | "movq %%mm2, %%mm3 \n\t" | ||
| 136 | "movq %%mm4, %%mm5 \n\t" | ||
| 137 | "punpcklbw %%mm7, %%mm2 \n\t" | ||
| 138 | "punpcklbw %%mm7, %%mm4 \n\t" | ||
| 139 | "punpckhbw %%mm7, %%mm3 \n\t" | ||
| 140 | "punpckhbw %%mm7, %%mm5 \n\t" | ||
| 141 | "paddusw %%mm2, %%mm4 \n\t" | ||
| 142 | "paddusw %%mm3, %%mm5 \n\t" | ||
| 143 | "paddusw %%mm6, %%mm0 \n\t" | ||
| 144 | "paddusw %%mm6, %%mm1 \n\t" | ||
| 145 | "paddusw %%mm4, %%mm0 \n\t" | ||
| 146 | "paddusw %%mm5, %%mm1 \n\t" | ||
| 147 | "psrlw $2, %%mm0 \n\t" | ||
| 148 | "psrlw $2, %%mm1 \n\t" | ||
| 149 | "movq (%2, %%"FF_REG_a"), %%mm3 \n\t" | ||
| 150 | "packuswb %%mm1, %%mm0 \n\t" | ||
| 151 | "pcmpeqd %%mm2, %%mm2 \n\t" | ||
| 152 | "paddb %%mm2, %%mm2 \n\t" | ||
| 153 | PAVGB_MMX(%%mm3, %%mm0, %%mm1, %%mm2) | ||
| 154 | "movq %%mm1, (%2, %%"FF_REG_a") \n\t" | ||
| 155 | "add %3, %%"FF_REG_a" \n\t" | ||
| 156 | |||
| 157 | "subl $2, %0 \n\t" | ||
| 158 | "jnz 1b \n\t" | ||
| 159 | :"+g"(h), "+S"(pixels) | ||
| 160 | :"D"(block), "r"((x86_reg)line_size) | ||
| 161 | :FF_REG_a, "memory"); | ||
| 162 | } | ||
| 163 | |||
| 164 | static void put_no_rnd_pixels8_x2_mmx(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h) | ||
| 165 | { | ||
| 166 | MOVQ_BFE(mm6); | ||
| 167 | __asm__ volatile( | ||
| 168 | "lea (%3, %3), %%"FF_REG_a" \n\t" | ||
| 169 | ".p2align 3 \n\t" | ||
| 170 | "1: \n\t" | ||
| 171 | "movq (%1), %%mm0 \n\t" | ||
| 172 | "movq 1(%1), %%mm1 \n\t" | ||
| 173 | "movq (%1, %3), %%mm2 \n\t" | ||
| 174 | "movq 1(%1, %3), %%mm3 \n\t" | ||
| 175 | PAVGBP_MMX_NO_RND(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5) | ||
| 176 | "movq %%mm4, (%2) \n\t" | ||
| 177 | "movq %%mm5, (%2, %3) \n\t" | ||
| 178 | "add %%"FF_REG_a", %1 \n\t" | ||
| 179 | "add %%"FF_REG_a", %2 \n\t" | ||
| 180 | "movq (%1), %%mm0 \n\t" | ||
| 181 | "movq 1(%1), %%mm1 \n\t" | ||
| 182 | "movq (%1, %3), %%mm2 \n\t" | ||
| 183 | "movq 1(%1, %3), %%mm3 \n\t" | ||
| 184 | PAVGBP_MMX_NO_RND(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5) | ||
| 185 | "movq %%mm4, (%2) \n\t" | ||
| 186 | "movq %%mm5, (%2, %3) \n\t" | ||
| 187 | "add %%"FF_REG_a", %1 \n\t" | ||
| 188 | "add %%"FF_REG_a", %2 \n\t" | ||
| 189 | "subl $4, %0 \n\t" | ||
| 190 | "jnz 1b \n\t" | ||
| 191 | :"+g"(h), "+S"(pixels), "+D"(block) | ||
| 192 | :"r"((x86_reg)line_size) | ||
| 193 | :FF_REG_a, "memory"); | ||
| 194 | } | ||
| 195 | |||
| 196 | static void put_no_rnd_pixels16_x2_mmx(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h) | ||
| 197 | { | ||
| 198 | MOVQ_BFE(mm6); | ||
| 199 | __asm__ volatile( | ||
