| 1 | /***************************************************************************** |
| 2 | * Copyright (C) 2013 x265 project |
| 3 | * |
| 4 | * Authors: Min Chen <chenm003@163.com> |
| 5 | * |
| 6 | * This program is free software; you can redistribute it and/or modify |
| 7 | * it under the terms of the GNU General Public License as published by |
| 8 | * the Free Software Foundation; either version 2 of the License, or |
| 9 | * (at your option) any later version. |
| 10 | * |
| 11 | * This program is distributed in the hope that it will be useful, |
| 12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 14 | * GNU General Public License for more details. |
| 15 | * |
| 16 | * You should have received a copy of the GNU General Public License |
| 17 | * along with this program; if not, write to the Free Software |
| 18 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA. |
| 19 | * |
| 20 | * This program is also available under a commercial proprietary license. |
| 21 | * For more information, contact us at license @ x265.com. |
| 22 | *****************************************************************************/ |
| 23 | |
| 24 | #include "common.h" |
| 25 | #include "predict.h" |
| 26 | #include "intrapredharness.h" |
| 27 | |
| 28 | using namespace x265; |
| 29 | |
| 30 | IntraPredHarness::IntraPredHarness() |
| 31 | { |
| 32 | for (int i = 0; i < INPUT_SIZE; i++) |
| 33 | pixel_buff[i] = rand() % PIXEL_MAX; |
| 34 | } |
| 35 | |
| 36 | bool IntraPredHarness::check_dc_primitive(intra_pred_t ref, intra_pred_t opt, int width) |
| 37 | { |
| 38 | int j = Predict::ADI_BUF_STRIDE; |
| 39 | intptr_t stride = FENC_STRIDE; |
| 40 | |
| 41 | #if _DEBUG |
| 42 | memset(pixel_out_vec, 0xCD, OUTPUT_SIZE); |
| 43 | memset(pixel_out_c, 0xCD, OUTPUT_SIZE); |
| 44 | #endif |
| 45 | |
| 46 | for (int i = 0; i <= 100; i++) |
| 47 | { |
| 48 | int rand_filter = rand() & 1; |
| 49 | if (width > 16) |
| 50 | rand_filter = 0; |
| 51 | |
| 52 | pixel left[MAX_CU_SIZE * 2 + 1]; |
| 53 | for (int k = 0; k < width * 2 + 1; k++) |
| 54 | { |
| 55 | left[k] = pixel_buff[j - 1 + k * Predict::ADI_BUF_STRIDE]; |
| 56 | } |
| 57 | |
| 58 | ref(pixel_out_c, stride, pixel_buff + j - Predict::ADI_BUF_STRIDE, left + 1, 0, rand_filter); |
| 59 | checked(opt, pixel_out_vec, stride, pixel_buff + j - Predict::ADI_BUF_STRIDE, left + 1, 0, rand_filter); |
| 60 | |
| 61 | for (int k = 0; k < width; k++) |
| 62 | { |
| 63 | if (memcmp(pixel_out_vec + k * FENC_STRIDE, pixel_out_c + k * FENC_STRIDE, width * sizeof(pixel))) |
| 64 | return false; |
| 65 | } |
| 66 | |
| 67 | reportfail(); |
| 68 | j += FENC_STRIDE; |
| 69 | } |
| 70 | |
| 71 | return true; |
| 72 | } |
| 73 | |
| 74 | bool IntraPredHarness::check_planar_primitive(intra_pred_t ref, intra_pred_t opt, int width) |
| 75 | { |
| 76 | int j = Predict::ADI_BUF_STRIDE; |
| 77 | intptr_t stride = FENC_STRIDE; |
| 78 | |
| 79 | #if _DEBUG |
| 80 | memset(pixel_out_vec, 0xCD, OUTPUT_SIZE); |
| 81 | memset(pixel_out_c, 0xCD, OUTPUT_SIZE); |
| 82 | #endif |
| 83 | |
| 84 | for (int i = 0; i <= 100; i++) |
| 85 | { |
| 86 | pixel left[MAX_CU_SIZE * 2 + 1]; |
| 87 | for (int k = 0; k < width * 2 + 1; k++) |
| 88 | { |
| 89 | left[k] = pixel_buff[j - 1 + k * Predict::ADI_BUF_STRIDE]; |
| 90 | } |
| 91 | |
| 92 | ref(pixel_out_c, stride, pixel_buff + j - Predict::ADI_BUF_STRIDE, left + 1, 0, 0); |
| 93 | checked(opt, pixel_out_vec, stride, pixel_buff + j - Predict::ADI_BUF_STRIDE, left + 1, 0, 0); |
| 94 | |
| 95 | for (int k = 0; k < width; k++) |
| 96 | { |
| 97 | if (memcmp(pixel_out_vec + k * FENC_STRIDE, pixel_out_c + k * FENC_STRIDE, width * sizeof(pixel))) |
