postproc.c 42.3 KB
Newer Older
John Koleszar's avatar
John Koleszar committed
1
/*
2
 *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
John Koleszar's avatar
John Koleszar committed
3
 *
4
 *  Use of this source code is governed by a BSD-style license
5 6
 *  that can be found in the LICENSE file in the root of the source
 *  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.
John Koleszar's avatar
John Koleszar committed
9 10 11
 */


12
#include "vpx_config.h"
John Koleszar's avatar
John Koleszar committed
13 14
#include "vpx_scale/yv12config.h"
#include "postproc.h"
15 16
#include "common.h"
#include "recon.h"
John Koleszar's avatar
John Koleszar committed
17 18 19
#include "vpx_scale/yv12extend.h"
#include "vpx_scale/vpxscale.h"
#include "systemdependent.h"
20
#include "../encoder/variance.h"
John Koleszar's avatar
John Koleszar committed
21 22 23 24

#include <math.h>
#include <stdlib.h>
#include <stdio.h>
25 26 27 28 29 30

#define RGB_TO_YUV(t)                                                                       \
    ( (0.257*(float)(t>>16)) + (0.504*(float)(t>>8&0xff)) + (0.098*(float)(t&0xff)) + 16),  \
    (-(0.148*(float)(t>>16)) - (0.291*(float)(t>>8&0xff)) + (0.439*(float)(t&0xff)) + 128), \
    ( (0.439*(float)(t>>16)) - (0.368*(float)(t>>8&0xff)) - (0.071*(float)(t&0xff)) + 128)

31
/* global constants */
32
#define MFQE_PRECISION 4
33
#if CONFIG_POSTPROC_VISUALIZER
34 35
static const unsigned char MB_PREDICTION_MODE_colors[MB_MODE_COUNT][3] =
{
36 37 38 39 40 41 42 43 44 45
    { RGB_TO_YUV(0x98FB98) },   /* PaleGreen */
    { RGB_TO_YUV(0x00FF00) },   /* Green */
    { RGB_TO_YUV(0xADFF2F) },   /* GreenYellow */
    { RGB_TO_YUV(0x228B22) },   /* ForestGreen */
    { RGB_TO_YUV(0x006400) },   /* DarkGreen */
    { RGB_TO_YUV(0x98F5FF) },   /* Cadet Blue */
    { RGB_TO_YUV(0x6CA6CD) },   /* Sky Blue */
    { RGB_TO_YUV(0x00008B) },   /* Dark blue */
    { RGB_TO_YUV(0x551A8B) },   /* Purple */
    { RGB_TO_YUV(0xFF0000) }    /* Red */
46 47
};

48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
static const unsigned char B_PREDICTION_MODE_colors[B_MODE_COUNT][3] =
{
    { RGB_TO_YUV(0x6633ff) },   /* Purple */
    { RGB_TO_YUV(0xcc33ff) },   /* Magenta */
    { RGB_TO_YUV(0xff33cc) },   /* Pink */
    { RGB_TO_YUV(0xff3366) },   /* Coral */
    { RGB_TO_YUV(0x3366ff) },   /* Blue */
    { RGB_TO_YUV(0xed00f5) },   /* Dark Blue */
    { RGB_TO_YUV(0x2e00b8) },   /* Dark Purple */
    { RGB_TO_YUV(0xff6633) },   /* Orange */
    { RGB_TO_YUV(0x33ccff) },   /* Light Blue */
    { RGB_TO_YUV(0x8ab800) },   /* Green */
    { RGB_TO_YUV(0xffcc33) },   /* Light Orange */
    { RGB_TO_YUV(0x33ffcc) },   /* Aqua */
    { RGB_TO_YUV(0x66ff33) },   /* Light Green */
    { RGB_TO_YUV(0xccff33) },   /* Yellow */
};

66
static const unsigned char MV_REFERENCE_FRAME_colors[MAX_REF_FRAMES][3] =
67
{
68 69 70 71
    { RGB_TO_YUV(0x00ff00) },   /* Blue */
    { RGB_TO_YUV(0x0000ff) },   /* Green */
    { RGB_TO_YUV(0xffff00) },   /* Yellow */
    { RGB_TO_YUV(0xff0000) },   /* Red */
72
};
73
#endif
74

John Koleszar's avatar
John Koleszar committed
75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128
static const short kernel5[] =
{
    1, 1, 4, 1, 1
};

const short vp8_rv[] =
{
    8, 5, 2, 2, 8, 12, 4, 9, 8, 3,
    0, 3, 9, 0, 0, 0, 8, 3, 14, 4,
    10, 1, 11, 14, 1, 14, 9, 6, 12, 11,
    8, 6, 10, 0, 0, 8, 9, 0, 3, 14,
    8, 11, 13, 4, 2, 9, 0, 3, 9, 6,
    1, 2, 3, 14, 13, 1, 8, 2, 9, 7,
    3, 3, 1, 13, 13, 6, 6, 5, 2, 7,
    11, 9, 11, 8, 7, 3, 2, 0, 13, 13,
    14, 4, 12, 5, 12, 10, 8, 10, 13, 10,
    4, 14, 4, 10, 0, 8, 11, 1, 13, 7,
    7, 14, 6, 14, 13, 2, 13, 5, 4, 4,
    0, 10, 0, 5, 13, 2, 12, 7, 11, 13,
    8, 0, 4, 10, 7, 2, 7, 2, 2, 5,
    3, 4, 7, 3, 3, 14, 14, 5, 9, 13,
    3, 14, 3, 6, 3, 0, 11, 8, 13, 1,
    13, 1, 12, 0, 10, 9, 7, 6, 2, 8,
    5, 2, 13, 7, 1, 13, 14, 7, 6, 7,
    9, 6, 10, 11, 7, 8, 7, 5, 14, 8,
    4, 4, 0, 8, 7, 10, 0, 8, 14, 11,
    3, 12, 5, 7, 14, 3, 14, 5, 2, 6,
    11, 12, 12, 8, 0, 11, 13, 1, 2, 0,
    5, 10, 14, 7, 8, 0, 4, 11, 0, 8,
    0, 3, 10, 5, 8, 0, 11, 6, 7, 8,
    10, 7, 13, 9, 2, 5, 1, 5, 10, 2,
    4, 3, 5, 6, 10, 8, 9, 4, 11, 14,
    0, 10, 0, 5, 13, 2, 12, 7, 11, 13,
    8, 0, 4, 10, 7, 2, 7, 2, 2, 5,
    3, 4, 7, 3, 3, 14, 14, 5, 9, 13,
    3, 14, 3, 6, 3, 0, 11, 8, 13, 1,
    13, 1, 12, 0, 10, 9, 7, 6, 2, 8,
    5, 2, 13, 7, 1, 13, 14, 7, 6, 7,
    9, 6, 10, 11, 7, 8, 7, 5, 14, 8,
    4, 4, 0, 8, 7, 10, 0, 8, 14, 11,
    3, 12, 5, 7, 14, 3, 14, 5, 2, 6,
    11, 12, 12, 8, 0, 11, 13, 1, 2, 0,
    5, 10, 14, 7, 8, 0, 4, 11, 0, 8,
    0, 3, 10, 5, 8, 0, 11, 6, 7, 8,
    10, 7, 13, 9, 2, 5, 1, 5, 10, 2,
    4, 3, 5, 6, 10, 8, 9, 4, 11, 14,
    3, 8, 3, 7, 8, 5, 11, 4, 12, 3,
    11, 9, 14, 8, 14, 13, 4, 3, 1, 2,
    14, 6, 5, 4, 4, 11, 4, 6, 2, 1,
    5, 8, 8, 12, 13, 5, 14, 10, 12, 13,
    0, 9, 5, 5, 11, 10, 13, 9, 10, 13,
};