| 200 | "lea (%3, %3), %%"FF_REG_a" \n\t" | ||
| 201 | ".p2align 3 \n\t" | ||
| 202 | "1: \n\t" | ||
| 203 | "movq (%1), %%mm0 \n\t" | ||
| 204 | "movq 1(%1), %%mm1 \n\t" | ||
| 205 | "movq (%1, %3), %%mm2 \n\t" | ||
| 206 | "movq 1(%1, %3), %%mm3 \n\t" | ||
| 207 | PAVGBP_MMX_NO_RND(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5) | ||
| 208 | "movq %%mm4, (%2) \n\t" | ||
| 209 | "movq %%mm5, (%2, %3) \n\t" | ||
| 210 | "movq 8(%1), %%mm0 \n\t" | ||
| 211 | "movq 9(%1), %%mm1 \n\t" | ||
| 212 | "movq 8(%1, %3), %%mm2 \n\t" | ||
| 213 | "movq 9(%1, %3), %%mm3 \n\t" | ||
| 214 | PAVGBP_MMX_NO_RND(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5) | ||
| 215 | "movq %%mm4, 8(%2) \n\t" | ||
| 216 | "movq %%mm5, 8(%2, %3) \n\t" | ||
| 217 | "add %%"FF_REG_a", %1 \n\t" | ||
| 218 | "add %%"FF_REG_a", %2 \n\t" | ||
| 219 | "movq (%1), %%mm0 \n\t" | ||
| 220 | "movq 1(%1), %%mm1 \n\t" | ||
| 221 | "movq (%1, %3), %%mm2 \n\t" | ||
| 222 | "movq 1(%1, %3), %%mm3 \n\t" | ||
| 223 | PAVGBP_MMX_NO_RND(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5) | ||
| 224 | "movq %%mm4, (%2) \n\t" | ||
| 225 | "movq %%mm5, (%2, %3) \n\t" | ||
| 226 | "movq 8(%1), %%mm0 \n\t" | ||
| 227 | "movq 9(%1), %%mm1 \n\t" | ||
| 228 | "movq 8(%1, %3), %%mm2 \n\t" | ||
| 229 | "movq 9(%1, %3), %%mm3 \n\t" | ||
| 230 | PAVGBP_MMX_NO_RND(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5) | ||
| 231 | "movq %%mm4, 8(%2) \n\t" | ||
| 232 | "movq %%mm5, 8(%2, %3) \n\t" | ||
| 233 | "add %%"FF_REG_a", %1 \n\t" | ||
| 234 | "add %%"FF_REG_a", %2 \n\t" | ||
| 235 | "subl $4, %0 \n\t" | ||
| 236 | "jnz 1b \n\t" | ||
| 237 | :"+g"(h), "+S"(pixels), "+D"(block) | ||
| 238 | :"r"((x86_reg)line_size) | ||
| 239 | :FF_REG_a, "memory"); | ||
| 240 | } | ||
| 241 | |||
| 242 | static void put_no_rnd_pixels8_y2_mmx(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h) | ||
| 243 | { | ||
| 244 | MOVQ_BFE(mm6); | ||
| 245 | __asm__ volatile( | ||
| 246 | "lea (%3, %3), %%"FF_REG_a" \n\t" | ||
| 247 | "movq (%1), %%mm0 \n\t" | ||
| 248 | ".p2align 3 \n\t" | ||
| 249 | "1: \n\t" | ||
| 250 | "movq (%1, %3), %%mm1 \n\t" | ||
| 251 | "movq (%1, %%"FF_REG_a"),%%mm2\n\t" | ||
| 252 | PAVGBP_MMX_NO_RND(%%mm1, %%mm0, %%mm4, %%mm2, %%mm1, %%mm5) | ||
| 253 | "movq %%mm4, (%2) \n\t" | ||
| 254 | "movq %%mm5, (%2, %3) \n\t" | ||
| 255 | "add %%"FF_REG_a", %1 \n\t" | ||
| 256 | "add %%"FF_REG_a", %2 \n\t" | ||
| 257 | "movq (%1, %3), %%mm1 \n\t" | ||
| 258 | "movq (%1, %%"FF_REG_a"),%%mm0\n\t" | ||
| 259 | PAVGBP_MMX_NO_RND(%%mm1, %%mm2, %%mm4, %%mm0, %%mm1, %%mm5) | ||
| 260 | "movq %%mm4, (%2) \n\t" | ||
