1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/media/libvpx/vp9/common/vp9_postproc.c Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,1014 @@
1.4 +/*
1.5 + * Copyright (c) 2010 The WebM project authors. All Rights Reserved.
1.6 + *
1.7 + * Use of this source code is governed by a BSD-style license
1.8 + * that can be found in the LICENSE file in the root of the source
1.9 + * tree. An additional intellectual property rights grant can be found
1.10 + * in the file PATENTS. All contributing project authors may
1.11 + * be found in the AUTHORS file in the root of the source tree.
1.12 + */
1.13 +
1.14 +#include <math.h>
1.15 +#include <stdlib.h>
1.16 +#include <stdio.h>
1.17 +
1.18 +#include "./vpx_config.h"
1.19 +#include "vpx_scale/yv12config.h"
1.20 +#include "vp9/common/vp9_postproc.h"
1.21 +#include "vp9/common/vp9_textblit.h"
1.22 +#include "vpx_scale/vpx_scale.h"
1.23 +#include "vp9/common/vp9_systemdependent.h"
1.24 +#include "./vp9_rtcd.h"
1.25 +#include "./vpx_scale_rtcd.h"
1.26 +
1.27 +#define RGB_TO_YUV(t) \
1.28 + ( (0.257*(float)(t >> 16)) + (0.504*(float)(t >> 8 & 0xff)) + \
1.29 + (0.098*(float)(t & 0xff)) + 16), \
1.30 + (-(0.148*(float)(t >> 16)) - (0.291*(float)(t >> 8 & 0xff)) + \
1.31 + (0.439*(float)(t & 0xff)) + 128), \
1.32 + ( (0.439*(float)(t >> 16)) - (0.368*(float)(t >> 8 & 0xff)) - \
1.33 + (0.071*(float)(t & 0xff)) + 128)
1.34 +
1.35 +/* global constants */
1.36 +#if 0 && CONFIG_POSTPROC_VISUALIZER
1.37 +static const unsigned char MB_PREDICTION_MODE_colors[MB_MODE_COUNT][3] = {
1.38 + { RGB_TO_YUV(0x98FB98) }, /* PaleGreen */
1.39 + { RGB_TO_YUV(0x00FF00) }, /* Green */
1.40 + { RGB_TO_YUV(0xADFF2F) }, /* GreenYellow */
1.41 + { RGB_TO_YUV(0x8F0000) }, /* Dark Red */
1.42 + { RGB_TO_YUV(0x008F8F) }, /* Dark Cyan */
1.43 + { RGB_TO_YUV(0x008F8F) }, /* Dark Cyan */
1.44 + { RGB_TO_YUV(0x008F8F) }, /* Dark Cyan */
1.45 + { RGB_TO_YUV(0x8F0000) }, /* Dark Red */
1.46 + { RGB_TO_YUV(0x8F0000) }, /* Dark Red */
1.47 + { RGB_TO_YUV(0x228B22) }, /* ForestGreen */
1.48 + { RGB_TO_YUV(0x006400) }, /* DarkGreen */
1.49 + { RGB_TO_YUV(0x98F5FF) }, /* Cadet Blue */
1.50 + { RGB_TO_YUV(0x6CA6CD) }, /* Sky Blue */
1.51 + { RGB_TO_YUV(0x00008B) }, /* Dark blue */
1.52 + { RGB_TO_YUV(0x551A8B) }, /* Purple */
1.53 + { RGB_TO_YUV(0xFF0000) } /* Red */
1.54 + { RGB_TO_YUV(0xCC33FF) }, /* Magenta */
1.55 +};
1.56 +
1.57 +static const unsigned char B_PREDICTION_MODE_colors[INTRA_MODES][3] = {
1.58 + { RGB_TO_YUV(0x6633ff) }, /* Purple */
1.59 + { RGB_TO_YUV(0xcc33ff) }, /* Magenta */
1.60 + { RGB_TO_YUV(0xff33cc) }, /* Pink */
1.61 + { RGB_TO_YUV(0xff3366) }, /* Coral */
1.62 + { RGB_TO_YUV(0x3366ff) }, /* Blue */
1.63 + { RGB_TO_YUV(0xed00f5) }, /* Dark Blue */
1.64 + { RGB_TO_YUV(0x2e00b8) }, /* Dark Purple */
1.65 + { RGB_TO_YUV(0xff6633) }, /* Orange */
1.66 + { RGB_TO_YUV(0x33ccff) }, /* Light Blue */
1.67 + { RGB_TO_YUV(0x8ab800) }, /* Green */
1.68 + { RGB_TO_YUV(0xffcc33) }, /* Light Orange */
1.69 + { RGB_TO_YUV(0x33ffcc) }, /* Aqua */
1.70 + { RGB_TO_YUV(0x66ff33) }, /* Light Green */
1.71 + { RGB_TO_YUV(0xccff33) }, /* Yellow */
1.72 +};
1.73 +
1.74 +static const unsigned char MV_REFERENCE_FRAME_colors[MAX_REF_FRAMES][3] = {
1.75 + { RGB_TO_YUV(0x00ff00) }, /* Blue */
1.76 + { RGB_TO_YUV(0x0000ff) }, /* Green */
1.77 + { RGB_TO_YUV(0xffff00) }, /* Yellow */
1.78 + { RGB_TO_YUV(0xff0000) }, /* Red */
1.79 +};
1.80 +#endif
1.81 +
1.82 +static const short kernel5[] = {
1.83 + 1, 1, 4, 1, 1
1.84 +};
1.85 +
1.86 +const short vp9_rv[] = {
1.87 + 8, 5, 2, 2, 8, 12, 4, 9, 8, 3,
1.88 + 0, 3, 9, 0, 0, 0, 8, 3, 14, 4,
1.89 + 10, 1, 11, 14, 1, 14, 9, 6, 12, 11,
1.90 + 8, 6, 10, 0, 0, 8, 9, 0, 3, 14,
1.91 + 8, 11, 13, 4, 2, 9, 0, 3, 9, 6,
1.92 + 1, 2, 3, 14, 13, 1, 8, 2, 9, 7,
1.93 + 3, 3, 1, 13, 13, 6, 6, 5, 2, 7,
1.94 + 11, 9, 11, 8, 7, 3, 2, 0, 13, 13,
1.95 + 14, 4, 12, 5, 12, 10, 8, 10, 13, 10,
1.96 + 4, 14, 4, 10, 0, 8, 11, 1, 13, 7,
1.97 + 7, 14, 6, 14, 13, 2, 13, 5, 4, 4,
1.98 + 0, 10, 0, 5, 13, 2, 12, 7, 11, 13,
1.99 + 8, 0, 4, 10, 7, 2, 7, 2, 2, 5,
1.100 + 3, 4, 7, 3, 3, 14, 14, 5, 9, 13,
1.101 + 3, 14, 3, 6, 3, 0, 11, 8, 13, 1,
1.102 + 13, 1, 12, 0, 10, 9, 7, 6, 2, 8,
1.103 + 5, 2, 13, 7, 1, 13, 14, 7, 6, 7,
