michael@0: /*
michael@0:  *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
michael@0:  *
michael@0:  *  Use of this source code is governed by a BSD-style license
michael@0:  *  that can be found in the LICENSE file in the root of the source
michael@0:  *  tree. An additional intellectual property rights grant can be found
michael@0:  *  in the file PATENTS.  All contributing project authors may
michael@0:  *  be found in the AUTHORS file in the root of the source tree.
michael@0:  */
michael@0: 
michael@0: #include <math.h>
michael@0: #include <stdlib.h>
michael@0: #include <stdio.h>
michael@0: 
michael@0: #include "./vpx_config.h"
michael@0: #include "vpx_scale/yv12config.h"
michael@0: #include "vp9/common/vp9_postproc.h"
michael@0: #include "vp9/common/vp9_textblit.h"
michael@0: #include "vpx_scale/vpx_scale.h"
michael@0: #include "vp9/common/vp9_systemdependent.h"
michael@0: #include "./vp9_rtcd.h"
michael@0: #include "./vpx_scale_rtcd.h"
michael@0: 
michael@0: #define RGB_TO_YUV(t)                                            \
michael@0:   ( (0.257*(float)(t >> 16))  + (0.504*(float)(t >> 8 & 0xff)) + \
michael@0:     (0.098*(float)(t & 0xff)) + 16),                             \
michael@0:   (-(0.148*(float)(t >> 16))  - (0.291*(float)(t >> 8 & 0xff)) + \
michael@0:     (0.439*(float)(t & 0xff)) + 128),                            \
michael@0:   ( (0.439*(float)(t >> 16))  - (0.368*(float)(t >> 8 & 0xff)) - \
michael@0:     (0.071*(float)(t & 0xff)) + 128)
michael@0: 
michael@0: /* global constants */
michael@0: #if 0 && CONFIG_POSTPROC_VISUALIZER
michael@0: static const unsigned char MB_PREDICTION_MODE_colors[MB_MODE_COUNT][3] = {
michael@0:   { RGB_TO_YUV(0x98FB98) },   /* PaleGreen */
michael@0:   { RGB_TO_YUV(0x00FF00) },   /* Green */
michael@0:   { RGB_TO_YUV(0xADFF2F) },   /* GreenYellow */
michael@0:   { RGB_TO_YUV(0x8F0000) },   /* Dark Red */
michael@0:   { RGB_TO_YUV(0x008F8F) },   /* Dark Cyan */
michael@0:   { RGB_TO_YUV(0x008F8F) },   /* Dark Cyan */
michael@0:   { RGB_TO_YUV(0x008F8F) },   /* Dark Cyan */
michael@0:   { RGB_TO_YUV(0x8F0000) },   /* Dark Red */
michael@0:   { RGB_TO_YUV(0x8F0000) },   /* Dark Red */
michael@0:   { RGB_TO_YUV(0x228B22) },   /* ForestGreen */
michael@0:   { RGB_TO_YUV(0x006400) },   /* DarkGreen */
michael@0:   { RGB_TO_YUV(0x98F5FF) },   /* Cadet Blue */
michael@0:   { RGB_TO_YUV(0x6CA6CD) },   /* Sky Blue */
michael@0:   { RGB_TO_YUV(0x00008B) },   /* Dark blue */
michael@0:   { RGB_TO_YUV(0x551A8B) },   /* Purple */
michael@0:   { RGB_TO_YUV(0xFF0000) }    /* Red */
michael@0:   { RGB_TO_YUV(0xCC33FF) },   /* Magenta */
michael@0: };
michael@0: 
michael@0: static const unsigned char B_PREDICTION_MODE_colors[INTRA_MODES][3] = {
michael@0:   { RGB_TO_YUV(0x6633ff) },   /* Purple */
michael@0:   { RGB_TO_YUV(0xcc33ff) },   /* Magenta */
michael@0:   { RGB_TO_YUV(0xff33cc) },   /* Pink */
michael@0:   { RGB_TO_YUV(0xff3366) },   /* Coral */
michael@0:   { RGB_TO_YUV(0x3366ff) },   /* Blue */
michael@0:   { RGB_TO_YUV(0xed00f5) },   /* Dark Blue */
michael@0:   { RGB_TO_YUV(0x2e00b8) },   /* Dark Purple */
michael@0:   { RGB_TO_YUV(0xff6633) },   /* Orange */
michael@0:   { RGB_TO_YUV(0x33ccff) },   /* Light Blue */
michael@0:   { RGB_TO_YUV(0x8ab800) },   /* Green */
michael@0:   { RGB_TO_YUV(0xffcc33) },   /* Light Orange */
michael@0:   { RGB_TO_YUV(0x33ffcc) },   /* Aqua */
michael@0:   { RGB_TO_YUV(0x66ff33) },   /* Light Green */
michael@0:   { RGB_TO_YUV(0xccff33) },   /* Yellow */
michael@0: };
michael@0: 
michael@0: static const unsigned char MV_REFERENCE_FRAME_colors[MAX_REF_FRAMES][3] = {
michael@0:   { RGB_TO_YUV(0x00ff00) },   /* Blue */
michael@0:   { RGB_TO_YUV(0x0000ff) },   /* Green */
michael@0:   { RGB_TO_YUV(0xffff00) },   /* Yellow */
michael@0:   { RGB_TO_YUV(0xff0000) },   /* Red */
michael@0: };
michael@0: #endif
michael@0: 
michael@0: static const short kernel5[] = {
michael@0:   1, 1, 4, 1, 1
michael@0: };
michael@0: 
michael@0: const short vp9_rv[] = {
michael@0:   8, 5, 2, 2, 8, 12, 4, 9, 8, 3,
michael@0:   0, 3, 9, 0, 0, 0, 8, 3, 14, 4,
michael@0:   10, 1, 11, 14, 1, 14, 9, 6, 12, 11,
michael@0:   8, 6, 10, 0, 0, 8, 9, 0, 3, 14,
michael@0:   8, 11, 13, 4, 2, 9, 0, 3, 9, 6,
michael@0:   1, 2, 3, 14, 13, 1, 8, 2, 9, 7,
michael@0:   3, 3, 1, 13, 13, 6, 6, 5, 2, 7,
michael@0:   11, 9, 11, 8, 7, 3, 2, 0, 13, 13,
michael@0:   14, 4, 12, 5, 12, 10, 8, 10, 13, 10,
michael@0:   4, 14, 4, 10, 0, 8, 11, 1, 13, 7,
michael@0:   7, 14, 6, 14, 13, 2, 13, 5, 4, 4,
michael@0:   0, 10, 0, 5, 13, 2, 12, 7, 11, 13,
michael@0:   8, 0, 4, 10, 7, 2, 7, 2, 2, 5,
michael@0:   3, 4, 7, 3, 3, 14, 14, 5, 9, 13,
michael@0:   3, 14, 3, 6, 3, 0, 11, 8, 13, 1,
michael@0:   13, 1, 12, 0, 10, 9, 7, 6, 2, 8,
michael@0:   5, 2, 13, 7, 1, 13, 14, 7, 6, 7,
michael@0:   9, 6, 10, 11, 7, 8, 7, 5, 14, 8,
michael@0:   4, 4, 0, 8, 7, 10, 0, 8, 14, 11,
michael@0:   3, 12, 5, 7, 14, 3, 14, 5, 2, 6,
michael@0:   11, 12, 12, 8, 0, 11, 13, 1, 2, 0,
michael@0:   5, 10, 14, 7, 8, 0, 4, 11, 0, 8,
michael@0:   0, 3, 10, 5, 8, 0, 11, 6, 7, 8,
michael@0:   10, 7, 13, 9, 2, 5, 1, 5, 10, 2,
michael@0:   4, 3, 5, 6, 10, 8, 9, 4, 11, 14,
michael@0:   0, 10, 0, 5, 13, 2, 12, 7, 11, 13,
michael@0:   8, 0, 4, 10, 7, 2, 7, 2, 2, 5,
michael@0:   3, 4, 7, 3, 3, 14, 14, 5, 9, 13,
michael@0:   3, 14, 3, 6, 3, 0, 11, 8, 13, 1,