| 261 | "movq %%mm5, (%2, %3) \n\t" | ||
| 262 | "add %%"FF_REG_a", %1 \n\t" | ||
| 263 | "add %%"FF_REG_a", %2 \n\t" | ||
| 264 | "subl $4, %0 \n\t" | ||
| 265 | "jnz 1b \n\t" | ||
| 266 | :"+g"(h), "+S"(pixels), "+D"(block) | ||
| 267 | :"r"((x86_reg)line_size) | ||
| 268 | :FF_REG_a, "memory"); | ||
| 269 | } | ||
| 270 | |||
| 271 | static void avg_no_rnd_pixels16_x2_mmx(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h) | ||
| 272 | { | ||
| 273 | MOVQ_BFE(mm6); | ||
| 274 | __asm__ volatile( | ||
| 275 | ".p2align 3 \n\t" | ||
| 276 | "1: \n\t" | ||
| 277 | "movq (%1), %%mm0 \n\t" | ||
| 278 | "movq 1(%1), %%mm1 \n\t" | ||
| 279 | "movq (%2), %%mm3 \n\t" | ||
| 280 | PAVGB_MMX_NO_RND(%%mm0, %%mm1, %%mm2, %%mm6) | ||
| 281 | PAVGB_MMX(%%mm3, %%mm2, %%mm0, %%mm6) | ||
| 282 | "movq %%mm0, (%2) \n\t" | ||
| 283 | "movq 8(%1), %%mm0 \n\t" | ||
| 284 | "movq 9(%1), %%mm1 \n\t" | ||
| 285 | "movq 8(%2), %%mm3 \n\t" | ||
| 286 | PAVGB_MMX_NO_RND(%%mm0, %%mm1, %%mm2, %%mm6) | ||
| 287 | PAVGB_MMX(%%mm3, %%mm2, %%mm0, %%mm6) | ||
| 288 | "movq %%mm0, 8(%2) \n\t" | ||
| 289 | "add %3, %1 \n\t" | ||
| 290 | "add %3, %2 \n\t" | ||
| 291 | "subl $1, %0 \n\t" | ||
| 292 | "jnz 1b \n\t" | ||
| 293 | :"+g"(h), "+S"(pixels), "+D"(block) | ||
| 294 | :"r"((x86_reg)line_size) | ||
| 295 | :"memory"); | ||
| 296 | } | ||
| 297 | |||
| 298 | static void avg_no_rnd_pixels8_y2_mmx(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h) | ||
| 299 | { | ||
| 300 | MOVQ_BFE(mm6); | ||
| 301 | __asm__ volatile( | ||
| 302 | "lea (%3, %3), %%"FF_REG_a" \n\t" | ||
| 303 | "movq (%1), %%mm0 \n\t" | ||
| 304 | ".p2align 3 \n\t" | ||
| 305 | "1: \n\t" | ||
| 306 | "movq (%1, %3), %%mm1 \n\t" | ||
| 307 | "movq (%1, %%"FF_REG_a"), %%mm2 \n\t" | ||
| 308 | PAVGBP_MMX_NO_RND(%%mm1, %%mm0, %%mm4, %%mm2, %%mm1, %%mm5) | ||
| 309 | "movq (%2), %%mm3 \n\t" | ||
| 310 | PAVGB_MMX(%%mm3, %%mm4, %%mm0, %%mm6) | ||
| 311 | "movq (%2, %3), %%mm3 \n\t" | ||
| 312 | PAVGB_MMX(%%mm3, %%mm5, %%mm1, %%mm6) | ||
| 313 | "movq %%mm0, (%2) \n\t" | ||
| 314 | "movq %%mm1, (%2, %3) \n\t" | ||
| 315 | "add %%"FF_REG_a", %1 \n\t" | ||
| 316 | "add %%"FF_REG_a", %2 \n\t" | ||
| 317 | |||
| 318 | "movq (%1, %3), %%mm1 \n\t" | ||
| 319 | "movq (%1, %%"FF_REG_a"), %%mm0 \n\t" | ||
| 320 | PAVGBP_MMX_NO_RND(%%mm1, %%mm2, %%mm4, %%mm0, %%mm1, %%mm5) | ||
| 321 | "movq (%2), %%mm3 \n\t" | ||
| 322 | PAVGB_MMX(%%mm3, %%mm4, %%mm2, %%mm6) | ||
| 323 | "movq (%2, %3), %%mm3 \n\t" | ||
| 324 | PAVGB_MMX(%%mm3, %%mm5, %%mm1, %%mm6) | ||
| 325 | "movq %%mm2, (%2) \n\t" | ||
