Details | Last modification | View Log | RSS feed
| Rev | Author | Line No. | Line |
|---|---|---|---|
| 6148 | serge | 1 | /* |
| 2 | * Copyright (c) 2000,2001 Fabrice Bellard |
||
| 3 | * Copyright (c) 2002-2004 Michael Niedermayer |
||
| 4 | * |
||
| 5 | * 4MV & hq & B-frame encoding stuff by Michael Niedermayer |
||
| 6 | * |
||
| 7 | * This file is part of FFmpeg. |
||
| 8 | * |
||
| 9 | * FFmpeg is free software; you can redistribute it and/or |
||
| 10 | * modify it under the terms of the GNU Lesser General Public |
||
| 11 | * License as published by the Free Software Foundation; either |
||
| 12 | * version 2.1 of the License, or (at your option) any later version. |
||
| 13 | * |
||
| 14 | * FFmpeg is distributed in the hope that it will be useful, |
||
| 15 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
||
| 16 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
||
| 17 | * Lesser General Public License for more details. |
||
| 18 | * |
||
| 19 | * You should have received a copy of the GNU Lesser General Public |
||
| 20 | * License along with FFmpeg; if not, write to the Free Software |
||
| 21 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
||
| 22 | */ |
||
| 23 | |||
| 24 | #include |
||
| 25 | |||
| 26 | #include "libavutil/avassert.h" |
||
| 27 | #include "libavutil/internal.h" |
||
| 28 | #include "avcodec.h" |
||
| 29 | #include "dsputil.h" |
||
| 30 | #include "h261.h" |
||
| 31 | #include "mpegvideo.h" |
||
| 32 | #include "mjpegenc.h" |
||
| 33 | #include "msmpeg4.h" |
||
| 34 | #include |
||
| 35 | |||
| 36 | static void gmc1_motion(MpegEncContext *s, |
||
| 37 | uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr, |
||
| 38 | uint8_t **ref_picture) |
||
| 39 | { |
||
| 40 | uint8_t *ptr; |
||
| 41 | int src_x, src_y, motion_x, motion_y; |
||
| 42 | ptrdiff_t offset, linesize, uvlinesize; |
||
| 43 | int emu = 0; |
||
| 44 | |||
| 45 | motion_x = s->sprite_offset[0][0]; |
||
| 46 | motion_y = s->sprite_offset[0][1]; |
||
| 47 | src_x = s->mb_x * 16 + (motion_x >> (s->sprite_warping_accuracy + 1)); |
||
| 48 | src_y = s->mb_y * 16 + (motion_y >> (s->sprite_warping_accuracy + 1)); |
||
| 49 | motion_x <<= (3 - s->sprite_warping_accuracy); |
||
| 50 | motion_y <<= (3 - s->sprite_warping_accuracy); |
||
| 51 | src_x = av_clip(src_x, -16, s->width); |
||
| 52 | if (src_x == s->width) |
||
| 53 | motion_x = 0; |
||
| 54 | src_y = av_clip(src_y, -16, s->height); |
||
| 55 | if (src_y == s->height) |
||
| 56 | motion_y = 0; |
||
| 57 | |||
| 58 | linesize = s->linesize; |
||
| 59 | uvlinesize = s->uvlinesize; |
||
| 60 | |||
| 61 | ptr = ref_picture[0] + src_y * linesize + src_x; |
||
| 62 | |||
| 63 | if ((unsigned)src_x >= FFMAX(s->h_edge_pos - 17, 0) || |
||
| 64 | (unsigned)src_y >= FFMAX(s->v_edge_pos - 17, 0)) { |
||
| 65 | s->vdsp.emulated_edge_mc(s->edge_emu_buffer, linesize, ptr, |
||
| 66 | linesize, |
||
| 67 | 17, 17, |
||
| 68 | src_x, src_y, |
||
| 69 | s->h_edge_pos, s->v_edge_pos); |
||
| 70 | ptr = s->edge_emu_buffer; |
||
| 71 | } |
||
| 72 | |||
| 73 | if ((motion_x | motion_y) & 7) { |
||
| 74 | s->dsp.gmc1(dest_y, ptr, linesize, 16, |
||
| 75 | motion_x & 15, motion_y & 15, 128 - s->no_rounding); |
||
| 76 | s->dsp.gmc1(dest_y + 8, ptr + 8, linesize, 16, |
||
| 77 | motion_x & 15, motion_y & 15, 128 - s->no_rounding); |
||
| 78 | } else { |
||
| 79 | int dxy; |
||
| 80 | |||
| 81 | dxy = ((motion_x >> 3) & 1) | ((motion_y >> 2) & 2); |
||
| 82 | if (s->no_rounding) { |
||
| 83 | s->hdsp.put_no_rnd_pixels_tab[0][dxy](dest_y, ptr, linesize, 16); |
||
| 84 | } else { |
||
| 85 | s->hdsp.put_pixels_tab[0][dxy](dest_y, ptr, linesize, 16); |
||
| 86 | } |
||
| 87 | } |
||
| 88 | |||
| 89 | if (CONFIG_GRAY && s->flags & CODEC_FLAG_GRAY) |
||
| 90 | return; |
||
| 91 | |||
| 92 | motion_x = s->sprite_offset[1][0]; |
||
| 93 | motion_y = s->sprite_offset[1][1]; |
||
| 94 | src_x = s->mb_x * 8 + (motion_x >> (s->sprite_warping_accuracy + 1)); |
||
| 95 | src_y = s->mb_y * 8 + (motion_y >> (s->sprite_warping_accuracy + 1)); |
||
| 96 | motion_x <<= (3 - s->sprite_warping_accuracy); |
||
| 97 | motion_y <<= (3 - s->sprite_warping_accuracy); |
||
| 98 | src_x = av_clip(src_x, -8, s->width >> 1); |
||
| 99 | if (src_x == s->width >> 1) |
||
| 100 | motion_x = 0; |
||
| 101 | src_y = av_clip(src_y, -8, s->height >> 1); |
||
| 102 | if (src_y == s->height >> 1) |
||
| 103 | motion_y = 0; |
||
| 104 | |||
| 105 | offset = (src_y * uvlinesize) + src_x; |
||
| 106 | ptr = ref_picture[1] + offset; |
||
| 107 | if ((unsigned)src_x >= FFMAX((s->h_edge_pos >> 1) - 9, 0) || |
||
| 108 | (unsigned)src_y >= FFMAX((s->v_edge_pos >> 1) - 9, 0)) { |
||
| 109 | s->vdsp.emulated_edge_mc(s->edge_emu_buffer, uvlinesize, |
||
| 110 | ptr, uvlinesize, |
||
| 111 | 9, 9, |
||
| 112 | src_x, src_y, |
||
| 113 | s->h_edge_pos >> 1, s->v_edge_pos >> 1); |
||
| 114 | ptr = s->edge_emu_buffer; |
||
| 115 | emu = 1; |
||
| 116 | } |
||
| 117 | s->dsp.gmc1(dest_cb, ptr, uvlinesize, 8, |
||
