projects / libav.git / blame
| Commit | Line | Data |
|---|---|---|
| 68d8238c | 1 | /* |
| 2539cf1c DB |
2 | * Half-pel DSP functions |
| 3 | * | |
| 68d8238c RB |
4 | * Copyright (c) 2000, 2001 Fabrice Bellard |
| 5 | * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at> | |
| 6 | * | |
| 7 | * gmc & q-pel & 32/64 bit based MC by Michael Niedermayer <michaelni@gmx.at> | |
| 8 | * | |
| 9 | * This file is part of Libav. | |
| 10 | * | |
| 11 | * Libav is free software; you can redistribute it and/or | |
| 12 | * modify it under the terms of the GNU Lesser General Public | |
| 13 | * License as published by the Free Software Foundation; either | |
| 14 | * version 2.1 of the License, or (at your option) any later version. | |
| 15 | * | |
| 16 | * Libav is distributed in the hope that it will be useful, | |
| 17 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
| 18 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
| 19 | * Lesser General Public License for more details. | |
| 20 | * | |
| 21 | * You should have received a copy of the GNU Lesser General Public | |
| 22 | * License along with Libav; if not, write to the Free Software | |
| 23 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | |
| 24 | */ | |
| 25 | ||
| 26 | /** | |
| 27 | * @file | |
| 2539cf1c | 28 | * Half-pel DSP functions |
| 68d8238c RB |
29 | */ |
| 30 | ||
| 31 | #include "bit_depth_template.c" | |
| 32 | ||
| 33 | #include "hpel_template.c" | |
| 34 | ||
| 2539cf1c DB |
35 | #define PIXOP2(OPNAME, OP) \ |
| 36 | static inline void FUNC(OPNAME ## _no_rnd_pixels8_l2)(uint8_t *dst, \ | |
| 37 | const uint8_t *src1, \ | |
| 38 | const uint8_t *src2, \ | |
| 39 | int dst_stride, \ | |
| 40 | int src_stride1, \ | |
| 41 | int src_stride2, \ | |
| 42 | int h) \ | |
| 43 | { \ | |
| 44 | int i; \ | |
| 45 | \ | |
| 46 | for (i = 0; i < h; i++) { \ | |
| 47 | pixel4 a, b; \ | |
| 48 | a = AV_RN4P(&src1[i * src_stride1]); \ | |
| 49 | b = AV_RN4P(&src2[i * src_stride2]); \ | |
| 50 | OP(*((pixel4 *) &dst[i * dst_stride]), \ | |
| 51 | no_rnd_avg_pixel4(a, b)); \ | |
| 52 | a = AV_RN4P(&src1[i * src_stride1 + 4 * sizeof(pixel)]); \ | |
| 53 | b = AV_RN4P(&src2[i * src_stride2 + 4 * sizeof(pixel)]); \ | |
| 54 | OP(*((pixel4 *) &dst[i * dst_stride + 4 * sizeof(pixel)]), \ | |
| 55 | no_rnd_avg_pixel4(a, b)); \ | |
| 56 | } \ | |
| 57 | } \ | |
| 58 | \ | |
| 59 | static inline void FUNCC(OPNAME ## _no_rnd_pixels8_x2)(uint8_t *block, \ | |
| 60 | const uint8_t *pixels, \ | |
| 61 | ptrdiff_t line_size, \ | |
| 62 | int h) \ | |
| 63 | { \ | |