| 326 | "movq %%mm1, (%2, %3) \n\t" | ||
| 327 | "add %%"FF_REG_a", %1 \n\t" | ||
| 328 | "add %%"FF_REG_a", %2 \n\t" | ||
| 329 | |||
| 330 | "subl $4, %0 \n\t" | ||
| 331 | "jnz 1b \n\t" | ||
| 332 | :"+g"(h), "+S"(pixels), "+D"(block) | ||
| 333 | :"r"((x86_reg)line_size) | ||
| 334 | :FF_REG_a, "memory"); | ||
| 335 | } | ||
| 336 | |||
| 90 | #if HAVE_MMX | 337 | #if HAVE_MMX |
| 91 | CALL_2X_PIXELS(avg_no_rnd_pixels16_y2_mmx, avg_no_rnd_pixels8_y2_mmx, 8) | 338 | CALL_2X_PIXELS(avg_no_rnd_pixels16_y2_mmx, avg_no_rnd_pixels8_y2_mmx, 8) |
| 92 | CALL_2X_PIXELS(put_no_rnd_pixels16_y2_mmx, put_no_rnd_pixels8_y2_mmx, 8) | 339 | CALL_2X_PIXELS(put_no_rnd_pixels16_y2_mmx, put_no_rnd_pixels8_y2_mmx, 8) |
| @@ -101,7 +348,6 @@ CALL_2X_PIXELS(put_no_rnd_pixels16_xy2_mmx, put_no_rnd_pixels8_xy2_mmx, 8) · 8 unmodified lines | |||
| 101 | #define SET_RND MOVQ_WTWO | 348 | #define SET_RND MOVQ_WTWO |
| 102 | #define DEF(x, y) ff_ ## x ## _ ## y ## _mmx | 349 | #define DEF(x, y) ff_ ## x ## _ ## y ## _mmx |
| 103 | #define STATIC | 350 | #define STATIC |
| 104 | #define NO_AVG | ||
| 105 | 351 | ||
| 106 | #include "rnd_template.c" | 352 | #include "rnd_template.c" |
| 107 | 353 | ||
| @@ -1,202 +0,0 @@ | |||
| 1 | /* | ||
| 2 | * SIMD-optimized halfpel functions are compiled twice for rnd/no_rnd | ||
| 3 | * Copyright (c) 2000, 2001 Fabrice Bellard | ||
| 4 | * Copyright (c) 2003-2004 Michael Niedermayer <michaelni@gmx.at> | ||
| 5 | * | ||
| 6 | * MMX optimization by Nick Kurshev <nickols_k@mail.ru> | ||
| 7 | * mostly rewritten by Michael Niedermayer <michaelni@gmx.at> | ||
| 8 | * and improved by Zdenek Kabelac <kabi@users.sf.net> | ||
| 9 | * | ||
| 10 | * This file is part of FFmpeg. | ||
| 11 | * | ||
| 12 | * FFmpeg is free software; you can redistribute it and/or | ||
| 13 | * modify it under the terms of the GNU Lesser General Public | ||
| 14 | * License as published by the Free Software Foundation; either | ||
| 15 | * version 2.1 of the License, or (at your option) any later version. | ||
| 16 | * | ||
| 17 | * FFmpeg is distributed in the hope that it will be useful, | ||
| 18 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
| 19 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | ||
| 20 | * Lesser General Public License for more details. | ||
| 21 | * | ||
| 22 | * You should have received a copy of the GNU Lesser General Public | ||
| 23 | * License along with FFmpeg; if not, write to the Free Software | ||
| 24 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | ||
| 25 | */ | ||
| 26 | |||
| 27 | #include <stddef.h> | ||