1.104 + 9, 6, 10, 11, 7, 8, 7, 5, 14, 8,
1.105 + 4, 4, 0, 8, 7, 10, 0, 8, 14, 11,
1.106 + 3, 12, 5, 7, 14, 3, 14, 5, 2, 6,
1.107 + 11, 12, 12, 8, 0, 11, 13, 1, 2, 0,
1.108 + 5, 10, 14, 7, 8, 0, 4, 11, 0, 8,
1.109 + 0, 3, 10, 5, 8, 0, 11, 6, 7, 8,
1.110 + 10, 7, 13, 9, 2, 5, 1, 5, 10, 2,
1.111 + 4, 3, 5, 6, 10, 8, 9, 4, 11, 14,
1.112 + 0, 10, 0, 5, 13, 2, 12, 7, 11, 13,
1.113 + 8, 0, 4, 10, 7, 2, 7, 2, 2, 5,
1.114 + 3, 4, 7, 3, 3, 14, 14, 5, 9, 13,
1.115 + 3, 14, 3, 6, 3, 0, 11, 8, 13, 1,
1.116 + 13, 1, 12, 0, 10, 9, 7, 6, 2, 8,
1.117 + 5, 2, 13, 7, 1, 13, 14, 7, 6, 7,
1.118 + 9, 6, 10, 11, 7, 8, 7, 5, 14, 8,
1.119 + 4, 4, 0, 8, 7, 10, 0, 8, 14, 11,
1.120 + 3, 12, 5, 7, 14, 3, 14, 5, 2, 6,
1.121 + 11, 12, 12, 8, 0, 11, 13, 1, 2, 0,
1.122 + 5, 10, 14, 7, 8, 0, 4, 11, 0, 8,
1.123 + 0, 3, 10, 5, 8, 0, 11, 6, 7, 8,
1.124 + 10, 7, 13, 9, 2, 5, 1, 5, 10, 2,
1.125 + 4, 3, 5, 6, 10, 8, 9, 4, 11, 14,
1.126 + 3, 8, 3, 7, 8, 5, 11, 4, 12, 3,
1.127 + 11, 9, 14, 8, 14, 13, 4, 3, 1, 2,
1.128 + 14, 6, 5, 4, 4, 11, 4, 6, 2, 1,
1.129 + 5, 8, 8, 12, 13, 5, 14, 10, 12, 13,
1.130 + 0, 9, 5, 5, 11, 10, 13, 9, 10, 13,
1.131 +};
1.132 +
1.133 +
1.134 +/****************************************************************************
1.135 + */
1.136 +void vp9_post_proc_down_and_across_c(const uint8_t *src_ptr,
1.137 + uint8_t *dst_ptr,
1.138 + int src_pixels_per_line,
1.139 + int dst_pixels_per_line,
1.140 + int rows,
1.141 + int cols,
1.142 + int flimit) {
1.143 + uint8_t const *p_src;
1.144 + uint8_t *p_dst;
1.145 + int row;
1.146 + int col;
1.147 + int i;
1.148 + int v;
1.149 + int pitch = src_pixels_per_line;
1.150 + uint8_t d[8];
1.151 + (void)dst_pixels_per_line;
1.152 +
1.153 + for (row = 0; row < rows; row++) {
1.154 + /* post_proc_down for one row */
1.155 + p_src = src_ptr;
1.156 + p_dst = dst_ptr;
1.157 +
1.158 + for (col = 0; col < cols; col++) {
1.159 + int kernel = 4;
1.160 + int v = p_src[col];
1.161 +
1.162 + for (i = -2; i <= 2; i++) {
1.163 + if (abs(v - p_src[col + i * pitch]) > flimit)
1.164 + goto down_skip_convolve;
1.165 +
1.166 + kernel += kernel5[2 + i] * p_src[col + i * pitch];
1.167 + }
1.168 +
1.169 + v = (kernel >> 3);
1.170 + down_skip_convolve:
1.171 + p_dst[col] = v;
1.172 + }
1.173 +
1.174 + /* now post_proc_across */
1.175 + p_src = dst_ptr;
1.176 + p_dst = dst_ptr;
1.177 +
1.178 + for (i = 0; i < 8; i++)
1.179 + d[i] = p_src[i];
1.180 +
1.181 + for (col = 0; col < cols; col++) {
1.182 + int kernel = 4;
1.183 + v = p_src[col];
1.184 +
1.185 + d[col & 7] = v;
1.186 +
1.187 + for (i = -2; i <= 2; i++) {
1.188 + if (abs(v - p_src[col + i]) > flimit)
1.189 + goto across_skip_convolve;
1.190 +
1.191 + kernel += kernel5[2 + i] * p_src[col + i];
1.192 + }
1.193 +
1.194 + d[col & 7] = (kernel >> 3);
1.195 + across_skip_convolve:
1.196 +
1.197 + if (col >= 2)
1.198 + p_dst[col - 2] = d[(col - 2) & 7];
1.199 + }
1.200 +
1.201 + /* handle the last two pixels */
1.202 + p_dst[col - 2] = d[(col - 2) & 7];
1.203 + p_dst[col - 1] = d[(col - 1) & 7];
1.204 +
1.205 +
1.206 + /* next row */
1.207 + src_ptr += pitch;
1.208 + dst_ptr += pitch;
1.209 + }
1.210 +}
1.211 +
1.212 +static int q2mbl(int x) {
1.213 + if (x < 20) x = 20;
1.214 +
1.215 + x = 50 + (x - 50) * 10 / 8;
1.216 + return x * x / 3;
1.217 +}
1.218 +
1.219 +void vp9_mbpost_proc_across_ip_c(uint8_t *src, int pitch,
1.220 + int rows, int cols, int flimit) {
1.221 + int r, c, i;
1.222 +
1.223 + uint8_t *s = src;
1.224 + uint8_t d[16];
1.225 +
1.226 +
1.227 + for (r = 0; r < rows; r++) {
1.228 + int sumsq = 0;
1.229 + int sum = 0;
1.230 +
1.231 + for (i = -8; i <= 6; i++) {
1.232 + sumsq += s[i] * s[i];
1.233 + sum += s[i];
1.234 + d[i + 8] = 0;
1.235 + }
1.236 +
1.237 + for (c = 0; c < cols + 8; c++) {
1.238 + int x = s[c + 7] - s[c - 8];
1.239 + int y = s[c + 7] + s[c - 8];
1.240 +
1.241 + sum += x;
1.242 + sumsq += x * y;
1.243 +
1.244 + d[c & 15] = s[c];
1.245 +
1.246 + if (sumsq * 15 - sum * sum < flimit) {
1.247 + d[c & 15] = (8 + sum + s[c]) >> 4;
1.248 + }
1.249 +
1.250 + s[c - 8] = d[(c - 8) & 15];
1.251 + }
1.252 +
1.253 + s += pitch;
1.254 + }
1.255 +}
1.256 +
1.257 +void vp9_mbpost_proc_down_c(uint8_t *dst, int pitch,
1.258 + int rows, int cols, int flimit) {
1.259 + int r, c, i;