| 64 | FUNC(OPNAME ## _no_rnd_pixels8_l2)(block, pixels, pixels + sizeof(pixel), \ | |
| 65 | line_size, line_size, line_size, h); \ | |
| 66 | } \ | |
| 67 | \ | |
| 68 | static inline void FUNCC(OPNAME ## _pixels8_x2)(uint8_t *block, \ | |
| 69 | const uint8_t *pixels, \ | |
| 70 | ptrdiff_t line_size, \ | |
| 71 | int h) \ | |
| 72 | { \ | |
| 73 | FUNC(OPNAME ## _pixels8_l2)(block, pixels, pixels + sizeof(pixel), \ | |
| 74 | line_size, line_size, line_size, h); \ | |
| 75 | } \ | |
| 76 | \ | |
| 77 | static inline void FUNCC(OPNAME ## _no_rnd_pixels8_y2)(uint8_t *block, \ | |
| 78 | const uint8_t *pixels, \ | |
| 79 | ptrdiff_t line_size, \ | |
| 80 | int h) \ | |
| 81 | { \ | |
| 82 | FUNC(OPNAME ## _no_rnd_pixels8_l2)(block, pixels, pixels + line_size, \ | |
| 83 | line_size, line_size, line_size, h); \ | |
| 84 | } \ | |
| 85 | \ | |
| 86 | static inline void FUNCC(OPNAME ## _pixels8_y2)(uint8_t *block, \ | |
| 87 | const uint8_t *pixels, \ | |
| 88 | ptrdiff_t line_size, \ | |
| 89 | int h) \ | |
| 90 | { \ | |
| 91 | FUNC(OPNAME ## _pixels8_l2)(block, pixels, pixels + line_size, \ | |
| 92 | line_size, line_size, line_size, h); \ | |
| 93 | } \ | |
| 94 | \ | |
| 95 | static inline void FUNCC(OPNAME ## _pixels4_x2)(uint8_t *block, \ | |
| 96 | const uint8_t *pixels, \ | |
| 97 | ptrdiff_t line_size, \ | |
| 98 | int h) \ | |
| 99 | { \ | |
| 100 | FUNC(OPNAME ## _pixels4_l2)(block, pixels, pixels + sizeof(pixel), \ | |
| 101 | line_size, line_size, line_size, h); \ | |
| 102 | } \ | |
| 103 | \ | |
| 104 | static inline void FUNCC(OPNAME ## _pixels4_y2)(uint8_t *block, \ | |
| 105 | const uint8_t *pixels, \ | |
| 106 | ptrdiff_t line_size, \ | |
| 107 | int h) \ | |
| 108 | { \ | |
| 109 | FUNC(OPNAME ## _pixels4_l2)(block, pixels, pixels + line_size, \ | |
| 110 | line_size, line_size, line_size, h); \ | |
| 111 | } \ | |
| 112 | \ | |
| 113 | static inline void FUNCC(OPNAME ## _pixels2_x2)(uint8_t *block, \ | |
| 114 | const uint8_t *pixels, \ | |
| 115 | ptrdiff_t line_size, \ | |
| 116 | int h) \ | |
| 117 | { \ | |
| 118 | FUNC(OPNAME ## _pixels2_l2)(block, pixels, pixels + sizeof(pixel), \ | |
| 119 | line_size, line_size, line_size, h); \ | |
| 120 | } \ | |
| 121 | \ | |
| 122 | static inline void FUNCC(OPNAME ## _pixels2_y2)(uint8_t *block, \ | |
| 123 | const uint8_t *pixels, \ | |
| 124 | ptrdiff_t line_size, \ | |
| 125 | int h) \ | |
| 126 | { \ | |
| 127 | FUNC(OPNAME ## _pixels2_l2)(block, pixels, pixels + line_size, \ | |
| 128 | line_size, line_size, line_size, h); \ | |
| 129 | } \ | |
| 130 | \ | |