extern void vp8_blit_text(const char *msg, unsigned char *address, const int pitch);
129
extern void vp8_blit_line(int x0, int x1, int y0, int y1, unsigned char *image, const int pitch);
John Koleszar's avatar
John Koleszar committed
130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153
/***********************************************************************************************************
 */
void vp8_post_proc_down_and_across_c
(
    unsigned char *src_ptr,
    unsigned char *dst_ptr,
    int src_pixels_per_line,
    int dst_pixels_per_line,
    int rows,
    int cols,
    int flimit
)
{
    unsigned char *p_src, *p_dst;
    int row;
    int col;
    int i;
    int v;
    int pitch = src_pixels_per_line;
    unsigned char d[8];
    (void)dst_pixels_per_line;

    for (row = 0; row < rows; row++)
    {
154
        /* post_proc_down for one row */
John Koleszar's avatar
John Koleszar committed
155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176
        p_src = src_ptr;
        p_dst = dst_ptr;

        for (col = 0; col < cols; col++)
        {

            int kernel = 4;
            int v = p_src[col];

            for (i = -2; i <= 2; i++)
            {
                if (abs(v - p_src[col+i*pitch]) > flimit)
                    goto down_skip_convolve;

                kernel += kernel5[2+i] * p_src[col+i*pitch];
            }

            v = (kernel >> 3);
        down_skip_convolve:
            p_dst[col] = v;
        }

177
        /* now post_proc_across */
John Koleszar's avatar
John Koleszar committed
178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205
        p_src = dst_ptr;
        p_dst = dst_ptr;

        for (i = 0; i < 8; i++)
            d[i] = p_src[i];

        for (col = 0; col < cols; col++)
        {
            int kernel = 4;
            v = p_src[col];

            d[col&7] = v;

            for (i = -2; i <= 2; i++)
            {
                if (abs(v - p_src[col+i]) > flimit)
                    goto across_skip_convolve;

                kernel += kernel5[2+i] * p_src[col+i];
            }

            d[col&7] = (kernel >> 3);
        across_skip_convolve:

            if (col >= 2)
                p_dst[col-2] = d[(col-2)&7];
        }

206
        /* handle the last two pixels */
John Koleszar's avatar
John Koleszar committed
207 208 209 210
        p_dst[col-2] = d[(col-2)&7];
        p_dst[col-1] = d[(col-1)&7];


211
        /* next row */
John Koleszar's avatar
John Koleszar committed
212 213 214 215 216
        src_ptr += pitch;
        dst_ptr += pitch;
    }
}

217
static int q2mbl(int x)
John Koleszar's avatar
John Koleszar committed
218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319
{
    if (x < 20) x = 20;

    x = 50 + (x - 50) * 10 / 8;
    return x * x / 3;
}
void vp8_mbpost_proc_across_ip_c(unsigned char *src, int pitch, int rows, int cols, int flimit)
{
    int r, c, i;

    unsigned char *s = src;
    unsigned char d[16];


    for (r = 0; r < rows; r++)
    {
        int sumsq = 0;
        int sum   = 0;

        for (i = -8; i <= 6; i++)
        {
            sumsq += s[i] * s[i];
            sum   += s[i];
            d[i+8] = 0;
        }

        for (c = 0; c < cols + 8; c++)
        {
            int x = s[c+7] - s[c-8];
            int y = s[c+7] + s[c-8];

            sum  += x;
            sumsq += x * y;

            d[c&15] = s[c];

            if (sumsq * 15 - sum * sum < flimit)
            {
                d[c&15] = (8 + sum + s[c]) >> 4;
            }

            s[c-8] = d[(c-8)&15];
        }

        s += pitch;
    }
}





void vp8_mbpost_proc_down_c(unsigned char *dst, int pitch, int rows, int cols, int flimit)
{
    int r, c, i;
    const short *rv3 = &vp8_rv[63&rand()];

    for (c = 0; c < cols; c++)
    {
        unsigned char *s = &dst[c];
        int sumsq = 0;
        int sum   = 0;
        unsigned char d[16];
        const short *rv2 = rv3 + ((c * 17) & 127);

        for (i = -8; i <= 6; i++)
        {
            sumsq += s[i*pitch] * s[i*pitch];
            sum   += s[i*pitch];
        }

        for (r = 0; r < rows + 8; r++)
        {
            sumsq += s[7*pitch] * s[ 7*pitch] - s[-8*pitch] * s[-8*pitch];
            sum  += s[7*pitch] - s[-8*pitch];
            d[r&15] = s[0];

            if (sumsq * 15 - sum * sum < flimit)
            {
                d[r&15] = (rv2[r&127] + sum + s[0]) >> 4;
            }

            s[-8*pitch] = d[(r-8)&15];
            s += pitch;
        }
    }
}


static void vp8_deblock_and_de_macro_block(YV12_BUFFER_CONFIG         *source,
        YV12_BUFFER_CONFIG         *post,
        int                         q,
        int                         low_var_thresh,
        int                         flag,
        vp8_postproc_rtcd_vtable_t *rtcd)
{
    double level = 6.0e-05 * q * q * q - .0067 * q * q + .306 * q + .0065;
    int ppl = (int)(level + .5);
    (void) low_var_thresh;
    (void) flag;