michael@0:   13, 1, 12, 0, 10, 9, 7, 6, 2, 8,
michael@0:   5, 2, 13, 7, 1, 13, 14, 7, 6, 7,
michael@0:   9, 6, 10, 11, 7, 8, 7, 5, 14, 8,
michael@0:   4, 4, 0, 8, 7, 10, 0, 8, 14, 11,
michael@0:   3, 12, 5, 7, 14, 3, 14, 5, 2, 6,
michael@0:   11, 12, 12, 8, 0, 11, 13, 1, 2, 0,
michael@0:   5, 10, 14, 7, 8, 0, 4, 11, 0, 8,
michael@0:   0, 3, 10, 5, 8, 0, 11, 6, 7, 8,
michael@0:   10, 7, 13, 9, 2, 5, 1, 5, 10, 2,
michael@0:   4, 3, 5, 6, 10, 8, 9, 4, 11, 14,
michael@0:   3, 8, 3, 7, 8, 5, 11, 4, 12, 3,
michael@0:   11, 9, 14, 8, 14, 13, 4, 3, 1, 2,
michael@0:   14, 6, 5, 4, 4, 11, 4, 6, 2, 1,
michael@0:   5, 8, 8, 12, 13, 5, 14, 10, 12, 13,
michael@0:   0, 9, 5, 5, 11, 10, 13, 9, 10, 13,
michael@0: };
michael@0: 
michael@0: 
michael@0: /****************************************************************************
michael@0:  */
michael@0: void vp9_post_proc_down_and_across_c(const uint8_t *src_ptr,
michael@0:                                      uint8_t *dst_ptr,
michael@0:                                      int src_pixels_per_line,
michael@0:                                      int dst_pixels_per_line,
michael@0:                                      int rows,
michael@0:                                      int cols,
michael@0:                                      int flimit) {
michael@0:   uint8_t const *p_src;
michael@0:   uint8_t *p_dst;
michael@0:   int row;
michael@0:   int col;
michael@0:   int i;
michael@0:   int v;
michael@0:   int pitch = src_pixels_per_line;
michael@0:   uint8_t d[8];
michael@0:   (void)dst_pixels_per_line;
michael@0: 
michael@0:   for (row = 0; row < rows; row++) {
michael@0:     /* post_proc_down for one row */
michael@0:     p_src = src_ptr;
michael@0:     p_dst = dst_ptr;
michael@0: 
michael@0:     for (col = 0; col < cols; col++) {
michael@0:       int kernel = 4;
michael@0:       int v = p_src[col];
michael@0: 
michael@0:       for (i = -2; i <= 2; i++) {
michael@0:         if (abs(v - p_src[col + i * pitch]) > flimit)
michael@0:           goto down_skip_convolve;
michael@0: 
michael@0:         kernel += kernel5[2 + i] * p_src[col + i * pitch];
michael@0:       }
michael@0: 
michael@0:       v = (kernel >> 3);
michael@0:     down_skip_convolve:
michael@0:       p_dst[col] = v;
michael@0:     }
michael@0: 
michael@0:     /* now post_proc_across */
michael@0:     p_src = dst_ptr;
michael@0:     p_dst = dst_ptr;
michael@0: 
michael@0:     for (i = 0; i < 8; i++)
michael@0:       d[i] = p_src[i];
michael@0: 
michael@0:     for (col = 0; col < cols; col++) {
michael@0:       int kernel = 4;
michael@0:       v = p_src[col];
michael@0: 
michael@0:       d[col & 7] = v;
michael@0: 
michael@0:       for (i = -2; i <= 2; i++) {
michael@0:         if (abs(v - p_src[col + i]) > flimit)
michael@0:           goto across_skip_convolve;
michael@0: 
michael@0:         kernel += kernel5[2 + i] * p_src[col + i];
michael@0:       }
michael@0: 
michael@0:       d[col & 7] = (kernel >> 3);
michael@0:     across_skip_convolve:
michael@0: 
michael@0:       if (col >= 2)
michael@0:         p_dst[col - 2] = d[(col - 2) & 7];
michael@0:     }
michael@0: 
michael@0:     /* handle the last two pixels */
michael@0:     p_dst[col - 2] = d[(col - 2) & 7];
michael@0:     p_dst[col - 1] = d[(col - 1) & 7];
michael@0: 
michael@0: 
michael@0:     /* next row */
michael@0:     src_ptr += pitch;
michael@0:     dst_ptr += pitch;
michael@0:   }
michael@0: }
michael@0: 
michael@0: static int q2mbl(int x) {
michael@0:   if (x < 20) x = 20;
michael@0: 
michael@0:   x = 50 + (x - 50) * 10 / 8;
michael@0:   return x * x / 3;
michael@0: }
michael@0: 
michael@0: void vp9_mbpost_proc_across_ip_c(uint8_t *src, int pitch,
michael@0:                                  int rows, int cols, int flimit) {
michael@0:   int r, c, i;
michael@0: 
michael@0:   uint8_t *s = src;
michael@0:   uint8_t d[16];
michael@0: 
michael@0: 
michael@0:   for (r = 0; r < rows; r++) {
michael@0:     int sumsq = 0;
michael@0:     int sum   = 0;
michael@0: 
michael@0:     for (i = -8; i <= 6; i++) {
michael@0:       sumsq += s[i] * s[i];
michael@0:       sum   += s[i];
michael@0:       d[i + 8] = 0;
michael@0:     }
michael@0: 
michael@0:     for (c = 0; c < cols + 8; c++) {
michael@0:       int x = s[c + 7] - s[c - 8];
michael@0:       int y = s[c + 7] + s[c - 8];
michael@0: 
michael@0:       sum  += x;
michael@0:       sumsq += x * y;
michael@0: 
michael@0:       d[c & 15] = s[c];
michael@0: 
michael@0:       if (sumsq * 15 - sum * sum < flimit) {
michael@0:         d[c & 15] = (8 + sum + s[c]) >> 4;
michael@0:       }
michael@0: 
michael@0:       s[c - 8] = d[(c - 8) & 15];
michael@0:     }
michael@0: 
michael@0:     s += pitch;
michael@0:   }
michael@0: }
michael@0: 
michael@0: void vp9_mbpost_proc_down_c(uint8_t *dst, int pitch,
michael@0:                             int rows, int cols, int flimit) {
michael@0:   int r, c, i;
michael@0:   const short *rv3 = &vp9_rv[63 & rand()]; // NOLINT
michael@0: 
michael@0:   for (c = 0; c < cols; c++) {
michael@0:     uint8_t *s = &dst[c];
michael@0:     int sumsq = 0;
michael@0:     int sum   = 0;
michael@0:     uint8_t d[16];
michael@0:     const short *rv2 = rv3 + ((c * 17) & 127);
michael@0: 
michael@0:     for (i = -8; i <= 6; i++) {
michael@0:       sumsq += s[i * pitch] * s[i * pitch];