1.260 + const short *rv3 = &vp9_rv[63 & rand()]; // NOLINT
1.261 +
1.262 + for (c = 0; c < cols; c++) {
1.263 + uint8_t *s = &dst[c];
1.264 + int sumsq = 0;
1.265 + int sum = 0;
1.266 + uint8_t d[16];
1.267 + const short *rv2 = rv3 + ((c * 17) & 127);
1.268 +
1.269 + for (i = -8; i <= 6; i++) {
1.270 + sumsq += s[i * pitch] * s[i * pitch];
1.271 + sum += s[i * pitch];
1.272 + }
1.273 +
1.274 + for (r = 0; r < rows + 8; r++) {
1.275 + sumsq += s[7 * pitch] * s[ 7 * pitch] - s[-8 * pitch] * s[-8 * pitch];
1.276 + sum += s[7 * pitch] - s[-8 * pitch];
1.277 + d[r & 15] = s[0];
1.278 +
1.279 + if (sumsq * 15 - sum * sum < flimit) {
1.280 + d[r & 15] = (rv2[r & 127] + sum + s[0]) >> 4;
1.281 + }
1.282 +
1.283 + s[-8 * pitch] = d[(r - 8) & 15];
1.284 + s += pitch;
1.285 + }
1.286 + }
1.287 +}
1.288 +
1.289 +static void deblock_and_de_macro_block(YV12_BUFFER_CONFIG *source,
1.290 + YV12_BUFFER_CONFIG *post,
1.291 + int q,
1.292 + int low_var_thresh,
1.293 + int flag) {
1.294 + double level = 6.0e-05 * q * q * q - .0067 * q * q + .306 * q + .0065;
1.295 + int ppl = (int)(level + .5);
1.296 + (void) low_var_thresh;
1.297 + (void) flag;
1.298 +
1.299 + vp9_post_proc_down_and_across(source->y_buffer, post->y_buffer,
1.300 + source->y_stride, post->y_stride,
1.301 + source->y_height, source->y_width, ppl);
1.302 +
1.303 + vp9_mbpost_proc_across_ip(post->y_buffer, post->y_stride, post->y_height,
1.304 + post->y_width, q2mbl(q));
1.305 +
1.306 + vp9_mbpost_proc_down(post->y_buffer, post->y_stride, post->y_height,
1.307 + post->y_width, q2mbl(q));
1.308 +
1.309 + vp9_post_proc_down_and_across(source->u_buffer, post->u_buffer,
1.310 + source->uv_stride, post->uv_stride,
1.311 + source->uv_height, source->uv_width, ppl);
1.312 + vp9_post_proc_down_and_across(source->v_buffer, post->v_buffer,
1.313 + source->uv_stride, post->uv_stride,
1.314 + source->uv_height, source->uv_width, ppl);
1.315 +}
1.316 +
1.317 +void vp9_deblock(const YV12_BUFFER_CONFIG *src, YV12_BUFFER_CONFIG *dst,
1.318 + int q) {
1.319 + const int ppl = (int)(6.0e-05 * q * q * q - 0.0067 * q * q + 0.306 * q
1.320 + + 0.0065 + 0.5);
1.321 + int i;
1.322 +
1.323 + const uint8_t *const srcs[4] = {src->y_buffer, src->u_buffer, src->v_buffer,
1.324 + src->alpha_buffer};
1.325 + const int src_strides[4] = {src->y_stride, src->uv_stride, src->uv_stride,
1.326 + src->alpha_stride};
1.327 + const int src_widths[4] = {src->y_width, src->uv_width, src->uv_width,
1.328 + src->alpha_width};
1.329 + const int src_heights[4] = {src->y_height, src->uv_height, src->uv_height,
1.330 + src->alpha_height};
1.331 +
1.332 + uint8_t *const dsts[4] = {dst->y_buffer, dst->u_buffer, dst->v_buffer,
1.333 + dst->alpha_buffer};
1.334 + const int dst_strides[4] = {dst->y_stride, dst->uv_stride, dst->uv_stride,
1.335 + dst->alpha_stride};
1.336 +
1.337 + for (i = 0; i < MAX_MB_PLANE; ++i)
1.338 + vp9_post_proc_down_and_across(srcs[i], dsts[i],
1.339 + src_strides[i], dst_strides[i],
1.340 + src_heights[i], src_widths[i], ppl);
1.341 +}
1.342 +
1.343 +void vp9_denoise(const YV12_BUFFER_CONFIG *src, YV12_BUFFER_CONFIG *dst,
1.344 + int q) {
1.345 + const int ppl = (int)(6.0e-05 * q * q * q - 0.0067 * q * q + 0.306 * q
1.346 + + 0.0065 + 0.5);
1.347 + int i;
1.348 +
1.349 + const uint8_t *const srcs[4] = {src->y_buffer, src->u_buffer, src->v_buffer,
1.350 + src->alpha_buffer};
1.351 + const int src_strides[4] = {src->y_stride, src->uv_stride, src->uv_stride,
1.352 + src->alpha_stride};
1.353 + const int src_widths[4] = {src->y_width, src->uv_width, src->uv_width,
1.354 + src->alpha_width};
1.355 + const int src_heights[4] = {src->y_height, src->uv_height, src->uv_height,
1.356 + src->alpha_height};
1.357 +
1.358 + uint8_t *const dsts[4] = {dst->y_buffer, dst->u_buffer, dst->v_buffer,
1.359 + dst->alpha_buffer};
1.360 + const int dst_strides[4] = {dst->y_stride, dst->uv_stride, dst->uv_stride,
1.361 + dst->alpha_stride};
1.362 +
1.363 + for (i = 0; i < MAX_MB_PLANE; ++i) {
1.364 + const int src_stride = src_strides[i];
1.365 + const uint8_t *const src = srcs[i] + 2 * src_stride + 2;
1.366 + const int src_width = src_widths[i] - 4;
1.367 + const int src_height = src_heights[i] - 4;
1.368 +
1.369 + const int dst_stride = dst_strides[i];