| 131 | static inline void FUNCC(OPNAME ## _pixels2_xy2)(uint8_t *_block, \ | |
| 132 | const uint8_t *_pixels, \ | |
| 133 | ptrdiff_t line_size, \ | |
| 134 | int h) \ | |
| 135 | { \ | |
| 2539cf1c DB |
136 | pixel *block = (pixel *) _block; \ |
| 137 | const pixel *pixels = (const pixel *) _pixels; \ | |
| cce791b1 DB |
138 | int i, a1, b1; \ |
| 139 | int a0 = pixels[0]; \ | |
| 140 | int b0 = pixels[1] + 2; \ | |
| 141 | \ | |
| 2539cf1c DB |
142 | a0 += b0; \ |
| 143 | b0 += pixels[2]; \ | |
| 2539cf1c DB |
144 | line_size /= sizeof(pixel); \ |
| 145 | pixels += line_size; \ | |
| 146 | for (i = 0; i < h; i += 2) { \ | |
| 147 | a1 = pixels[0]; \ | |
| 148 | b1 = pixels[1]; \ | |
| 149 | a1 += b1; \ | |
| 150 | b1 += pixels[2]; \ | |
| 151 | \ | |
| 152 | block[0] = (a1 + a0) >> 2; /* FIXME non put */ \ | |
| 153 | block[1] = (b1 + b0) >> 2; \ | |
| 154 | \ | |
| 155 | pixels += line_size; \ | |
| 156 | block += line_size; \ | |
| 157 | \ | |
| 158 | a0 = pixels[0]; \ | |
| 159 | b0 = pixels[1] + 2; \ | |
| 160 | a0 += b0; \ | |
| 161 | b0 += pixels[2]; \ | |
| 162 | \ | |
| 163 | block[0] = (a1 + a0) >> 2; \ | |
| 164 | block[1] = (b1 + b0) >> 2; \ | |
| 165 | pixels += line_size; \ | |
| 166 | block += line_size; \ | |
| 167 | } \ | |
| 168 | } \ | |
| 169 | \ | |
| 170 | static inline void FUNCC(OPNAME ## _pixels4_xy2)(uint8_t *block, \ | |
| 171 | const uint8_t *pixels, \ | |
| 172 | ptrdiff_t line_size, \ | |
| 173 | int h) \ | |
| 174 | { \ | |
| 175 | /* FIXME HIGH BIT DEPTH */ \ | |
| 176 | int i; \ | |
| 177 | const uint32_t a = AV_RN32(pixels); \ | |
| 178 | const uint32_t b = AV_RN32(pixels + 1); \ | |
| 179 | uint32_t l0 = (a & 0x03030303UL) + \ | |
| 180 | (b & 0x03030303UL) + \ | |
| 181 | 0x02020202UL; \ | |
| 182 | uint32_t h0 = ((a & 0xFCFCFCFCUL) >> 2) + \ | |
| 183 | ((b & 0xFCFCFCFCUL) >> 2); \ | |
| 184 | uint32_t l1, h1; \ | |
| 185 | \ | |
| 186 | pixels += line_size; \ | |
| 187 | for (i = 0; i < h; i += 2) { \ | |
| 188 | uint32_t a = AV_RN32(pixels); \ | |
| 189 | uint32_t b = AV_RN32(pixels + 1); \ | |
| 190 | l1 = (a & 0x03030303UL) + \ | |
| 191 | (b & 0x03030303UL); \ | |
| 192 | h1 = ((a & 0xFCFCFCFCUL) >> 2) + \ | |
| 193 | ((b & 0xFCFCFCFCUL) >> 2); \ | |
| 194 | OP(*((uint32_t *) block), h0 + h1 + \ | |
| 195 | (((l0 + l1) >> 2) & 0x0F0F0F0FUL)); \ | |
| 196 | pixels += line_size; \ | |
| 197 | block += line_size; \ | |
| 198 | a = AV_RN32(pixels); \ | |
| 199 | b = AV_RN32(pixels + 1); \ | |
| 200 | l0 = (a & 0x03030303UL) + \ | |
| 201 | (b & 0x03030303UL) + \ | |
| 202 | 0x02020202UL; \ | |
| 203 | h0 = ((a & 0xFCFCFCFCUL) >> 2) + \ | |