| 118 | motion_x & 15, motion_y & 15, 128 - s->no_rounding); |
||
| 119 | |||
| 120 | ptr = ref_picture[2] + offset; |
||
| 121 | if (emu) { |
||
| 122 | s->vdsp.emulated_edge_mc(s->edge_emu_buffer, uvlinesize, |
||
| 123 | ptr, uvlinesize, |
||
| 124 | 9, 9, |
||
| 125 | src_x, src_y, |
||
| 126 | s->h_edge_pos >> 1, s->v_edge_pos >> 1); |
||
| 127 | ptr = s->edge_emu_buffer; |
||
| 128 | } |
||
| 129 | s->dsp.gmc1(dest_cr, ptr, uvlinesize, 8, |
||
| 130 | motion_x & 15, motion_y & 15, 128 - s->no_rounding); |
||
| 131 | } |
||
| 132 | |||
| 133 | static void gmc_motion(MpegEncContext *s, |
||
| 134 | uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr, |
||
| 135 | uint8_t **ref_picture) |
||
| 136 | { |
||
| 137 | uint8_t *ptr; |
||
| 138 | int linesize, uvlinesize; |
||
| 139 | const int a = s->sprite_warping_accuracy; |
||
| 140 | int ox, oy; |
||
| 141 | |||
| 142 | linesize = s->linesize; |
||
| 143 | uvlinesize = s->uvlinesize; |
||
| 144 | |||
| 145 | ptr = ref_picture[0]; |
||
| 146 | |||
| 147 | ox = s->sprite_offset[0][0] + s->sprite_delta[0][0] * s->mb_x * 16 + |
||
| 148 | s->sprite_delta[0][1] * s->mb_y * 16; |
||
| 149 | oy = s->sprite_offset[0][1] + s->sprite_delta[1][0] * s->mb_x * 16 + |
||
| 150 | s->sprite_delta[1][1] * s->mb_y * 16; |
||
| 151 | |||
| 152 | s->dsp.gmc(dest_y, ptr, linesize, 16, |
||
| 153 | ox, oy, |
||
| 154 | s->sprite_delta[0][0], s->sprite_delta[0][1], |
||
| 155 | s->sprite_delta[1][0], s->sprite_delta[1][1], |
||
| 156 | a + 1, (1 << (2 * a + 1)) - s->no_rounding, |
||
| 157 | s->h_edge_pos, s->v_edge_pos); |
||
| 158 | s->dsp.gmc(dest_y + 8, ptr, linesize, 16, |
||
| 159 | ox + s->sprite_delta[0][0] * 8, |
||
| 160 | oy + s->sprite_delta[1][0] * 8, |
||
| 161 | s->sprite_delta[0][0], s->sprite_delta[0][1], |
||
| 162 | s->sprite_delta[1][0], s->sprite_delta[1][1], |
||
| 163 | a + 1, (1 << (2 * a + 1)) - s->no_rounding, |
||
| 164 | s->h_edge_pos, s->v_edge_pos); |
||
| 165 | |||
| 166 | if (CONFIG_GRAY && s->flags & CODEC_FLAG_GRAY) |
||
| 167 | return; |
||
| 168 | |||
| 169 | ox = s->sprite_offset[1][0] + s->sprite_delta[0][0] * s->mb_x * 8 + |
||
| 170 | s->sprite_delta[0][1] * s->mb_y * 8; |
||
| 171 | oy = s->sprite_offset[1][1] + s->sprite_delta[1][0] * s->mb_x * 8 + |
||
| 172 | s->sprite_delta[1][1] * s->mb_y * 8; |
||
| 173 | |||
| 174 | ptr = ref_picture[1]; |
||
| 175 | s->dsp.gmc(dest_cb, ptr, uvlinesize, 8, |
||
| 176 | ox, oy, |
||
| 177 | s->sprite_delta[0][0], s->sprite_delta[0][1], |
||
| 178 | s->sprite_delta[1][0], s->sprite_delta[1][1], |
||
| 179 | a + 1, (1 << (2 * a + 1)) - s->no_rounding, |
||
| 180 | s->h_edge_pos >> 1, s->v_edge_pos >> 1); |
||
| 181 | |||
| 182 | ptr = ref_picture[2]; |
||
| 183 | s->dsp.gmc(dest_cr, ptr, uvlinesize, 8, |
||
| 184 | ox, oy, |
||
| 185 | s->sprite_delta[0][0], s->sprite_delta[0][1], |
||
| 186 | s->sprite_delta[1][0], s->sprite_delta[1][1], |
||
| 187 | a + 1, (1 << (2 * a + 1)) - s->no_rounding, |
||
| 188 | s->h_edge_pos >> 1, s->v_edge_pos >> 1); |
||
| 189 | } |
||
| 190 | |||
| 191 | static inline int hpel_motion(MpegEncContext *s, |
||
| 192 | uint8_t *dest, uint8_t *src, |
||
| 193 | int src_x, int src_y, |
||
| 194 | op_pixels_func *pix_op, |
||
| 195 | int motion_x, int motion_y) |
||
| 196 | { |
||
| 197 | int dxy = 0; |
||
| 198 | int emu = 0; |
||
| 199 | |||
| 200 | src_x += motion_x >> 1; |
||
| 201 | src_y += motion_y >> 1; |
||
| 202 | |||
| 203 | /* WARNING: do no forget half pels */ |
||
| 204 | src_x = av_clip(src_x, -16, s->width); // FIXME unneeded for emu? |
||
| 205 | if (src_x != s->width) |
||
| 206 | dxy |= motion_x & 1; |
||
| 207 | src_y = av_clip(src_y, -16, s->height); |
||
| 208 | if (src_y != s->height) |
||
| 209 | dxy |= (motion_y & 1) << 1; |
||
| 210 | src += src_y * s->linesize + src_x; |
||
| 211 | |||
| 212 | if (s->unrestricted_mv && (s->flags & CODEC_FLAG_EMU_EDGE)) { |
||
| 213 | if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 1) - 8, 0) || |
||
| 214 | (unsigned)src_y > FFMAX(s->v_edge_pos - (motion_y & 1) - 8, 0)) { |
||
| 215 | s->vdsp.emulated_edge_mc(s->edge_emu_buffer, s->linesize, src, |
||
| 216 | s->linesize, |
||
| 217 | 9, 9, |
||
| 218 | src_x, src_y, |
||
| 219 | s->h_edge_pos, s->v_edge_pos); |
||
| 220 | src = s->edge_emu_buffer; |
||
| 221 | emu = 1; |
||
| 222 | } |
||
| 223 | } |
||
| 224 | pix_op[dxy](dest, src, s->linesize, 8); |
||
| 225 | return emu; |
||
| 226 | } |
||
| 227 | |||
| 228 | static av_always_inline |
||
| 229 | void mpeg_motion_internal(MpegEncContext *s, |
||
| 230 | uint8_t *dest_y, |
||
| 231 | uint8_t *dest_cb, |
||
| 232 | uint8_t *dest_cr, |
||
| 233 | int field_based, |
||
| 234 | int bottom_field, |
||
| 235 | int field_select, |
||
| 236 | uint8_t **ref_picture, |
||
| 237 | op_pixels_func (*pix_op)[4], |
||
| 238 | int motion_x, |
||
| 239 | int motion_y, |
||
| 240 | int h, |
||
| 241 | int is_mpeg12, |
||
| 242 | int mb_y) |
||
| 243 | { |
||
| 244 | uint8_t *ptr_y, *ptr_cb, *ptr_cr; |
||
| 245 | int dxy, uvdxy, mx, my, src_x, src_y, |
||
