1 /* 2 * Copyright (c) 2016, Alliance for Open Media. All rights reserved 3 * 4 * This source code is subject to the terms of the BSD 2 Clause License and 5 * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License 6 * was not distributed with this source code in the LICENSE file, you can 7 * obtain it at www.aomedia.org/license/software. If the Alliance for Open 8 * Media Patent License 1.0 was not distributed with this source code in the 9 * PATENTS file, you can obtain it at www.aomedia.org/license/patent. 10 */ 11 12 #include <assert.h> 13 #include <math.h> 14 15 #include "config/aom_dsp_rtcd.h" 16 17 #include "aom_dsp/psnr.h" 18 #include "aom_scale/yv12config.h" 19 20 double aom_sse_to_psnr(double samples, double peak, double sse) { 21 if (sse > 0.0) { 22 const double psnr = 10.0 * log10(samples * peak * peak / sse); 23 return psnr > MAX_PSNR ? MAX_PSNR : psnr; 24 } else { 25 return MAX_PSNR; 26 } 27 } 28 29 /* TODO(yaowu): The block_variance calls the unoptimized versions of variance() 30 * and highbd_8_variance(). It should not. 31 */ 32 static void encoder_variance(const uint8_t *a, int a_stride, const uint8_t *b, 33 int b_stride, int w, int h, unsigned int *sse, 34 int *sum) { 35 int i, j; 36 37 *sum = 0; 38 *sse = 0; 39 40 for (i = 0; i < h; i++) { 41 for (j = 0; j < w; j++) { 42 const int diff = a[j] - b[j]; 43 *sum += diff; 44 *sse += diff * diff; 45 } 46 47 a += a_stride; 48 b += b_stride; 49 } 50 } 51 52 static void encoder_highbd_variance64(const uint8_t *a8, int a_stride, 53 const uint8_t *b8, int b_stride, int w, 54 int h, uint64_t *sse, int64_t *sum) { 55 const uint16_t *a = CONVERT_TO_SHORTPTR(a8); 56 const uint16_t *b = CONVERT_TO_SHORTPTR(b8); 57 int64_t tsum = 0; 58 uint64_t tsse = 0; 59 for (int i = 0; i < h; ++i) { 60 int32_t lsum = 0; 61 for (int j = 0; j < w; ++j) { 62 const int diff = a[j] - b[j]; 63 lsum += diff; 64 tsse += (uint32_t)(diff * diff); 65 } 66 tsum += lsum; 67 a += a_stride; 68 b += b_stride; 69 } 70 *sum = tsum; 71 *sse = tsse; 72 } 73 74 static void encoder_highbd_8_variance(const uint8_t *a8, int a_stride, 75 const uint8_t *b8, int b_stride, int w, 76 int h, unsigned int *sse, int *sum) { 77 uint64_t sse_long = 0; 78 int64_t sum_long = 0; 79 encoder_highbd_variance64(a8, a_stride, b8, b_stride, w, h, &sse_long, 80 &sum_long); 81 *sse = (unsigned int)sse_long; 82 *sum = (int)sum_long; 83 } 84 85 static int64_t get_sse(const uint8_t *a, int a_stride, const uint8_t *b, 86 int b_stride, int width, int height) { 87 const int dw = width % 16; 88 const int dh = height % 16; 89 int64_t total_sse = 0; 90 unsigned int sse = 0; 91 int sum = 0; 92 int x, y; 93 94 if (dw > 0) { 95 encoder_variance(&a[width - dw], a_stride, &b[width - dw], b_stride, dw, 96 height, &sse, &sum); 97 total_sse += sse; 98 } 99 100 if (dh > 0) { 101 encoder_variance(&a[(height - dh) * a_stride], a_stride, 102 &b[(height - dh) * b_stride], b_stride, width - dw, dh, 103 &sse, &sum); 104 total_sse += sse; 105 } 106 107 for (y = 0; y < height / 16; ++y) { 108 const uint8_t *pa = a; 109 const uint8_t *pb = b; 110 for (x = 0; x < width / 16; ++x) { 111 aom_mse16x16(pa, a_stride, pb, b_stride, &sse); 112 total_sse += sse; 113 114 pa += 16; 115 pb += 16; 116 } 117 118 a += 16 * a_stride; 119 b += 16 * b_stride; 120 } 121 122 return total_sse; 123 } 124 125 static int64_t highbd_get_sse_shift(const uint8_t *a8, int a_stride, 126 const uint8_t *b8, int b_stride, int width, 127 int height, unsigned int input_shift) { 128 const uint16_t *a = CONVERT_TO_SHORTPTR(a8); 129 const uint16_t *b = CONVERT_TO_SHORTPTR(b8); 130 int64_t total_sse = 0; 131 int x, y; 132 for (y = 0; y < height; ++y) { 133 for (x = 0; x < width; ++x) { 134 int64_t diff; 135 diff = (a[x] >> input_shift) - (b[x] >> input_shift); 136 total_sse += diff * diff; 137 } 138 a += a_stride; 139 b += b_stride; 140 } 141 return total_sse; 142 } 143 144 static int64_t highbd_get_sse(const uint8_t *a, int a_stride, const uint8_t *b, 145 int b_stride, int width, int height) { 146 int64_t total_sse = 0; 147 int x, y; 148 const int dw = width % 16; 149 const int dh = height % 16; 150 unsigned int sse = 0; 151 int sum = 0; 152 if (dw > 0) { 153 encoder_highbd_8_variance(&a[width - dw], a_stride, &b[width - dw], 154 b_stride, dw, height, &sse, &sum); 155 total_sse += sse; 156 } 157 if (dh > 0) { 158 encoder_highbd_8_variance(&a[(height - dh) * a_stride], a_stride, 159 &b[(height - dh) * b_stride], b_stride, 160 width - dw, dh, &sse, &sum); 161 total_sse += sse; 162 } 163 for (y = 0; y < height / 16; ++y) { 164 const uint8_t *pa = a; 165 const uint8_t *pb = b; 166 for (x = 0; x < width / 16; ++x) { 167 aom_highbd_8_mse16x16(pa, a_stride, pb, b_stride, &sse); 168 total_sse += sse; 169 pa += 16; 170 pb += 16; 171 } 172 a += 16 * a_stride; 173 b += 16 * b_stride; 174 } 175 return total_sse; 176 } 177 178 int64_t aom_get_y_sse_part(const YV12_BUFFER_CONFIG *a, 179 const YV12_BUFFER_CONFIG *b, int hstart, int width, 180 int vstart, int height) { 181 return get_sse(a->y_buffer + vstart * a->y_stride + hstart, a->y_stride, 182 b->y_buffer + vstart * b->y_stride + hstart, b->y_stride, 183 width, height); 184 } 185 186 int64_t aom_get_y_sse(const YV12_BUFFER_CONFIG *a, 187 const YV12_BUFFER_CONFIG *b) { 188 assert(a->y_crop_width == b->y_crop_width); 189 assert(a->y_crop_height == b->y_crop_height); 190 191 return get_sse(a->y_buffer, a->y_stride, b->y_buffer, b->y_stride, 192 a->y_crop_width, a->y_crop_height); 193 } 194 195 int64_t aom_get_u_sse_part(const YV12_BUFFER_CONFIG *a, 196 const YV12_BUFFER_CONFIG *b, int hstart, int width, 197 int vstart, int height) { 198 return get_sse(a->u_buffer + vstart * a->uv_stride + hstart, a->uv_stride, 199 b->u_buffer + vstart * b->uv_stride + hstart, b->uv_stride, 200 width, height); 201 } 202 203 int64_t aom_get_u_sse(const YV12_BUFFER_CONFIG *a, 204 const YV12_BUFFER_CONFIG *b) { 205 assert(a->uv_crop_width == b->uv_crop_width); 206 assert(a->uv_crop_height == b->uv_crop_height); 207 208 return get_sse(a->u_buffer, a->uv_stride, b->u_buffer, b->uv_stride, 209 a->uv_crop_width, a->uv_crop_height); 210 } 211 212 int64_t aom_get_v_sse_part(const YV12_BUFFER_CONFIG *a, 213 const YV12_BUFFER_CONFIG *b, int hstart, int width, 214 int vstart, int height) { 215 return get_sse(a->v_buffer + vstart * a->uv_stride + hstart, a->uv_stride, 216 b->v_buffer + vstart * b->uv_stride + hstart, b->uv_stride, 217 width, height); 218 } 219 220 int64_t aom_get_v_sse(const YV12_BUFFER_CONFIG *a, 221 const YV12_BUFFER_CONFIG *b) { 222 assert(a->uv_crop_width == b->uv_crop_width); 223 assert(a->uv_crop_height == b->uv_crop_height); 224 225 return get_sse(a->v_buffer, a->uv_stride, b->v_buffer, b->uv_stride, 226 a->uv_crop_width, a->uv_crop_height); 227 } 228 229 int64_t aom_highbd_get_y_sse_part(const YV12_BUFFER_CONFIG *a, 230 const YV12_BUFFER_CONFIG *b, int hstart, 231 int width, int vstart, int height) { 232 return highbd_get_sse( 233 a->y_buffer + vstart * a->y_stride + hstart, a->y_stride, 234 b->y_buffer + vstart * b->y_stride + hstart, b->y_stride, width, height); 235 } 236 237 int64_t aom_highbd_get_y_sse(const YV12_BUFFER_CONFIG *a, 238 const YV12_BUFFER_CONFIG *b) { 239 assert(a->y_crop_width == b->y_crop_width); 240 assert(a->y_crop_height == b->y_crop_height); 241 assert((a->flags & YV12_FLAG_HIGHBITDEPTH) != 0); 242 assert((b->flags & YV12_FLAG_HIGHBITDEPTH) != 0); 243 244 return highbd_get_sse(a->y_buffer, a->y_stride, b->y_buffer, b->y_stride, 245 a->y_crop_width, a->y_crop_height); 246 } 247 248 int64_t aom_highbd_get_u_sse_part(const YV12_BUFFER_CONFIG *a, 249 const YV12_BUFFER_CONFIG *b, int hstart, 250 int width, int vstart, int height) { 251 return highbd_get_sse(a->u_buffer + vstart * a->uv_stride + hstart, 252 a->uv_stride, 253 b->u_buffer + vstart * b->uv_stride + hstart, 254 b->uv_stride, width, height); 255 } 256 257 int64_t aom_highbd_get_u_sse(const YV12_BUFFER_CONFIG *a, 258 const YV12_BUFFER_CONFIG *b) { 259 assert(a->uv_crop_width == b->uv_crop_width); 260 assert(a->uv_crop_height == b->uv_crop_height); 261 assert((a->flags & YV12_FLAG_HIGHBITDEPTH) != 0); 262 assert((b->flags & YV12_FLAG_HIGHBITDEPTH) != 0); 263 264 return highbd_get_sse(a->u_buffer, a->uv_stride, b->u_buffer, b->uv_stride, 265 a->uv_crop_width, a->uv_crop_height); 266 } 267 268 int64_t aom_highbd_get_v_sse_part(const YV12_BUFFER_CONFIG *a, 269 const YV12_BUFFER_CONFIG *b, int hstart, 270 int width, int vstart, int height) { 271 return highbd_get_sse(a->v_buffer + vstart * a->uv_stride + hstart, 272 a->uv_stride, 273 b->v_buffer + vstart * b->uv_stride + hstart, 274 b->uv_stride, width, height); 275 } 276 277 int64_t aom_highbd_get_v_sse(const YV12_BUFFER_CONFIG *a, 278 const YV12_BUFFER_CONFIG *b) { 279 assert(a->uv_crop_width == b->uv_crop_width); 280 assert(a->uv_crop_height == b->uv_crop_height); 281 assert((a->flags & YV12_FLAG_HIGHBITDEPTH) != 0); 282 assert((b->flags & YV12_FLAG_HIGHBITDEPTH) != 0); 283 284 return highbd_get_sse(a->v_buffer, a->uv_stride, b->v_buffer, b->uv_stride, 285 a->uv_crop_width, a->uv_crop_height); 286 } 287 288 int64_t aom_get_sse_plane(const YV12_BUFFER_CONFIG *a, 289 const YV12_BUFFER_CONFIG *b, int plane, int highbd) { 290 if (highbd) { 291 switch (plane) { 292 case 0: return aom_highbd_get_y_sse(a, b); 293 