michael@0:       sum   += s[i * pitch];
michael@0:     }
michael@0: 
michael@0:     for (r = 0; r < rows + 8; r++) {
michael@0:       sumsq += s[7 * pitch] * s[ 7 * pitch] - s[-8 * pitch] * s[-8 * pitch];
michael@0:       sum  += s[7 * pitch] - s[-8 * pitch];
michael@0:       d[r & 15] = s[0];
michael@0: 
michael@0:       if (sumsq * 15 - sum * sum < flimit) {
michael@0:         d[r & 15] = (rv2[r & 127] + sum + s[0]) >> 4;
michael@0:       }
michael@0: 
michael@0:       s[-8 * pitch] = d[(r - 8) & 15];
michael@0:       s += pitch;
michael@0:     }
michael@0:   }
michael@0: }
michael@0: 
michael@0: static void deblock_and_de_macro_block(YV12_BUFFER_CONFIG   *source,
michael@0:                                        YV12_BUFFER_CONFIG   *post,
michael@0:                                        int                   q,
michael@0:                                        int                   low_var_thresh,
michael@0:                                        int                   flag) {
michael@0:   double level = 6.0e-05 * q * q * q - .0067 * q * q + .306 * q + .0065;
michael@0:   int ppl = (int)(level + .5);
michael@0:   (void) low_var_thresh;
michael@0:   (void) flag;
michael@0: 
michael@0:   vp9_post_proc_down_and_across(source->y_buffer, post->y_buffer,
michael@0:                                 source->y_stride, post->y_stride,
michael@0:                                 source->y_height, source->y_width, ppl);
michael@0: 
michael@0:   vp9_mbpost_proc_across_ip(post->y_buffer, post->y_stride, post->y_height,
michael@0:                             post->y_width, q2mbl(q));
michael@0: 
michael@0:   vp9_mbpost_proc_down(post->y_buffer, post->y_stride, post->y_height,
michael@0:                        post->y_width, q2mbl(q));
michael@0: 
michael@0:   vp9_post_proc_down_and_across(source->u_buffer, post->u_buffer,
michael@0:                                 source->uv_stride, post->uv_stride,
michael@0:                                 source->uv_height, source->uv_width, ppl);
michael@0:   vp9_post_proc_down_and_across(source->v_buffer, post->v_buffer,
michael@0:                                 source->uv_stride, post->uv_stride,
michael@0:                                 source->uv_height, source->uv_width, ppl);
michael@0: }
michael@0: 
michael@0: void vp9_deblock(const YV12_BUFFER_CONFIG *src, YV12_BUFFER_CONFIG *dst,
michael@0:                  int q) {
michael@0:   const int ppl = (int)(6.0e-05 * q * q * q - 0.0067 * q * q + 0.306 * q
michael@0:                         + 0.0065 + 0.5);
michael@0:   int i;
michael@0: 
michael@0:   const uint8_t *const srcs[4] = {src->y_buffer, src->u_buffer, src->v_buffer,
michael@0:                                   src->alpha_buffer};
michael@0:   const int src_strides[4] = {src->y_stride, src->uv_stride, src->uv_stride,
michael@0:                               src->alpha_stride};
michael@0:   const int src_widths[4] = {src->y_width, src->uv_width, src->uv_width,
michael@0:                              src->alpha_width};
michael@0:   const int src_heights[4] = {src->y_height, src->uv_height, src->uv_height,
michael@0:                               src->alpha_height};
michael@0: 
michael@0:   uint8_t *const dsts[4] = {dst->y_buffer, dst->u_buffer, dst->v_buffer,
michael@0:                             dst->alpha_buffer};
michael@0:   const int dst_strides[4] = {dst->y_stride, dst->uv_stride, dst->uv_stride,
michael@0:                               dst->alpha_stride};
michael@0: 
michael@0:   for (i = 0; i < MAX_MB_PLANE; ++i)
michael@0:     vp9_post_proc_down_and_across(srcs[i], dsts[i],
michael@0:                                   src_strides[i], dst_strides[i],
michael@0:                                   src_heights[i], src_widths[i], ppl);
michael@0: }
michael@0: 
michael@0: void vp9_denoise(const YV12_BUFFER_CONFIG *src, YV12_BUFFER_CONFIG *dst,
michael@0:                  int q) {
michael@0:   const int ppl = (int)(6.0e-05 * q * q * q - 0.0067 * q * q + 0.306 * q
michael@0:                         + 0.0065 + 0.5);
michael@0:   int i;
michael@0: 
michael@0:   const uint8_t *const srcs[4] = {src->y_buffer, src->u_buffer, src->v_buffer,
michael@0:                                   src->alpha_buffer};
michael@0:   const int src_strides[4] = {src->y_stride, src->uv_stride, src->uv_stride,
michael@0:                               src->alpha_stride};
michael@0:   const int src_widths[4] = {src->y_width, src->uv_width, src->uv_width,
michael@0:                              src->alpha_width};
michael@0:   const int src_heights[4] = {src->y_height, src->uv_height, src->uv_height,
michael@0:                               src->alpha_height};
michael@0: 
michael@0:   uint8_t *const dsts[4] = {dst->y_buffer, dst->u_buffer, dst->v_buffer,
michael@0:                             dst->alpha_buffer};
michael@0:   const int dst_strides[4] = {dst->y_stride, dst->uv_stride, dst->uv_stride,
michael@0:                               dst->alpha_stride};
michael@0: 
michael@0:   for (i = 0; i < MAX_MB_PLANE; ++i) {
michael@0:     const int src_stride = src_strides[i];
michael@0:     const uint8_t *const src = srcs[i] + 2 * src_stride + 2;
michael@0:     const int src_width = src_widths[i] - 4;
michael@0:     const int src_height = src_heights[i] - 4;
michael@0: 
michael@0:     const int dst_stride = dst_strides[i];