| 28 | #include <stdint.h> | ||
| 29 | |||
| 30 | // put_pixels | ||
| 31 | av_unused static void DEF(put, pixels8_x2)(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h) | ||
| 32 | { | ||
| 33 | MOVQ_BFE(mm6); | ||
| 34 | __asm__ volatile( | ||
| 35 | "lea (%3, %3), %%"FF_REG_a" \n\t" | ||
| 36 | ".p2align 3 \n\t" | ||
| 37 | "1: \n\t" | ||
| 38 | "movq (%1), %%mm0 \n\t" | ||
| 39 | "movq 1(%1), %%mm1 \n\t" | ||
| 40 | "movq (%1, %3), %%mm2 \n\t" | ||
| 41 | "movq 1(%1, %3), %%mm3 \n\t" | ||
| 42 | PAVGBP(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5) | ||
| 43 | "movq %%mm4, (%2) \n\t" | ||
| 44 | "movq %%mm5, (%2, %3) \n\t" | ||
| 45 | "add %%"FF_REG_a", %1 \n\t" | ||
| 46 | "add %%"FF_REG_a", %2 \n\t" | ||
| 47 | "movq (%1), %%mm0 \n\t" | ||
| 48 | "movq 1(%1), %%mm1 \n\t" | ||
| 49 | "movq (%1, %3), %%mm2 \n\t" | ||
| 50 | "movq 1(%1, %3), %%mm3 \n\t" | ||
| 51 | PAVGBP(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5) | ||
| 52 | "movq %%mm4, (%2) \n\t" | ||
| 53 | "movq %%mm5, (%2, %3) \n\t" | ||
| 54 | "add %%"FF_REG_a", %1 \n\t" | ||
| 55 | "add %%"FF_REG_a", %2 \n\t" | ||
| 56 | "subl $4, %0 \n\t" | ||
| 57 | "jnz 1b \n\t" | ||
| 58 | :"+g"(h), "+S"(pixels), "+D"(block) | ||
| 59 | :"r"((x86_reg)line_size) | ||
| 60 | :FF_REG_a, "memory"); | ||
| 61 | } | ||
| 62 | |||
| 63 | av_unused static void DEF(put, pixels16_x2)(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h) | ||
| 64 | { | ||
| 65 | MOVQ_BFE(mm6); | ||
| 66 | __asm__ volatile( | ||
| 67 | "lea (%3, %3), %%"FF_REG_a" \n\t" | ||
| 68 | ".p2align 3 \n\t" | ||
| 69 | "1: \n\t" | ||
| 70 | "movq (%1), %%mm0 \n\t" | ||
| 71 | "movq 1(%1), %%mm1 \n\t" | ||
| 72 | "movq (%1, %3), %%mm2 \n\t" | ||
| 73 | "movq 1(%1, %3), %%mm3 \n\t" | ||
| 74 | PAVGBP(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5) | ||
| 75 | "movq %%mm4, (%2) \n\t" | ||
| 76 | "movq %%mm5, (%2, %3) \n\t" | ||
| 77 | "movq 8(%1), %%mm0 \n\t" | ||
| 78 | "movq 9(%1), %%mm1 \n\t" | ||
| 79 | "movq 8(%1, %3), %%mm2 \n\t" | ||
| 80 | "movq 9(%1, %3), %%mm3 \n\t" | ||
| 81 | PAVGBP(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5) | ||
| 82 | "movq %%mm4, 8(%2) \n\t" | ||
| 83 | "movq %%mm5, 8(%2, %3) \n\t" | ||
| 84 | "add %%"FF_REG_a", %1 \n\t" | ||
| 85 | "add %%"FF_REG_a", %2 \n\t" | ||
| 86 | "movq (%1), %%mm0 \n\t" | ||
| 87 | "movq 1(%1), %%mm1 \n\t" | ||
| 88 | "movq (%1, %3), %%mm2 \n\t" | ||
| 89 | "movq 1(%1, %3), %%mm3 \n\t" | ||
| 90 | PAVGBP(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5) | ||
| 91 | "movq %%mm4, (%2) \n\t" | ||
| 92 | "movq %%mm5, (%2, %3) \n\t" | ||