    POSTPROC_INVOKE(rtcd, downacross)(source->y_buffer, post->y_buffer, source->y_stride,  post->y_stride, source->y_height, source->y_width,  ppl);
320 321
    POSTPROC_INVOKE(rtcd, across)(post->y_buffer, post->y_stride, post->y_height, post->y_width, q2mbl(q));
    POSTPROC_INVOKE(rtcd, down)(post->y_buffer, post->y_stride, post->y_height, post->y_width, q2mbl(q));
John Koleszar's avatar
John Koleszar committed
322 323 324 325 326 327

    POSTPROC_INVOKE(rtcd, downacross)(source->u_buffer, post->u_buffer, source->uv_stride, post->uv_stride, source->uv_height, source->uv_width, ppl);
    POSTPROC_INVOKE(rtcd, downacross)(source->v_buffer, post->v_buffer, source->uv_stride, post->uv_stride, source->uv_height, source->uv_width, ppl);

}

Luca Barbato's avatar
Luca Barbato committed
328
void vp8_deblock(YV12_BUFFER_CONFIG         *source,
329 330 331 332 333
                 YV12_BUFFER_CONFIG         *post,
                 int                         q,
                 int                         low_var_thresh,
                 int                         flag,
                 vp8_postproc_rtcd_vtable_t *rtcd)
John Koleszar's avatar
John Koleszar committed
334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403
{
    double level = 6.0e-05 * q * q * q - .0067 * q * q + .306 * q + .0065;
    int ppl = (int)(level + .5);
    (void) low_var_thresh;
    (void) flag;

    POSTPROC_INVOKE(rtcd, downacross)(source->y_buffer, post->y_buffer, source->y_stride,  post->y_stride, source->y_height, source->y_width,   ppl);
    POSTPROC_INVOKE(rtcd, downacross)(source->u_buffer, post->u_buffer, source->uv_stride, post->uv_stride,  source->uv_height, source->uv_width, ppl);
    POSTPROC_INVOKE(rtcd, downacross)(source->v_buffer, post->v_buffer, source->uv_stride, post->uv_stride, source->uv_height, source->uv_width, ppl);
}

void vp8_de_noise(YV12_BUFFER_CONFIG         *source,
                  YV12_BUFFER_CONFIG         *post,
                  int                         q,
                  int                         low_var_thresh,
                  int                         flag,
                  vp8_postproc_rtcd_vtable_t *rtcd)
{
    double level = 6.0e-05 * q * q * q - .0067 * q * q + .306 * q + .0065;
    int ppl = (int)(level + .5);
    (void) post;
    (void) low_var_thresh;
    (void) flag;

    POSTPROC_INVOKE(rtcd, downacross)(
        source->y_buffer + 2 * source->y_stride + 2,
        source->y_buffer + 2 * source->y_stride + 2,
        source->y_stride,
        source->y_stride,
        source->y_height - 4,
        source->y_width - 4,
        ppl);
    POSTPROC_INVOKE(rtcd, downacross)(
        source->u_buffer + 2 * source->uv_stride + 2,
        source->u_buffer + 2 * source->uv_stride + 2,
        source->uv_stride,
        source->uv_stride,
        source->uv_height - 4,
        source->uv_width - 4, ppl);
    POSTPROC_INVOKE(rtcd, downacross)(
        source->v_buffer + 2 * source->uv_stride + 2,
        source->v_buffer + 2 * source->uv_stride + 2,
        source->uv_stride,
        source->uv_stride,
        source->uv_height - 4,
        source->uv_width - 4, ppl);

}

double vp8_gaussian(double sigma, double mu, double x)
{
    return 1 / (sigma * sqrt(2.0 * 3.14159265)) *
           (exp(-(x - mu) * (x - mu) / (2 * sigma * sigma)));
}

extern void (*vp8_clear_system_state)(void);


static void fillrd(struct postproc_state *state, int q, int a)
{
    char char_dist[300];

    double sigma;
    int ai = a, qi = q, i;

    vp8_clear_system_state();


    sigma = ai + .5 + .6 * (63 - qi) / 63.0;

404 405 406
    /* set up a lookup table of 256 entries that matches
     * a gaussian distribution with sigma determined by q.
     */
John Koleszar's avatar
John Koleszar committed
407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496
    {
        double i;
        int next, j;

        next = 0;

        for (i = -32; i < 32; i++)
        {
            int a = (int)(.5 + 256 * vp8_gaussian(sigma, 0, i));

            if (a)
            {
                for (j = 0; j < a; j++)
                {
                    char_dist[next+j] = (char) i;
                }

                next = next + j;
            }

        }

        for (next = next; next < 256; next++)
            char_dist[next] = 0;

    }

    for (i = 0; i < 3072; i++)
    {
        state->noise[i] = char_dist[rand() & 0xff];
    }

    for (i = 0; i < 16; i++)
    {
        state->blackclamp[i] = -char_dist[0];
        state->whiteclamp[i] = -char_dist[0];
        state->bothclamp[i] = -2 * char_dist[0];
    }

    state->last_q = q;
    state->last_noise = a;
}

/****************************************************************************
 *
 *  ROUTINE       : plane_add_noise_c
 *
 *  INPUTS        : unsigned char *Start    starting address of buffer to add gaussian
 *                                  noise to
 *                  unsigned int Width    width of plane
 *                  unsigned int Height   height of plane
 *                  int  Pitch    distance between subsequent lines of frame
 *                  int  q        quantizer used to determine amount of noise
 *                                  to add
 *
 *  OUTPUTS       : None.
 *
 *  RETURNS       : void.
 *
 *  FUNCTION      : adds gaussian noise to a plane of pixels
 *
 *  SPECIAL NOTES : None.
 *
 ****************************************************************************/
void vp8_plane_add_noise_c(unsigned char *Start, char *noise,
                           char blackclamp[16],
                           char whiteclamp[16],
                           char bothclamp[16],
                           unsigned int Width, unsigned int Height, int Pitch)
{
    unsigned int i, j;

    for (i = 0; i < Height; i++)
    {
        unsigned char *Pos = Start + i * Pitch;
        char  *Ref = (char *)(noise + (rand() & 0xff));

        for (j = 0; j < Width; j++)
        {
            if (Pos[j] < blackclamp[0])
                Pos[j] = blackclamp[0];

            if (Pos[j] > 255 + whiteclamp[0])
                Pos[j] = 255 + whiteclamp[0];

            Pos[j] += Ref[j];
        }
    }
}

497 498 499 500
/* Blend the macro block with a solid colored square.  Leave the
 * edges unblended to give distinction to macro blocks in areas
 * filled with the same color block.
 */
501
void vp8_blend_mb_inner_c (unsigned char *y, unsigned char *u, unsigned char *v,
502 503 504 505 506 507 508
                        int y1, int u1, int v1, int alpha, int stride)
{
    int i, j;
    int y1_const = y1*((1<<16)-alpha);
    int u1_const = u1*((1<<16)-alpha);
    int v1_const = v1*((1<<16)-alpha);