michael@0:     uint8_t *const dst = dsts[i] + 2 * dst_stride + 2;
michael@0: 
michael@0:     vp9_post_proc_down_and_across(src, dst, src_stride, dst_stride,
michael@0:                                   src_height, src_width, ppl);
michael@0:   }
michael@0: }
michael@0: 
michael@0: double vp9_gaussian(double sigma, double mu, double x) {
michael@0:   return 1 / (sigma * sqrt(2.0 * 3.14159265)) *
michael@0:          (exp(-(x - mu) * (x - mu) / (2 * sigma * sigma)));
michael@0: }
michael@0: 
michael@0: static void fillrd(struct postproc_state *state, int q, int a) {
michael@0:   char char_dist[300];
michael@0: 
michael@0:   double sigma;
michael@0:   int ai = a, qi = q, i;
michael@0: 
michael@0:   vp9_clear_system_state();
michael@0: 
michael@0:   sigma = ai + .5 + .6 * (63 - qi) / 63.0;
michael@0: 
michael@0:   /* set up a lookup table of 256 entries that matches
michael@0:    * a gaussian distribution with sigma determined by q.
michael@0:    */
michael@0:   {
michael@0:     double i;
michael@0:     int next, j;
michael@0: 
michael@0:     next = 0;
michael@0: 
michael@0:     for (i = -32; i < 32; i++) {
michael@0:       int a = (int)(.5 + 256 * vp9_gaussian(sigma, 0, i));
michael@0: 
michael@0:       if (a) {
michael@0:         for (j = 0; j < a; j++) {
michael@0:           char_dist[next + j] = (char) i;
michael@0:         }
michael@0: 
michael@0:         next = next + j;
michael@0:       }
michael@0:     }
michael@0: 
michael@0:     for (; next < 256; next++)
michael@0:       char_dist[next] = 0;
michael@0:   }
michael@0: 
michael@0:   for (i = 0; i < 3072; i++) {
michael@0:     state->noise[i] = char_dist[rand() & 0xff];  // NOLINT
michael@0:   }
michael@0: 
michael@0:   for (i = 0; i < 16; i++) {
michael@0:     state->blackclamp[i] = -char_dist[0];
michael@0:     state->whiteclamp[i] = -char_dist[0];
michael@0:     state->bothclamp[i] = -2 * char_dist[0];
michael@0:   }
michael@0: 
michael@0:   state->last_q = q;
michael@0:   state->last_noise = a;
michael@0: }
michael@0: 
michael@0: /****************************************************************************
michael@0:  *
michael@0:  *  ROUTINE       : plane_add_noise_c
michael@0:  *
michael@0:  *  INPUTS        : unsigned char *Start  starting address of buffer to
michael@0:  *                                        add gaussian noise to
michael@0:  *                  unsigned int width    width of plane
michael@0:  *                  unsigned int height   height of plane
michael@0:  *                  int  pitch    distance between subsequent lines of frame
michael@0:  *                  int  q        quantizer used to determine amount of noise
michael@0:  *                                  to add
michael@0:  *
michael@0:  *  OUTPUTS       : None.
michael@0:  *
michael@0:  *  RETURNS       : void.
michael@0:  *
michael@0:  *  FUNCTION      : adds gaussian noise to a plane of pixels
michael@0:  *
michael@0:  *  SPECIAL NOTES : None.
michael@0:  *
michael@0:  ****************************************************************************/
michael@0: void vp9_plane_add_noise_c(uint8_t *start, char *noise,
michael@0:                            char blackclamp[16],
michael@0:                            char whiteclamp[16],
michael@0:                            char bothclamp[16],
michael@0:                            unsigned int width, unsigned int height, int pitch) {
michael@0:   unsigned int i, j;
michael@0: 
michael@0:   for (i = 0; i < height; i++) {
michael@0:     uint8_t *pos = start + i * pitch;
michael@0:     char  *ref = (char *)(noise + (rand() & 0xff));  // NOLINT
michael@0: 
michael@0:     for (j = 0; j < width; j++) {
michael@0:       if (pos[j] < blackclamp[0])
michael@0:         pos[j] = blackclamp[0];
michael@0: 
michael@0:       if (pos[j] > 255 + whiteclamp[0])
michael@0:         pos[j] = 255 + whiteclamp[0];
michael@0: 
michael@0:       pos[j] += ref[j];
michael@0:     }
michael@0:   }
michael@0: }
michael@0: 
michael@0: /* Blend the macro block with a solid colored square.  Leave the
michael@0:  * edges unblended to give distinction to macro blocks in areas
michael@0:  * filled with the same color block.
michael@0:  */
michael@0: void vp9_blend_mb_inner_c(uint8_t *y, uint8_t *u, uint8_t *v,
michael@0:                           int y1, int u1, int v1, int alpha, int stride) {
michael@0:   int i, j;
michael@0:   int y1_const = y1 * ((1 << 16) - alpha);
michael@0:   int u1_const = u1 * ((1 << 16) - alpha);
michael@0:   int v1_const = v1 * ((1 << 16) - alpha);
michael@0: 
michael@0:   y += 2 * stride + 2;
michael@0:   for (i = 0; i < 12; i++) {
michael@0:     for (j = 0; j < 12; j++) {
michael@0:       y[j] = (y[j] * alpha + y1_const) >> 16;
michael@0:     }
michael@0:     y += stride;
michael@0:   }
michael@0: 
michael@0:   stride >>= 1;
michael@0: 
michael@0:   u += stride + 1;
michael@0:   v += stride + 1;
michael@0: 
michael@0:   for (i = 0; i < 6; i++) {
michael@0:     for (j = 0; j < 6; j++) {
michael@0:       u[j] = (u[j] * alpha + u1_const) >> 16;
michael@0:       v[j] = (v[j] * alpha + v1_const) >> 16;
michael@0:     }
michael@0:     u += stride;
michael@0:     v += stride;
michael@0:   }
michael@0: }
michael@0: 
michael@0: /* Blend only the edge of the macro block.  Leave center
michael@0:  * unblended to allow for other visualizations to be layered.