1.370 + uint8_t *const dst = dsts[i] + 2 * dst_stride + 2;
1.371 +
1.372 + vp9_post_proc_down_and_across(src, dst, src_stride, dst_stride,
1.373 + src_height, src_width, ppl);
1.374 + }
1.375 +}
1.376 +
1.377 +double vp9_gaussian(double sigma, double mu, double x) {
1.378 + return 1 / (sigma * sqrt(2.0 * 3.14159265)) *
1.379 + (exp(-(x - mu) * (x - mu) / (2 * sigma * sigma)));
1.380 +}
1.381 +
1.382 +static void fillrd(struct postproc_state *state, int q, int a) {
1.383 + char char_dist[300];
1.384 +
1.385 + double sigma;
1.386 + int ai = a, qi = q, i;
1.387 +
1.388 + vp9_clear_system_state();
1.389 +
1.390 + sigma = ai + .5 + .6 * (63 - qi) / 63.0;
1.391 +
1.392 + /* set up a lookup table of 256 entries that matches
1.393 + * a gaussian distribution with sigma determined by q.
1.394 + */
1.395 + {
1.396 + double i;
1.397 + int next, j;
1.398 +
1.399 + next = 0;
1.400 +
1.401 + for (i = -32; i < 32; i++) {
1.402 + int a = (int)(.5 + 256 * vp9_gaussian(sigma, 0, i));
1.403 +
1.404 + if (a) {
1.405 + for (j = 0; j < a; j++) {
1.406 + char_dist[next + j] = (char) i;
1.407 + }
1.408 +
1.409 + next = next + j;
1.410 + }
1.411 + }
1.412 +
1.413 + for (; next < 256; next++)
1.414 + char_dist[next] = 0;
1.415 + }
1.416 +
1.417 + for (i = 0; i < 3072; i++) {
1.418 + state->noise[i] = char_dist[rand() & 0xff]; // NOLINT
1.419 + }
1.420 +
1.421 + for (i = 0; i < 16; i++) {
1.422 + state->blackclamp[i] = -char_dist[0];
1.423 + state->whiteclamp[i] = -char_dist[0];
1.424 + state->bothclamp[i] = -2 * char_dist[0];
1.425 + }
1.426 +
1.427 + state->last_q = q;
1.428 + state->last_noise = a;
1.429 +}
1.430 +
1.431 +/****************************************************************************
1.432 + *
1.433 + * ROUTINE : plane_add_noise_c
1.434 + *
1.435 + * INPUTS : unsigned char *Start starting address of buffer to
1.436 + * add gaussian noise to
1.437 + * unsigned int width width of plane
1.438 + * unsigned int height height of plane
1.439 + * int pitch distance between subsequent lines of frame
1.440 + * int q quantizer used to determine amount of noise
1.441 + * to add
1.442 + *
1.443 + * OUTPUTS : None.
1.444 + *
1.445 + * RETURNS : void.
1.446 + *
1.447 + * FUNCTION : adds gaussian noise to a plane of pixels
1.448 + *
1.449 + * SPECIAL NOTES : None.
1.450 + *
1.451 + ****************************************************************************/
1.452 +void vp9_plane_add_noise_c(uint8_t *start, char *noise,
1.453 + char blackclamp[16],
1.454 + char whiteclamp[16],
1.455 + char bothclamp[16],
1.456 + unsigned int width, unsigned int height, int pitch) {
1.457 + unsigned int i, j;
1.458 +
1.459 + for (i = 0; i < height; i++) {
1.460 + uint8_t *pos = start + i * pitch;
1.461 + char *ref = (char *)(noise + (rand() & 0xff)); // NOLINT
1.462 +
1.463 + for (j = 0; j < width; j++) {
1.464 + if (pos[j] < blackclamp[0])
1.465 + pos[j] = blackclamp[0];
1.466 +
1.467 + if (pos[j] > 255 + whiteclamp[0])
1.468 + pos[j] = 255 + whiteclamp[0];
1.469 +
1.470 + pos[j] += ref[j];
1.471 + }
1.472 + }
1.473 +}
1.474 +
1.475 +/* Blend the macro block with a solid colored square. Leave the
1.476 + * edges unblended to give distinction to macro blocks in areas
1.477 + * filled with the same color block.
1.478 + */
1.479 +void vp9_blend_mb_inner_c(uint8_t *y, uint8_t *u, uint8_t *v,
1.480 + int y1, int u1, int v1, int alpha, int stride) {
1.481 + int i, j;
1.482 + int y1_const = y1 * ((1 << 16) - alpha);
1.483 + int u1_const = u1 * ((1 << 16) - alpha);
1.484 + int v1_const = v1 * ((1 << 16) - alpha);
1.485 +
1.486 + y += 2 * stride + 2;
1.487 + for (i = 0; i < 12; i++) {
1.488 + for (j = 0; j < 12; j++) {
1.489 + y[j] = (y[j] * alpha + y1_const) >> 16;
1.490 + }
1.491 + y += stride;
1.492 + }
1.493 +
1.494 + stride >>= 1;
1.495 +
1.496 + u += stride + 1;
1.497 + v += stride + 1;
1.498 +
1.499 + for (i = 0; i < 6; i++) {
1.500 + for (j = 0; j < 6; j++) {
1.501 + u[j] = (u[j] * alpha + u1_const) >> 16;
1.502 + v[j] = (v[j] * alpha + v1_const) >> 16;
1.503 + }
1.504 + u += stride;
1.505 + v += stride;
1.506 + }
1.507 +}
1.508 +
1.509 +/* Blend only the edge of the macro block. Leave center
1.510 + * unblended to allow for other visualizations to be layered.