509 510
    y += 2*stride + 2;
    for (i = 0; i < 12; i++)
511
    {
512
        for (j = 0; j < 12; j++)
513 514 515 516 517 518
        {
            y[j] = (y[j]*alpha + y1_const)>>16;
        }
        y += stride;
    }

519 520 521 522 523 524
    stride >>= 1;

    u += stride + 1;
    v += stride + 1;

    for (i = 0; i < 6; i++)
525
    {
526
        for (j = 0; j < 6; j++)
527 528 529 530
        {
            u[j] = (u[j]*alpha + u1_const)>>16;
            v[j] = (v[j]*alpha + v1_const)>>16;
        }
531 532
        u += stride;
        v += stride;
533 534
    }
}
John Koleszar's avatar
John Koleszar committed
535

536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633
/* Blend only the edge of the macro block.  Leave center
 * unblended to allow for other visualizations to be layered.
 */
void vp8_blend_mb_outer_c (unsigned char *y, unsigned char *u, unsigned char *v,
                        int y1, int u1, int v1, int alpha, int stride)
{
    int i, j;
    int y1_const = y1*((1<<16)-alpha);
    int u1_const = u1*((1<<16)-alpha);
    int v1_const = v1*((1<<16)-alpha);

    for (i = 0; i < 2; i++)
    {
        for (j = 0; j < 16; j++)
        {
            y[j] = (y[j]*alpha + y1_const)>>16;
        }
        y += stride;
    }

    for (i = 0; i < 12; i++)
    {
        y[0]  = (y[0]*alpha  + y1_const)>>16;
        y[1]  = (y[1]*alpha  + y1_const)>>16;
        y[14] = (y[14]*alpha + y1_const)>>16;
        y[15] = (y[15]*alpha + y1_const)>>16;
        y += stride;
    }

    for (i = 0; i < 2; i++)
    {
        for (j = 0; j < 16; j++)
        {
            y[j] = (y[j]*alpha + y1_const)>>16;
        }
        y += stride;
    }

    stride >>= 1;

    for (j = 0; j < 8; j++)
    {
        u[j] = (u[j]*alpha + u1_const)>>16;
        v[j] = (v[j]*alpha + v1_const)>>16;
    }
    u += stride;
    v += stride;

    for (i = 0; i < 6; i++)
    {
        u[0] = (u[0]*alpha + u1_const)>>16;
        v[0] = (v[0]*alpha + v1_const)>>16;

        u[7] = (u[7]*alpha + u1_const)>>16;
        v[7] = (v[7]*alpha + v1_const)>>16;

        u += stride;
        v += stride;
    }

    for (j = 0; j < 8; j++)
    {
        u[j] = (u[j]*alpha + u1_const)>>16;
        v[j] = (v[j]*alpha + v1_const)>>16;
    }
}

void vp8_blend_b_c (unsigned char *y, unsigned char *u, unsigned char *v,
                        int y1, int u1, int v1, int alpha, int stride)
{
    int i, j;
    int y1_const = y1*((1<<16)-alpha);
    int u1_const = u1*((1<<16)-alpha);
    int v1_const = v1*((1<<16)-alpha);

    for (i = 0; i < 4; i++)
    {
        for (j = 0; j < 4; j++)
        {
            y[j] = (y[j]*alpha + y1_const)>>16;
        }
        y += stride;
    }

    stride >>= 1;

    for (i = 0; i < 2; i++)
    {
        for (j = 0; j < 2; j++)
        {
            u[j] = (u[j]*alpha + u1_const)>>16;
            v[j] = (v[j]*alpha + v1_const)>>16;
        }
        u += stride;
        v += stride;
    }
}

634 635
static void constrain_line (int x0, int *x1, int y0, int *y1, int width, int height)
{
636 637
    int dx;
    int dy;
638 639 640

    if (*x1 > width)
    {
641 642 643
        dx = *x1 - x0;
        dy = *y1 - y0;

644
        *x1 = width;
645
        if (dx)
646 647 648 649
            *y1 = ((width-x0)*dy)/dx + y0;
    }
    if (*x1 < 0)
    {
650 651 652
        dx = *x1 - x0;
        dy = *y1 - y0;

653
        *x1 = 0;
654
        if (dx)
655 656 657 658
            *y1 = ((0-x0)*dy)/dx + y0;
    }
    if (*y1 > height)
    {
659 660 661
        dx = *x1 - x0;
        dy = *y1 - y0;

662
        *y1 = height;
663
        if (dy)
664 665 666 667
            *x1 = ((height-y0)*dx)/dy + x0;
    }
    if (*y1 < 0)
    {
668 669 670
        dx = *x1 - x0;
        dy = *y1 - y0;

671
        *y1 = 0;
672
        if (dy)
673 674 675 676
            *x1 = ((0-y0)*dx)/dy + x0;
    }
}

677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694
int vp8_references_buffer( VP8_COMMON *oci, int ref_frame )
{
    const MODE_INFO *mi = oci->mi;
    int mb_row, mb_col;

    for (mb_row = 0; mb_row < oci->mb_rows; mb_row++)
    {
        for (mb_col = 0; mb_col < oci->mb_cols; mb_col++,mi++)
        {
            if( mi->mbmi.ref_frame == ref_frame)
              return 1;
        }
        mi++;
    }
    return 0;

}

695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748
static void multiframe_quality_enhance_block
(
    int blksize, /* Currently only values supported are 16, 8, 4 */
    int qcurr,
    int qprev,
    unsigned char *y,
    unsigned char *u,
    unsigned char *v,
    int y_stride,
    int uv_stride,
    unsigned char *yd,
    unsigned char *ud,
    unsigned char *vd,
    int yd_stride,
    int uvd_stride
)
{
    static const unsigned char VP8_ZEROS[16]=
    {
         0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
    };
    int blksizeby2 = blksize >> 1;
    int blksizesq = blksize * blksize;

    int i, j;
    unsigned char *yp;
    unsigned char *ydp;
    unsigned char *up;
    unsigned char *udp;
    unsigned char *vp;
    unsigned char *vdp;

    unsigned int act, sse, sad, thr;
    if (blksize == 16)
    {
        act = vp8_variance_var16x16(y, y_stride, VP8_ZEROS, 0, &sse);
        sad = vp8_variance_sad16x16(y, y_stride, yd, yd_stride, 0);
    }
    else if (blksize == 8)
    {
        act = vp8_variance_var8x8(y, y_stride, VP8_ZEROS, 0, &sse);
        sad = vp8_variance_sad8x8(y, y_stride, yd, yd_stride, 0);
    }
    else
    {
        act = vp8_variance_var4x4(y, y_stride, VP8_ZEROS, 0, &sse);
        sad = vp8_variance_sad4x4(y, y_stride, yd, yd_stride, 0);
    }