| 98 | return false; |
| 99 | } |
| 100 | |
| 101 | reportfail(); |
| 102 | j += FENC_STRIDE; |
| 103 | } |
| 104 | |
| 105 | return true; |
| 106 | } |
| 107 | |
| 108 | bool IntraPredHarness::check_angular_primitive(const intra_pred_t ref[][NUM_TR_SIZE], const intra_pred_t opt[][NUM_TR_SIZE]) |
| 109 | { |
| 110 | int j = Predict::ADI_BUF_STRIDE; |
| 111 | intptr_t stride = FENC_STRIDE; |
| 112 | |
| 113 | #if _DEBUG |
| 114 | memset(pixel_out_vec, 0xCD, OUTPUT_SIZE); |
| 115 | memset(pixel_out_c, 0xCD, OUTPUT_SIZE); |
| 116 | #endif |
| 117 | |
| 118 | for (int size = 2; size <= 5; size++) |
| 119 | { |
| 120 | int width = (1 << size); |
| 121 | for (int i = 0; i <= 100; i++) |
| 122 | { |
| 123 | int bFilter = (width <= 16) && (rand() % 2); |
| 124 | for (int pmode = 2; pmode <= 34; pmode++) |
| 125 | { |
| 126 | if (!opt[pmode][size - 2]) |
| 127 | continue; |
| 128 | |
| 129 | pixel * refAbove = pixel_buff + j; |
| 130 | pixel * refLeft = refAbove + 3 * width; |
| 131 | refLeft[0] = refAbove[0]; |
| 132 | |
| 133 | checked(opt[pmode][size - 2], pixel_out_vec, stride, refLeft, refAbove, pmode, bFilter); |
| 134 | ref[pmode][size - 2](pixel_out_c, stride, refLeft, refAbove, pmode, bFilter); |
| 135 | |
| 136 | for (int k = 0; k < width; k++) |
| 137 | { |
| 138 | if (memcmp(pixel_out_vec + k * FENC_STRIDE, pixel_out_c + k * FENC_STRIDE, width * sizeof(pixel))) |
| 139 | return false; |
| 140 | } |
| 141 | |
| 142 | reportfail(); |
| 143 | } |
| 144 | |
| 145 | j += FENC_STRIDE; |
| 146 | } |
| 147 | } |
| 148 | |
| 149 | return true; |
| 150 | } |
| 151 | |
| 152 | bool IntraPredHarness::check_allangs_primitive(const intra_allangs_t ref[], const intra_allangs_t opt[]) |
| 153 | { |
| 154 | int j = Predict::ADI_BUF_STRIDE; |
| 155 | int isLuma; |
| 156 | |
| 157 | #if _DEBUG |
| 158 | memset(pixel_out_33_vec, 0xCD, OUTPUT_SIZE_33); |
| 159 | memset(pixel_out_33_c, 0xCD, OUTPUT_SIZE_33); |
| 160 | #endif |
| 161 | |
| 162 | for (int size = 2; size <= 5; size++) |
| 163 | { |
| 164 | if (opt[size - 2] == NULL) continue; |
| 165 | |
| 166 | const int width = (1 << size); |
| 167 | |
| 168 | for (int i = 0; i <= 100; i++) |
| 169 | { |
| 170 | isLuma = (width <= 16) ? true : false; // bFilter is true for 4x4, 8x8, 16x16 and false for 32x32 |
| 171 | |
| 172 | pixel * refAbove0 = pixel_buff + j; |
| 173 | pixel * refLeft0 = refAbove0 + 3 * width; |
| 174 | |
| 175 | pixel * refAbove1 = pixel_buff + j + 3 * FENC_STRIDE; // keep this offset, since vector code may broken input buffer range [-(width-1), 0] |
| 176 | pixel * refLeft1 = refAbove1 + 3 * width + FENC_STRIDE; |
| 177 | refLeft0[0] = refAbove0[0] = refLeft1[0] = refAbove1[0]; |
| 178 | |
| 179 | ref[size - 2](pixel_out_33_c, refAbove0, refLeft0, refAbove1, refLeft1, isLuma); |
| 180 | checked(opt[size - 2], pixel_out_33_vec, refAbove0, refLeft0, refAbove1, refLeft1, isLuma); |
| 181 | |
| 182 | for (int p = 2 - 2; p <= 34 - 2; p++) |
| 183 | { |
| 184 | for (int k = 0; k < width; k++) |
| 185 | { |
| 186 | if (memcmp(pixel_out_33_c + p * (width * width) + k * width, pixel_out_33_vec + p * (width * width) + k * width, width * sizeof(pixel))) |
| 187 | { |
| 188 | printf("\nFailed: (%dx%d) Mode(%2d), Line[%2d], bfilter=%d\n", width, width, p + 2, k, isLuma); |
| 189 | opt[size - 2](pixel_out_33_vec, refAbove0, refLeft0, refAbove1, refLeft1, isLuma); |
| 190 | return false; |
| 191 | } |
| 192 | } |
| 193 | } |
| 194 | |
| 195 | reportfail(); |
| 196 | j += FENC_STRIDE; |
| 197 | } |
| 198 | } |
| 199 | |
| 200 | return true; |
| 201 | } |