1.511 + */
1.512 +void vp9_blend_mb_outer_c(uint8_t *y, uint8_t *u, uint8_t *v,
1.513 + int y1, int u1, int v1, int alpha, int stride) {
1.514 + int i, j;
1.515 + int y1_const = y1 * ((1 << 16) - alpha);
1.516 + int u1_const = u1 * ((1 << 16) - alpha);
1.517 + int v1_const = v1 * ((1 << 16) - alpha);
1.518 +
1.519 + for (i = 0; i < 2; i++) {
1.520 + for (j = 0; j < 16; j++) {
1.521 + y[j] = (y[j] * alpha + y1_const) >> 16;
1.522 + }
1.523 + y += stride;
1.524 + }
1.525 +
1.526 + for (i = 0; i < 12; i++) {
1.527 + y[0] = (y[0] * alpha + y1_const) >> 16;
1.528 + y[1] = (y[1] * alpha + y1_const) >> 16;
1.529 + y[14] = (y[14] * alpha + y1_const) >> 16;
1.530 + y[15] = (y[15] * alpha + y1_const) >> 16;
1.531 + y += stride;
1.532 + }
1.533 +
1.534 + for (i = 0; i < 2; i++) {
1.535 + for (j = 0; j < 16; j++) {
1.536 + y[j] = (y[j] * alpha + y1_const) >> 16;
1.537 + }
1.538 + y += stride;
1.539 + }
1.540 +
1.541 + stride >>= 1;
1.542 +
1.543 + for (j = 0; j < 8; j++) {
1.544 + u[j] = (u[j] * alpha + u1_const) >> 16;
1.545 + v[j] = (v[j] * alpha + v1_const) >> 16;
1.546 + }
1.547 + u += stride;
1.548 + v += stride;
1.549 +
1.550 + for (i = 0; i < 6; i++) {
1.551 + u[0] = (u[0] * alpha + u1_const) >> 16;
1.552 + v[0] = (v[0] * alpha + v1_const) >> 16;
1.553 +
1.554 + u[7] = (u[7] * alpha + u1_const) >> 16;
1.555 + v[7] = (v[7] * alpha + v1_const) >> 16;
1.556 +
1.557 + u += stride;
1.558 + v += stride;
1.559 + }
1.560 +
1.561 + for (j = 0; j < 8; j++) {
1.562 + u[j] = (u[j] * alpha + u1_const) >> 16;
1.563 + v[j] = (v[j] * alpha + v1_const) >> 16;
1.564 + }
1.565 +}
1.566 +
1.567 +void vp9_blend_b_c(uint8_t *y, uint8_t *u, uint8_t *v,
1.568 + int y1, int u1, int v1, int alpha, int stride) {
1.569 + int i, j;
1.570 + int y1_const = y1 * ((1 << 16) - alpha);
1.571 + int u1_const = u1 * ((1 << 16) - alpha);
1.572 + int v1_const = v1 * ((1 << 16) - alpha);
1.573 +
1.574 + for (i = 0; i < 4; i++) {
1.575 + for (j = 0; j < 4; j++) {
1.576 + y[j] = (y[j] * alpha + y1_const) >> 16;
1.577 + }
1.578 + y += stride;
1.579 + }
1.580 +
1.581 + stride >>= 1;
1.582 +
1.583 + for (i = 0; i < 2; i++) {
1.584 + for (j = 0; j < 2; j++) {
1.585 + u[j] = (u[j] * alpha + u1_const) >> 16;
1.586 + v[j] = (v[j] * alpha + v1_const) >> 16;
1.587 + }
1.588 + u += stride;
1.589 + v += stride;
1.590 + }
1.591 +}
1.592 +
1.593 +static void constrain_line(int x0, int *x1, int y0, int *y1,
1.594 + int width, int height) {
1.595 + int dx;
1.596 + int dy;
1.597 +
1.598 + if (*x1 > width) {
1.599 + dx = *x1 - x0;
1.600 + dy = *y1 - y0;
1.601 +
1.602 + *x1 = width;
1.603 + if (dx)
1.604 + *y1 = ((width - x0) * dy) / dx + y0;
1.605 + }
1.606 + if (*x1 < 0) {
1.607 + dx = *x1 - x0;
1.608 + dy = *y1 - y0;
1.609 +
1.610 + *x1 = 0;
1.611 + if (dx)
1.612 + *y1 = ((0 - x0) * dy) / dx + y0;
1.613 + }
1.614 + if (*y1 > height) {
1.615 + dx = *x1 - x0;
1.616 + dy = *y1 - y0;
1.617 +
1.618 + *y1 = height;
1.619 + if (dy)
1.620 + *x1 = ((height - y0) * dx) / dy + x0;
1.621 + }
1.622 + if (*y1 < 0) {
1.623 + dx = *x1 - x0;
1.624 + dy = *y1 - y0;
1.625 +
1.626 + *y1 = 0;
1.627 + if (dy)
1.628 + *x1 = ((0 - y0) * dx) / dy + x0;
1.629 + }
1.630 +}
1.631 +
1.632 +int vp9_post_proc_frame(struct VP9Common *cm,
1.633 + YV12_BUFFER_CONFIG *dest, vp9_ppflags_t *ppflags) {
1.634 + int q = cm->lf.filter_level * 10 / 6;
1.635 + int flags = ppflags->post_proc_flag;
1.636 + int deblock_level = ppflags->deblocking_level;
1.637 + int noise_level = ppflags->noise_level;
1.638 +
1.639 + if (!cm->frame_to_show)
1.640 + return -1;
1.641 +
1.642 + if (q > 63)
1.643 + q = 63;
1.644 +
1.645 + if (!flags) {
1.646 + *dest = *cm->frame_to_show;
1.647 + return 0;
1.648 + }
1.649 +
1.650 +#if ARCH_X86||ARCH_X86_64
1.651 + vpx_reset_mmx_state();
1.652 +#endif
1.653 +
1.654 + if (flags & VP9D_DEMACROBLOCK) {
1.655 + deblock_and_de_macro_block(cm->frame_to_show, &cm->post_proc_buffer,
1.656 + q + (deblock_level - 5) * 10, 1, 0);
