simple_idct_template.c (11724B)
1 /* 2 * Simple IDCT 3 * 4 * Copyright (c) 2001 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 23 /** 24 * @file 25 * simpleidct in C. 26 */ 27 28 /* Based upon some commented-out C code from mpeg2dec (idct_mmx.c 29 * written by Aaron Holtzman <aholtzma@ess.engr.uvic.ca>). */ 30 31 #include "simple_idct.h" 32 33 #include "bit_depth_template.c" 34 35 #undef W1 36 #undef W2 37 #undef W3 38 #undef W4 39 #undef W5 40 #undef W6 41 #undef W7 42 #undef ROW_SHIFT 43 #undef COL_SHIFT 44 #undef DC_SHIFT 45 #undef MUL 46 #undef MAC 47 48 #if BIT_DEPTH == 8 49 50 #define W1 22725 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5 51 #define W2 21407 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5 52 #define W3 19266 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5 53 #define W4 16383 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5 54 #define W5 12873 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5 55 #define W6 8867 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5 56 #define W7 4520 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5 57 58 #define ROW_SHIFT 11 59 #define COL_SHIFT 20 60 #define DC_SHIFT 3 61 62 #define MUL(a, b) MUL16(a, b) 63 #define MAC(a, b, c) MAC16(a, b, c) 64 65 #elif BIT_DEPTH == 10 || BIT_DEPTH == 12 66 67 # if BIT_DEPTH == 10 68 #define W1 22725 // 90901 69 #define W2 21407 // 85627 70 #define W3 19265 // 77062 71 #define W4 16384 // 65535 72 #define W5 12873 // 51491 73 #define W6 8867 // 35468 74 #define W7 4520 // 18081 75 76 # ifdef EXTRA_SHIFT 77 #define ROW_SHIFT 13 78 #define COL_SHIFT 18 79 #define DC_SHIFT 1 80 # elif IN_IDCT_DEPTH == 32 81 #define ROW_SHIFT 13 82 #define COL_SHIFT 21 83 #define DC_SHIFT 2 84 # else 85 #define ROW_SHIFT 12 86 #define COL_SHIFT 19 87 #define DC_SHIFT 2 88 # endif 89 90 # else 91 #define W1 45451 92 #define W2 42813 93 #define W3 38531 94 #define W4 32767 95 #define W5 25746 96 #define W6 17734 97 #define W7 9041 98 99 #define ROW_SHIFT 16 100 #define COL_SHIFT 17 101 #define DC_SHIFT -1 102 # endif 103 104 #define MUL(a, b) ((int)((SUINT)(a) * (b))) 105 #define MAC(a, b, c) ((a) += (SUINT)(b) * (c)) 106 107 #else 108 109 #error "Unsupported bitdepth" 110 111 #endif 112 113 #ifdef EXTRA_SHIFT 114 static inline void FUNC(idctRowCondDC_extrashift)(int16_t *row, int extra_shift) 115 #else 116 static inline void FUNC6(idctRowCondDC)(idctin *row, int extra_shift) 117 #endif 118 { 119 SUINT a0, a1, a2, a3, b0, b1, b2, b3; 120 121 // TODO: Add DC-only support for int32_t input 122 #if IN_IDCT_DEPTH == 16 123 #if HAVE_FAST_64BIT 124 #define ROW0_MASK (0xffffULL << 48 * HAVE_BIGENDIAN) 125 if (((AV_RN64A(row) & ~ROW0_MASK) | AV_RN64A(row+4)) == 0) { 126 uint64_t temp; 127 if (DC_SHIFT - extra_shift >= 0) { 128 temp = (row[0] * (1 << (DC_SHIFT - extra_shift))) & 0xffff; 129 } else { 130 temp = ((row[0] + (1<<(extra_shift - DC_SHIFT-1))) >> (extra_shift - DC_SHIFT)) & 0xffff; 131 } 132 temp += temp * (1 << 16); 133 temp += temp * ((uint64_t) 1 << 32); 134 AV_WN64A(row, temp); 135 AV_WN64A(row + 4, temp); 136 return; 137 } 138 #else 139 if (!