| 246 | uvsrc_x, uvsrc_y, v_edge_pos; |
||
| 247 | ptrdiff_t uvlinesize, linesize; |
||
| 248 | |||
| 249 | #if 0 |
||
| 250 | if (s->quarter_sample) { |
||
| 251 | motion_x >>= 1; |
||
| 252 | motion_y >>= 1; |
||
| 253 | } |
||
| 254 | #endif |
||
| 255 | |||
| 256 | v_edge_pos = s->v_edge_pos >> field_based; |
||
| 257 | linesize = s->current_picture.f.linesize[0] << field_based; |
||
| 258 | uvlinesize = s->current_picture.f.linesize[1] << field_based; |
||
| 259 | |||
| 260 | dxy = ((motion_y & 1) << 1) | (motion_x & 1); |
||
| 261 | src_x = s->mb_x * 16 + (motion_x >> 1); |
||
| 262 | src_y = (mb_y << (4 - field_based)) + (motion_y >> 1); |
||
| 263 | |||
| 264 | if (!is_mpeg12 && s->out_format == FMT_H263) { |
||
| 265 | if ((s->workaround_bugs & FF_BUG_HPEL_CHROMA) && field_based) { |
||
| 266 | mx = (motion_x >> 1) | (motion_x & 1); |
||
| 267 | my = motion_y >> 1; |
||
| 268 | uvdxy = ((my & 1) << 1) | (mx & 1); |
||
| 269 | uvsrc_x = s->mb_x * 8 + (mx >> 1); |
||
| 270 | uvsrc_y = (mb_y << (3 - field_based)) + (my >> 1); |
||
| 271 | } else { |
||
| 272 | uvdxy = dxy | (motion_y & 2) | ((motion_x & 2) >> 1); |
||
| 273 | uvsrc_x = src_x >> 1; |
||
| 274 | uvsrc_y = src_y >> 1; |
||
| 275 | } |
||
| 276 | // Even chroma mv's are full pel in H261 |
||
| 277 | } else if (!is_mpeg12 && s->out_format == FMT_H261) { |
||
| 278 | mx = motion_x / 4; |
||
| 279 | my = motion_y / 4; |
||
| 280 | uvdxy = 0; |
||
| 281 | uvsrc_x = s->mb_x * 8 + mx; |
||
| 282 | uvsrc_y = mb_y * 8 + my; |
||
| 283 | } else { |
||
| 284 | if (s->chroma_y_shift) { |
||
| 285 | mx = motion_x / 2; |
||
| 286 | my = motion_y / 2; |
||
| 287 | uvdxy = ((my & 1) << 1) | (mx & 1); |
||
| 288 | uvsrc_x = s->mb_x * 8 + (mx >> 1); |
||
| 289 | uvsrc_y = (mb_y << (3 - field_based)) + (my >> 1); |
||
| 290 | } else { |
||
| 291 | if (s->chroma_x_shift) { |
||
| 292 | // Chroma422 |
||
| 293 | mx = motion_x / 2; |
||
| 294 | uvdxy = ((motion_y & 1) << 1) | (mx & 1); |
||
| 295 | uvsrc_x = s->mb_x * 8 + (mx >> 1); |
||
| 296 | uvsrc_y = src_y; |
||
| 297 | } else { |
||
| 298 | // Chroma444 |
||
| 299 | uvdxy = dxy; |
||
| 300 | uvsrc_x = src_x; |
||
| 301 | uvsrc_y = src_y; |
||
| 302 | } |
||
| 303 | } |
||
| 304 | } |
||
| 305 | |||
| 306 | ptr_y = ref_picture[0] + src_y * linesize + src_x; |
||
| 307 | ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x; |
||
| 308 | ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x; |
||
| 309 | |||
| 310 | if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 1) - 16, 0) || |
||
| 311 | (unsigned)src_y > FFMAX( v_edge_pos - (motion_y & 1) - h , 0)) { |
||
| 312 | if (is_mpeg12 || |
||
| 313 | s->codec_id == AV_CODEC_ID_MPEG2VIDEO || |
||
| 314 | s->codec_id == AV_CODEC_ID_MPEG1VIDEO) { |
||
| 315 | av_log(s->avctx, AV_LOG_DEBUG, |
||
| 316 | "MPEG motion vector out of boundary (%d %d)\n", src_x, |
||
| 317 | src_y); |
||
| 318 | return; |
||
| 319 | } |
||
| 320 | s->vdsp.emulated_edge_mc(s->edge_emu_buffer, s->linesize, ptr_y, |
||
| 321 | s->linesize, |
||
| 322 | 17, 17 + field_based, |
||
| 323 | src_x, src_y << field_based, |
||
| 324 | s->h_edge_pos, s->v_edge_pos); |
||
| 325 | ptr_y = s->edge_emu_buffer; |
||
| 326 | if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) { |
||
| 327 | uint8_t *uvbuf = s->edge_emu_buffer + 18 * s->linesize; |
||
| 328 | s->vdsp.emulated_edge_mc(uvbuf, s->uvlinesize, ptr_cb, |
||
| 329 | s->uvlinesize, |
||
| 330 | 9, 9 + field_based, |
||
| 331 | uvsrc_x, uvsrc_y << field_based, |
||
| 332 | s->h_edge_pos >> 1, s->v_edge_pos >> 1); |
||
| 333 | s->vdsp.emulated_edge_mc(uvbuf + 16, s->uvlinesize, ptr_cr, |
||
| 334 | s->uvlinesize, |
||
| 335 | 9, 9 + field_based, |
||
| 336 | uvsrc_x, uvsrc_y << field_based, |
||
| 337 | s->h_edge_pos >> 1, s->v_edge_pos >> 1); |
||
| 338 | ptr_cb = uvbuf; |
||
| 339 | ptr_cr = uvbuf + 16; |
||
| 340 | } |
||
| 341 | } |
||
| 342 | |||
| 343 | /* FIXME use this for field pix too instead of the obnoxious hack which |
||
| 344 | * changes picture.data */ |
||
| 345 | if (bottom_field) { |
||
| 346 | dest_y += s->linesize; |
||
| 347 | dest_cb += s->uvlinesize; |
||
| 348 | dest_cr += s->uvlinesize; |
||
| 349 | } |
||
| 350 | |||
| 351 | if (field_select) { |
||
| 352 | ptr_y += s->linesize; |
||
| 353 | ptr_cb += s->uvlinesize; |
||
| 354 | ptr_cr += s->uvlinesize; |
||
| 355 | } |
||
| 356 | |||
| 357 | pix_op[0][dxy](dest_y, ptr_y, linesize, h); |
||
| 358 | |||
| 359 | if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) { |
||
| 360 | pix_op[s->chroma_x_shift][uvdxy] |
||
| 361 | (dest_cb, ptr_cb, uvlinesize, h >> s->chroma_y_shift); |
||
| 362 | pix_op[s->chroma_x_shift][uvdxy] |
||
| 363 | (dest_cr, ptr_cr, uvlinesize, h >> s->chroma_y_shift); |
||
| 364 | } |
||
| 365 | if (!is_mpeg12 && (CONFIG_H261_ENCODER || CONFIG_H261_DECODER) && |
||
| 366 | s->out_format == FMT_H261) { |
||
| 367 | ff_h261_loop_filter(s); |
||
| 368 | } |
||
| 369 | } |
||
| 370 | /* apply one mpeg motion vector to the three components */ |
||