1.657 + } else if (flags & VP9D_DEBLOCK) {
1.658 + vp9_deblock(cm->frame_to_show, &cm->post_proc_buffer, q);
1.659 + } else {
1.660 + vp8_yv12_copy_frame(cm->frame_to_show, &cm->post_proc_buffer);
1.661 + }
1.662 +
1.663 + if (flags & VP9D_ADDNOISE) {
1.664 + if (cm->postproc_state.last_q != q
1.665 + || cm->postproc_state.last_noise != noise_level) {
1.666 + fillrd(&cm->postproc_state, 63 - q, noise_level);
1.667 + }
1.668 +
1.669 + vp9_plane_add_noise(cm->post_proc_buffer.y_buffer,
1.670 + cm->postproc_state.noise,
1.671 + cm->postproc_state.blackclamp,
1.672 + cm->postproc_state.whiteclamp,
1.673 + cm->postproc_state.bothclamp,
1.674 + cm->post_proc_buffer.y_width,
1.675 + cm->post_proc_buffer.y_height,
1.676 + cm->post_proc_buffer.y_stride);
1.677 + }
1.678 +
1.679 +#if 0 && CONFIG_POSTPROC_VISUALIZER
1.680 + if (flags & VP9D_DEBUG_TXT_FRAME_INFO) {
1.681 + char message[512];
1.682 + snprintf(message, sizeof(message) -1,
1.683 + "F%1dG%1dQ%3dF%3dP%d_s%dx%d",
1.684 + (cm->frame_type == KEY_FRAME),
1.685 + cm->refresh_golden_frame,
1.686 + cm->base_qindex,
1.687 + cm->filter_level,
1.688 + flags,
1.689 + cm->mb_cols, cm->mb_rows);
1.690 + vp9_blit_text(message, cm->post_proc_buffer.y_buffer,
1.691 + cm->post_proc_buffer.y_stride);
1.692 + }
1.693 +
1.694 + if (flags & VP9D_DEBUG_TXT_MBLK_MODES) {
1.695 + int i, j;
1.696 + uint8_t *y_ptr;
1.697 + YV12_BUFFER_CONFIG *post = &cm->post_proc_buffer;
1.698 + int mb_rows = post->y_height >> 4;
1.699 + int mb_cols = post->y_width >> 4;
1.700 + int mb_index = 0;
1.701 + MODE_INFO *mi = cm->mi;
1.702 +
1.703 + y_ptr = post->y_buffer + 4 * post->y_stride + 4;
1.704 +
1.705 + /* vp9_filter each macro block */
1.706 + for (i = 0; i < mb_rows; i++) {
1.707 + for (j = 0; j < mb_cols; j++) {
1.708 + char zz[4];
1.709 +
1.710 + snprintf(zz, sizeof(zz) - 1, "%c", mi[mb_index].mbmi.mode + 'a');
1.711 +
1.712 + vp9_blit_text(zz, y_ptr, post->y_stride);
1.713 + mb_index++;
1.714 + y_ptr += 16;
1.715 + }
1.716 +
1.717 + mb_index++; /* border */
1.718 + y_ptr += post->y_stride * 16 - post->y_width;
1.719 + }
1.720 + }
1.721 +
1.722 + if (flags & VP9D_DEBUG_TXT_DC_DIFF) {
1.723 + int i, j;
1.724 + uint8_t *y_ptr;
1.725 + YV12_BUFFER_CONFIG *post = &cm->post_proc_buffer;
1.726 + int mb_rows = post->y_height >> 4;
1.727 + int mb_cols = post->y_width >> 4;
1.728 + int mb_index = 0;
1.729 + MODE_INFO *mi = cm->mi;
1.730 +
1.731 + y_ptr = post->y_buffer + 4 * post->y_stride + 4;
1.732 +
1.733 + /* vp9_filter each macro block */
1.734 + for (i = 0; i < mb_rows; i++) {
1.735 + for (j = 0; j < mb_cols; j++) {
1.736 + char zz[4];
1.737 + int dc_diff = !(mi[mb_index].mbmi.mode != I4X4_PRED &&
1.738 + mi[mb_index].mbmi.mode != SPLITMV &&
1.739 + mi[mb_index].mbmi.skip_coeff);
1.740 +
1.741 + if (cm->frame_type == KEY_FRAME)
1.742 + snprintf(zz, sizeof(zz) - 1, "a");
1.743 + else
1.744 + snprintf(zz, sizeof(zz) - 1, "%c", dc_diff + '0');
1.745 +
1.746 + vp9_blit_text(zz, y_ptr, post->y_stride);
1.747 + mb_index++;
1.748 + y_ptr += 16;
1.749 + }
1.750 +
1.751 + mb_index++; /* border */
1.752 + y_ptr += post->y_stride * 16 - post->y_width;
1.753 + }
1.754 + }
1.755 +
1.756 + if (flags & VP9D_DEBUG_TXT_RATE_INFO) {
1.757 + char message[512];
1.758 + snprintf(message, sizeof(message),
1.759 + "Bitrate: %10.2f framerate: %10.2f ",
1.760 + cm->bitrate, cm->framerate);
1.761 + vp9_blit_text(message, cm->post_proc_buffer.y_buffer,
1.762 + cm->post_proc_buffer.y_stride);
1.763 + }
1.764 +
1.765 + /* Draw motion vectors */
1.766 + if ((flags & VP9D_DEBUG_DRAW_MV) && ppflags->display_mv_flag) {
1.767 + YV12_BUFFER_CONFIG *post = &cm->post_proc_buffer;
1.768 + int width = post->y_width;
1.769 + int height = post->y_height;
1.770 + uint8_t *y_buffer = cm->post_proc_buffer.y_buffer;
1.771 + int y_stride = cm->post_proc_buffer.y_stride;
1.772 + MODE_INFO *mi = cm->mi;
1.773 + int x0, y0;
1.774 +
1.775 + for (y0 = 0; y0 < height; y0 += 16) {