    thr = 6 * blksizesq + (act >> 3);
    if (thr > 12 * blksizesq) thr = 12 * blksizesq;
    // These thresholds should be adapted later based on qcurr and qprev
    if (sad < thr)
    {
749 750
        static const int roundoff = (1 << (MFQE_PRECISION - 1));
        int ifactor = (sad << MFQE_PRECISION) / thr;
751 752 753
        // TODO: SIMD optimize this section
        if (ifactor)
        {
754
            int icfactor = (1 << MFQE_PRECISION) - ifactor;
755 756 757
            for (yp = y, ydp = yd, i = 0; i < blksize; ++i, yp += y_stride, ydp += yd_stride)
            {
                for (j = 0; j < blksize; ++j)
758
                    ydp[j] = (int)((yp[j] * ifactor + ydp[j] * icfactor + roundoff) >> MFQE_PRECISION);
759 760 761 762
            }
            for (up = u, udp = ud, i = 0; i < blksizeby2; ++i, up += uv_stride, udp += uvd_stride)
            {
                for (j = 0; j < blksizeby2; ++j)
763
                    udp[j] = (int)((up[j] * ifactor + udp[j] * icfactor + roundoff) >> MFQE_PRECISION);
764 765 766 767
            }
            for (vp = v, vdp = vd, i = 0; i < blksizeby2; ++i, vp += uv_stride, vdp += uvd_stride)
            {
                for (j = 0; j < blksizeby2; ++j)
768
                    vdp[j] = (int)((vp[j] * ifactor + vdp[j] * icfactor + roundoff) >> MFQE_PRECISION);
769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799
            }
        }
    }
    else
    {
        if (blksize == 16)
        {
            vp8_recon_copy16x16(y, y_stride, yd, yd_stride);
            vp8_recon_copy8x8(u, uv_stride, ud, uvd_stride);
            vp8_recon_copy8x8(v, uv_stride, vd, uvd_stride);
        }
        else if (blksize == 8)
        {
            vp8_recon_copy8x8(y, y_stride, yd, yd_stride);
            for (up = u, udp = ud, i = 0; i < blksizeby2; ++i, up += uv_stride, udp += uvd_stride)
                vpx_memcpy(udp, up, blksizeby2);
            for (vp = v, vdp = vd, i = 0; i < blksizeby2; ++i, vp += uv_stride, vdp += uvd_stride)
                vpx_memcpy(vdp, vp, blksizeby2);
        }
        else
        {
            for (yp = y, ydp = yd, i = 0; i < blksize; ++i, yp += y_stride, ydp += yd_stride)
                vpx_memcpy(ydp, yp, blksize);
            for (up = u, udp = ud, i = 0; i < blksizeby2; ++i, up += uv_stride, udp += uvd_stride)
                vpx_memcpy(udp, up, blksizeby2);
            for (vp = v, vdp = vd, i = 0; i < blksizeby2; ++i, vp += uv_stride, vdp += uvd_stride)
                vpx_memcpy(vdp, vp, blksizeby2);
        }
    }
}

800 801 802 803 804 805
#if CONFIG_RUNTIME_CPU_DETECT
#define RTCD_VTABLE(oci) (&(oci)->rtcd.postproc)
#else
#define RTCD_VTABLE(oci) NULL
#endif

806 807 808 809 810 811 812 813 814 815 816 817 818
void vp8_multiframe_quality_enhance
(
    VP8_COMMON *cm
)
{
    YV12_BUFFER_CONFIG *show = cm->frame_to_show;
    YV12_BUFFER_CONFIG *dest = &cm->post_proc_buffer;

    FRAME_TYPE frame_type = cm->frame_type;
    /* Point at base of Mb MODE_INFO list has motion vectors etc */
    const MODE_INFO *mode_info_context = cm->mi;
    int mb_row;
    int mb_col;
819 820
    int qcurr = cm->base_qindex;
    int qprev = cm->postproc_state.last_base_qindex;
821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902

    unsigned char *y_ptr, *u_ptr, *v_ptr;
    unsigned char *yd_ptr, *ud_ptr, *vd_ptr;

    /* Set up the buffer pointers */
    y_ptr = show->y_buffer;
    u_ptr = show->u_buffer;
    v_ptr = show->v_buffer;
    yd_ptr = dest->y_buffer;
    ud_ptr = dest->u_buffer;
    vd_ptr = dest->v_buffer;

    /* postprocess each macro block */
    for (mb_row = 0; mb_row < cm->mb_rows; mb_row++)
    {
        for (mb_col = 0; mb_col < cm->mb_cols; mb_col++)
        {
            /* if motion is high there will likely be no benefit */
            if (((frame_type == INTER_FRAME &&
                  abs(mode_info_context->mbmi.mv.as_mv.row) <= 10 &&
                  abs(mode_info_context->mbmi.mv.as_mv.col) <= 10) ||
                 (frame_type == KEY_FRAME)) &&
                mode_info_context->mbmi.mode != B_PRED)
            {
                multiframe_quality_enhance_block(16,
                                                 qcurr,
                                                 qprev,
                                                 y_ptr,
                                                 u_ptr,
                                                 v_ptr,
                                                 show->y_stride,
                                                 show->uv_stride,
                                                 yd_ptr,
                                                 ud_ptr,
                                                 vd_ptr,
                                                 dest->y_stride,
                                                 dest->uv_stride);
            }
            else if (mode_info_context->mbmi.mode == B_PRED)
            {
                int i, j;
                for (i=0; i<2; ++i)
                    for (j=0; j<2; ++j)
                        multiframe_quality_enhance_block(8,
                                                         qcurr,
                                                         qprev,
                                                         y_ptr + 8*(i*show->y_stride+j),
                                                         u_ptr + 4*(i*show->uv_stride+j),
                                                         v_ptr + 4*(i*show->uv_stride+j),
                                                         show->y_stride,
                                                         show->uv_stride,
                                                         yd_ptr + 8*(i*dest->y_stride+j),
                                                         ud_ptr + 4*(i*dest->uv_stride+j),
                                                         vd_ptr + 4*(i*dest->uv_stride+j),
                                                         dest->y_stride,
                                                         dest->uv_stride);
            }
            else
            {
                vp8_recon_copy16x16(y_ptr, show->y_stride, yd_ptr, dest->y_stride);
                vp8_recon_copy8x8(u_ptr, show->uv_stride, ud_ptr, dest->uv_stride);
                vp8_recon_copy8x8(v_ptr, show->uv_stride, vd_ptr, dest->uv_stride);
            }
            y_ptr += 16;
            u_ptr += 8;
            v_ptr += 8;
            yd_ptr += 16;
            ud_ptr += 8;
            vd_ptr += 8;
            mode_info_context++;     /* step to next MB */
        }