| 93 | "movq 8(%1), %%mm0 \n\t" | ||
| 94 | "movq 9(%1), %%mm1 \n\t" | ||
| 95 | "movq 8(%1, %3), %%mm2 \n\t" | ||
| 96 | "movq 9(%1, %3), %%mm3 \n\t" | ||
| 97 | PAVGBP(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5) | ||
| 98 | "movq %%mm4, 8(%2) \n\t" | ||
| 99 | "movq %%mm5, 8(%2, %3) \n\t" | ||
| 100 | "add %%"FF_REG_a", %1 \n\t" | ||
| 101 | "add %%"FF_REG_a", %2 \n\t" | ||
| 102 | "subl $4, %0 \n\t" | ||
| 103 | "jnz 1b \n\t" | ||
| 104 | :"+g"(h), "+S"(pixels), "+D"(block) | ||
| 105 | :"r"((x86_reg)line_size) | ||
| 106 | :FF_REG_a, "memory"); | ||
| 107 | } | ||
| 108 | |||
| 109 | av_unused static void DEF(put, pixels8_y2)(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h) | ||
| 110 | { | ||
| 111 | MOVQ_BFE(mm6); | ||
| 112 | __asm__ volatile( | ||
| 113 | "lea (%3, %3), %%"FF_REG_a" \n\t" | ||
| 114 | "movq (%1), %%mm0 \n\t" | ||
| 115 | ".p2align 3 \n\t" | ||
| 116 | "1: \n\t" | ||
| 117 | "movq (%1, %3), %%mm1 \n\t" | ||
| 118 | "movq (%1, %%"FF_REG_a"),%%mm2\n\t" | ||
| 119 | PAVGBP(%%mm1, %%mm0, %%mm4, %%mm2, %%mm1, %%mm5) | ||
| 120 | "movq %%mm4, (%2) \n\t" | ||
| 121 | "movq %%mm5, (%2, %3) \n\t" | ||
| 122 | "add %%"FF_REG_a", %1 \n\t" | ||
| 123 | "add %%"FF_REG_a", %2 \n\t" | ||
| 124 | "movq (%1, %3), %%mm1 \n\t" | ||
| 125 | "movq (%1, %%"FF_REG_a"),%%mm0\n\t" | ||
| 126 | PAVGBP(%%mm1, %%mm2, %%mm4, %%mm0, %%mm1, %%mm5) | ||
| 127 | "movq %%mm4, (%2) \n\t" | ||
| 128 | "movq %%mm5, (%2, %3) \n\t" | ||
| 129 | "add %%"FF_REG_a", %1 \n\t" | ||
| 130 | "add %%"FF_REG_a", %2 \n\t" | ||
| 131 | "subl $4, %0 \n\t" | ||
| 132 | "jnz 1b \n\t" | ||
| 133 | :"+g"(h), "+S"(pixels), "+D"(block) | ||
| 134 | :"r"((x86_reg)line_size) | ||
| 135 | :FF_REG_a, "memory"); | ||
| 136 | } | ||
| 137 | |||
| 138 | av_unused static void DEF(avg, pixels16_x2)(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h) | ||
| 139 | { | ||
| 140 | MOVQ_BFE(mm6); | ||
| 141 | __asm__ volatile( | ||
| 142 | ".p2align 3 \n\t" | ||
| 143 | "1: \n\t" | ||
| 144 | "movq (%1), %%mm0 \n\t" | ||
| 145 | "movq 1(%1), %%mm1 \n\t" | ||
| 146 | "movq (%2), %%mm3 \n\t" | ||
| 147 | PAVGB(%%mm0, %%mm1, %%mm2, %%mm6) | ||
| 148 | PAVGB_MMX(%%mm3, %%mm2, %%mm0, %%mm6) | ||
| 149 | "movq %%mm0, (%2) \n\t" | ||
| 150 | "movq 8(%1), %%mm0 \n\t" | ||
| 151 | "movq 9(%1), %%mm1 \n\t" | ||
| 152 | "movq 8(%2), %%mm3 \n\t" | ||
| 153 | PAVGB(%%mm0, %%mm1, %%mm2, %%mm6) | ||
| 154 | PAVGB_MMX(%%mm3, %%mm2, %%mm0, %%mm6) | ||
| 155 | "movq %%mm0, 8(%2) \n\t" | ||
| 156 | "add %3, %1 \n\t" | ||
| 157 | "add %3, %2 \n\t" | ||
| 158 | "subl $1, %0 \n\t" | ||