1.776 + for (x0 = 0; x0 < width; x0 += 16) {
1.777 + int x1, y1;
1.778 +
1.779 + if (!(ppflags->display_mv_flag & (1 << mi->mbmi.mode))) {
1.780 + mi++;
1.781 + continue;
1.782 + }
1.783 +
1.784 + if (mi->mbmi.mode == SPLITMV) {
1.785 + switch (mi->mbmi.partitioning) {
1.786 + case PARTITIONING_16X8 : { /* mv_top_bottom */
1.787 + union b_mode_info *bmi = &mi->bmi[0];
1.788 + MV *mv = &bmi->mv.as_mv;
1.789 +
1.790 + x1 = x0 + 8 + (mv->col >> 3);
1.791 + y1 = y0 + 4 + (mv->row >> 3);
1.792 +
1.793 + constrain_line(x0 + 8, &x1, y0 + 4, &y1, width, height);
1.794 + vp9_blit_line(x0 + 8, x1, y0 + 4, y1, y_buffer, y_stride);
1.795 +
1.796 + bmi = &mi->bmi[8];
1.797 +
1.798 + x1 = x0 + 8 + (mv->col >> 3);
1.799 + y1 = y0 + 12 + (mv->row >> 3);
1.800 +
1.801 + constrain_line(x0 + 8, &x1, y0 + 12, &y1, width, height);
1.802 + vp9_blit_line(x0 + 8, x1, y0 + 12, y1, y_buffer, y_stride);
1.803 +
1.804 + break;
1.805 + }
1.806 + case PARTITIONING_8X16 : { /* mv_left_right */
1.807 + union b_mode_info *bmi = &mi->bmi[0];
1.808 + MV *mv = &bmi->mv.as_mv;
1.809 +
1.810 + x1 = x0 + 4 + (mv->col >> 3);
1.811 + y1 = y0 + 8 + (mv->row >> 3);
1.812 +
1.813 + constrain_line(x0 + 4, &x1, y0 + 8, &y1, width, height);
1.814 + vp9_blit_line(x0 + 4, x1, y0 + 8, y1, y_buffer, y_stride);
1.815 +
1.816 + bmi = &mi->bmi[2];
1.817 +
1.818 + x1 = x0 + 12 + (mv->col >> 3);
1.819 + y1 = y0 + 8 + (mv->row >> 3);
1.820 +
1.821 + constrain_line(x0 + 12, &x1, y0 + 8, &y1, width, height);
1.822 + vp9_blit_line(x0 + 12, x1, y0 + 8, y1, y_buffer, y_stride);
1.823 +
1.824 + break;
1.825 + }
1.826 + case PARTITIONING_8X8 : { /* mv_quarters */
1.827 + union b_mode_info *bmi = &mi->bmi[0];
1.828 + MV *mv = &bmi->mv.as_mv;
1.829 +
1.830 + x1 = x0 + 4 + (mv->col >> 3);
1.831 + y1 = y0 + 4 + (mv->row >> 3);
1.832 +
1.833 + constrain_line(x0 + 4, &x1, y0 + 4, &y1, width, height);
1.834 + vp9_blit_line(x0 + 4, x1, y0 + 4, y1, y_buffer, y_stride);
1.835 +
1.836 + bmi = &mi->bmi[2];
1.837 +
1.838 + x1 = x0 + 12 + (mv->col >> 3);
1.839 + y1 = y0 + 4 + (mv->row >> 3);
1.840 +
1.841 + constrain_line(x0 + 12, &x1, y0 + 4, &y1, width, height);
1.842 + vp9_blit_line(x0 + 12, x1, y0 + 4, y1, y_buffer, y_stride);
1.843 +
1.844 + bmi = &mi->bmi[8];
1.845 +
1.846 + x1 = x0 + 4 + (mv->col >> 3);
1.847 + y1 = y0 + 12 + (mv->row >> 3);
1.848 +
1.849 + constrain_line(x0 + 4, &x1, y0 + 12, &y1, width, height);
1.850 + vp9_blit_line(x0 + 4, x1, y0 + 12, y1, y_buffer, y_stride);
1.851 +
1.852 + bmi = &mi->bmi[10];
1.853 +
1.854 + x1 = x0 + 12 + (mv->col >> 3);
1.855 + y1 = y0 + 12 + (mv->row >> 3);
1.856 +
1.857 + constrain_line(x0 + 12, &x1, y0 + 12, &y1, width, height);
1.858 + vp9_blit_line(x0 + 12, x1, y0 + 12, y1, y_buffer, y_stride);
1.859 + break;
1.860 + }
1.861 + case PARTITIONING_4X4:
1.862 + default : {
1.863 + union b_mode_info *bmi = mi->bmi;
1.864 + int bx0, by0;
1.865 +
1.866 + for (by0 = y0; by0 < (y0 + 16); by0 += 4) {
1.867 + for (bx0 = x0; bx0 < (x0 + 16); bx0 += 4) {
1.868 + MV *mv = &bmi->mv.as_mv;
1.869 +
1.870 + x1 = bx0 + 2 + (mv->col >> 3);
1.871 + y1 = by0 + 2 + (mv->row >> 3);
1.872 +
1.873 + constrain_line(bx0 + 2, &x1, by0 + 2, &y1, width, height);
1.874 + vp9_blit_line(bx0 + 2, x1, by0 + 2, y1, y_buffer, y_stride);
1.875 +
1.876 + bmi++;
1.877 + }
1.878 + }
1.879 + }
1.880 + }
1.881 + } else if (is_inter_mode(mi->mbmi.mode)) {
1.882 + MV *mv = &mi->mbmi.mv.as_mv;
1.883 + const int lx0 = x0 + 8;
1.884 + const int ly0 = y0 + 8;
1.885 +
1.886 + x1 = lx0 + (mv->col >> 3);
1.887 + y1 = ly0 + (mv->row >> 3);
1.888 +
1.889 + if (x1 != lx0 && y1 != ly0) {
1.890 + constrain_line(lx0, &x1, ly0 - 1, &y1, width, height);
1.891 + vp9_blit_line(lx0, x1, ly0 - 1, y1, y_buffer, y_stride);
1.892 +
1.893 + constrain_line(lx0, &x1, ly0 + 1, &y1, width, height);
1.894 + vp9_blit_line(lx0, x1, ly0 + 1, y1, y_buffer, y_stride);
1.895 + } else {
1.896 + vp9_blit_line(lx0, x1, ly0, y1, y_buffer, y_stride);
