1 /* 2 * Copyright (c) 2010 The WebM project authors. All Rights Reserved. 3 * 4 * Use of this source code is governed by a BSD-style license 5 * that can be found in the LICENSE file in the root of the source 6 * tree. An additional intellectual property rights grant can be found 7 * in the file PATENTS. All contributing project authors may 8 * be found in the AUTHORS file in the root of the source tree. 9 */ 10 11 12 #include "vpx_ports/config.h" 13 #include "vpx_scale/yv12config.h" 14 #include "postproc.h" 15 #include "vpx_scale/yv12extend.h" 16 #include "vpx_scale/vpxscale.h" 17 #include "systemdependent.h" 18 19 #include <math.h> 20 #include <stdlib.h> 21 #include <stdio.h> 22 // global constants 23 24 static const short kernel5[] = 25 { 26 1, 1, 4, 1, 1 27 }; 28 29 const short vp8_rv[] = 30 { 31 8, 5, 2, 2, 8, 12, 4, 9, 8, 3, 32 0, 3, 9, 0, 0, 0, 8, 3, 14, 4, 33 10, 1, 11, 14, 1, 14, 9, 6, 12, 11, 34 8, 6, 10, 0, 0, 8, 9, 0, 3, 14, 35 8, 11, 13, 4, 2, 9, 0, 3, 9, 6, 36 1, 2, 3, 14, 13, 1, 8, 2, 9, 7, 37 3, 3, 1, 13, 13, 6, 6, 5, 2, 7, 38 11, 9, 11, 8, 7, 3, 2, 0, 13, 13, 39 14, 4, 12, 5, 12, 10, 8, 10, 13, 10, 40 4, 14, 4, 10, 0, 8, 11, 1, 13, 7, 41 7, 14, 6, 14, 13, 2, 13, 5, 4, 4, 42 0, 10, 0, 5, 13, 2, 12, 7, 11, 13, 43 8, 0, 4, 10, 7, 2, 7, 2, 2, 5, 44 3, 4, 7, 3, 3, 14, 14, 5, 9, 13, 45 3, 14, 3, 6, 3, 0, 11, 8, 13, 1, 46 13, 1, 12, 0, 10, 9, 7, 6, 2, 8, 47 5, 2, 13, 7, 1, 13, 14, 7, 6, 7, 48 9, 6, 10, 11, 7, 8, 7, 5, 14, 8, 49 4, 4, 0, 8, 7, 10, 0, 8, 14, 11, 50 3, 12, 5, 7, 14, 3, 14, 5, 2, 6, 51 11, 12, 12, 8, 0, 11, 13, 1, 2, 0, 52 5, 10, 14, 7, 8, 0, 4, 11, 0, 8, 53 0, 3, 10, 5, 8, 0, 11, 6, 7, 8, 54 10, 7, 13, 9, 2, 5, 1, 5, 10, 2, 55 4, 3, 5, 6, 10, 8, 9, 4, 11, 14, 56 0, 10, 0, 5, 13, 2, 12, 7, 11, 13, 57 8, 0, 4, 10, 7, 2, 7, 2, 2, 5, 58 3, 4, 7, 3, 3, 14, 14, 5, 9, 13, 59 3, 14, 3, 6, 3, 0, 11, 8, 13, 1, 60 13, 1, 12, 0, 10, 9, 7, 6, 2, 8, 61 5, 2, 13, 7, 1, 13, 14, 7, 6, 7, 62 9, 6, 10, 11, 7, 8, 7, 5, 14, 8, 63 4, 4, 0, 8, 7, 10, 0, 8, 14, 11, 64 3, 12, 5, 7, 14, 3, 14, 5, 2, 6, 65 11, 12, 12, 8, 0, 11, 13, 1, 2, 0, 66 5, 10, 14, 7, 8, 0, 4, 11, 0, 8, 67 0, 3, 10, 5, 8, 0, 11, 6, 7, 8, 68 10, 7, 13, 9, 2, 5, 1, 5, 10, 2, 69 4, 3, 5, 6, 10, 8, 9, 4, 11, 14, 70 3, 8, 3, 7, 8, 5, 11, 4, 12, 3, 71 11, 9, 14, 8, 14, 13, 4, 3, 1, 2, 72 14, 6, 5, 4, 4, 11, 4, 6, 2, 1, 73 5, 8, 8, 12, 13, 5, 14, 10, 12, 13, 74 0, 9, 5, 5, 11, 10, 13, 9, 10, 13, 75 }; 76 77 78 extern void vp8_blit_text(const char *msg, unsigned char *address, const int pitch); 79 80 /*********************************************************************************************************** 81 */ 82 void vp8_post_proc_down_and_across_c 