| 371 | static void mpeg_motion(MpegEncContext *s, |
||
| 372 | uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr, |
||
| 373 | int field_select, uint8_t **ref_picture, |
||
| 374 | op_pixels_func (*pix_op)[4], |
||
| 375 | int motion_x, int motion_y, int h, int mb_y) |
||
| 376 | { |
||
| 377 | #if !CONFIG_SMALL |
||
| 378 | if (s->out_format == FMT_MPEG1) |
||
| 379 | mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 0, 0, |
||
| 380 | field_select, ref_picture, pix_op, |
||
| 381 | motion_x, motion_y, h, 1, mb_y); |
||
| 382 | else |
||
| 383 | #endif |
||
| 384 | mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 0, 0, |
||
| 385 | field_select, ref_picture, pix_op, |
||
| 386 | motion_x, motion_y, h, 0, mb_y); |
||
| 387 | } |
||
| 388 | |||
| 389 | static void mpeg_motion_field(MpegEncContext *s, uint8_t *dest_y, |
||
| 390 | uint8_t *dest_cb, uint8_t *dest_cr, |
||
| 391 | int bottom_field, int field_select, |
||
| 392 | uint8_t **ref_picture, |
||
| 393 | op_pixels_func (*pix_op)[4], |
||
| 394 | int motion_x, int motion_y, int h, int mb_y) |
||
| 395 | { |
||
| 396 | #if !CONFIG_SMALL |
||
| 397 | if (s->out_format == FMT_MPEG1) |
||
| 398 | mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 1, |
||
| 399 | bottom_field, field_select, ref_picture, pix_op, |
||
| 400 | motion_x, motion_y, h, 1, mb_y); |
||
| 401 | else |
||
| 402 | #endif |
||
| 403 | mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 1, |
||
| 404 | bottom_field, field_select, ref_picture, pix_op, |
||
| 405 | motion_x, motion_y, h, 0, mb_y); |
||
| 406 | } |
||
| 407 | |||
| 408 | // FIXME move to dsputil, avg variant, 16x16 version |
||
| 409 | static inline void put_obmc(uint8_t *dst, uint8_t *src[5], int stride) |
||
| 410 | { |
||
| 411 | int x; |
||
| 412 | uint8_t *const top = src[1]; |
||
| 413 | uint8_t *const left = src[2]; |
||
| 414 | uint8_t *const mid = src[0]; |
||
| 415 | uint8_t *const right = src[3]; |
||
| 416 | uint8_t *const bottom = src[4]; |
||
| 417 | #define OBMC_FILTER(x, t, l, m, r, b)\ |
||
| 418 | dst[x]= (t*top[x] + l*left[x] + m*mid[x] + r*right[x] + b*bottom[x] + 4)>>3 |
||
| 419 | #define OBMC_FILTER4(x, t, l, m, r, b)\ |
||
| 420 | OBMC_FILTER(x , t, l, m, r, b);\ |
||
| 421 | OBMC_FILTER(x+1 , t, l, m, r, b);\ |
||
| 422 | OBMC_FILTER(x +stride, t, l, m, r, b);\ |
||
| 423 | OBMC_FILTER(x+1+stride, t, l, m, r, b); |
||
| 424 | |||
| 425 | x = 0; |
||
| 426 | OBMC_FILTER (x , 2, 2, 4, 0, 0); |
||
| 427 | OBMC_FILTER (x + 1, 2, 1, 5, 0, 0); |
||
| 428 | OBMC_FILTER4(x + 2, 2, 1, 5, 0, 0); |
||
| 429 | OBMC_FILTER4(x + 4, 2, 0, 5, 1, 0); |
||
| 430 | OBMC_FILTER (x + 6, 2, 0, 5, 1, 0); |
||
| 431 | OBMC_FILTER (x + 7, 2, 0, 4, 2, 0); |
||
| 432 | x += stride; |
||
| 433 | OBMC_FILTER (x , 1, 2, 5, 0, 0); |
||
| 434 | OBMC_FILTER (x + 1, 1, 2, 5, 0, 0); |
||
| 435 | OBMC_FILTER (x + 6, 1, 0, 5, 2, 0); |
||
| 436 | OBMC_FILTER (x + 7, 1, 0, 5, 2, 0); |
||
| 437 | x += stride; |
||
| 438 | OBMC_FILTER4(x , 1, 2, 5, 0, 0); |
||
| 439 | OBMC_FILTER4(x + 2, 1, 1, 6, 0, 0); |
||
| 440 | OBMC_FILTER4(x + 4, 1, 0, 6, 1, 0); |
||
| 441 | OBMC_FILTER4(x + 6, 1, 0, 5, 2, 0); |
||
| 442 | x += 2 * stride; |
||
| 443 | OBMC_FILTER4(x , 0, 2, 5, 0, 1); |
||
| 444 | OBMC_FILTER4(x + 2, 0, 1, 6, 0, 1); |
||
| 445 | OBMC_FILTER4(x + 4, 0, 0, 6, 1, 1); |
||
| 446 | OBMC_FILTER4(x + 6, 0, 0, 5, 2, 1); |
||
| 447 | x += 2*stride; |
||
| 448 | OBMC_FILTER (x , 0, 2, 5, 0, 1); |
||
| 449 | OBMC_FILTER (x + 1, 0, 2, 5, 0, 1); |
||
| 450 | OBMC_FILTER4(x + 2, 0, 1, 5, 0, 2); |
||
| 451 | OBMC_FILTER4(x + 4, 0, 0, 5, 1, 2); |
||
| 452 | OBMC_FILTER (x + 6, 0, 0, 5, 2, 1); |
||
| 453 | OBMC_FILTER (x + 7, 0, 0, 5, 2, 1); |
||
| 454 | x += stride; |
||
| 455 | OBMC_FILTER (x , 0, 2, 4, 0, 2); |
||
| 456 | OBMC_FILTER (x + 1, 0, 1, 5, 0, 2); |
||
| 457 | OBMC_FILTER (x + 6, 0, 0, 5, 1, 2); |
||
| 458 | OBMC_FILTER (x + 7, 0, 0, 4, 2, 2); |
||
| 459 | } |
||
| 460 | |||
| 461 | /* obmc for 1 8x8 luma block */ |
||
| 462 | static inline void obmc_motion(MpegEncContext *s, |
||
| 463 | uint8_t *dest, uint8_t *src, |
||
| 464 | int src_x, int src_y, |
||
| 465 | op_pixels_func *pix_op, |
||
| 466 | int16_t mv[5][2] /* mid top left right bottom */) |
||
| 467 | #define MID 0 |
||
| 468 | { |
||
| 469 | int i; |
||
| 470 | uint8_t *ptr[5]; |
||
| 471 | |||
| 472 | av_assert2(s->quarter_sample == 0); |
||
| 473 | |||
| 474 | for (i = 0; i < 5; i++) { |
||
| 475 | if (i && mv[i][0] == mv[MID][0] && mv[i][1] == mv[MID][1]) { |
||
| 476 | ptr[i] = ptr[MID]; |
||
| 477 | } else { |
||
| 478 | ptr[i] = s->obmc_scratchpad + 8 * (i & 1) + |
||
| 479 | s->linesize * 8 * (i >> 1); |
||
| 480 | hpel_motion(s, ptr[i], src, src_x, src_y, pix_op, |
||
| 481 | mv[i][0], mv[i][1]); |
||
| 482 | } |
||
| 483 | } |
||
| 484 | |||
| 485 | put_obmc(dest, ptr, s->linesize); |
||
| 486 | } |
||
| 487 | |||
| 488 | static inline void qpel_motion(MpegEncContext *s, |
||
| 489 | uint8_t *dest_y, |
||
| 490 | uint8_t *dest_cb, |
||
| 491 | uint8_t *dest_cr, |
||