michael@0:  */
michael@0: void vp9_blend_mb_outer_c(uint8_t *y, uint8_t *u, uint8_t *v,
michael@0:                           int y1, int u1, int v1, int alpha, int stride) {
michael@0:   int i, j;
michael@0:   int y1_const = y1 * ((1 << 16) - alpha);
michael@0:   int u1_const = u1 * ((1 << 16) - alpha);
michael@0:   int v1_const = v1 * ((1 << 16) - alpha);
michael@0: 
michael@0:   for (i = 0; i < 2; i++) {
michael@0:     for (j = 0; j < 16; j++) {
michael@0:       y[j] = (y[j] * alpha + y1_const) >> 16;
michael@0:     }
michael@0:     y += stride;
michael@0:   }
michael@0: 
michael@0:   for (i = 0; i < 12; i++) {
michael@0:     y[0]  = (y[0] * alpha  + y1_const) >> 16;
michael@0:     y[1]  = (y[1] * alpha  + y1_const) >> 16;
michael@0:     y[14] = (y[14] * alpha + y1_const) >> 16;
michael@0:     y[15] = (y[15] * alpha + y1_const) >> 16;
michael@0:     y += stride;
michael@0:   }
michael@0: 
michael@0:   for (i = 0; i < 2; i++) {
michael@0:     for (j = 0; j < 16; j++) {
michael@0:       y[j] = (y[j] * alpha + y1_const) >> 16;
michael@0:     }
michael@0:     y += stride;
michael@0:   }
michael@0: 
michael@0:   stride >>= 1;
michael@0: 
michael@0:   for (j = 0; j < 8; j++) {
michael@0:     u[j] = (u[j] * alpha + u1_const) >> 16;
michael@0:     v[j] = (v[j] * alpha + v1_const) >> 16;
michael@0:   }
michael@0:   u += stride;
michael@0:   v += stride;
michael@0: 
michael@0:   for (i = 0; i < 6; i++) {
michael@0:     u[0] = (u[0] * alpha + u1_const) >> 16;
michael@0:     v[0] = (v[0] * alpha + v1_const) >> 16;
michael@0: 
michael@0:     u[7] = (u[7] * alpha + u1_const) >> 16;
michael@0:     v[7] = (v[7] * alpha + v1_const) >> 16;
michael@0: 
michael@0:     u += stride;
michael@0:     v += stride;
michael@0:   }
michael@0: 
michael@0:   for (j = 0; j < 8; j++) {
michael@0:     u[j] = (u[j] * alpha + u1_const) >> 16;
michael@0:     v[j] = (v[j] * alpha + v1_const) >> 16;
michael@0:   }
michael@0: }
michael@0: 
michael@0: void vp9_blend_b_c(uint8_t *y, uint8_t *u, uint8_t *v,
michael@0:                    int y1, int u1, int v1, int alpha, int stride) {
michael@0:   int i, j;
michael@0:   int y1_const = y1 * ((1 << 16) - alpha);
michael@0:   int u1_const = u1 * ((1 << 16) - alpha);
michael@0:   int v1_const = v1 * ((1 << 16) - alpha);
michael@0: 
michael@0:   for (i = 0; i < 4; i++) {
michael@0:     for (j = 0; j < 4; j++) {
michael@0:       y[j] = (y[j] * alpha + y1_const) >> 16;
michael@0:     }
michael@0:     y += stride;
michael@0:   }
michael@0: 
michael@0:   stride >>= 1;
michael@0: 
michael@0:   for (i = 0; i < 2; i++) {
michael@0:     for (j = 0; j < 2; j++) {
michael@0:       u[j] = (u[j] * alpha + u1_const) >> 16;
michael@0:       v[j] = (v[j] * alpha + v1_const) >> 16;
michael@0:     }
michael@0:     u += stride;
michael@0:     v += stride;
michael@0:   }
michael@0: }
michael@0: 
michael@0: static void constrain_line(int x0, int *x1, int y0, int *y1,
michael@0:                            int width, int height) {
michael@0:   int dx;
michael@0:   int dy;
michael@0: 
michael@0:   if (*x1 > width) {
michael@0:     dx = *x1 - x0;
michael@0:     dy = *y1 - y0;
michael@0: 
michael@0:     *x1 = width;
michael@0:     if (dx)
michael@0:       *y1 = ((width - x0) * dy) / dx + y0;
michael@0:   }
michael@0:   if (*x1 < 0) {
michael@0:     dx = *x1 - x0;
michael@0:     dy = *y1 - y0;
michael@0: 
michael@0:     *x1 = 0;
michael@0:     if (dx)
michael@0:       *y1 = ((0 - x0) * dy) / dx + y0;
michael@0:   }
michael@0:   if (*y1 > height) {
michael@0:     dx = *x1 - x0;
michael@0:     dy = *y1 - y0;
michael@0: 
michael@0:     *y1 = height;
michael@0:     if (dy)
michael@0:       *x1 = ((height - y0) * dx) / dy + x0;
michael@0:   }
michael@0:   if (*y1 < 0) {
michael@0:     dx = *x1 - x0;
michael@0:     dy = *y1 - y0;
michael@0: 
michael@0:     *y1 = 0;
michael@0:     if (dy)
michael@0:       *x1 = ((0 - y0) * dx) / dy + x0;
michael@0:   }
michael@0: }
michael@0: 
michael@0: int vp9_post_proc_frame(struct VP9Common *cm,
michael@0:                         YV12_BUFFER_CONFIG *dest, vp9_ppflags_t *ppflags) {
michael@0:   int q = cm->lf.filter_level * 10 / 6;
michael@0:   int flags = ppflags->post_proc_flag;
michael@0:   int deblock_level = ppflags->deblocking_level;
michael@0:   int noise_level = ppflags->noise_level;
michael@0: 
michael@0:   if (!cm->frame_to_show)
michael@0:     return -1;
michael@0: 
michael@0:   if (q > 63)
michael@0:     q = 63;
michael@0: 
michael@0:   if (!flags) {
michael@0:     *dest = *cm->frame_to_show;
michael@0:     return 0;
michael@0:   }
michael@0: 
michael@0: #if ARCH_X86||ARCH_X86_64
michael@0:   vpx_reset_mmx_state();
michael@0: #endif
michael@0: 
michael@0:   if (flags & VP9D_DEMACROBLOCK) {
michael@0:     deblock_and_de_macro_block(cm->frame_to_show, &cm->post_proc_buffer,
michael@0:                                q + (deblock_level - 5) * 10, 1, 0);
michael@0:   } else if (flags & VP9D_DEBLOCK) {
michael@0:     vp9_deblock(cm->frame_to_show, &cm->post_proc_buffer, q);
michael@0:   } else {
michael@0:     vp8_yv12_copy_frame(cm->frame_to_show, &cm->post_proc_buffer);
michael@0:   }
michael@0: 