| 159 | "jnz 1b \n\t" | ||
| 160 | :"+g"(h), "+S"(pixels), "+D"(block) | ||
| 161 | :"r"((x86_reg)line_size) | ||
| 162 | :"memory"); | ||
| 163 | } | ||
| 164 | |||
| 165 | av_unused static void DEF(avg, pixels8_y2)(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h) | ||
| 166 | { | ||
| 167 | MOVQ_BFE(mm6); | ||
| 168 | __asm__ volatile( | ||
| 169 | "lea (%3, %3), %%"FF_REG_a" \n\t" | ||
| 170 | "movq (%1), %%mm0 \n\t" | ||
| 171 | ".p2align 3 \n\t" | ||
| 172 | "1: \n\t" | ||
| 173 | "movq (%1, %3), %%mm1 \n\t" | ||
| 174 | "movq (%1, %%"FF_REG_a"), %%mm2 \n\t" | ||
| 175 | PAVGBP(%%mm1, %%mm0, %%mm4, %%mm2, %%mm1, %%mm5) | ||
| 176 | "movq (%2), %%mm3 \n\t" | ||
| 177 | PAVGB_MMX(%%mm3, %%mm4, %%mm0, %%mm6) | ||
| 178 | "movq (%2, %3), %%mm3 \n\t" | ||
| 179 | PAVGB_MMX(%%mm3, %%mm5, %%mm1, %%mm6) | ||
| 180 | "movq %%mm0, (%2) \n\t" | ||
| 181 | "movq %%mm1, (%2, %3) \n\t" | ||
| 182 | "add %%"FF_REG_a", %1 \n\t" | ||
| 183 | "add %%"FF_REG_a", %2 \n\t" | ||
| 184 | |||
| 185 | "movq (%1, %3), %%mm1 \n\t" | ||
| 186 | "movq (%1, %%"FF_REG_a"), %%mm0 \n\t" | ||
| 187 | PAVGBP(%%mm1, %%mm2, %%mm4, %%mm0, %%mm1, %%mm5) | ||
| 188 | "movq (%2), %%mm3 \n\t" | ||
| 189 | PAVGB_MMX(%%mm3, %%mm4, %%mm2, %%mm6) | ||
| 190 | "movq (%2, %3), %%mm3 \n\t" | ||
| 191 | PAVGB_MMX(%%mm3, %%mm5, %%mm1, %%mm6) | ||
| 192 | "movq %%mm2, (%2) \n\t" | ||
| 193 | "movq %%mm1, (%2, %3) \n\t" | ||
| 194 | "add %%"FF_REG_a", %1 \n\t" | ||
| 195 | "add %%"FF_REG_a", %2 \n\t" | ||
| 196 | |||
| 197 | "subl $4, %0 \n\t" | ||
| 198 | "jnz 1b \n\t" | ||
| 199 | :"+g"(h), "+S"(pixels), "+D"(block) | ||
| 200 | :"r"((x86_reg)line_size) | ||
| 201 | :FF_REG_a, "memory"); | ||
| 202 | } | ||
| @@ -96,82 +96,3 @@ av_unused STATIC void DEF(put, pixels8_xy2)(uint8_t *block, const uint8_t *pixel · 95 unmodified lines | |||
| 96 | :"D"(block), "r"((x86_reg)line_size) | 96 | :"D"(block), "r"((x86_reg)line_size) |
| 97 | :FF_REG_a, "memory"); | 97 | :FF_REG_a, "memory"); |
| 98 | } | 98 | } |
| 99 | |||
| 100 | #ifndef NO_AVG | ||
| 101 | // avg_pixels | ||
| 102 | // this routine is 'slightly' suboptimal but mostly unused | ||
| 103 | av_unused STATIC void DEF(avg, pixels8_xy2)(uint8_t *block, const uint8_t *pixels, | ||
| 104 | ptrdiff_t line_size, int h) | ||
| 105 | { | ||
| 106 | MOVQ_ZERO(mm7); | ||
| 107 | SET_RND(mm6); // =2 for rnd and =1 for no_rnd version | ||
| 108 | __asm__ volatile( | ||
| 109 | "movq (%1), %%mm0 \n\t" | ||
| 110 | "movq 1(%1), %%mm4 \n\t" | ||
| 111 | "movq %%mm0, %%mm1 \n\t" | ||
| 112 | "movq %%mm4, %%mm5 \n\t" | ||
| 113 | "punpcklbw %%mm7, %%mm0 \n\t" | ||