| 204 | ((b & 0xFCFCFCFCUL) >> 2); \ | |
| 205 | OP(*((uint32_t *) block), h0 + h1 + \ | |
| 206 | (((l0 + l1) >> 2) & 0x0F0F0F0FUL)); \ | |
| 207 | pixels += line_size; \ | |
| 208 | block += line_size; \ | |
| 209 | } \ | |
| 210 | } \ | |
| 211 | \ | |
| 212 | static inline void FUNCC(OPNAME ## _pixels8_xy2)(uint8_t *block, \ | |
| 213 | const uint8_t *pixels, \ | |
| 214 | ptrdiff_t line_size, \ | |
| 215 | int h) \ | |
| 216 | { \ | |
| 217 | /* FIXME HIGH BIT DEPTH */ \ | |
| 218 | int j; \ | |
| 219 | \ | |
| 220 | for (j = 0; j < 2; j++) { \ | |
| 221 | int i; \ | |
| 222 | const uint32_t a = AV_RN32(pixels); \ | |
| 223 | const uint32_t b = AV_RN32(pixels + 1); \ | |
| 224 | uint32_t l0 = (a & 0x03030303UL) + \ | |
| 225 | (b & 0x03030303UL) + \ | |
| 226 | 0x02020202UL; \ | |
| 227 | uint32_t h0 = ((a & 0xFCFCFCFCUL) >> 2) + \ | |
| 228 | ((b & 0xFCFCFCFCUL) >> 2); \ | |
| 229 | uint32_t l1, h1; \ | |
| 230 | \ | |
| 231 | pixels += line_size; \ | |
| 232 | for (i = 0; i < h; i += 2) { \ | |
| 233 | uint32_t a = AV_RN32(pixels); \ | |
| 234 | uint32_t b = AV_RN32(pixels + 1); \ | |
| 235 | l1 = (a & 0x03030303UL) + \ | |
| 236 | (b & 0x03030303UL); \ | |
| 237 | h1 = ((a & 0xFCFCFCFCUL) >> 2) + \ | |
| 238 | ((b & 0xFCFCFCFCUL) >> 2); \ | |
| 239 | OP(*((uint32_t *) block), h0 + h1 + \ | |
| 240 | (((l0 + l1) >> 2) & 0x0F0F0F0FUL)); \ | |
| 241 | pixels += line_size; \ | |
| 242 | block += line_size; \ | |
| 243 | a = AV_RN32(pixels); \ | |
| 244 | b = AV_RN32(pixels + 1); \ | |
| 245 | l0 = (a & 0x03030303UL) + \ | |
| 246 | (b & 0x03030303UL) + \ | |
| 247 | 0x02020202UL; \ | |
| 248 | h0 = ((a & 0xFCFCFCFCUL) >> 2) + \ | |
| 249 | ((b & 0xFCFCFCFCUL) >> 2); \ | |
| 250 | OP(*((uint32_t *) block), h0 + h1 + \ | |
| 251 | (((l0 + l1) >> 2) & 0x0F0F0F0FUL)); \ | |
| 252 | pixels += line_size; \ | |
| 253 | block += line_size; \ | |
| 254 | } \ | |
| 255 | pixels += 4 - line_size * (h + 1); \ | |
| 256 | block += 4 - line_size * h; \ | |
| 257 | } \ | |
| 258 | } \ | |
| 259 | \ | |
| 260 | static inline void FUNCC(OPNAME ## _no_rnd_pixels8_xy2)(uint8_t *block, \ | |
| 261 | const uint8_t *pixels, \ | |
| 262 | ptrdiff_t line_size, \ | |
| 263 | int h) \ | |
| 264 | { \ | |
| 265 | /* FIXME HIGH BIT DEPTH */ \ | |
| 266 | int j; \ | |
| 267 | \ | |
| 268 | for (j = 0; j < 2; j++) { \ | |
| 269 | int i; \ | |
| 270 | const uint32_t a = AV_RN32(pixels); \ | |
| 271 | const uint32_t b = AV_RN32(pixels + 1); \ | |
| 272 | uint32_t l0 = (a & 0x03030303UL) + \ | |
| 273 | (b & 0x03030303UL) + \ | |
| 274 | 0x01010101UL; \ | |