michael@0:   if (flags & VP9D_ADDNOISE) {
michael@0:     if (cm->postproc_state.last_q != q
michael@0:         || cm->postproc_state.last_noise != noise_level) {
michael@0:       fillrd(&cm->postproc_state, 63 - q, noise_level);
michael@0:     }
michael@0: 
michael@0:     vp9_plane_add_noise(cm->post_proc_buffer.y_buffer,
michael@0:                         cm->postproc_state.noise,
michael@0:                         cm->postproc_state.blackclamp,
michael@0:                         cm->postproc_state.whiteclamp,
michael@0:                         cm->postproc_state.bothclamp,
michael@0:                         cm->post_proc_buffer.y_width,
michael@0:                         cm->post_proc_buffer.y_height,
michael@0:                         cm->post_proc_buffer.y_stride);
michael@0:   }
michael@0: 
michael@0: #if 0 && CONFIG_POSTPROC_VISUALIZER
michael@0:   if (flags & VP9D_DEBUG_TXT_FRAME_INFO) {
michael@0:     char message[512];
michael@0:     snprintf(message, sizeof(message) -1,
michael@0:              "F%1dG%1dQ%3dF%3dP%d_s%dx%d",
michael@0:              (cm->frame_type == KEY_FRAME),
michael@0:              cm->refresh_golden_frame,
michael@0:              cm->base_qindex,
michael@0:              cm->filter_level,
michael@0:              flags,
michael@0:              cm->mb_cols, cm->mb_rows);
michael@0:     vp9_blit_text(message, cm->post_proc_buffer.y_buffer,
michael@0:                   cm->post_proc_buffer.y_stride);
michael@0:   }
michael@0: 
michael@0:   if (flags & VP9D_DEBUG_TXT_MBLK_MODES) {
michael@0:     int i, j;
michael@0:     uint8_t *y_ptr;
michael@0:     YV12_BUFFER_CONFIG *post = &cm->post_proc_buffer;
michael@0:     int mb_rows = post->y_height >> 4;
michael@0:     int mb_cols = post->y_width  >> 4;
michael@0:     int mb_index = 0;
michael@0:     MODE_INFO *mi = cm->mi;
michael@0: 
michael@0:     y_ptr = post->y_buffer + 4 * post->y_stride + 4;
michael@0: 
michael@0:     /* vp9_filter each macro block */
michael@0:     for (i = 0; i < mb_rows; i++) {
michael@0:       for (j = 0; j < mb_cols; j++) {
michael@0:         char zz[4];
michael@0: 
michael@0:         snprintf(zz, sizeof(zz) - 1, "%c", mi[mb_index].mbmi.mode + 'a');
michael@0: 
michael@0:         vp9_blit_text(zz, y_ptr, post->y_stride);
michael@0:         mb_index++;
michael@0:         y_ptr += 16;
michael@0:       }
michael@0: 
michael@0:       mb_index++; /* border */
michael@0:       y_ptr += post->y_stride  * 16 - post->y_width;
michael@0:     }
michael@0:   }
michael@0: 
michael@0:   if (flags & VP9D_DEBUG_TXT_DC_DIFF) {
michael@0:     int i, j;
michael@0:     uint8_t *y_ptr;
michael@0:     YV12_BUFFER_CONFIG *post = &cm->post_proc_buffer;
michael@0:     int mb_rows = post->y_height >> 4;
michael@0:     int mb_cols = post->y_width  >> 4;
michael@0:     int mb_index = 0;
michael@0:     MODE_INFO *mi = cm->mi;
michael@0: 
michael@0:     y_ptr = post->y_buffer + 4 * post->y_stride + 4;
michael@0: 
michael@0:     /* vp9_filter each macro block */
michael@0:     for (i = 0; i < mb_rows; i++) {
michael@0:       for (j = 0; j < mb_cols; j++) {
michael@0:         char zz[4];
michael@0:         int dc_diff = !(mi[mb_index].mbmi.mode != I4X4_PRED &&
michael@0:                         mi[mb_index].mbmi.mode != SPLITMV &&
michael@0:                         mi[mb_index].mbmi.skip_coeff);
michael@0: 
michael@0:         if (cm->frame_type == KEY_FRAME)
michael@0:           snprintf(zz, sizeof(zz) - 1, "a");
michael@0:         else
michael@0:           snprintf(zz, sizeof(zz) - 1, "%c", dc_diff + '0');
michael@0: 
michael@0:         vp9_blit_text(zz, y_ptr, post->y_stride);
michael@0:         mb_index++;
michael@0:         y_ptr += 16;
michael@0:       }
michael@0: 
michael@0:       mb_index++; /* border */
michael@0:       y_ptr += post->y_stride  * 16 - post->y_width;
michael@0:     }
michael@0:   }
michael@0: 
michael@0:   if (flags & VP9D_DEBUG_TXT_RATE_INFO) {
michael@0:     char message[512];
michael@0:     snprintf(message, sizeof(message),
michael@0:              "Bitrate: %10.2f framerate: %10.2f ",
michael@0:              cm->bitrate, cm->framerate);
michael@0:     vp9_blit_text(message, cm->post_proc_buffer.y_buffer,
michael@0:                   cm->post_proc_buffer.y_stride);
michael@0:   }
michael@0: 
michael@0:   /* Draw motion vectors */
michael@0:   if ((flags & VP9D_DEBUG_DRAW_MV) && ppflags->display_mv_flag) {
michael@0:     YV12_BUFFER_CONFIG *post = &cm->post_proc_buffer;
michael@0:     int width  = post->y_width;
michael@0:     int height = post->y_height;
michael@0:     uint8_t *y_buffer = cm->post_proc_buffer.y_buffer;
michael@0:     int y_stride = cm->post_proc_buffer.y_stride;
michael@0:     MODE_INFO *mi = cm->mi;
michael@0:     int x0, y0;
michael@0: 
michael@0:     for (y0 = 0; y0 < height; y0 += 16) {
michael@0:       for (x0 = 0; x0 < width; x0 += 16) {
michael@0:         int x1, y1;
michael@0: 
michael@0:         if (!(ppflags->display_mv_flag & (1 << mi->mbmi.mode))) {
michael@0:           mi++;
michael@0:           continue;
michael@0:         }
michael@0: 
michael@0:         if (mi->mbmi.mode == SPLITMV) {