1.897 + }
1.898 + }
1.899 +
1.900 + mi++;
1.901 + }
1.902 + mi++;
1.903 + }
1.904 + }
1.905 +
1.906 + /* Color in block modes */
1.907 + if ((flags & VP9D_DEBUG_CLR_BLK_MODES)
1.908 + && (ppflags->display_mb_modes_flag || ppflags->display_b_modes_flag)) {
1.909 + int y, x;
1.910 + YV12_BUFFER_CONFIG *post = &cm->post_proc_buffer;
1.911 + int width = post->y_width;
1.912 + int height = post->y_height;
1.913 + uint8_t *y_ptr = cm->post_proc_buffer.y_buffer;
1.914 + uint8_t *u_ptr = cm->post_proc_buffer.u_buffer;
1.915 + uint8_t *v_ptr = cm->post_proc_buffer.v_buffer;
1.916 + int y_stride = cm->post_proc_buffer.y_stride;
1.917 + MODE_INFO *mi = cm->mi;
1.918 +
1.919 + for (y = 0; y < height; y += 16) {
1.920 + for (x = 0; x < width; x += 16) {
1.921 + int Y = 0, U = 0, V = 0;
1.922 +
1.923 + if (mi->mbmi.mode == I4X4_PRED &&
1.924 + ((ppflags->display_mb_modes_flag & I4X4_PRED) ||
1.925 + ppflags->display_b_modes_flag)) {
1.926 + int by, bx;
1.927 + uint8_t *yl, *ul, *vl;
1.928 + union b_mode_info *bmi = mi->bmi;
1.929 +
1.930 + yl = y_ptr + x;
1.931 + ul = u_ptr + (x >> 1);
1.932 + vl = v_ptr + (x >> 1);
1.933 +
1.934 + for (by = 0; by < 16; by += 4) {
1.935 + for (bx = 0; bx < 16; bx += 4) {
1.936 + if ((ppflags->display_b_modes_flag & (1 << mi->mbmi.mode))
1.937 + || (ppflags->display_mb_modes_flag & I4X4_PRED)) {
1.938 + Y = B_PREDICTION_MODE_colors[bmi->as_mode][0];
1.939 + U = B_PREDICTION_MODE_colors[bmi->as_mode][1];
1.940 + V = B_PREDICTION_MODE_colors[bmi->as_mode][2];
1.941 +
1.942 + vp9_blend_b(yl + bx, ul + (bx >> 1), vl + (bx >> 1), Y, U, V,
1.943 + 0xc000, y_stride);
1.944 + }
1.945 + bmi++;
1.946 + }
1.947 +
1.948 + yl += y_stride * 4;
1.949 + ul += y_stride * 1;
1.950 + vl += y_stride * 1;
1.951 + }
1.952 + } else if (ppflags->display_mb_modes_flag & (1 << mi->mbmi.mode)) {
1.953 + Y = MB_PREDICTION_MODE_colors[mi->mbmi.mode][0];
1.954 + U = MB_PREDICTION_MODE_colors[mi->mbmi.mode][1];
1.955 + V = MB_PREDICTION_MODE_colors[mi->mbmi.mode][2];
1.956 +
1.957 + vp9_blend_mb_inner(y_ptr + x, u_ptr + (x >> 1), v_ptr + (x >> 1),
1.958 + Y, U, V, 0xc000, y_stride);
1.959 + }
1.960 +
1.961 + mi++;
1.962 + }
1.963 + y_ptr += y_stride * 16;
1.964 + u_ptr += y_stride * 4;
1.965 + v_ptr += y_stride * 4;
1.966 +
1.967 + mi++;
1.968 + }
1.969 + }
1.970 +
1.971 + /* Color in frame reference blocks */
1.972 + if ((flags & VP9D_DEBUG_CLR_FRM_REF_BLKS) &&
1.973 + ppflags->display_ref_frame_flag) {
1.974 + int y, x;
1.975 + YV12_BUFFER_CONFIG *post = &cm->post_proc_buffer;
1.976 + int width = post->y_width;
1.977 + int height = post->y_height;
1.978 + uint8_t *y_ptr = cm->post_proc_buffer.y_buffer;
1.979 + uint8_t *u_ptr = cm->post_proc_buffer.u_buffer;
1.980 + uint8_t *v_ptr = cm->post_proc_buffer.v_buffer;
1.981 + int y_stride = cm->post_proc_buffer.y_stride;
1.982 + MODE_INFO *mi = cm->mi;
1.983 +
1.984 + for (y = 0; y < height; y += 16) {
1.985 + for (x = 0; x < width; x += 16) {
1.986 + int Y = 0, U = 0, V = 0;
1.987 +
1.988 + if (ppflags->display_ref_frame_flag & (1 << mi->mbmi.ref_frame)) {
1.989 + Y = MV_REFERENCE_FRAME_colors[mi->mbmi.ref_frame][0];
1.990 + U = MV_REFERENCE_FRAME_colors[mi->mbmi.ref_frame][1];
1.991 + V = MV_REFERENCE_FRAME_colors[mi->mbmi.ref_frame][2];
1.992 +
1.993 + vp9_blend_mb_outer(y_ptr + x, u_ptr + (x >> 1), v_ptr + (x >> 1),
1.994 + Y, U, V, 0xc000, y_stride);
1.995 + }
1.996 +
1.997 + mi++;
1.998 + }
1.999 + y_ptr += y_stride * 16;
1.1000 + u_ptr += y_stride * 4;
1.1001 + v_ptr += y_stride * 4;
1.1002 +
1.1003 + mi++;
1.1004 + }
1.1005 + }
1.1006 +#endif
1.1007 +
1.1008 + *dest = cm->post_proc_buffer;
1.1009 +
1.1010 + /* handle problem with extending borders */
1.1011 + dest->y_width = cm->width;
1.1012 + dest->y_height = cm->height;
1.1013 + dest->uv_width = dest->y_width >> cm->subsampling_x;
1.1014 + dest->uv_height = dest->y_height >> cm->subsampling_y;
1.1015 +
1.1016 + return 0;
1.1017 +}