case 1: return aom_highbd_get_u_sse(a, b); 294 case 2: return aom_highbd_get_v_sse(a, b); 295 default: assert(plane >= 0 && plane <= 2); return 0; 296 } 297 } 298 switch (plane) { 299 case 0: return aom_get_y_sse(a, b); 300 case 1: return aom_get_u_sse(a, b); 301 case 2: return aom_get_v_sse(a, b); 302 default: assert(plane >= 0 && plane <= 2); return 0; 303 } 304 } 305 306 void aom_calc_highbd_psnr(const YV12_BUFFER_CONFIG *a, 307 const YV12_BUFFER_CONFIG *b, PSNR_STATS *psnr, 308 uint32_t bit_depth, uint32_t in_bit_depth) { 309 const int widths[3] = { a->y_crop_width, a->uv_crop_width, a->uv_crop_width }; 310 const int heights[3] = { a->y_crop_height, a->uv_crop_height, 311 a->uv_crop_height }; 312 const int a_strides[3] = { a->y_stride, a->uv_stride, a->uv_stride }; 313 const int b_strides[3] = { b->y_stride, b->uv_stride, b->uv_stride }; 314 int i; 315 uint64_t total_sse = 0; 316 uint32_t total_samples = 0; 317 const double peak = (double)((1 << in_bit_depth) - 1); 318 const unsigned int input_shift = bit_depth - in_bit_depth; 319 320 for (i = 0; i < 3; ++i) { 321 const int w = widths[i]; 322 const int h = heights[i]; 323 const uint32_t samples = w * h; 324 uint64_t sse; 325 if (a->flags & YV12_FLAG_HIGHBITDEPTH) { 326 if (input_shift) { 327 sse = highbd_get_sse_shift(a->buffers[i], a_strides[i], b->buffers[i], 328 b_strides[i], w, h, input_shift); 329 } else { 330 sse = highbd_get_sse(a->buffers[i], a_strides[i], b->buffers[i], 331 b_strides[i], w, h); 332 } 333 } else { 334 sse = get_sse(a->buffers[i], a_strides[i], b->buffers[i], b_strides[i], w, 335 h); 336 } 337 psnr->sse[1 + i] = sse; 338 psnr->samples[1 + i] = samples; 339 psnr->psnr[1 + i] = aom_sse_to_psnr(samples, peak, (double)sse); 340 341 total_sse += sse; 342 total_samples += samples; 343 } 344 345 psnr->sse[0] = total_sse; 346 psnr->samples[0] = total_samples; 347 psnr->psnr[0] = 348 aom_sse_to_psnr((double)total_samples, peak, (double)total_sse); 349 } 350 351 void aom_calc_psnr(const YV12_BUFFER_CONFIG *a, const YV12_BUFFER_CONFIG *b, 352 PSNR_STATS *psnr) { 353 static const double peak = 255.0; 354 const int widths[3] = { a->y_crop_width, a->uv_crop_width, a->uv_crop_width }; 355 const int heights[3] = { a->y_crop_height, a->uv_crop_height, 356 a->uv_crop_height }; 357 const int a_strides[3] = { a->y_stride, a->uv_stride, a->uv_stride }; 358 const int b_strides[3] = { b->y_stride, b->uv_stride, b->uv_stride }; 359 int i; 360 uint64_t total_sse = 0; 361 uint32_t total_samples = 0; 362 363 for (i = 0; i < 3; ++i) { 364 const int w = widths[i]; 365 const int h = heights[i]; 366 const uint32_t samples = w * h; 367 const uint64_t sse = 368 get_sse(a->buffers[i], a_strides[i], b->buffers[i], b_strides[i], w, h); 369 psnr->sse[1 + i] = sse; 370 psnr->samples[1 + i] = samples; 371 psnr->psnr[1 + i] = aom_sse_to_psnr(samples, peak, (double)sse); 372 373 total_sse += sse; 374 total_samples += samples; 375 } 376 377 psnr->sse[0] = total_sse; 378 psnr->samples[0] = total_samples; 379 psnr->psnr[0] = 380 aom_sse_to_psnr((double)total_samples, peak, (double)total_sse); 381 } 382