michael@0:           switch (mi->mbmi.partitioning) {
michael@0:             case PARTITIONING_16X8 : {  /* mv_top_bottom */
michael@0:               union b_mode_info *bmi = &mi->bmi[0];
michael@0:               MV *mv = &bmi->mv.as_mv;
michael@0: 
michael@0:               x1 = x0 + 8 + (mv->col >> 3);
michael@0:               y1 = y0 + 4 + (mv->row >> 3);
michael@0: 
michael@0:               constrain_line(x0 + 8, &x1, y0 + 4, &y1, width, height);
michael@0:               vp9_blit_line(x0 + 8,  x1, y0 + 4,  y1, y_buffer, y_stride);
michael@0: 
michael@0:               bmi = &mi->bmi[8];
michael@0: 
michael@0:               x1 = x0 + 8 + (mv->col >> 3);
michael@0:               y1 = y0 + 12 + (mv->row >> 3);
michael@0: 
michael@0:               constrain_line(x0 + 8, &x1, y0 + 12, &y1, width, height);
michael@0:               vp9_blit_line(x0 + 8,  x1, y0 + 12,  y1, y_buffer, y_stride);
michael@0: 
michael@0:               break;
michael@0:             }
michael@0:             case PARTITIONING_8X16 : {  /* mv_left_right */
michael@0:               union b_mode_info *bmi = &mi->bmi[0];
michael@0:               MV *mv = &bmi->mv.as_mv;
michael@0: 
michael@0:               x1 = x0 + 4 + (mv->col >> 3);
michael@0:               y1 = y0 + 8 + (mv->row >> 3);
michael@0: 
michael@0:               constrain_line(x0 + 4, &x1, y0 + 8, &y1, width, height);
michael@0:               vp9_blit_line(x0 + 4,  x1, y0 + 8,  y1, y_buffer, y_stride);
michael@0: 
michael@0:               bmi = &mi->bmi[2];
michael@0: 
michael@0:               x1 = x0 + 12 + (mv->col >> 3);
michael@0:               y1 = y0 + 8 + (mv->row >> 3);
michael@0: 
michael@0:               constrain_line(x0 + 12, &x1, y0 + 8, &y1, width, height);
michael@0:               vp9_blit_line(x0 + 12,  x1, y0 + 8,  y1, y_buffer, y_stride);
michael@0: 
michael@0:               break;
michael@0:             }
michael@0:             case PARTITIONING_8X8 : {  /* mv_quarters   */
michael@0:               union b_mode_info *bmi = &mi->bmi[0];
michael@0:               MV *mv = &bmi->mv.as_mv;
michael@0: 
michael@0:               x1 = x0 + 4 + (mv->col >> 3);
michael@0:               y1 = y0 + 4 + (mv->row >> 3);
michael@0: 
michael@0:               constrain_line(x0 + 4, &x1, y0 + 4, &y1, width, height);
michael@0:               vp9_blit_line(x0 + 4,  x1, y0 + 4,  y1, y_buffer, y_stride);
michael@0: 
michael@0:               bmi = &mi->bmi[2];
michael@0: 
michael@0:               x1 = x0 + 12 + (mv->col >> 3);
michael@0:               y1 = y0 + 4 + (mv->row >> 3);
michael@0: 
michael@0:               constrain_line(x0 + 12, &x1, y0 + 4, &y1, width, height);
michael@0:               vp9_blit_line(x0 + 12,  x1, y0 + 4,  y1, y_buffer, y_stride);
michael@0: 
michael@0:               bmi = &mi->bmi[8];
michael@0: 
michael@0:               x1 = x0 + 4 + (mv->col >> 3);
michael@0:               y1 = y0 + 12 + (mv->row >> 3);
michael@0: 
michael@0:               constrain_line(x0 + 4, &x1, y0 + 12, &y1, width, height);
michael@0:               vp9_blit_line(x0 + 4,  x1, y0 + 12,  y1, y_buffer, y_stride);
michael@0: 
michael@0:               bmi = &mi->bmi[10];
michael@0: 
michael@0:               x1 = x0 + 12 + (mv->col >> 3);
michael@0:               y1 = y0 + 12 + (mv->row >> 3);
michael@0: 
michael@0:               constrain_line(x0 + 12, &x1, y0 + 12, &y1, width, height);
michael@0:               vp9_blit_line(x0 + 12,  x1, y0 + 12,  y1, y_buffer, y_stride);
michael@0:               break;
michael@0:             }
michael@0:             case PARTITIONING_4X4:
michael@0:             default : {
michael@0:               union b_mode_info *bmi = mi->bmi;
michael@0:               int bx0, by0;
michael@0: 
michael@0:               for (by0 = y0; by0 < (y0 + 16); by0 += 4) {
michael@0:                 for (bx0 = x0; bx0 < (x0 + 16); bx0 += 4) {
michael@0:                   MV *mv = &bmi->mv.as_mv;
michael@0: 
michael@0:                   x1 = bx0 + 2 + (mv->col >> 3);
michael@0:                   y1 = by0 + 2 + (mv->row >> 3);
michael@0: 
michael@0:                   constrain_line(bx0 + 2, &x1, by0 + 2, &y1, width, height);
michael@0:                   vp9_blit_line(bx0 + 2,  x1, by0 + 2,  y1, y_buffer, y_stride);
michael@0: 
michael@0:                   bmi++;
michael@0:                 }
michael@0:               }
michael@0:             }
michael@0:           }
michael@0:         } else if (is_inter_mode(mi->mbmi.mode)) {
michael@0:           MV *mv = &mi->mbmi.mv.as_mv;
michael@0:           const int lx0 = x0 + 8;
michael@0:           const int ly0 = y0 + 8;
michael@0: 
michael@0:           x1 = lx0 + (mv->col >> 3);
michael@0:           y1 = ly0 + (mv->row >> 3);
michael@0: 
michael@0:           if (x1 != lx0 && y1 != ly0) {
michael@0:             constrain_line(lx0, &x1, ly0 - 1, &y1, width, height);
michael@0:             vp9_blit_line(lx0,  x1, ly0 - 1,  y1, y_buffer, y_stride);
michael@0: 
michael@0:             constrain_line(lx0, &x1, ly0 + 1, &y1, width, height);
michael@0:             vp9_blit_line(lx0,  x1, ly0 + 1,  y1, y_buffer, y_stride);
michael@0:           } else {