| 492 | int field_based, int bottom_field, |
||
| 493 | int field_select, uint8_t **ref_picture, |
||
| 494 | op_pixels_func (*pix_op)[4], |
||
| 495 | qpel_mc_func (*qpix_op)[16], |
||
| 496 | int motion_x, int motion_y, int h) |
||
| 497 | { |
||
| 498 | uint8_t *ptr_y, *ptr_cb, *ptr_cr; |
||
| 499 | int dxy, uvdxy, mx, my, src_x, src_y, uvsrc_x, uvsrc_y, v_edge_pos; |
||
| 500 | ptrdiff_t linesize, uvlinesize; |
||
| 501 | |||
| 502 | dxy = ((motion_y & 3) << 2) | (motion_x & 3); |
||
| 503 | |||
| 504 | src_x = s->mb_x * 16 + (motion_x >> 2); |
||
| 505 | src_y = s->mb_y * (16 >> field_based) + (motion_y >> 2); |
||
| 506 | |||
| 507 | v_edge_pos = s->v_edge_pos >> field_based; |
||
| 508 | linesize = s->linesize << field_based; |
||
| 509 | uvlinesize = s->uvlinesize << field_based; |
||
| 510 | |||
| 511 | if (field_based) { |
||
| 512 | mx = motion_x / 2; |
||
| 513 | my = motion_y >> 1; |
||
| 514 | } else if (s->workaround_bugs & FF_BUG_QPEL_CHROMA2) { |
||
| 515 | static const int rtab[8] = { 0, 0, 1, 1, 0, 0, 0, 1 }; |
||
| 516 | mx = (motion_x >> 1) + rtab[motion_x & 7]; |
||
| 517 | my = (motion_y >> 1) + rtab[motion_y & 7]; |
||
| 518 | } else if (s->workaround_bugs & FF_BUG_QPEL_CHROMA) { |
||
| 519 | mx = (motion_x >> 1) | (motion_x & 1); |
||
| 520 | my = (motion_y >> 1) | (motion_y & 1); |
||
| 521 | } else { |
||
| 522 | mx = motion_x / 2; |
||
| 523 | my = motion_y / 2; |
||
| 524 | } |
||
| 525 | mx = (mx >> 1) | (mx & 1); |
||
| 526 | my = (my >> 1) | (my & 1); |
||
| 527 | |||
| 528 | uvdxy = (mx & 1) | ((my & 1) << 1); |
||
| 529 | mx >>= 1; |
||
| 530 | my >>= 1; |
||
| 531 | |||
| 532 | uvsrc_x = s->mb_x * 8 + mx; |
||
| 533 | uvsrc_y = s->mb_y * (8 >> field_based) + my; |
||
| 534 | |||
| 535 | ptr_y = ref_picture[0] + src_y * linesize + src_x; |
||
| 536 | ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x; |
||
| 537 | ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x; |
||
| 538 | |||
| 539 | if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 3) - 16, 0) || |
||
| 540 | (unsigned)src_y > FFMAX( v_edge_pos - (motion_y & 3) - h, 0)) { |
||
| 541 | s->vdsp.emulated_edge_mc(s->edge_emu_buffer, |
||
| 542 | s->linesize, |
||
| 543 | ptr_y, s->linesize, |
||
| 544 | 17, 17 + field_based, |
||
| 545 | src_x, src_y << field_based, |
||
| 546 | s->h_edge_pos, s->v_edge_pos); |
||
| 547 | ptr_y = s->edge_emu_buffer; |
||
| 548 | if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) { |
||
| 549 | uint8_t *uvbuf = s->edge_emu_buffer + 18 * s->linesize; |
||
| 550 | s->vdsp.emulated_edge_mc(uvbuf, s->uvlinesize, |
||
| 551 | ptr_cb, |
||
| 552 | s->uvlinesize, |
||
| 553 | 9, 9 + field_based, |
||
| 554 | uvsrc_x, uvsrc_y << field_based, |
||
| 555 | s->h_edge_pos >> 1, s->v_edge_pos >> 1); |
||
| 556 | s->vdsp.emulated_edge_mc(uvbuf + 16, s->uvlinesize, |
||
| 557 | ptr_cr, |
||
| 558 | s->uvlinesize, |
||
| 559 | 9, 9 + field_based, |
||
| 560 | uvsrc_x, uvsrc_y << field_based, |
||
| 561 | s->h_edge_pos >> 1, s->v_edge_pos >> 1); |
||
| 562 | ptr_cb = uvbuf; |
||
| 563 | ptr_cr = uvbuf + 16; |
||
| 564 | } |
||
| 565 | } |
||
| 566 | |||
| 567 | if (!field_based) |
||
| 568 | qpix_op[0][dxy](dest_y, ptr_y, linesize); |
||
| 569 | else { |
||
| 570 | if (bottom_field) { |
||
| 571 | dest_y += s->linesize; |
||
| 572 | dest_cb += s->uvlinesize; |
||
| 573 | dest_cr += s->uvlinesize; |
||
| 574 | } |
||
| 575 | |||
| 576 | if (field_select) { |
||
| 577 | ptr_y += s->linesize; |
||
| 578 | ptr_cb += s->uvlinesize; |
||
| 579 | ptr_cr += s->uvlinesize; |
||
| 580 | } |
||
| 581 | // damn interlaced mode |
||
| 582 | // FIXME boundary mirroring is not exactly correct here |
||
| 583 | qpix_op[1][dxy](dest_y, ptr_y, linesize); |
||
| 584 | qpix_op[1][dxy](dest_y + 8, ptr_y + 8, linesize); |
||
| 585 | } |
||
| 586 | if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) { |
||
| 587 | pix_op[1][uvdxy](dest_cr, ptr_cr, uvlinesize, h >> 1); |
||
| 588 | pix_op[1][uvdxy](dest_cb, ptr_cb, uvlinesize, h >> 1); |
||
| 589 | } |
||
| 590 | } |
||
| 591 | |||
| 592 | /** |
||
| 593 | * h263 chroma 4mv motion compensation. |
||
| 594 | */ |
||
| 595 | static void chroma_4mv_motion(MpegEncContext *s, |
||
| 596 | uint8_t *dest_cb, uint8_t *dest_cr, |
||
| 597 | uint8_t **ref_picture, |
||
| 598 | op_pixels_func *pix_op, |
||
| 599 | int mx, int my) |
||
| 600 | { |
||
| 601 | uint8_t *ptr; |
||
| 602 | int src_x, src_y, dxy, emu = 0; |
||
| 603 | ptrdiff_t offset; |
||
| 604 | |||
| 605 | /* In case of 8X8, we construct a single chroma motion vector |
||
| 606 | * with a special rounding */ |
||
| 607 | mx = ff_h263_round_chroma(mx); |
||
| 608 | my = ff_h263_round_chroma(my); |
||
| 609 | |||
| 610 | dxy = ((my & 1) << 1) | (mx & 1); |
||
| 611 | mx >>= 1; |
||
| 612 | my >>= 1; |
||
| 613 | |||
| 614 | src_x = s->mb_x * 8 + mx; |
||
| 615 | src_y = s->mb_y * 8 + my; |
||
| 616 | src_x = av_clip(src_x, -8, (s->width >> 1)); |
||
| 617 | if (src_x == (s->width >> 1)) |
||
| 618 | dxy &= ~1; |
||