        y_ptr += show->y_stride  * 16 - 16 * cm->mb_cols;
        u_ptr += show->uv_stride *  8 - 8 * cm->mb_cols;
        v_ptr += show->uv_stride *  8 - 8 * cm->mb_cols;
        yd_ptr += dest->y_stride  * 16 - 16 * cm->mb_cols;
        ud_ptr += dest->uv_stride *  8 - 8 * cm->mb_cols;
        vd_ptr += dest->uv_stride *  8 - 8 * cm->mb_cols;

        mode_info_context++;         /* Skip border mb */
    }
}
903

904
int vp8_post_proc_frame(VP8_COMMON *oci, YV12_BUFFER_CONFIG *dest, vp8_ppflags_t *ppflags)
John Koleszar's avatar
John Koleszar committed
905 906
{
    int q = oci->filter_level * 10 / 6;
907 908 909
    int flags = ppflags->post_proc_flag;
    int deblock_level = ppflags->deblocking_level;
    int noise_level = ppflags->noise_level;
John Koleszar's avatar
John Koleszar committed
910 911 912 913 914 915 916 917 918 919 920

    if (!oci->frame_to_show)
        return -1;

    if (q > 63)
        q = 63;

    if (!flags)
    {
        *dest = *oci->frame_to_show;

921
        /* handle problem with extending borders */
John Koleszar's avatar
John Koleszar committed
922 923 924
        dest->y_width = oci->Width;
        dest->y_height = oci->Height;
        dest->uv_height = dest->y_height / 2;
925
        oci->postproc_state.last_base_qindex = oci->base_qindex;
John Koleszar's avatar
John Koleszar committed
926 927 928
        return 0;
    }

929 930 931 932 933 934 935 936 937 938 939 940
    /* Allocate post_proc_buffer_int if needed */
    if ((flags & VP8D_MFQE) && !oci->post_proc_buffer_int_used)
    {
        if ((flags & VP8D_DEBLOCK) || (flags & VP8D_DEMACROBLOCK))
        {
            if (vp8_yv12_alloc_frame_buffer(&oci->post_proc_buffer_int, oci->Width, oci->Height, VP8BORDERINPIXELS) >= 0)
            {
                oci->post_proc_buffer_int_used = 1;
            }
        }
    }

John Koleszar's avatar
John Koleszar committed
941 942 943 944
#if ARCH_X86||ARCH_X86_64
    vpx_reset_mmx_state();
#endif

945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966
    if ((flags & VP8D_MFQE) &&
         oci->current_video_frame >= 2 &&
         oci->base_qindex - oci->postproc_state.last_base_qindex >= 10)
    {
        vp8_multiframe_quality_enhance(oci);
        if (((flags & VP8D_DEBLOCK) || (flags & VP8D_DEMACROBLOCK)) &&
            oci->post_proc_buffer_int_used)
        {
            vp8_yv12_copy_frame_ptr(&oci->post_proc_buffer, &oci->post_proc_buffer_int);
            if (flags & VP8D_DEMACROBLOCK)
            {
                vp8_deblock_and_de_macro_block(&oci->post_proc_buffer_int, &oci->post_proc_buffer,
                                               q + (deblock_level - 5) * 10, 1, 0, RTCD_VTABLE(oci));
            }
            else if (flags & VP8D_DEBLOCK)
            {
                vp8_deblock(&oci->post_proc_buffer_int, &oci->post_proc_buffer,
                            q, 1, 0, RTCD_VTABLE(oci));
            }
        }
    }
    else if (flags & VP8D_DEMACROBLOCK)
John Koleszar's avatar
John Koleszar committed
967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998
    {
        vp8_deblock_and_de_macro_block(oci->frame_to_show, &oci->post_proc_buffer,
                                       q + (deblock_level - 5) * 10, 1, 0, RTCD_VTABLE(oci));
    }
    else if (flags & VP8D_DEBLOCK)
    {
        vp8_deblock(oci->frame_to_show, &oci->post_proc_buffer,
                    q, 1, 0, RTCD_VTABLE(oci));
    }
    else
    {
        vp8_yv12_copy_frame_ptr(oci->frame_to_show, &oci->post_proc_buffer);
    }

    if (flags & VP8D_ADDNOISE)
    {
        if (oci->postproc_state.last_q != q
            || oci->postproc_state.last_noise != noise_level)
        {
            fillrd(&oci->postproc_state, 63 - q, noise_level);
        }

        POSTPROC_INVOKE(RTCD_VTABLE(oci), addnoise)
        (oci->post_proc_buffer.y_buffer,
         oci->postproc_state.noise,
         oci->postproc_state.blackclamp,
         oci->postproc_state.whiteclamp,
         oci->postproc_state.bothclamp,
         oci->post_proc_buffer.y_width, oci->post_proc_buffer.y_height,
         oci->post_proc_buffer.y_stride);
    }

999 1000
#if CONFIG_POSTPROC_VISUALIZER
    if (flags & VP8D_DEBUG_TXT_FRAME_INFO)
John Koleszar's avatar
John Koleszar committed
1001
    {
Henrik Lundin's avatar
Henrik Lundin committed
1002
        char message[512];
John Koleszar's avatar
John Koleszar committed
1003 1004 1005 1006 1007 1008 1009 1010 1011
        sprintf(message, "F%1dG%1dQ%3dF%3dP%d_s%dx%d",
                (oci->frame_type == KEY_FRAME),
                oci->refresh_golden_frame,
                oci->base_qindex,
                oci->filter_level,
                flags,
                oci->mb_cols, oci->mb_rows);
        vp8_blit_text(message, oci->post_proc_buffer.y_buffer, oci->post_proc_buffer.y_stride);
    }
1012

1013
    if (flags & VP8D_DEBUG_TXT_MBLK_MODES)
John Koleszar's avatar
John Koleszar committed
1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024
    {
        int i, j;
        unsigned char *y_ptr;
        YV12_BUFFER_CONFIG *post = &oci->post_proc_buffer;
        int mb_rows = post->y_height >> 4;
        int mb_cols = post->y_width  >> 4;
        int mb_index = 0;
        MODE_INFO *mi = oci->mi;

        y_ptr = post->y_buffer + 4 * post->y_stride + 4;

