| 1 | /* |
| 2 | * Half-pel DSP functions. |
| 3 | * Copyright (c) 2000, 2001 Fabrice Bellard |
| 4 | * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at> |
| 5 | * |
| 6 | * gmc & q-pel & 32/64 bit based MC by Michael Niedermayer <michaelni@gmx.at> |
| 7 | * |
| 8 | * This file is part of FFmpeg. |
| 9 | * |
| 10 | * FFmpeg is free software; you can redistribute it and/or |
| 11 | * modify it under the terms of the GNU Lesser General Public |
| 12 | * License as published by the Free Software Foundation; either |
| 13 | * version 2.1 of the License, or (at your option) any later version. |
| 14 | * |
| 15 | * FFmpeg is distributed in the hope that it will be useful, |
| 16 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 17 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 18 | * Lesser General Public License for more details. |
| 19 | * |
| 20 | * You should have received a copy of the GNU Lesser General Public |
| 21 | * License along with FFmpeg; if not, write to the Free Software |
| 22 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
| 23 | */ |
| 24 | |
| 25 | /** |
| 26 | * @file |
| 27 | * Half-pel DSP functions. |
| 28 | */ |
| 29 | |
| 30 | #include "libavutil/attributes.h" |
| 31 | #include "libavutil/intreadwrite.h" |
| 32 | #include "hpeldsp.h" |
| 33 | |
| 34 | #define BIT_DEPTH 8 |
| 35 | #include "hpel_template.c" |
| 36 | #include "pel_template.c" |
| 37 | |
| 38 | #define PIXOP2(OPNAME, OP) \ |
| 39 | static inline void OPNAME ## _no_rnd_pixels8_l2_8(uint8_t *dst, \ |
| 40 | const uint8_t *src1, \ |
| 41 | const uint8_t *src2, \ |
| 42 | int dst_stride, \ |
| 43 | int src_stride1, \ |
| 44 | int src_stride2, \ |
| 45 | int h) \ |
| 46 | { \ |
| 47 | int i; \ |
| 48 | \ |
| 49 | for (i = 0; i < h; i++) { \ |
| 50 | uint32_t a, b; \ |
| 51 | a = AV_RN32(&src1[i * src_stride1]); \ |
| 52 | b = AV_RN32(&src2[i * src_stride2]); \ |
| 53 | OP(*((uint32_t *) &dst[i * dst_stride]), \ |
| 54 | no_rnd_avg32(a, b)); \ |
| 55 | a = AV_RN32(&src1[i * src_stride1 + 4]); \ |
| 56 | b = AV_RN32(&src2[i * src_stride2 + 4]); \ |
| 57 | OP(*((uint32_t *) &dst[i * dst_stride + 4]), \ |
| 58 | no_rnd_avg32(a, b)); \ |
| 59 | } \ |
| 60 | } \ |
| 61 | \ |
| 62 | static inline void OPNAME ## _no_rnd_pixels8_x2_8_c(uint8_t *block, \ |
| 63 | const uint8_t *pixels, \ |
| 64 | ptrdiff_t line_size, \ |
| 65 | int h) \ |
| 66 | { \ |
| 67 | OPNAME ## _no_rnd_pixels8_l2_8(block, pixels, pixels + 1, \ |
| 68 | line_size, line_size, line_size, h); \ |
| 69 | } \ |
| 70 | \ |
| 71 | static inline void OPNAME ## _pixels8_x2_8_c(uint8_t *block, \ |
| 72 | const uint8_t *pixels, \ |
| 73 | ptrdiff_t line_size, \ |
| 74 | int h) \ |
| 75 | { \ |
| 76 | OPNAME ## _pixels8_l2_8(block, pixels, pixels + 1, \ |
| 77 | line_size, line_size, line_size, h); \ |
| 78 | } \ |
| 79 | \ |
| 80 | static inline void OPNAME ## _no_rnd_pixels8_y2_8_c(uint8_t *block, \ |
| 81 | const uint8_t *pixels, \ |
| 82 | ptrdiff_t line_size, \ |
| 83 | int h) \ |
| 84 | { \ |
| 85 | OPNAME ## _no_rnd_pixels8_l2_8(block, pixels, pixels + line_size, \ |
| 86 | line_size, line_size, line_size, h); \ |
| 87 | } \ |
| 88 | \ |
| 89 | static inline void OPNAME ## _pixels8_y2_8_c(uint8_t *block, \ |
| 90 | const uint8_t *pixels, \ |
| 91 | ptrdiff_t line_size, \ |
| 92 | int h) \ |
| 93 | { \ |
| 94 | OPNAME ## _pixels8_l2_8(block, pixels, pixels + line_size, \ |
| 95 | line_size, line_size, line_size, h); \ |
| 96 | } \ |
| 97 | \ |
| 98 | static inline void OPNAME ## _pixels4_x2_8_c(uint8_t *block, \ |
| 99 | const uint8_t *pixels, \ |
| 100 | ptrdiff_t line_size, \ |
| 101 | int h) \ |
| 102 | { \ |
| 103 | OPNAME ## _pixels4_l2_8(block, pixels, pixels + 1, \ |
| 104 | line_size, line_size, line_size, h); \ |
| 105 | } \ |
| 106 | \ |
| 107 | static inline void OPNAME ## _pixels4_y2_8_c(uint8_t *block, \ |