| 202 | |
| 203 | bool IntraPredHarness::testCorrectness(const EncoderPrimitives& ref, const EncoderPrimitives& opt) |
| 204 | { |
| 205 | for (int i = BLOCK_4x4; i <= BLOCK_32x32; i++) |
| 206 | { |
| 207 | if (opt.intra_pred[1][i]) |
| 208 | { |
| 209 | const int size = (1 << (i + 2)); |
| 210 | if (!check_dc_primitive(ref.intra_pred[1][i], opt.intra_pred[1][i], size)) |
| 211 | { |
| 212 | printf("intra_dc %dx%d failed\n", size, size); |
| 213 | return false; |
| 214 | } |
| 215 | } |
| 216 | if (opt.intra_pred[0][i]) |
| 217 | { |
| 218 | const int size = (1 << (i + 2)); |
| 219 | if (!check_planar_primitive(ref.intra_pred[0][i], opt.intra_pred[0][i], size)) |
| 220 | { |
| 221 | printf("intra_planar %dx%d failed\n", size, size); |
| 222 | return false; |
| 223 | } |
| 224 | } |
| 225 | } |
| 226 | |
| 227 | // NOTE: always call since this function have check pointer in loop |
| 228 | if (!check_angular_primitive(ref.intra_pred, opt.intra_pred)) |
| 229 | { |
| 230 | printf("intra_angular failed\n"); |
| 231 | return false; |
| 232 | } |
| 233 | |
| 234 | if (opt.intra_pred_allangs[0]) |
| 235 | { |
| 236 | if (!check_allangs_primitive(ref.intra_pred_allangs, opt.intra_pred_allangs)) |
| 237 | { |
| 238 | printf("intra_allangs failed\n"); |
| 239 | return false; |
| 240 | } |
| 241 | } |
| 242 | |
| 243 | return true; |
| 244 | } |
| 245 | |
| 246 | void IntraPredHarness::measureSpeed(const EncoderPrimitives& ref, const EncoderPrimitives& opt) |
| 247 | { |
| 248 | int width = 64; |
| 249 | uint16_t srcStride = 96; |
| 250 | |
| 251 | for (int i = BLOCK_4x4; i <= BLOCK_32x32; i++) |
| 252 | { |
| 253 | const int size = (1 << (i + 2)); |
| 254 | if (opt.intra_pred[1][i]) |
| 255 | { |
| 256 | printf("intra_dc_%dx%d[f=0]", size, size); |
| 257 | REPORT_SPEEDUP(opt.intra_pred[1][i], ref.intra_pred[1][i], |
| 258 | pixel_out_vec, FENC_STRIDE, pixel_buff + srcStride, pixel_buff, 0, 0); |
| 259 | if (size <= 16) |
| 260 | { |
| 261 | printf("intra_dc_%dx%d[f=1]", size, size); |
| 262 | REPORT_SPEEDUP(opt.intra_pred[1][i], ref.intra_pred[1][i], |
| 263 | pixel_out_vec, FENC_STRIDE, pixel_buff + srcStride, pixel_buff, 0, 1); |
| 264 | } |
| 265 | } |
| 266 | if (opt.intra_pred[0][i]) |
| 267 | { |
| 268 | printf("intra_planar %2dx%d", size, size); |
| 269 | REPORT_SPEEDUP(opt.intra_pred[0][i], ref.intra_pred[0][i], |
| 270 | pixel_out_vec, FENC_STRIDE, pixel_buff + srcStride, pixel_buff, 0, 0); |
| 271 | } |
| 272 | if (opt.intra_pred_allangs[i]) |
| 273 | { |
| 274 | bool bFilter = (size <= 16); |
| 275 | pixel * refAbove = pixel_buff + srcStride; |
| 276 | pixel * refLeft = refAbove + 3 * size; |
| 277 | refLeft[0] = refAbove[0]; |
| 278 | printf("intra_allangs%dx%d", size, size); |
| 279 | REPORT_SPEEDUP(opt.intra_pred_allangs[i], ref.intra_pred_allangs[i], |
| 280 | pixel_out_33_vec, refAbove, refLeft, refAbove, refLeft, bFilter); |
| 281 | } |
| 282 | } |
| 283 | |
| 284 | for (int ii = 2; ii <= 5; ii++) |
| 285 | { |
| 286 | for (int p = 2; p <= 34; p += 1) |
| 287 | { |
| 288 | int pmode = p; //(rand()%33)+2; |
| 289 | if (opt.intra_pred[pmode][ii - 2]) |
| 290 | { |
| 291 | width = (1 << ii); |
| 292 | bool bFilter = (width <= 16); |
| 293 | pixel * refAbove = pixel_buff + srcStride; |
| 294 | pixel * refLeft = refAbove + 3 * width; |
| 295 | refLeft[0] = refAbove[0]; |
| 296 | printf("intra_ang%dx%d[%2d]", width, width, pmode); |
| 297 | REPORT_SPEEDUP(opt.intra_pred[pmode][ii - 2], ref.intra_pred[pmode][ii - 2], |
| 298 | pixel_out_vec, FENC_STRIDE, refAbove, refLeft, pmode, bFilter); |
| 299 | } |
| 300 | } |
| 301 | } |
| 302 | } |