| 619 | src_y = av_clip(src_y, -8, (s->height >> 1)); |
||
| 620 | if (src_y == (s->height >> 1)) |
||
| 621 | dxy &= ~2; |
||
| 622 | |||
| 623 | offset = src_y * s->uvlinesize + src_x; |
||
| 624 | ptr = ref_picture[1] + offset; |
||
| 625 | if (s->flags & CODEC_FLAG_EMU_EDGE) { |
||
| 626 | if ((unsigned)src_x > FFMAX((s->h_edge_pos >> 1) - (dxy & 1) - 8, 0) || |
||
| 627 | (unsigned)src_y > FFMAX((s->v_edge_pos >> 1) - (dxy >> 1) - 8, 0)) { |
||
| 628 | s->vdsp.emulated_edge_mc(s->edge_emu_buffer, s->uvlinesize, |
||
| 629 | ptr, s->uvlinesize, |
||
| 630 | 9, 9, src_x, src_y, |
||
| 631 | s->h_edge_pos >> 1, s->v_edge_pos >> 1); |
||
| 632 | ptr = s->edge_emu_buffer; |
||
| 633 | emu = 1; |
||
| 634 | } |
||
| 635 | } |
||
| 636 | pix_op[dxy](dest_cb, ptr, s->uvlinesize, 8); |
||
| 637 | |||
| 638 | ptr = ref_picture[2] + offset; |
||
| 639 | if (emu) { |
||
| 640 | s->vdsp.emulated_edge_mc(s->edge_emu_buffer, s->uvlinesize, |
||
| 641 | ptr, s->uvlinesize, |
||
| 642 | 9, 9, src_x, src_y, |
||
| 643 | s->h_edge_pos >> 1, s->v_edge_pos >> 1); |
||
| 644 | ptr = s->edge_emu_buffer; |
||
| 645 | } |
||
| 646 | pix_op[dxy](dest_cr, ptr, s->uvlinesize, 8); |
||
| 647 | } |
||
| 648 | |||
| 649 | static inline void prefetch_motion(MpegEncContext *s, uint8_t **pix, int dir) |
||
| 650 | { |
||
| 651 | /* fetch pixels for estimated mv 4 macroblocks ahead |
||
| 652 | * optimized for 64byte cache lines */ |
||
| 653 | const int shift = s->quarter_sample ? 2 : 1; |
||
| 654 | const int mx = (s->mv[dir][0][0] >> shift) + 16 * s->mb_x + 8; |
||
| 655 | const int my = (s->mv[dir][0][1] >> shift) + 16 * s->mb_y; |
||
| 656 | int off = mx + (my + (s->mb_x & 3) * 4) * s->linesize + 64; |
||
| 657 | |||
| 658 | s->vdsp.prefetch(pix[0] + off, s->linesize, 4); |
||
| 659 | off = (mx >> 1) + ((my >> 1) + (s->mb_x & 7)) * s->uvlinesize + 64; |
||
| 660 | s->vdsp.prefetch(pix[1] + off, pix[2] - pix[1], 2); |
||
| 661 | } |
||
| 662 | |||
| 663 | static inline void apply_obmc(MpegEncContext *s, |
||
| 664 | uint8_t *dest_y, |
||
| 665 | uint8_t *dest_cb, |
||
| 666 | uint8_t *dest_cr, |
||
| 667 | uint8_t **ref_picture, |
||
| 668 | op_pixels_func (*pix_op)[4]) |
||
| 669 | { |
||
| 670 | LOCAL_ALIGNED_8(int16_t, mv_cache, [4], [4][2]); |
||
| 671 | Picture *cur_frame = &s->current_picture; |
||
| 672 | int mb_x = s->mb_x; |
||
| 673 | int mb_y = s->mb_y; |
||
| 674 | const int xy = mb_x + mb_y * s->mb_stride; |
||
| 675 | const int mot_stride = s->b8_stride; |
||
| 676 | const int mot_xy = mb_x * 2 + mb_y * 2 * mot_stride; |
||
| 677 | int mx, my, i; |
||
| 678 | |||
| 679 | av_assert2(!s->mb_skipped); |
||
| 680 | |||
| 681 | AV_COPY32(mv_cache[1][1], cur_frame->motion_val[0][mot_xy]); |
||
| 682 | AV_COPY32(mv_cache[1][2], cur_frame->motion_val[0][mot_xy + 1]); |
||
| 683 | |||
| 684 | AV_COPY32(mv_cache[2][1], |
||
| 685 | cur_frame->motion_val[0][mot_xy + mot_stride]); |
||
| 686 | AV_COPY32(mv_cache[2][2], |
||
| 687 | cur_frame->motion_val[0][mot_xy + mot_stride + 1]); |
||
| 688 | |||
| 689 | AV_COPY32(mv_cache[3][1], |
||
| 690 | cur_frame->motion_val[0][mot_xy + mot_stride]); |
||
| 691 | AV_COPY32(mv_cache[3][2], |
||
| 692 | cur_frame->motion_val[0][mot_xy + mot_stride + 1]); |
||
| 693 | |||
| 694 | if (mb_y == 0 || IS_INTRA(cur_frame->mb_type[xy - s->mb_stride])) { |
||
| 695 | AV_COPY32(mv_cache[0][1], mv_cache[1][1]); |
||
| 696 | AV_COPY32(mv_cache[0][2], mv_cache[1][2]); |
||
| 697 | } else { |
||
| 698 | AV_COPY32(mv_cache[0][1], |
||
| 699 | cur_frame->motion_val[0][mot_xy - mot_stride]); |
||
| 700 | AV_COPY32(mv_cache[0][2], |
||
| 701 | cur_frame->motion_val[0][mot_xy - mot_stride + 1]); |
||
| 702 | } |
||
| 703 | |||
| 704 | if (mb_x == 0 || IS_INTRA(cur_frame->mb_type[xy - 1])) { |
||
| 705 | AV_COPY32(mv_cache[1][0], mv_cache[1][1]); |
||
| 706 | AV_COPY32(mv_cache[2][0], mv_cache[2][1]); |
||
| 707 | } else { |
||
| 708 | AV_COPY32(mv_cache[1][0], cur_frame->motion_val[0][mot_xy - 1]); |
||
| 709 | AV_COPY32(mv_cache[2][0], |
||
| 710 | cur_frame->motion_val[0][mot_xy - 1 + mot_stride]); |
||
| 711 | } |
||
| 712 | |||
| 713 | if (mb_x + 1 >= s->mb_width || IS_INTRA(cur_frame->mb_type[xy + 1])) { |
||
| 714 | AV_COPY32(mv_cache[1][3], mv_cache[1][2]); |
||
| 715 | AV_COPY32(mv_cache[2][3], mv_cache[2][2]); |
||
| 716 | } else { |
||
| 717 | AV_COPY32(mv_cache[1][3], cur_frame->motion_val[0][mot_xy + 2]); |
||
| 718 | AV_COPY32(mv_cache[2][3], |
||
| 719 | cur_frame->motion_val[0][mot_xy + 2 + mot_stride]); |
||
| 720 | } |
||
| 721 | |||
| 722 | mx = 0; |
||
| 723 | my = 0; |
||
| 724 | for (i = 0; i < 4; i++) { |
||
| 725 | const int x = (i & 1) + 1; |
||
| 726 | const int y = (i >> 1) + 1; |
||
| 727 | int16_t mv[5][2] = { |
||
| 728 | { mv_cache[y][x][0], mv_cache[y][x][1] }, |
||
| 729 | { mv_cache[y - 1][x][0], mv_cache[y - 1][x][1] }, |
||
| 730 | { mv_cache[y][x - 1][0], mv_cache[y][x - 1][1] }, |
||
| 731 | { mv_cache[y][x + 1][0], mv_cache[y][x + 1][1] }, |
||
| 732 | { mv_cache[y + 1][x][0], mv_cache[y + 1][x][1] } |
||
| 733 | }; |