1025
        /* vp8_filter each macro block */
John Koleszar's avatar
John Koleszar committed
1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038
        for (i = 0; i < mb_rows; i++)
        {
            for (j = 0; j < mb_cols; j++)
            {
                char zz[4];

                sprintf(zz, "%c", mi[mb_index].mbmi.mode + 'a');

                vp8_blit_text(zz, y_ptr, post->y_stride);
                mb_index ++;
                y_ptr += 16;
            }

1039
            mb_index ++; /* border */
John Koleszar's avatar
John Koleszar committed
1040 1041 1042 1043
            y_ptr += post->y_stride  * 16 - post->y_width;

        }
    }
1044

1045
    if (flags & VP8D_DEBUG_TXT_DC_DIFF)
John Koleszar's avatar
John Koleszar committed
1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056
    {
        int i, j;
        unsigned char *y_ptr;
        YV12_BUFFER_CONFIG *post = &oci->post_proc_buffer;
        int mb_rows = post->y_height >> 4;
        int mb_cols = post->y_width  >> 4;
        int mb_index = 0;
        MODE_INFO *mi = oci->mi;

        y_ptr = post->y_buffer + 4 * post->y_stride + 4;

1057
        /* vp8_filter each macro block */
John Koleszar's avatar
John Koleszar committed
1058 1059 1060 1061 1062
        for (i = 0; i < mb_rows; i++)
        {
            for (j = 0; j < mb_cols; j++)
            {
                char zz[4];
1063 1064
                int dc_diff = !(mi[mb_index].mbmi.mode != B_PRED &&
                              mi[mb_index].mbmi.mode != SPLITMV &&
1065
                              mi[mb_index].mbmi.mb_skip_coeff);
John Koleszar's avatar
John Koleszar committed
1066 1067 1068 1069

                if (oci->frame_type == KEY_FRAME)
                    sprintf(zz, "a");
                else
1070
                    sprintf(zz, "%c", dc_diff + '0');
John Koleszar's avatar
John Koleszar committed
1071 1072 1073 1074 1075 1076

                vp8_blit_text(zz, y_ptr, post->y_stride);
                mb_index ++;
                y_ptr += 16;
            }

1077
            mb_index ++; /* border */
John Koleszar's avatar
John Koleszar committed
1078 1079 1080 1081
            y_ptr += post->y_stride  * 16 - post->y_width;

        }
    }
1082

1083
    if (flags & VP8D_DEBUG_TXT_RATE_INFO)
John Koleszar's avatar
John Koleszar committed
1084
    {
Henrik Lundin's avatar
Henrik Lundin committed
1085
        char message[512];
John Koleszar's avatar
John Koleszar committed
1086 1087 1088
        sprintf(message, "Bitrate: %10.2f frame_rate: %10.2f ", oci->bitrate, oci->framerate);
        vp8_blit_text(message, oci->post_proc_buffer.y_buffer, oci->post_proc_buffer.y_stride);
    }
1089

1090
    /* Draw motion vectors */
1091
    if ((flags & VP8D_DEBUG_DRAW_MV) && ppflags->display_mv_flag)
1092 1093 1094 1095 1096 1097 1098 1099 1100
    {
        YV12_BUFFER_CONFIG *post = &oci->post_proc_buffer;
        int width  = post->y_width;
        int height = post->y_height;
        unsigned char *y_buffer = oci->post_proc_buffer.y_buffer;
        int y_stride = oci->post_proc_buffer.y_stride;
        MODE_INFO *mi = oci->mi;
        int x0, y0;

1101
        for (y0 = 0; y0 < height; y0 += 16)
1102
        {
1103
            for (x0 = 0; x0 < width; x0 += 16)
1104
            {
1105 1106
                int x1, y1;

1107 1108 1109 1110 1111 1112
                if (!(ppflags->display_mv_flag & (1<<mi->mbmi.mode)))
                {
                    mi++;
                    continue;
                }

1113 1114
                if (mi->mbmi.mode == SPLITMV)
                {
1115 1116 1117 1118
                    switch (mi->mbmi.partitioning)
                    {
                        case 0 :    /* mv_top_bottom */
                        {
1119
                            union b_mode_info *bmi = &mi->bmi[0];
1120
                            MV *mv = &bmi->mv.as_mv;
1121

1122 1123
                            x1 = x0 + 8 + (mv->col >> 3);
                            y1 = y0 + 4 + (mv->row >> 3);
1124

1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139
                            constrain_line (x0+8, &x1, y0+4, &y1, width, height);
                            vp8_blit_line  (x0+8,  x1, y0+4,  y1, y_buffer, y_stride);

                            bmi = &mi->bmi[8];

                            x1 = x0 + 8 + (mv->col >> 3);
                            y1 = y0 +12 + (mv->row >> 3);

                            constrain_line (x0+8, &x1, y0+12, &y1, width, height);
                            vp8_blit_line  (x0+8,  x1, y0+12,  y1, y_buffer, y_stride);

                            break;
                        }
                        case 1 :    /* mv_left_right */
                        {
1140
                            union b_mode_info *bmi = &mi->bmi[0];
1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159
                            MV *mv = &bmi->mv.as_mv;

                            x1 = x0 + 4 + (mv->col >> 3);
                            y1 = y0 + 8 + (mv->row >> 3);

                            constrain_line (x0+4, &x1, y0+8, &y1, width, height);
                            vp8_blit_line  (x0+4,  x1, y0+8,  y1, y_buffer, y_stride);

                            bmi = &mi->bmi[2];

                            x1 = x0 +12 + (mv->col >> 3);
                            y1 = y0 + 8 + (mv->row >> 3);

                            constrain_line (x0+12, &x1, y0+8, &y1, width, height);
                            vp8_blit_line  (x0+12,  x1, y0+8,  y1, y_buffer, y_stride);

                            break;
                        }
                        case 2 :    /* mv_quarters   */
1160
                        {
1161
                            union b_mode_info *bmi = &mi->bmi[0];
1162
                            MV *mv = &bmi->mv.as_mv;
1163

1164 1165
                            x1 = x0 + 4 + (mv->col >> 3);
                            y1 = y0 + 4 + (mv->row >> 3);
1166

1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196
                            constrain_line (x0+4, &x1, y0+4, &y1, width, height);
                            vp8_blit_line  (x0+4,  x1, y0+4,  y1, y_buffer, y_stride);

                            bmi = &mi->bmi[2];

                            x1 = x0 +12 + (mv->col >> 3);
                            y1 = y0 + 4 + (mv->row >> 3);

                            constrain_line (x0+12, &x1, y0+4, &y1, width, height);
                            vp8_blit_line  (x0+12,  x1, y0+4,  y1, y_buffer, y_stride);

