1/* 2 * Copyright 2011 The LibYuv Project Authors. All rights reserved. 3 * 4 * Use of this source code is governed by a BSD-style license 5 * that can be found in the LICENSE file in the root of the source 6 * tree. An additional intellectual property rights grant can be found 7 * in the file PATENTS. All contributing project authors may 8 * be found in the AUTHORS file in the root of the source tree. 9 */ 10 11#include "libyuv/compare.h" 12 13#include <float.h> 14#include <math.h> 15#ifdef _OPENMP 16#include <omp.h> 17#endif 18 19#include "libyuv/basic_types.h" 20#include "libyuv/compare_row.h" 21#include "libyuv/cpu_id.h" 22#include "libyuv/row.h" 23#include "libyuv/video_common.h" 24 25#ifdef __cplusplus 26namespace libyuv { 27extern "C" { 28#endif 29 30// hash seed of 5381 recommended. 31LIBYUV_API 32uint32 HashDjb2(const uint8* src, uint64 count, uint32 seed) { 33 const int kBlockSize = 1 << 15; // 32768; 34 int remainder; 35 uint32 (*HashDjb2_SSE)(const uint8* src, int count, uint32 seed) = 36 HashDjb2_C; 37#if defined(HAS_HASHDJB2_SSE41) 38 if (TestCpuFlag(kCpuHasSSE41)) { 39 HashDjb2_SSE = HashDjb2_SSE41; 40 } 41#endif 42#if defined(HAS_HASHDJB2_AVX2) 43 if (TestCpuFlag(kCpuHasAVX2)) { 44 HashDjb2_SSE = HashDjb2_AVX2; 45 } 46#endif 47 48 while (count >= (uint64)(kBlockSize)) { 49 seed = HashDjb2_SSE(src, kBlockSize, seed); 50 src += kBlockSize; 51 count -= kBlockSize; 52 } 53 remainder = (int)(count) & ~15; 54 if (remainder) { 55 seed = HashDjb2_SSE(src, remainder, seed); 56 src += remainder; 57 count -= remainder; 58 } 59 remainder = (int)(count) & 15; 60 if (remainder) { 61 seed = HashDjb2_C(src, remainder, seed); 62 } 63 return seed; 64} 65 66static uint32 ARGBDetectRow_C(const uint8* argb, int width) { 67 int x; 68 for (x = 0; x < width - 1; x += 2) { 69 if (argb[0] != 255) { // First byte is not Alpha of 255, so not ARGB. 70 return FOURCC_BGRA; 71 } 72 if (argb[3] != 255) { // 4th byte is not Alpha of 255, so not BGRA. 73 return FOURCC_ARGB; 74 } 75 if (argb[4] != 255) { // Second pixel first byte is not Alpha of 255. 76 return FOURCC_BGRA; 77 } 78 if (argb[7] != 255) { // Second pixel 4th byte is not Alpha of 255. 79 return FOURCC_ARGB; 80 } 81 argb += 8; 82 } 83 if (width & 1) { 84 if (argb[0] != 255) { // First byte is not Alpha of 255, so not ARGB. 85 return FOURCC_BGRA; 86 } 87 if (argb[3] != 255) { // 4th byte is not Alpha of 255, so not BGRA. 88 return FOURCC_ARGB; 89 } 90 } 91 return 0; 92} 93 94// Scan an opaque argb image and return fourcc based on alpha offset. 95// Returns FOURCC_ARGB, FOURCC_BGRA, or 0 if unknown. 96LIBYUV_API 97uint32 ARGBDetect(const uint8* argb, int stride_argb, int width, int height) { 98 uint32 fourcc = 0; 99 int h; 100 101 // Coalesce rows. 102 if (stride_argb == width * 4) { 103 width *= height; 104 height = 1; 105 stride_argb = 0; 106 } 107 for (h = 0; h < height && fourcc == 0; ++h) { 108 fourcc = ARGBDetectRow_C(argb, width); 109 argb += stride_argb; 110 } 111 return fourcc; 112} 113 114// TODO(fbarchard): Refactor into row function. 115LIBYUV_API 116uint64 ComputeSumSquareError(const uint8* src_a, const uint8* src_b, 117 int count) { 118 // SumSquareError returns values 0 to 65535 for each squared difference. 119 // Up to 65536 of those can be summed and remain within a uint32. 