83 ( 84 unsigned char *src_ptr, 85 unsigned char *dst_ptr, 86 int src_pixels_per_line, 87 int dst_pixels_per_line, 88 int rows, 89 int cols, 90 int flimit 91 ) 92 { 93 unsigned char *p_src, *p_dst; 94 int row; 95 int col; 96 int i; 97 int v; 98 int pitch = src_pixels_per_line; 99 unsigned char d[8]; 100 (void)dst_pixels_per_line; 101 102 for (row = 0; row < rows; row++) 103 { 104 // post_proc_down for one row 105 p_src = src_ptr; 106 p_dst = dst_ptr; 107 108 for (col = 0; col < cols; col++) 109 { 110 111 int kernel = 4; 112 int v = p_src[col]; 113 114 for (i = -2; i <= 2; i++) 115 { 116 if (abs(v - p_src[col+i*pitch]) > flimit) 117 goto down_skip_convolve; 118 119 kernel += kernel5[2+i] * p_src[col+i*pitch]; 120 } 121 122 v = (kernel >> 3); 123 down_skip_convolve: 124 p_dst[col] = v; 125 } 126 127 // now post_proc_across 128 p_src = dst_ptr; 129 p_dst = dst_ptr; 130 131 for (i = 0; i < 8; i++) 132 d[i] = p_src[i]; 133 134 for (col = 0; col < cols; col++) 135 { 136 int kernel = 4; 137 v = p_src[col]; 138 139 d[col&7] = v; 140 141 for (i = -2; i <= 2; i++) 142 { 143 if (abs(v - p_src[col+i]) > flimit) 144 goto across_skip_convolve; 145 146 kernel += kernel5[2+i] * p_src[col+i]; 147 } 148 149 d[col&7] = (kernel >> 3); 150 across_skip_convolve: 151 152 if (col >= 2) 153 p_dst[col-2] = d[(col-2)&7]; 154 } 155 156 //handle the last two pixels 157 p_dst[col-2] = d[(col-2)&7]; 158 p_dst[col-1] = d[(col-1)&7]; 159 160 161 //next row 162 src_ptr += pitch; 163 dst_ptr += pitch; 164 } 165 } 166 167 int vp8_q2mbl(int x) 168 { 169 if (x < 20) x = 20; 170 171 x = 50 + (x - 50) * 10 / 8; 172 return x * x / 3; 173 } 174 void vp8_mbpost_proc_across_ip_c(unsigned char *src, int pitch, int rows, int cols, int flimit) 175 { 176 int r, c, i; 177 178 unsigned char *s = src; 179 unsigned char d[16]; 180 181 182 for (r = 0; r < rows; r++) 183 { 184 int sumsq = 0; 185 int sum = 0; 186 187 for (i = -8; i <= 6; i++) 188 { 189 sumsq += s[i] * s[i]; 190 sum += s[i]; 191 d[i+8] = 0; 192 } 193 194 for (c = 0; c < cols + 8; c++) 195 { 196 int x = s[c+7] - s[c-8]; 197 int y = s[c+7] + s[c-8]; 198 199 sum += x; 200 sumsq += x * y; 201 202 d[c&15] = s[c]; 203 204 if (sumsq * 15 - sum * sum < flimit) 205 { 206 d[c&15] = (8 + sum + s[c]) >> 4; 207 } 208 209 s[c-8] = d[(c-8)&15]; 210 } 211 212 s += pitch; 213 } 214 } 215 216 217 218 219 220 void vp8_mbpost_proc_down_c(unsigned char *dst, int pitch, int rows, int cols, int flimit) 221 { 222 int r, c, i; 223 const short *rv3 = &vp8_rv[63&rand()]; 224 225 for (c = 0; c < cols; c++) 226 { 227 unsigned char *s = &dst[c]; 228 int sumsq = 0; 229 int sum = 0; 230 unsigned char d[16]; 231 const short *rv2 = rv3 + ((c * 17) & 127); 232 233 for (i = -8; i <= 6; i++) 234 { 235 sumsq += s[i*pitch] * s[i*pitch]; 236 sum += s[i*pitch]; 237 } 238 239 for (r = 0; r < rows + 8; r++) 240 { 241 sumsq += s[7*pitch] * s[ 7*pitch] - s[-8*pitch] * s[-8*pitch]; 242 sum += s[7*pitch] - s[-8*pitch]; 243 d[r&15] = s[0]; 244 245 if (sumsq * 15 - sum * sum < flimit) 246 { 247 d[r&15] = (rv2[r&127] + sum + s[0]) >> 4; 248 } 249 250 s[-8*pitch] = d[(r-8)&15]; 251 s += pitch; 252 } 253 } 254 } 255 256 257 static void vp8_deblock_and_de_macro_block(YV12_BUFFER_CONFIG *source, 258 YV12_BUFFER_CONFIG *post, 259 int q, 260 int low_var_thresh, 261 int flag, 262 vp8_postproc_rtcd_vtable_t *rtcd) 263 { 264 double level = 6.0e-05 * q * q * q - .0067 * q * q + .306 * q + .0065; 265 int ppl = (int)(level + .5); 266 (void) low_var_thresh; 267 (void) flag; 268 269 POSTPROC_INVOKE(rtcd, downacross)(source->y_buffer, post->y_buffer, source->y_stride, post->y_stride, source->y_height, source->y_width, ppl); 270 POSTPROC_INVOKE(rtcd, across)(post->y_buffer, post->y_stride, post->y_height, post->y_width, vp8_q2mbl(q)); 271 POSTPROC_INVOKE(rtcd, down)(post->y_buffer, post->y_stride, post->y_height, post->y_width, vp8_q2mbl(q)); 272 273 POSTPROC_INVOKE(rtcd, downacross)(source->u_buffer, post->u_buffer, source->uv_stride, post->uv_stride, source->uv_height, source->uv_width, ppl); 274 POSTPROC_INVOKE(rtcd, downacross)(source->v_buffer, post->v_buffer, source->uv_stride, post->uv_stride, source->uv_height, source->uv_width, ppl); 275 276 } 277 278 void vp8_deblock(YV12_BUFFER_CONFIG *source, 279 YV12_BUFFER_CONFIG *post, 280 int q, 281 int low_var_thresh, 282 int flag, 283 vp8_postproc_rtcd_vtable_t *rtcd) 284 { 285 double level = 6.0e-05 * q * q * q - .0067 * q * q + .306 * q + .0065; 286 int ppl = (int)(level + .5); 287 (void) low_var_thresh; 288 (void) flag; 289 290 POSTPROC_INVOKE(rtcd, downacross)(source->y_buffer, post->y_buffer, source->y_stride, post->y_stride, source->y_height, source->y_width, ppl); 291 POSTPROC_INVOKE(rtcd, downacross)(source->u_buffer, post->u_buffer, source->uv_stride, post->uv_stride, source->uv_height, source->uv_width, ppl); 292 POSTPROC_INVOKE(rtcd, downacross)(source->v_buffer, post->v_buffer, source->uv_stride, post->uv_stride, source->uv_height, source->uv_width, ppl); 293 } 294 295 void vp8_de_noise(YV12_BUFFER_CONFIG *source, 296 YV12_BUFFER_CONFIG *post, 297 int q, 298 int low_var_thresh, 299 int flag, 300 vp8_postproc_rtcd_vtable_t *rtcd) 301 { 302 double level = 6.0e-05 * q * q * q - .0067 * q * q + .306 * q + .0065; 303 int ppl = (int)(level + .5); 304 (void) post; 305 (void) low_var_thresh; 306 (void) flag; 