                            bmi = &mi->bmi[8];

                            x1 = x0 + 4 + (mv->col >> 3);
                            y1 = y0 +12 + (mv->row >> 3);

                            constrain_line (x0+4, &x1, y0+12, &y1, width, height);
                            vp8_blit_line  (x0+4,  x1, y0+12,  y1, y_buffer, y_stride);

                            bmi = &mi->bmi[10];

                            x1 = x0 +12 + (mv->col >> 3);
                            y1 = y0 +12 + (mv->row >> 3);

                            constrain_line (x0+12, &x1, y0+12, &y1, width, height);
                            vp8_blit_line  (x0+12,  x1, y0+12,  y1, y_buffer, y_stride);
                            break;
                        }
                        default :
                        {
1197
                            union b_mode_info *bmi = mi->bmi;
1198 1199 1200
                            int bx0, by0;

                            for (by0 = y0; by0 < (y0+16); by0 += 4)
1201
                            {
1202 1203 1204 1205 1206 1207
                                for (bx0 = x0; bx0 < (x0+16); bx0 += 4)
                                {
                                    MV *mv = &bmi->mv.as_mv;

                                    x1 = bx0 + 2 + (mv->col >> 3);
                                    y1 = by0 + 2 + (mv->row >> 3);
1208

1209 1210 1211 1212 1213 1214
                                    constrain_line (bx0+2, &x1, by0+2, &y1, width, height);
                                    vp8_blit_line  (bx0+2,  x1, by0+2,  y1, y_buffer, y_stride);

                                    bmi++;
                                }
                            }
1215 1216 1217 1218
                        }
                    }
                }
                else if (mi->mbmi.mode >= NEARESTMV)
1219 1220
                {
                    MV *mv = &mi->mbmi.mv.as_mv;
1221 1222
                    const int lx0 = x0 + 8;
                    const int ly0 = y0 + 8;
John Koleszar's avatar
John Koleszar committed
1223

1224 1225
                    x1 = lx0 + (mv->col >> 3);
                    y1 = ly0 + (mv->row >> 3);
John Koleszar's avatar
John Koleszar committed
1226

1227
                    if (x1 != lx0 && y1 != ly0)
1228
                    {
1229 1230
                        constrain_line (lx0, &x1, ly0-1, &y1, width, height);
                        vp8_blit_line  (lx0,  x1, ly0-1,  y1, y_buffer, y_stride);
1231

1232 1233
                        constrain_line (lx0, &x1, ly0+1, &y1, width, height);
                        vp8_blit_line  (lx0,  x1, ly0+1,  y1, y_buffer, y_stride);
1234 1235
                    }
                    else
1236
                        vp8_blit_line  (lx0,  x1, ly0,  y1, y_buffer, y_stride);
1237
                }
1238

1239 1240 1241 1242 1243
                mi++;
            }
            mi++;
        }
    }
John Koleszar's avatar
John Koleszar committed
1244

1245
    /* Color in block modes */
1246 1247
    if ((flags & VP8D_DEBUG_CLR_BLK_MODES)
        && (ppflags->display_mb_modes_flag || ppflags->display_b_modes_flag))
1248
    {
1249
        int y, x;
1250 1251 1252 1253 1254 1255 1256 1257 1258
        YV12_BUFFER_CONFIG *post = &oci->post_proc_buffer;
        int width  = post->y_width;
        int height = post->y_height;
        unsigned char *y_ptr = oci->post_proc_buffer.y_buffer;
        unsigned char *u_ptr = oci->post_proc_buffer.u_buffer;
        unsigned char *v_ptr = oci->post_proc_buffer.v_buffer;
        int y_stride = oci->post_proc_buffer.y_stride;
        MODE_INFO *mi = oci->mi;

1259
        for (y = 0; y < height; y += 16)
1260
        {
1261
            for (x = 0; x < width; x += 16)
1262 1263 1264
            {
                int Y = 0, U = 0, V = 0;

1265 1266
                if (mi->mbmi.mode == B_PRED &&
                    ((ppflags->display_mb_modes_flag & B_PRED) || ppflags->display_b_modes_flag))
1267 1268 1269
                {
                    int by, bx;
                    unsigned char *yl, *ul, *vl;
1270
                    union b_mode_info *bmi = mi->bmi;
1271 1272 1273 1274

                    yl = y_ptr + x;
                    ul = u_ptr + (x>>1);
                    vl = v_ptr + (x>>1);
1275

1276 1277 1278 1279
                    for (by = 0; by < 16; by += 4)
                    {
                        for (bx = 0; bx < 16; bx += 4)
                        {
1280 1281 1282
                            if ((ppflags->display_b_modes_flag & (1<<mi->mbmi.mode))
                                || (ppflags->display_mb_modes_flag & B_PRED))
                            {
1283 1284 1285
                                Y = B_PREDICTION_MODE_colors[bmi->as_mode][0];
                                U = B_PREDICTION_MODE_colors[bmi->as_mode][1];
                                V = B_PREDICTION_MODE_colors[bmi->as_mode][2];
1286

1287 1288 1289
                                POSTPROC_INVOKE(RTCD_VTABLE(oci), blend_b)
                                    (yl+bx, ul+(bx>>1), vl+(bx>>1), Y, U, V, 0xc000, y_stride);
                            }
1290 1291 1292 1293 1294 1295 1296 1297
                            bmi++;
                        }

                        yl += y_stride*4;
                        ul += y_stride*1;
                        vl += y_stride*1;
                    }
                }
1298
                else if (ppflags->display_mb_modes_flag & (1<<mi->mbmi.mode))
1299 1300 1301 1302 1303 1304 1305 1306
                {
                    Y = MB_PREDICTION_MODE_colors[mi->mbmi.mode][0];
                    U = MB_PREDICTION_MODE_colors[mi->mbmi.mode][1];
                    V = MB_PREDICTION_MODE_colors[mi->mbmi.mode][2];

                    POSTPROC_INVOKE(RTCD_VTABLE(oci), blend_mb_inner)
                        (y_ptr+x, u_ptr+(x>>1), v_ptr+(x>>1), Y, U, V, 0xc000, y_stride);
                }
1307

1308 1309 1310 1311 1312 1313 1314 1315 1316 1317
                mi++;
            }
            y_ptr += y_stride*16;
            u_ptr += y_stride*4;
            v_ptr += y_stride*4;

            mi++;
        }
    }

1318
    /* Color in frame reference blocks */
1319
    if ((flags & VP8D_DEBUG_CLR_FRM_REF_BLKS) && ppflags->display_ref_frame_flag)
1320
    {