120 // After each block of 65536 pixels, accumulate into a uint64. 121 const int kBlockSize = 65536; 122 int remainder = count & (kBlockSize - 1) & ~31; 123 uint64 sse = 0; 124 int i; 125 uint32 (*SumSquareError)(const uint8* src_a, const uint8* src_b, int count) = 126 SumSquareError_C; 127#if defined(HAS_SUMSQUAREERROR_NEON) 128 if (TestCpuFlag(kCpuHasNEON)) { 129 SumSquareError = SumSquareError_NEON; 130 } 131#endif 132#if defined(HAS_SUMSQUAREERROR_SSE2) 133 if (TestCpuFlag(kCpuHasSSE2)) { 134 // Note only used for multiples of 16 so count is not checked. 135 SumSquareError = SumSquareError_SSE2; 136 } 137#endif 138#if defined(HAS_SUMSQUAREERROR_AVX2) 139 if (TestCpuFlag(kCpuHasAVX2)) { 140 // Note only used for multiples of 32 so count is not checked. 141 SumSquareError = SumSquareError_AVX2; 142 } 143#endif 144#ifdef _OPENMP 145#pragma omp parallel for reduction(+: sse) 146#endif 147 for (i = 0; i < (count - (kBlockSize - 1)); i += kBlockSize) { 148 sse += SumSquareError(src_a + i, src_b + i, kBlockSize); 149 } 150 src_a += count & ~(kBlockSize - 1); 151 src_b += count & ~(kBlockSize - 1); 152 if (remainder) { 153 sse += SumSquareError(src_a, src_b, remainder); 154 src_a += remainder; 155 src_b += remainder; 156 } 157 remainder = count & 31; 158 if (remainder) { 159 sse += SumSquareError_C(src_a, src_b, remainder); 160 } 161 return sse; 162} 163 164LIBYUV_API 165uint64 ComputeSumSquareErrorPlane(const uint8* src_a, int stride_a, 166 const uint8* src_b, int stride_b, 167 int width, int height) { 168 uint64 sse = 0; 169 int h; 170 // Coalesce rows. 171 if (stride_a == width && 172 stride_b == width) { 173 width *= height; 174 height = 1; 175 stride_a = stride_b = 0; 176 } 177 for (h = 0; h < height; ++h) { 178 sse += ComputeSumSquareError(src_a, src_b, width); 179 src_a += stride_a; 180 src_b += stride_b; 181 } 182 return sse; 183} 184 185LIBYUV_API 186double SumSquareErrorToPsnr(uint64 sse, uint64 count) { 187 double psnr; 188 if (sse > 0) { 189 double mse = (double)(count) / (double)(sse); 190 psnr = 10.0 * log10(255.0 * 255.0 * mse); 191 } else { 192 psnr = kMaxPsnr; // Limit to prevent divide by 0 193 } 194 195 if (psnr > kMaxPsnr) 196 psnr = kMaxPsnr; 197 198 return psnr; 199} 200 201LIBYUV_API 202double CalcFramePsnr(const uint8* src_a, int stride_a, 203 const uint8* src_b, int stride_b, 204 int width, int height) { 205 const uint64 samples = width * height; 206 const uint64 sse = ComputeSumSquareErrorPlane(src_a, stride_a, 207 src_b, stride_b, 208 width, height); 209 return SumSquareErrorToPsnr(sse, samples); 210} 211 212LIBYUV_API 213double I420Psnr(const uint8* src_y_a, int stride_y_a, 214 const uint8* src_u_a, int stride_u_a, 215 const uint8* src_v_a, int stride_v_a, 216 const uint8* src_y_b, int stride_y_b, 217 const uint8* src_u_b, int stride_u_b, 218 const uint8* src_v_b, int stride_v_b, 219 int width, int height) { 220 const uint64 sse_y = ComputeSumSquareErrorPlane(src_y_a, stride_y_a, 221 src_y_b, stride_y_b, 222 width, height); 223 const int width_uv = (width + 1) >> 1; 224 const int height_uv = (height + 1) >> 1; 225 const uint64 sse_u = ComputeSumSquareErrorPlane(src_u_a, stride_u_a, 226 src_u_b, stride_u_b, 227 width_uv, height_uv); 228 const uint64 sse_v = ComputeSumSquareErrorPlane(src_v_a, stride_v_a, 