||
| 734 | // FIXME cleanup |
||
| 735 | obmc_motion(s, dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize, |
||
| 736 | ref_picture[0], |
||
| 737 | mb_x * 16 + (i & 1) * 8, mb_y * 16 + (i >> 1) * 8, |
||
| 738 | pix_op[1], |
||
| 739 | mv); |
||
| 740 | |||
| 741 | mx += mv[0][0]; |
||
| 742 | my += mv[0][1]; |
||
| 743 | } |
||
| 744 | if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) |
||
| 745 | chroma_4mv_motion(s, dest_cb, dest_cr, |
||
| 746 | ref_picture, pix_op[1], |
||
| 747 | mx, my); |
||
| 748 | } |
||
| 749 | |||
| 750 | static inline void apply_8x8(MpegEncContext *s, |
||
| 751 | uint8_t *dest_y, |
||
| 752 | uint8_t *dest_cb, |
||
| 753 | uint8_t *dest_cr, |
||
| 754 | int dir, |
||
| 755 | uint8_t **ref_picture, |
||
| 756 | qpel_mc_func (*qpix_op)[16], |
||
| 757 | op_pixels_func (*pix_op)[4]) |
||
| 758 | { |
||
| 759 | int dxy, mx, my, src_x, src_y; |
||
| 760 | int i; |
||
| 761 | int mb_x = s->mb_x; |
||
| 762 | int mb_y = s->mb_y; |
||
| 763 | uint8_t *ptr, *dest; |
||
| 764 | |||
| 765 | mx = 0; |
||
| 766 | my = 0; |
||
| 767 | if (s->quarter_sample) { |
||
| 768 | for (i = 0; i < 4; i++) { |
||
| 769 | int motion_x = s->mv[dir][i][0]; |
||
| 770 | int motion_y = s->mv[dir][i][1]; |
||
| 771 | |||
| 772 | dxy = ((motion_y & 3) << 2) | (motion_x & 3); |
||
| 773 | src_x = mb_x * 16 + (motion_x >> 2) + (i & 1) * 8; |
||
| 774 | src_y = mb_y * 16 + (motion_y >> 2) + (i >> 1) * 8; |
||
| 775 | |||
| 776 | /* WARNING: do no forget half pels */ |
||
| 777 | src_x = av_clip(src_x, -16, s->width); |
||
| 778 | if (src_x == s->width) |
||
| 779 | dxy &= ~3; |
||
| 780 | src_y = av_clip(src_y, -16, s->height); |
||
| 781 | if (src_y == s->height) |
||
| 782 | dxy &= ~12; |
||
| 783 | |||
| 784 | ptr = ref_picture[0] + (src_y * s->linesize) + (src_x); |
||
| 785 | if (s->flags & CODEC_FLAG_EMU_EDGE) { |
||
| 786 | if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 3) - 8, 0) || |
||
| 787 | (unsigned)src_y > FFMAX(s->v_edge_pos - (motion_y & 3) - 8, 0)) { |
||
| 788 | s->vdsp.emulated_edge_mc(s->edge_emu_buffer, |
||
| 789 | s->linesize, ptr, |
||
| 790 | s->linesize, |
||
| 791 | 9, 9, |
||
| 792 | src_x, src_y, |
||
| 793 | s->h_edge_pos, |
||
| 794 | s->v_edge_pos); |
||
| 795 | ptr = s->edge_emu_buffer; |
||
| 796 | } |
||
| 797 | } |
||
| 798 | dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize; |
||
| 799 | qpix_op[1][dxy](dest, ptr, s->linesize); |
||
| 800 | |||
| 801 | mx += s->mv[dir][i][0] / 2; |
||
| 802 | my += s->mv[dir][i][1] / 2; |
||
| 803 | } |
||
| 804 | } else { |
||
| 805 | for (i = 0; i < 4; i++) { |
||
| 806 | hpel_motion(s, |
||
| 807 | dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize, |
||
| 808 | ref_picture[0], |
||
| 809 | mb_x * 16 + (i & 1) * 8, |
||
| 810 | mb_y * 16 + (i >> 1) * 8, |
||
| 811 | pix_op[1], |
||
| 812 | s->mv[dir][i][0], |
||
| 813 | s->mv[dir][i][1]); |
||
| 814 | |||
| 815 | mx += s->mv[dir][i][0]; |
||
| 816 | my += s->mv[dir][i][1]; |
||
| 817 | } |
||
| 818 | } |
||
| 819 | |||
| 820 | if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) |
||
| 821 | chroma_4mv_motion(s, dest_cb, dest_cr, |
||
| 822 | ref_picture, pix_op[1], mx, my); |
||
| 823 | } |
||
| 824 | |||
| 825 | /** |
||
| 826 | * motion compensation of a single macroblock |
||
| 827 | * @param s context |
||
| 828 | * @param dest_y luma destination pointer |
||
| 829 | * @param dest_cb chroma cb/u destination pointer |
||
| 830 | * @param dest_cr chroma cr/v destination pointer |
||
| 831 | * @param dir direction (0->forward, 1->backward) |
||
| 832 | * @param ref_picture array[3] of pointers to the 3 planes of the reference picture |
||
| 833 | * @param pix_op halfpel motion compensation function (average or put normally) |
||
| 834 | * @param qpix_op qpel motion compensation function (average or put normally) |
||
| 835 | * the motion vectors are taken from s->mv and the MV type from s->mv_type |
||
| 836 | */ |
||
| 837 | static av_always_inline void MPV_motion_internal(MpegEncContext *s, |
||
| 838 | uint8_t *dest_y, |
||
| 839 | uint8_t *dest_cb, |
||
| 840 | uint8_t *dest_cr, |
||
| 841 | int dir, |
||
| 842 | uint8_t **ref_picture, |
||
| 843 | op_pixels_func (*pix_op)[4], |
||
| 844 | qpel_mc_func (*qpix_op)[16], |
||
| 845 | int is_mpeg12) |
||
| 846 | { |
||
| 847 | int i; |
||
| 848 | int mb_y = s->mb_y; |
||
| 849 | |||
| 850 | prefetch_motion(s, ref_picture, dir); |
||
| 851 | |||
| 852 | if (!is_mpeg12 && s->obmc && s->pict_type != AV_PICTURE_TYPE_B) { |
||
| 853 | apply_obmc(s, dest_y, dest_cb, dest_cr, ref_picture, pix_op); |
||
| 854 | return; |
||
| 855 | } |
||
| 856 | |||
| 857 | switch (s->mv_type) { |
||
| 858 | case MV_TYPE_16X16: |
||
| 859 | if (s->mcsel) { |
||
| 860 | if (s->real_sprite_warping_points == 1) { |
||
| 861 | gmc1_motion(s, dest_y, dest_cb, dest_cr, |
||
| 862 | ref_picture); |
||
| 863 | } else { |
||
| 864 | gmc_motion(s, dest_y, dest_cb, dest_cr, |
||
| 865 | ref_picture); |
||
| 866 | } |
||