307 308 POSTPROC_INVOKE(rtcd, downacross)( 309 source->y_buffer + 2 * source->y_stride + 2, 310 source->y_buffer + 2 * source->y_stride + 2, 311 source->y_stride, 312 source->y_stride, 313 source->y_height - 4, 314 source->y_width - 4, 315 ppl); 316 POSTPROC_INVOKE(rtcd, downacross)( 317 source->u_buffer + 2 * source->uv_stride + 2, 318 source->u_buffer + 2 * source->uv_stride + 2, 319 source->uv_stride, 320 source->uv_stride, 321 source->uv_height - 4, 322 source->uv_width - 4, ppl); 323 POSTPROC_INVOKE(rtcd, downacross)( 324 source->v_buffer + 2 * source->uv_stride + 2, 325 source->v_buffer + 2 * source->uv_stride + 2, 326 source->uv_stride, 327 source->uv_stride, 328 source->uv_height - 4, 329 source->uv_width - 4, ppl); 330 331 } 332 333 double vp8_gaussian(double sigma, double mu, double x) 334 { 335 return 1 / (sigma * sqrt(2.0 * 3.14159265)) * 336 (exp(-(x - mu) * (x - mu) / (2 * sigma * sigma))); 337 } 338 339 extern void (*vp8_clear_system_state)(void); 340 341 342 static void fillrd(struct postproc_state *state, int q, int a) 343 { 344 char char_dist[300]; 345 346 double sigma; 347 int ai = a, qi = q, i; 348 349 vp8_clear_system_state(); 350 351 352 sigma = ai + .5 + .6 * (63 - qi) / 63.0; 353 354 // set up a lookup table of 256 entries that matches 355 // a gaussian distribution with sigma determined by q. 356 // 357 { 358 double i; 359 int next, j; 360 361 next = 0; 362 363 for (i = -32; i < 32; i++) 364 { 365 int a = (int)(.5 + 256 * vp8_gaussian(sigma, 0, i)); 366 367 if (a) 368 { 369 for (j = 0; j < a; j++) 370 { 371 char_dist[next+j] = (char) i; 372 } 373 374 next = next + j; 375 } 376 377 } 378 379 for (next = next; next < 256; next++) 380 char_dist[next] = 0; 381 382 } 383 384 for (i = 0; i < 3072; i++) 385 { 386 state->noise[i] = char_dist[rand() & 0xff]; 387 } 388 389 for (i = 0; i < 16; i++) 390 { 391 state->blackclamp[i] = -char_dist[0]; 392 state->whiteclamp[i] = -char_dist[0]; 393 state->bothclamp[i] = -2 * char_dist[0]; 394 } 395 396 state->last_q = q; 397 state->last_noise = a; 398 } 399 400 /**************************************************************************** 401 * 402 * ROUTINE : plane_add_noise_c 403 * 404 * INPUTS : unsigned char *Start starting address of buffer to add gaussian 405 * noise to 406 * unsigned int Width width of plane 407 * unsigned int Height height of plane 408 * int Pitch distance between subsequent lines of frame 409 * int q quantizer used to determine amount of noise 410 * to add 411 * 412 * OUTPUTS : None. 413 * 414 * RETURNS : void. 415 * 416 * FUNCTION : adds gaussian noise to a plane of pixels 417 * 418 * SPECIAL NOTES : None. 419 * 420 ****************************************************************************/ 421 void vp8_plane_add_noise_c(unsigned char *Start, char *noise, 422 char blackclamp[16], 423 char whiteclamp[16], 424 char bothclamp[16], 425 unsigned int Width, unsigned int Height, int Pitch) 426 { 427 unsigned int i, j; 428 429 for (i = 0; i < Height; i++) 430 { 431 unsigned char *Pos = Start + i * Pitch; 432 char *Ref = (char *)(noise + (rand() & 0xff)); 433 434 for (j = 0; j < Width; j++) 435 { 436 if (Pos[j] < blackclamp[0]) 437 Pos[j] = blackclamp[0]; 438 439 if (Pos[j] > 255 + whiteclamp[0]) 440 Pos[j] = 255 + whiteclamp[0]; 441 442 Pos[j] += Ref[j]; 443 } 444 } 445 } 446 447 #if CONFIG_RUNTIME_CPU_DETECT 448 #define RTCD_VTABLE(oci) (&(oci)->rtcd.postproc) 449 #else 450 #define RTCD_VTABLE(oci) NULL 451 #endif 452 453 int vp8_post_proc_frame(VP8_COMMON *oci, YV12_BUFFER_CONFIG *dest, int deblock_level, int noise_level, int flags) 454 { 455 char message[512]; 456 int q = oci->filter_level * 10 / 6; 457 458 if (!oci->frame_to_show) 459 return -1; 460 461 if (q > 63) 462 q = 63; 463 464 if (!flags) 465 { 466 *dest = *oci->frame_to_show; 467 468 // handle problem with extending borders 469 dest->y_width = oci->Width; 470 dest->y_height = oci->Height; 471 dest->uv_height = dest->y_height / 2; 472 return 0; 473 474 } 475 476 #if ARCH_X86||ARCH_X86_64 477 vpx_reset_mmx_state(); 478 #endif 479 480 if (flags & VP8D_DEMACROBLOCK) 481 { 482 vp8_deblock_and_de_macro_block(oci->frame_to_show, &oci->post_proc_buffer, 483 q + (deblock_level - 5) * 10, 1, 0, RTCD_VTABLE(oci)); 484 } 485 else if (flags & VP8D_DEBLOCK) 486 { 487 vp8_deblock(oci->frame_to_show, &oci->post_proc_buffer, 488 q, 1, 0, RTCD_VTABLE(oci)); 489 } 490 else 491 { 492 vp8_yv12_copy_frame_ptr(oci->frame_to_show, &oci->post_proc_buffer); 493 } 494 495 if (flags & VP8D_ADDNOISE) 496 { 497 if (oci->postproc_state.last_q != q 498 || oci->postproc_state.last_noise != noise_level) 499 { 500 fillrd(&oci->postproc_state, 63 - q, noise_level); 501 } 502 503 POSTPROC_INVOKE(RTCD_VTABLE(oci), addnoise) 504 (oci->post_proc_buffer.y_buffer, 505 oci->postproc_state.noise, 506 oci->postproc_state.blackclamp, 507 oci->postproc_state.whiteclamp, 508 oci->postproc_state.bothclamp, 509 oci->post_proc_buffer.y_width, oci->post_proc_buffer.y_height, 510 oci->post_proc_buffer.y_stride); 511 } 512 513 if (flags & VP8D_DEBUG_LEVEL1) 514 { 515 sprintf(message, "F%1dG%1dQ%3dF%3dP%d_s%dx%d", 516 (oci->frame_type == KEY_FRAME), 517 oci->refresh_golden_frame, 518 oci->base_qindex, 519 oci->filter_level, 520 flags, 521 oci->mb_cols, oci->mb_rows); 522 vp8_blit_text(message, oci->post_proc_buffer.y_buffer, oci->post_proc_buffer.y_stride); 