michael@0:             vp9_blit_line(lx0,  x1, ly0,  y1, y_buffer, y_stride);
michael@0:           }
michael@0:         }
michael@0: 
michael@0:         mi++;
michael@0:       }
michael@0:       mi++;
michael@0:     }
michael@0:   }
michael@0: 
michael@0:   /* Color in block modes */
michael@0:   if ((flags & VP9D_DEBUG_CLR_BLK_MODES)
michael@0:       && (ppflags->display_mb_modes_flag || ppflags->display_b_modes_flag)) {
michael@0:     int y, x;
michael@0:     YV12_BUFFER_CONFIG *post = &cm->post_proc_buffer;
michael@0:     int width  = post->y_width;
michael@0:     int height = post->y_height;
michael@0:     uint8_t *y_ptr = cm->post_proc_buffer.y_buffer;
michael@0:     uint8_t *u_ptr = cm->post_proc_buffer.u_buffer;
michael@0:     uint8_t *v_ptr = cm->post_proc_buffer.v_buffer;
michael@0:     int y_stride = cm->post_proc_buffer.y_stride;
michael@0:     MODE_INFO *mi = cm->mi;
michael@0: 
michael@0:     for (y = 0; y < height; y += 16) {
michael@0:       for (x = 0; x < width; x += 16) {
michael@0:         int Y = 0, U = 0, V = 0;
michael@0: 
michael@0:         if (mi->mbmi.mode == I4X4_PRED &&
michael@0:             ((ppflags->display_mb_modes_flag & I4X4_PRED) ||
michael@0:              ppflags->display_b_modes_flag)) {
michael@0:           int by, bx;
michael@0:           uint8_t *yl, *ul, *vl;
michael@0:           union b_mode_info *bmi = mi->bmi;
michael@0: 
michael@0:           yl = y_ptr + x;
michael@0:           ul = u_ptr + (x >> 1);
michael@0:           vl = v_ptr + (x >> 1);
michael@0: 
michael@0:           for (by = 0; by < 16; by += 4) {
michael@0:             for (bx = 0; bx < 16; bx += 4) {
michael@0:               if ((ppflags->display_b_modes_flag & (1 << mi->mbmi.mode))
michael@0:                   || (ppflags->display_mb_modes_flag & I4X4_PRED)) {
michael@0:                 Y = B_PREDICTION_MODE_colors[bmi->as_mode][0];
michael@0:                 U = B_PREDICTION_MODE_colors[bmi->as_mode][1];
michael@0:                 V = B_PREDICTION_MODE_colors[bmi->as_mode][2];
michael@0: 
michael@0:                 vp9_blend_b(yl + bx, ul + (bx >> 1), vl + (bx >> 1), Y, U, V,
michael@0:                     0xc000, y_stride);
michael@0:               }
michael@0:               bmi++;
michael@0:             }
michael@0: 
michael@0:             yl += y_stride * 4;
michael@0:             ul += y_stride * 1;
michael@0:             vl += y_stride * 1;
michael@0:           }
michael@0:         } else if (ppflags->display_mb_modes_flag & (1 << mi->mbmi.mode)) {
michael@0:           Y = MB_PREDICTION_MODE_colors[mi->mbmi.mode][0];
michael@0:           U = MB_PREDICTION_MODE_colors[mi->mbmi.mode][1];
michael@0:           V = MB_PREDICTION_MODE_colors[mi->mbmi.mode][2];
michael@0: 
michael@0:           vp9_blend_mb_inner(y_ptr + x, u_ptr + (x >> 1), v_ptr + (x >> 1),
michael@0:                              Y, U, V, 0xc000, y_stride);
michael@0:         }
michael@0: 
michael@0:         mi++;
michael@0:       }
michael@0:       y_ptr += y_stride * 16;
michael@0:       u_ptr += y_stride * 4;
michael@0:       v_ptr += y_stride * 4;
michael@0: 
michael@0:       mi++;
michael@0:     }
michael@0:   }
michael@0: 
michael@0:   /* Color in frame reference blocks */
michael@0:   if ((flags & VP9D_DEBUG_CLR_FRM_REF_BLKS) &&
michael@0:       ppflags->display_ref_frame_flag) {
michael@0:     int y, x;
michael@0:     YV12_BUFFER_CONFIG *post = &cm->post_proc_buffer;
michael@0:     int width  = post->y_width;
michael@0:     int height = post->y_height;
michael@0:     uint8_t *y_ptr = cm->post_proc_buffer.y_buffer;
michael@0:     uint8_t *u_ptr = cm->post_proc_buffer.u_buffer;
michael@0:     uint8_t *v_ptr = cm->post_proc_buffer.v_buffer;
michael@0:     int y_stride = cm->post_proc_buffer.y_stride;
michael@0:     MODE_INFO *mi = cm->mi;
michael@0: 
michael@0:     for (y = 0; y < height; y += 16) {
michael@0:       for (x = 0; x < width; x += 16) {
michael@0:         int Y = 0, U = 0, V = 0;
michael@0: 
michael@0:         if (ppflags->display_ref_frame_flag & (1 << mi->mbmi.ref_frame)) {
michael@0:           Y = MV_REFERENCE_FRAME_colors[mi->mbmi.ref_frame][0];
michael@0:           U = MV_REFERENCE_FRAME_colors[mi->mbmi.ref_frame][1];
michael@0:           V = MV_REFERENCE_FRAME_colors[mi->mbmi.ref_frame][2];
michael@0: 
michael@0:           vp9_blend_mb_outer(y_ptr + x, u_ptr + (x >> 1), v_ptr + (x >> 1),
michael@0:                              Y, U, V, 0xc000, y_stride);
michael@0:         }
michael@0: 
michael@0:         mi++;
michael@0:       }
michael@0:       y_ptr += y_stride * 16;
michael@0:       u_ptr += y_stride * 4;
michael@0:       v_ptr += y_stride * 4;
michael@0: 
michael@0:       mi++;
michael@0:     }
michael@0:   }
michael@0: #endif
michael@0: 
michael@0:   *dest = cm->post_proc_buffer;
michael@0: 
michael@0:   /* handle problem with extending borders */
michael@0:   dest->y_width = cm->width;
michael@0:   dest->y_height = cm->height;
michael@0:   dest->uv_width = dest->y_width >> cm->subsampling_x;
michael@0:   dest->uv_height = dest->y_height >> cm->subsampling_y;
michael@0: 
michael@0:   return 0;
michael@0: }