| 108 | const uint8_t *pixels, \ |
| 109 | ptrdiff_t line_size, \ |
| 110 | int h) \ |
| 111 | { \ |
| 112 | OPNAME ## _pixels4_l2_8(block, pixels, pixels + line_size, \ |
| 113 | line_size, line_size, line_size, h); \ |
| 114 | } \ |
| 115 | \ |
| 116 | static inline void OPNAME ## _pixels2_x2_8_c(uint8_t *block, \ |
| 117 | const uint8_t *pixels, \ |
| 118 | ptrdiff_t line_size, \ |
| 119 | int h) \ |
| 120 | { \ |
| 121 | OPNAME ## _pixels2_l2_8(block, pixels, pixels + 1, \ |
| 122 | line_size, line_size, line_size, h); \ |
| 123 | } \ |
| 124 | \ |
| 125 | static inline void OPNAME ## _pixels2_y2_8_c(uint8_t *block, \ |
| 126 | const uint8_t *pixels, \ |
| 127 | ptrdiff_t line_size, \ |
| 128 | int h) \ |
| 129 | { \ |
| 130 | OPNAME ## _pixels2_l2_8(block, pixels, pixels + line_size, \ |
| 131 | line_size, line_size, line_size, h); \ |
| 132 | } \ |
| 133 | \ |
| 134 | static inline void OPNAME ## _pixels2_xy2_8_c(uint8_t *block, \ |
| 135 | const uint8_t *pixels, \ |
| 136 | ptrdiff_t line_size, \ |
| 137 | int h) \ |
| 138 | { \ |
| 139 | int i, a1, b1; \ |
| 140 | int a0 = pixels[0]; \ |
| 141 | int b0 = pixels[1] + 2; \ |
| 142 | \ |
| 143 | a0 += b0; \ |
| 144 | b0 += pixels[2]; \ |
| 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 OPNAME ## _pixels4_xy2_8_c(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 OPNAME ## _pixels8_xy2_8_c(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 OPNAME ## _no_rnd_pixels8_xy2_8_c(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(OPNAME ## _pixels16_x2_8_c, \ |
| 309 | OPNAME ## _pixels8_x2_8_c, \ |
| 310 | 8) \ |
| 311 | CALL_2X_PIXELS(OPNAME ## _pixels16_y2_8_c, \ |
| 312 | OPNAME ## _pixels8_y2_8_c, \ |
| 313 | 8) \ |
| 314 | CALL_2X_PIXELS(OPNAME ## _pixels16_xy2_8_c, \ |
| 315 | OPNAME ## _pixels8_xy2_8_c, \ |
| 316 | 8) \ |
| 317 | CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_8_c, \ |
| 318 | OPNAME ## _pixels8_8_c, \ |
| 319 | 8) \ |
| 320 | CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_x2_8_c, \ |
| 321 | OPNAME ## _no_rnd_pixels8_x2_8_c, \ |
| 322 | 8) \ |
| 323 | CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_y2_8_c, \ |
| 324 | OPNAME ## _no_rnd_pixels8_y2_8_c, \ |
| 325 | 8) \ |
| 326 | CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_xy2_8_c, \ |
| 327 | OPNAME ## _no_rnd_pixels8_xy2_8_c, \ |
| 328 | 8) \ |
| 329 | |
| 330 | #define op_avg(a, b) a = rnd_avg32(a, b) |
| 331 | #define op_put(a, b) a = b |
| 332 | #define put_no_rnd_pixels8_8_c put_pixels8_8_c |
| 333 | PIXOP2(avg, op_avg) |
| 334 | PIXOP2(put, op_put) |
| 335 | #undef op_avg |
| 336 | #undef op_put |
| 337 | |
| 338 | av_cold void ff_hpeldsp_init(HpelDSPContext *c, int flags) |
| 339 | { |
| 340 | #define hpel_funcs(prefix, idx, num) \ |
| 341 | c->prefix ## _pixels_tab idx [0] = prefix ## _pixels ## num ## _8_c; \ |
| 342 | c->prefix ## _pixels_tab idx [1] = prefix ## _pixels ## num ## _x2_8_c; \ |
| 343 | c->prefix ## _pixels_tab idx [2] = prefix ## _pixels ## num ## _y2_8_c; \ |
| 344 | c->prefix ## _pixels_tab idx [3] = prefix ## _pixels ## num ## _xy2_8_c |
| 345 | |
| 346 | hpel_funcs(put, [0], 16); |
| 347 | hpel_funcs(put, [1], 8); |
| 348 | hpel_funcs(put, [2], 4); |
| 349 | hpel_funcs(put, [3], 2); |
| 350 | hpel_funcs(put_no_rnd, [0], 16); |
| 351 | hpel_funcs(put_no_rnd, [1], 8); |
| 352 | hpel_funcs(avg, [0], 16); |
| 353 | hpel_funcs(avg, [1], 8); |
| 354 | hpel_funcs(avg, [2], 4); |
| 355 | hpel_funcs(avg, [3], 2); |
| 356 | hpel_funcs(avg_no_rnd,, 16); |
| 357 | |
| 358 | if (ARCH_AARCH64) |
| 359 | ff_hpeldsp_init_aarch64(c, flags); |
| 360 | if (ARCH_ALPHA) |
| 361 | ff_hpeldsp_init_alpha(c, flags); |
| 362 | if (ARCH_ARM) |
| 363 | ff_hpeldsp_init_arm(c, flags); |
| 364 | if (ARCH_PPC) |
| 365 | ff_hpeldsp_init_ppc(c, flags); |
| 366 | if (ARCH_X86) |
| 367 | ff_hpeldsp_init_x86(c, flags); |
| 368 | } |