(AV_RN32A(row+2) | 140 AV_RN32A(row+4) | 141 AV_RN32A(row+6) | 142 row[1])) { 143 uint32_t temp; 144 if (DC_SHIFT - extra_shift >= 0) { 145 temp = (row[0] * (1 << (DC_SHIFT - extra_shift))) & 0xffff; 146 } else { 147 temp = ((row[0] + (1<<(extra_shift - DC_SHIFT-1))) >> (extra_shift - DC_SHIFT)) & 0xffff; 148 } 149 temp += temp * (1 << 16); 150 AV_WN32A(row, temp); 151 AV_WN32A(row+2, temp); 152 AV_WN32A(row+4, temp); 153 AV_WN32A(row+6, temp); 154 return; 155 } 156 #endif 157 #endif 158 159 a0 = ((SUINT)W4 * row[0]) + (1 << (ROW_SHIFT + extra_shift - 1)); 160 a1 = a0; 161 a2 = a0; 162 a3 = a0; 163 164 a0 += (SUINT)W2 * row[2]; 165 a1 += (SUINT)W6 * row[2]; 166 a2 -= (SUINT)W6 * row[2]; 167 a3 -= (SUINT)W2 * row[2]; 168 169 b0 = MUL(W1, row[1]); 170 MAC(b0, W3, row[3]); 171 b1 = MUL(W3, row[1]); 172 MAC(b1, -W7, row[3]); 173 b2 = MUL(W5, row[1]); 174 MAC(b2, -W1, row[3]); 175 b3 = MUL(W7, row[1]); 176 MAC(b3, -W5, row[3]); 177 178 #if IN_IDCT_DEPTH == 32 179 if (AV_RN64A(row + 4) | AV_RN64A(row + 6)) { 180 #else 181 if (AV_RN64A(row + 4)) { 182 #endif 183 a0 += (SUINT) W4*row[4] + (SUINT)W6*row[6]; 184 a1 += (SUINT)- W4*row[4] - (SUINT)W2*row[6]; 185 a2 += (SUINT)- W4*row[4] + (SUINT)W2*row[6]; 186 a3 += (SUINT) W4*row[4] - (SUINT)W6*row[6]; 187 188 MAC(b0, W5, row[5]); 189 MAC(b0, W7, row[7]); 190 191 MAC(b1, -W1, row[5]); 192 MAC(b1, -W5, row[7]); 193 194 MAC(b2, W7, row[5]); 195 MAC(b2, W3, row[7]); 196 197 MAC(b3, W3, row[5]); 198 MAC(b3, -W1, row[7]); 199 } 200 201 row[0] = (int)(a0 + b0) >> (ROW_SHIFT + extra_shift); 202 row[7] = (int)(a0 - b0) >> (ROW_SHIFT + extra_shift); 203 row[1] = (int)(a1 + b1) >> (ROW_SHIFT + extra_shift); 204 row[6] = (int)(a1 - b1) >> (ROW_SHIFT + extra_shift); 205 row[2] = (int)(a2 + b2) >> (ROW_SHIFT + extra_shift); 206 row[5] = (int)(a2 - b2) >> (ROW_SHIFT + extra_shift); 207 row[3] = (int)(a3 + b3) >> (ROW_SHIFT + extra_shift); 208 row[4] = (int)(a3 - b3) >> (ROW_SHIFT + extra_shift); 209 } 210 211 #define IDCT_COLS do { \ 212 a0 = (SUINT)W4 * (col[8*0] + ((1<<(COL_SHIFT-1))/W4)); \ 213 a1 = a0; \ 214 a2 = a0; \ 215 a3 = a0; \ 216 \ 217 a0 += (SUINT) W2*col[8*2]; \ 218 a1 += (SUINT) W6*col[8*2]; \ 219 a2 += (SUINT)-W6*col[8*2]; \ 220 a3 += (SUINT)-W2*col[8*2]; \ 221 \ 222 b0 = MUL(W1, col[8*1]); \ 223 b1 = MUL(W3, col[8*1]); \ 224 b2 = MUL(W5, col[8*1]); \ 225 b3 = MUL(W7, col[8*1]); \ 226 \ 227 MAC(b0, W3, col[8*3]); \ 228 MAC(b1, -W7, col[8*3]); \ 229 MAC(b2, -W1, col[8*3]); \ 230 MAC(b3, -W5, col[8*3]); \ 231 \ 232 if (col[8*4]) { \ 233 a0 += (SUINT) W4*col[8*4]; \ 234 a1 += (SUINT)-W4*col[8*4]; \ 235 a2 += (SUINT)-W4*col[8*4]; \ 236 a3 += (SUINT) W4*col[8*4]; \ 237 } \ 238 \ 239 if (col[8*5]) { \ 240 MAC(b0, W5, col[8*5]); \ 241 MAC(b1, -W1, col[8*5]); \ 242 MAC(b2, W7, col[8*5]); \ 243 MAC(b3, W3, col[8*5]); \ 244 } \ 245 \ 246 if (col[8*6]) { \ 247 a0 += (SUINT) W6*col[8*6]; \ 248 a1 += (SUINT)-W2*col[8*6]; \ 249 a2 += (SUINT) W2*col[8*6]; \ 250 a3 += (SUINT)-W6*col[8*6]; \ 251 } \ 252 \ 253 if (col[8*7]) { \ 254 MAC(b0, W7, col[8*7]); \ 255 MAC(b1, -W5, col[8*7]); \ 256 MAC(b2, W3, col[8*7]); \ 257 MAC(b3, -W1, col[8*7]); \ 258 } \ 259 } while (0) 260 261 #ifdef EXTRA_SHIFT 262 static inline void FUNC(idctSparseCol_extrashift)(int16_t *col) 263 #else 264 static inline void FUNC6(idctSparseColPut)(pixel *dest, ptrdiff_t line_size, 265 idctin *col) 266 { 267 SUINT a0, a1, a2, a3, b0, b1, b2, b3; 268 269 IDCT_COLS; 270 271 dest[0] = av_clip_pixel((int)(a0 + b0) >> COL_SHIFT); 272 dest += line_size; 273 dest[0] = av_clip_pixel((int)(a1 + b1) >> COL_SHIFT); 274 dest += line_size; 275 dest[0] = av_clip_pixel((int)(a2 + b2) >> COL_SHIFT); 276 dest += line_size; 277 dest[0] = av_clip_pixel((int)(a3 + b3) >> COL_SHIFT); 278 dest += line_size; 279 dest[0] = av_clip_pixel((int)(a3 - b3) >> COL_SHIFT); 280 dest += line_size; 281 dest[0] = av_clip_pixel((int)(a2 - b2) >> COL_SHIFT); 282 dest += line_size; 283 dest[0] = av_clip_pixel((int)(a1 - b1) >> COL_SHIFT); 284 dest += line_size; 285 dest[0] = av_clip_pixel((int)(a0 - b0) >> COL_SHIFT); 286 } 287 288 static inline void FUNC6(idctSparseColAdd)(pixel *dest, ptrdiff_t line_size, 289 idctin *col) 290 { 291 unsigned a0, a1, a2, a3, b0, b1, b2, b3; 292 293 IDCT_COLS; 294 295 dest[0] = av_clip_pixel(dest[0] + ((int)(a0 + b0) >> COL_SHIFT)); 296 dest += line_size; 297 dest[0] = av_clip_pixel(dest[0] + ((int)(a1 + b1) >> COL_SHIFT)); 298 dest += line_size; 299 dest[0] = av_clip_pixel(dest[0] + ((int)(a2 + b2) >> COL_SHIFT)); 300 dest += line_size; 301 dest[0] = av_clip_pixel(dest[0] + ((int)(a3 + b3) >> COL_SHIFT)); 302 dest += line_size; 303 dest[0] = av_clip_pixel(dest[0] + ((int)(a3 - b3) >> COL_SHIFT)); 304 dest += line_size; 305 dest[0] = av_clip_pixel(dest[0] + ((int)(a2 - b2) >> COL_SHIFT)); 306 dest += line_size; 307 dest[0] = av_clip_pixel(dest[0] + ((int)(a1 - b1) >> COL_SHIFT)); 308 dest += line_size; 309 dest[0] = av_clip_pixel(dest[0] + ((int)(a0 - b0) >> COL_SHIFT)); 310 } 311 312 static inline void FUNC6(idctSparseCol)(idctin *col) 313 #endif 314 { 315 unsigned a0, a1, a2, a3, b0, b1, b2, b3; 316 317 IDCT_COLS; 318 319 col[0 ] = ((int)(a0 + b0) >> COL_SHIFT); 320 col[8 ] = ((int)(a1 + b1) >> COL_SHIFT); 321 col[16] = ((int)(a2 + b2) >> COL_SHIFT); 322 col[24] = ((int)(a3 + b3) >> COL_SHIFT); 323 col[32] = ((int)(a3 - b3) >> COL_SHIFT); 324 col[40] = ((int)(a2 - b2) >> COL_SHIFT); 325 col[48] = ((int)(a1 - b1) >> COL_SHIFT); 326 col[56] = ((int)(a0 - b0) >> COL_SHIFT); 327 } 328 329 #ifndef EXTRA_SHIFT 330 void FUNC6(ff_simple_idct_put)(uint8_t *dest_, ptrdiff_t line_size, int16_t *block_) 331 { 332 idctin *block = (idctin *)block_; 333 pixel *dest = (pixel *)dest_; 334 int i; 335 336 line_size /= sizeof(pixel); 337 338 for (i = 0; i < 8; i++) 339 FUNC6(idctRowCondDC)(block + i*8, 0); 340 341 for (i = 0; i < 8; i++) 342 FUNC6(idctSparseColPut)(dest + i, line_size, block + i); 343 } 344 345 #if IN_IDCT_DEPTH == 16 346 void FUNC6(ff_simple_idct_add)(uint8_t *dest_, ptrdiff_t line_size, int16_t *block) 347 { 348 pixel *dest = (pixel *)dest_; 349 int i; 350 351 line_size /= sizeof(pixel); 352 353 for (i = 0; i < 8; i++) 354 FUNC6(idctRowCondDC)(block + i*8, 0); 355 356 for (i = 0; i < 8; i++) 357 FUNC6(idctSparseColAdd)(dest + i, line_size, block + i); 358 } 359 360 void FUNC6(ff_simple_idct)(int16_t *block) 361 { 362 int i; 363 364 for (i = 0; i < 8; i++) 365 FUNC6(idctRowCondDC)(block + i*8, 0); 366 367 for (i = 0; i < 8; i++) 368 FUNC6(idctSparseCol)(block + i); 369 } 370 #endif 371 #endif