523 } 524 else if (flags & VP8D_DEBUG_LEVEL2) 525 { 526 int i, j; 527 unsigned char *y_ptr; 528 YV12_BUFFER_CONFIG *post = &oci->post_proc_buffer; 529 int mb_rows = post->y_height >> 4; 530 int mb_cols = post->y_width >> 4; 531 int mb_index = 0; 532 MODE_INFO *mi = oci->mi; 533 534 y_ptr = post->y_buffer + 4 * post->y_stride + 4; 535 536 // vp8_filter each macro block 537 for (i = 0; i < mb_rows; i++) 538 { 539 for (j = 0; j < mb_cols; j++) 540 { 541 char zz[4]; 542 543 sprintf(zz, "%c", mi[mb_index].mbmi.mode + 'a'); 544 545 vp8_blit_text(zz, y_ptr, post->y_stride); 546 mb_index ++; 547 y_ptr += 16; 548 } 549 550 mb_index ++; //border 551 y_ptr += post->y_stride * 16 - post->y_width; 552 553 } 554 } 555 else if (flags & VP8D_DEBUG_LEVEL3) 556 { 557 int i, j; 558 unsigned char *y_ptr; 559 YV12_BUFFER_CONFIG *post = &oci->post_proc_buffer; 560 int mb_rows = post->y_height >> 4; 561 int mb_cols = post->y_width >> 4; 562 int mb_index = 0; 563 MODE_INFO *mi = oci->mi; 564 565 y_ptr = post->y_buffer + 4 * post->y_stride + 4; 566 567 // vp8_filter each macro block 568 for (i = 0; i < mb_rows; i++) 569 { 570 for (j = 0; j < mb_cols; j++) 571 { 572 char zz[4]; 573 574 if (oci->frame_type == KEY_FRAME) 575 sprintf(zz, "a"); 576 else 577 sprintf(zz, "%c", mi[mb_index].mbmi.dc_diff + '0'); 578 579 vp8_blit_text(zz, y_ptr, post->y_stride); 580 mb_index ++; 581 y_ptr += 16; 582 } 583 584 mb_index ++; //border 585 y_ptr += post->y_stride * 16 - post->y_width; 586 587 } 588 } 589 else if (flags & VP8D_DEBUG_LEVEL4) 590 { 591 sprintf(message, "Bitrate: %10.2f frame_rate: %10.2f ", oci->bitrate, oci->framerate); 592 vp8_blit_text(message, oci->post_proc_buffer.y_buffer, oci->post_proc_buffer.y_stride); 593 #if 0 594 int i, j; 595 unsigned char *y_ptr; 596 YV12_BUFFER_CONFIG *post = &oci->post_proc_buffer; 597 int mb_rows = post->y_height >> 4; 598 int mb_cols = post->y_width >> 4; 599 int mb_index = 0; 600 MODE_INFO *mi = oci->mi; 601 602 y_ptr = post->y_buffer + 4 * post->y_stride + 4; 603 604 // vp8_filter each macro block 605 for (i = 0; i < mb_rows; i++) 606 { 607 for (j = 0; j < mb_cols; j++) 608 { 609 char zz[4]; 610 611 sprintf(zz, "%c", mi[mb_index].mbmi.dc_diff + '0'); 612 vp8_blit_text(zz, y_ptr, post->y_stride); 613 mb_index ++; 614 y_ptr += 16; 615 } 616 617 mb_index ++; //border 618 y_ptr += post->y_stride * 16 - post->y_width; 619 620 } 621 622 #endif 623 624 } 625 626 627 628 *dest = oci->post_proc_buffer; 629 630 // handle problem with extending borders 631 dest->y_width = oci->Width; 632 dest->y_height = oci->Height; 633 dest->uv_height = dest->y_height / 2; 634 return 0; 635 } 636