229 src_v_b, stride_v_b, 230 width_uv, height_uv); 231 const uint64 samples = width * height + 2 * (width_uv * height_uv); 232 const uint64 sse = sse_y + sse_u + sse_v; 233 return SumSquareErrorToPsnr(sse, samples); 234} 235 236static const int64 cc1 = 26634; // (64^2*(.01*255)^2 237static const int64 cc2 = 239708; // (64^2*(.03*255)^2 238 239static double Ssim8x8_C(const uint8* src_a, int stride_a, 240 const uint8* src_b, int stride_b) { 241 int64 sum_a = 0; 242 int64 sum_b = 0; 243 int64 sum_sq_a = 0; 244 int64 sum_sq_b = 0; 245 int64 sum_axb = 0; 246 247 int i; 248 for (i = 0; i < 8; ++i) { 249 int j; 250 for (j = 0; j < 8; ++j) { 251 sum_a += src_a[j]; 252 sum_b += src_b[j]; 253 sum_sq_a += src_a[j] * src_a[j]; 254 sum_sq_b += src_b[j] * src_b[j]; 255 sum_axb += src_a[j] * src_b[j]; 256 } 257 258 src_a += stride_a; 259 src_b += stride_b; 260 } 261 262 { 263 const int64 count = 64; 264 // scale the constants by number of pixels 265 const int64 c1 = (cc1 * count * count) >> 12; 266 const int64 c2 = (cc2 * count * count) >> 12; 267 268 const int64 sum_a_x_sum_b = sum_a * sum_b; 269 270 const int64 ssim_n = (2 * sum_a_x_sum_b + c1) * 271 (2 * count * sum_axb - 2 * sum_a_x_sum_b + c2); 272 273 const int64 sum_a_sq = sum_a*sum_a; 274 const int64 sum_b_sq = sum_b*sum_b; 275 276 const int64 ssim_d = (sum_a_sq + sum_b_sq + c1) * 277 (count * sum_sq_a - sum_a_sq + 278 count * sum_sq_b - sum_b_sq + c2); 279 280 if (ssim_d == 0.0) { 281 return DBL_MAX; 282 } 283 return ssim_n * 1.0 / ssim_d; 284 } 285} 286 287// We are using a 8x8 moving window with starting location of each 8x8 window 288// on the 4x4 pixel grid. Such arrangement allows the windows to overlap 289// block boundaries to penalize blocking artifacts. 290LIBYUV_API 291double CalcFrameSsim(const uint8* src_a, int stride_a, 292 const uint8* src_b, int stride_b, 293 int width, int height) { 294 int samples = 0; 295 double ssim_total = 0; 296 double (*Ssim8x8)(const uint8* src_a, int stride_a, 297 const uint8* src_b, int stride_b) = Ssim8x8_C; 298 299 // sample point start with each 4x4 location 300 int i; 301 for (i = 0; i < height - 8; i += 4) { 302 int j; 303 for (j = 0; j < width - 8; j += 4) { 304 ssim_total += Ssim8x8(src_a + j, stride_a, src_b + j, stride_b); 305 samples++; 306 } 307 308 src_a += stride_a * 4; 309 src_b += stride_b * 4; 310 } 311 312 ssim_total /= samples; 313 return ssim_total; 314} 315 316LIBYUV_API 317double I420Ssim(const uint8* src_y_a, int stride_y_a, 318 const uint8* src_u_a, int stride_u_a, 319 const uint8* src_v_a, int stride_v_a, 320 const uint8* src_y_b, int stride_y_b, 321 const uint8* src_u_b, int stride_u_b, 322 const uint8* src_v_b, int stride_v_b, 323 int width, int height) { 324 const double ssim_y = CalcFrameSsim(src_y_a, stride_y_a, 325 src_y_b, stride_y_b, width, height); 326 const int width_uv = (width + 1) >> 1; 327 const int height_uv = (height + 1) >> 1; 328 const double ssim_u = CalcFrameSsim(src_u_a, stride_u_a, 329 src_u_b, stride_u_b, 330 width_uv, height_uv); 331 const double ssim_v = CalcFrameSsim(src_v_a, stride_v_a, 332 src_v_b, stride_v_b, 333 width_uv, height_uv); 334 return ssim_y * 0.8 + 0.1 * (ssim_u + ssim_v); 335} 336 337#ifdef __cplusplus 338} // extern "C" 339} // namespace libyuv 340#endif 341