| 275 | uint32_t h0 = ((a & 0xFCFCFCFCUL) >> 2) + \ | |
| 276 | ((b & 0xFCFCFCFCUL) >> 2); \ | |
| 277 | uint32_t l1, h1; \ | |
| 278 | \ | |
| 279 | pixels += line_size; \ | |
| 280 | for (i = 0; i < h; i += 2) { \ | |
| 281 | uint32_t a = AV_RN32(pixels); \ | |
| 282 | uint32_t b = AV_RN32(pixels + 1); \ | |
| 283 | l1 = (a & 0x03030303UL) + \ | |
| 284 | (b & 0x03030303UL); \ | |
| 285 | h1 = ((a & 0xFCFCFCFCUL) >> 2) + \ | |
| 286 | ((b & 0xFCFCFCFCUL) >> 2); \ | |
| 287 | OP(*((uint32_t *) block), h0 + h1 + \ | |
| 288 | (((l0 + l1) >> 2) & 0x0F0F0F0FUL)); \ | |
| 289 | pixels += line_size; \ | |
| 290 | block += line_size; \ | |
| 291 | a = AV_RN32(pixels); \ | |
| 292 | b = AV_RN32(pixels + 1); \ | |
| 293 | l0 = (a & 0x03030303UL) + \ | |
| 294 | (b & 0x03030303UL) + \ | |
| 295 | 0x01010101UL; \ | |
| 296 | h0 = ((a & 0xFCFCFCFCUL) >> 2) + \ | |
| 297 | ((b & 0xFCFCFCFCUL) >> 2); \ | |
| 298 | OP(*((uint32_t *) block), h0 + h1 + \ | |
| 299 | (((l0 + l1) >> 2) & 0x0F0F0F0FUL)); \ | |
| 300 | pixels += line_size; \ | |
| 301 | block += line_size; \ | |
| 302 | } \ | |
| 303 | pixels += 4 - line_size * (h + 1); \ | |
| 304 | block += 4 - line_size * h; \ | |
| 305 | } \ | |
| 306 | } \ | |
| 307 | \ | |
| 308 | CALL_2X_PIXELS(FUNCC(OPNAME ## _pixels16_x2), \ | |
| 309 | FUNCC(OPNAME ## _pixels8_x2), \ | |
| 310 | 8 * sizeof(pixel)) \ | |
| 311 | CALL_2X_PIXELS(FUNCC(OPNAME ## _pixels16_y2), \ | |
| 312 | FUNCC(OPNAME ## _pixels8_y2), \ | |
| 313 | 8 * sizeof(pixel)) \ | |
| 314 | CALL_2X_PIXELS(FUNCC(OPNAME ## _pixels16_xy2), \ | |
| 315 | FUNCC(OPNAME ## _pixels8_xy2), \ | |
| 316 | 8 * sizeof(pixel)) \ | |
| 317 | av_unused CALL_2X_PIXELS(FUNCC(OPNAME ## _no_rnd_pixels16), \ | |
| 318 | FUNCC(OPNAME ## _pixels8), \ | |
| 319 | 8 * sizeof(pixel)) \ | |
| 320 | CALL_2X_PIXELS(FUNCC(OPNAME ## _no_rnd_pixels16_x2), \ | |
| 321 | FUNCC(OPNAME ## _no_rnd_pixels8_x2), \ | |
| 322 | 8 * sizeof(pixel)) \ | |
| 323 | CALL_2X_PIXELS(FUNCC(OPNAME ## _no_rnd_pixels16_y2), \ | |
| 324 | FUNCC(OPNAME ## _no_rnd_pixels8_y2), \ | |
| 325 | 8 * sizeof(pixel)) \ | |
| 326 | CALL_2X_PIXELS(FUNCC(OPNAME ## _no_rnd_pixels16_xy2), \ | |
| 327 | FUNCC(OPNAME ## _no_rnd_pixels8_xy2), \ | |
| 328 | 8 * sizeof(pixel)) \ | |
| 68d8238c RB |
329 | |
| 330 | #define op_avg(a, b) a = rnd_avg_pixel4(a, b) | |
| 331 | #define op_put(a, b) a = b | |
| 332 | #if BIT_DEPTH == 8 | |
| 333 | #define put_no_rnd_pixels8_8_c put_pixels8_8_c | |
| 334 | PIXOP2(avg, op_avg) | |
| 335 | PIXOP2(put, op_put) | |
| 336 | #endif | |
| 337 | #undef op_avg | |
| 338 | #undef op_put |