| 867 | } else if (!is_mpeg12 && s->quarter_sample) { |
||
| 868 | qpel_motion(s, dest_y, dest_cb, dest_cr, |
||
| 869 | 0, 0, 0, |
||
| 870 | ref_picture, pix_op, qpix_op, |
||
| 871 | s->mv[dir][0][0], s->mv[dir][0][1], 16); |
||
| 872 | } else if (!is_mpeg12 && (CONFIG_WMV2_DECODER || CONFIG_WMV2_ENCODER) && |
||
| 873 | s->mspel && s->codec_id == AV_CODEC_ID_WMV2) { |
||
| 874 | ff_mspel_motion(s, dest_y, dest_cb, dest_cr, |
||
| 875 | ref_picture, pix_op, |
||
| 876 | s->mv[dir][0][0], s->mv[dir][0][1], 16); |
||
| 877 | } else { |
||
| 878 | mpeg_motion(s, dest_y, dest_cb, dest_cr, 0, |
||
| 879 | ref_picture, pix_op, |
||
| 880 | s->mv[dir][0][0], s->mv[dir][0][1], 16, mb_y); |
||
| 881 | } |
||
| 882 | break; |
||
| 883 | case MV_TYPE_8X8: |
||
| 884 | if (!is_mpeg12) |
||
| 885 | apply_8x8(s, dest_y, dest_cb, dest_cr, |
||
| 886 | dir, ref_picture, qpix_op, pix_op); |
||
| 887 | break; |
||
| 888 | case MV_TYPE_FIELD: |
||
| 889 | if (s->picture_structure == PICT_FRAME) { |
||
| 890 | if (!is_mpeg12 && s->quarter_sample) { |
||
| 891 | for (i = 0; i < 2; i++) |
||
| 892 | qpel_motion(s, dest_y, dest_cb, dest_cr, |
||
| 893 | 1, i, s->field_select[dir][i], |
||
| 894 | ref_picture, pix_op, qpix_op, |
||
| 895 | s->mv[dir][i][0], s->mv[dir][i][1], 8); |
||
| 896 | } else { |
||
| 897 | /* top field */ |
||
| 898 | mpeg_motion_field(s, dest_y, dest_cb, dest_cr, |
||
| 899 | 0, s->field_select[dir][0], |
||
| 900 | ref_picture, pix_op, |
||
| 901 | s->mv[dir][0][0], s->mv[dir][0][1], 8, mb_y); |
||
| 902 | /* bottom field */ |
||
| 903 | mpeg_motion_field(s, dest_y, dest_cb, dest_cr, |
||
| 904 | 1, s->field_select[dir][1], |
||
| 905 | ref_picture, pix_op, |
||
| 906 | s->mv[dir][1][0], s->mv[dir][1][1], 8, mb_y); |
||
| 907 | } |
||
| 908 | } else { |
||
| 909 | if ( s->picture_structure != s->field_select[dir][0] + 1 && s->pict_type != AV_PICTURE_TYPE_B && !s->first_field |
||
| 910 | || !ref_picture[0]) { |
||
| 911 | ref_picture = s->current_picture_ptr->f.data; |
||
| 912 | } |
||
| 913 | |||
| 914 | mpeg_motion(s, dest_y, dest_cb, dest_cr, |
||
| 915 | s->field_select[dir][0], |
||
| 916 | ref_picture, pix_op, |
||
| 917 | s->mv[dir][0][0], s->mv[dir][0][1], 16, mb_y >> 1); |
||
| 918 | } |
||
| 919 | break; |
||
| 920 | case MV_TYPE_16X8: |
||
| 921 | for (i = 0; i < 2; i++) { |
||
| 922 | uint8_t **ref2picture; |
||
| 923 | |||
| 924 | if ((s->picture_structure == s->field_select[dir][i] + 1 |
||
| 925 | || s->pict_type == AV_PICTURE_TYPE_B || s->first_field) && ref_picture[0]) { |
||
| 926 | ref2picture = ref_picture; |
||
| 927 | } else { |
||
| 928 | ref2picture = s->current_picture_ptr->f.data; |
||
| 929 | } |
||
| 930 | |||
| 931 | mpeg_motion(s, dest_y, dest_cb, dest_cr, |
||
| 932 | s->field_select[dir][i], |
||
| 933 | ref2picture, pix_op, |
||
| 934 | s->mv[dir][i][0], s->mv[dir][i][1] + 16 * i, |
||
| 935 | 8, mb_y >> 1); |
||
| 936 | |||
| 937 | dest_y += 16 * s->linesize; |
||
| 938 | dest_cb += (16 >> s->chroma_y_shift) * s->uvlinesize; |
||
| 939 | dest_cr += (16 >> s->chroma_y_shift) * s->uvlinesize; |
||
| 940 | } |
||
| 941 | break; |
||
| 942 | case MV_TYPE_DMV: |
||
| 943 | if (s->picture_structure == PICT_FRAME) { |
||
| 944 | for (i = 0; i < 2; i++) { |
||
| 945 | int j; |
||
| 946 | for (j = 0; j < 2; j++) |
||
| 947 | mpeg_motion_field(s, dest_y, dest_cb, dest_cr, |
||
| 948 | j, j ^ i, ref_picture, pix_op, |
||
| 949 | s->mv[dir][2 * i + j][0], |
||
| 950 | s->mv[dir][2 * i + j][1], 8, mb_y); |
||
| 951 | pix_op = s->hdsp.avg_pixels_tab; |
||
| 952 | } |
||
| 953 | } else { |
||
| 954 | if (!ref_picture[0]) { |
||
| 955 | ref_picture = s->current_picture_ptr->f.data; |
||
| 956 | } |
||
| 957 | for (i = 0; i < 2; i++) { |
||
| 958 | mpeg_motion(s, dest_y, dest_cb, dest_cr, |
||
| 959 | s->picture_structure != i + 1, |
||
| 960 | ref_picture, pix_op, |
||
| 961 | s->mv[dir][2 * i][0], s->mv[dir][2 * i][1], |
||
| 962 | 16, mb_y >> 1); |
||
| 963 | |||
| 964 | // after put we make avg of the same block |
||
| 965 | pix_op = s->hdsp.avg_pixels_tab; |
||
| 966 | |||
| 967 | /* opposite parity is always in the same frame if this is |
||
| 968 | * second field */ |
||
| 969 | if (!s->first_field) { |
||
| 970 | ref_picture = s->current_picture_ptr->f.data; |
||
| 971 | } |
||
| 972 | } |
||
| 973 | } |
||
| 974 | break; |
||
| 975 | default: av_assert2(0); |
||
| 976 | } |
||
| 977 | } |
||
| 978 | |||
| 979 | void ff_MPV_motion(MpegEncContext *s, |
||
| 980 | uint8_t *dest_y, uint8_t *dest_cb, |
||
| 981 | uint8_t *dest_cr, int dir, |
||
| 982 | uint8_t **ref_picture, |
||
| 983 | op_pixels_func (*pix_op)[4], |
||
| 984 | qpel_mc_func (*qpix_op)[16]) |
||
| 985 | { |
||
| 986 | #if !CONFIG_SMALL |
||
| 987 | if (s->out_format == FMT_MPEG1) |
||
| 988 | MPV_motion_internal(s, dest_y, dest_cb, dest_cr, dir, |
||
| 989 | ref_picture, pix_op, qpix_op, 1); |
||
| 990 | else |
||
| 991 | #endif |
||
| 992 | MPV_motion_internal(s, dest_y, dest_cb, dest_cr, dir, |
||
| 993 | ref_picture, pix_op, qpix_op, 0); |
||
| 994 | } |