| 114 | "punpcklbw %%mm7, %%mm4 \n\t" | ||
| 115 | "punpckhbw %%mm7, %%mm1 \n\t" | ||
| 116 | "punpckhbw %%mm7, %%mm5 \n\t" | ||
| 117 | "paddusw %%mm0, %%mm4 \n\t" | ||
| 118 | "paddusw %%mm1, %%mm5 \n\t" | ||
| 119 | "xor %%"FF_REG_a", %%"FF_REG_a" \n\t" | ||
| 120 | "add %3, %1 \n\t" | ||
| 121 | ".p2align 3 \n\t" | ||
| 122 | "1: \n\t" | ||
| 123 | "movq (%1, %%"FF_REG_a"), %%mm0 \n\t" | ||
| 124 | "movq 1(%1, %%"FF_REG_a"), %%mm2 \n\t" | ||
| 125 | "movq %%mm0, %%mm1 \n\t" | ||
| 126 | "movq %%mm2, %%mm3 \n\t" | ||
| 127 | "punpcklbw %%mm7, %%mm0 \n\t" | ||
| 128 | "punpcklbw %%mm7, %%mm2 \n\t" | ||
| 129 | "punpckhbw %%mm7, %%mm1 \n\t" | ||
| 130 | "punpckhbw %%mm7, %%mm3 \n\t" | ||
| 131 | "paddusw %%mm2, %%mm0 \n\t" | ||
| 132 | "paddusw %%mm3, %%mm1 \n\t" | ||
| 133 | "paddusw %%mm6, %%mm4 \n\t" | ||
| 134 | "paddusw %%mm6, %%mm5 \n\t" | ||
| 135 | "paddusw %%mm0, %%mm4 \n\t" | ||
| 136 | "paddusw %%mm1, %%mm5 \n\t" | ||
| 137 | "psrlw $2, %%mm4 \n\t" | ||
| 138 | "psrlw $2, %%mm5 \n\t" | ||
| 139 | "movq (%2, %%"FF_REG_a"), %%mm3 \n\t" | ||
| 140 | "packuswb %%mm5, %%mm4 \n\t" | ||
| 141 | "pcmpeqd %%mm2, %%mm2 \n\t" | ||
| 142 | "paddb %%mm2, %%mm2 \n\t" | ||
| 143 | PAVGB_MMX(%%mm3, %%mm4, %%mm5, %%mm2) | ||
| 144 | "movq %%mm5, (%2, %%"FF_REG_a") \n\t" | ||
| 145 | "add %3, %%"FF_REG_a" \n\t" | ||
| 146 | |||
| 147 | "movq (%1, %%"FF_REG_a"), %%mm2 \n\t" // 0 <-> 2 1 <-> 3 | ||
| 148 | "movq 1(%1, %%"FF_REG_a"), %%mm4 \n\t" | ||
| 149 | "movq %%mm2, %%mm3 \n\t" | ||
| 150 | "movq %%mm4, %%mm5 \n\t" | ||
| 151 | "punpcklbw %%mm7, %%mm2 \n\t" | ||
| 152 | "punpcklbw %%mm7, %%mm4 \n\t" | ||
| 153 | "punpckhbw %%mm7, %%mm3 \n\t" | ||
| 154 | "punpckhbw %%mm7, %%mm5 \n\t" | ||
| 155 | "paddusw %%mm2, %%mm4 \n\t" | ||
| 156 | "paddusw %%mm3, %%mm5 \n\t" | ||
| 157 | "paddusw %%mm6, %%mm0 \n\t" | ||
| 158 | "paddusw %%mm6, %%mm1 \n\t" | ||
| 159 | "paddusw %%mm4, %%mm0 \n\t" | ||
| 160 | "paddusw %%mm5, %%mm1 \n\t" | ||
| 161 | "psrlw $2, %%mm0 \n\t" | ||
| 162 | "psrlw $2, %%mm1 \n\t" | ||
| 163 | "movq (%2, %%"FF_REG_a"), %%mm3 \n\t" | ||
| 164 | "packuswb %%mm1, %%mm0 \n\t" | ||
| 165 | "pcmpeqd %%mm2, %%mm2 \n\t" | ||
| 166 | "paddb %%mm2, %%mm2 \n\t" | ||
| 167 | PAVGB_MMX(%%mm3, %%mm0, %%mm1, %%mm2) | ||
| 168 | "movq %%mm1, (%2, %%"FF_REG_a") \n\t" | ||
| 169 | "add %3, %%"FF_REG_a" \n\t" | ||
| 170 | |||
| 171 | "subl $2, %0 \n\t" | ||
| 172 | "jnz 1b \n\t" | ||
| 173 | :"+g"(h), "+S"(pixels) | ||
| 174 | :"D"(block), "r"((x86_reg)line_size) | ||
| 175 | :FF_REG_a, "memory"); | ||
| 176 | } | ||
| 177 | #endif | ||