| 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 "primitives.h" |
| 26 | |
| 27 | using namespace x265; |
| 28 | |
| 29 | namespace { |
| 30 | pixel dcPredValue(pixel* above, pixel* left, intptr_t width) |
| 31 | { |
| 32 | int w, sum = 0; |
| 33 | pixel pDcVal; |
| 34 | |
| 35 | for (w = 0; w < width; w++) |
| 36 | { |
| 37 | sum += above[w]; |
| 38 | } |
| 39 | |
| 40 | for (w = 0; w < width; w++) |
| 41 | { |
| 42 | sum += left[w]; |
| 43 | } |
| 44 | |
| 45 | pDcVal = (pixel)((sum + width) / (width + width)); |
| 46 | |
| 47 | return pDcVal; |
| 48 | } |
| 49 | |
| 50 | void dcPredFilter(pixel* above, pixel* left, pixel* dst, intptr_t dststride, int size) |
| 51 | { |
| 52 | // boundary pixels processing |
| 53 | dst[0] = (pixel)((above[0] + left[0] + 2 * dst[0] + 2) >> 2); |
| 54 | |
| 55 | for (int x = 1; x < size; x++) |
| 56 | { |
| 57 | dst[x] = (pixel)((above[x] + 3 * dst[x] + 2) >> 2); |
| 58 | } |
| 59 | |
| 60 | dst += dststride; |
| 61 | for (int y = 1; y < size; y++) |
| 62 | { |
| 63 | *dst = (pixel)((left[y] + 3 * *dst + 2) >> 2); |
| 64 | dst += dststride; |
| 65 | } |
| 66 | } |
| 67 | |
| 68 | template<int width> |
| 69 | void intra_pred_dc_c(pixel* dst, intptr_t dstStride, pixel* left, pixel* above, int /*dirMode*/, int bFilter) |
| 70 | { |
| 71 | int k, l; |
| 72 | |
| 73 | pixel dcval = dcPredValue(above + 1, left + 1, width); |
| 74 | |
| 75 | for (k = 0; k < width; k++) |
| 76 | { |
| 77 | for (l = 0; l < width; l++) |
| 78 | { |
| 79 | dst[k * dstStride + l] = dcval; |
| 80 | } |
| 81 | } |
| 82 | |
| 83 | if (bFilter) |
| 84 | { |
| 85 | dcPredFilter(above + 1, left + 1, dst, dstStride, width); |
| 86 | } |
| 87 | } |
| 88 | |
| 89 | template<int log2Size> |
| 90 | void planar_pred_c(pixel* dst, intptr_t dstStride, pixel* left, pixel* above, int /*dirMode*/, int /*bFilter*/) |
| 91 | { |
| 92 | above += 1; |
| 93 | left += 1; |
| 94 | int k, l; |
| 95 | pixel bottomLeft, topRight; |
| 96 | int horPred; |
| 97 | int32_t leftColumn[MAX_CU_SIZE + 1], topRow[MAX_CU_SIZE + 1]; |
| 98 | // CHECK_ME: dynamic range is 9 bits or 15 bits(I assume max input bit_depth is 14 bits) |
| 99 | int16_t bottomRow[MAX_CU_SIZE], rightColumn[MAX_CU_SIZE]; |
| 100 | const int blkSize = 1 << log2Size; |
| 101 | const int offset2D = blkSize; |
| 102 | const int shift1D = log2Size; |
| 103 | const int shift2D = shift1D + 1; |
| 104 | |
| 105 | // Get left and above reference column and row |
| 106 | for (k = 0; k < blkSize + 1; k++) |
| 107 | { |
| 108 | topRow[k] = above[k]; |
| 109 | leftColumn[k] = left[k]; |
| 110 | } |
| 111 | |
| 112 | // Prepare intermediate variables used in interpolation |
| 113 | bottomLeft = (pixel)leftColumn[blkSize]; |
| 114 | topRight = (pixel)topRow[blkSize]; |
| 115 | for (k = 0; k < blkSize; k++) |
| 116 | { |
| 117 | bottomRow[k] = (int16_t)(bottomLeft - topRow[k]); |
| 118 | rightColumn[k] = (int16_t)(topRight - leftColumn[k]); |
| 119 | topRow[k] <<= shift1D; |
| 120 | leftColumn[k] <<= shift1D; |
| 121 | } |
| 122 | |
| 123 | // Generate prediction signal |
| 124 | for (k = 0; k < blkSize; k++) |
| 125 | { |
| 126 | horPred = leftColumn[k] + offset2D; |
| 127 | for (l = 0; l < blkSize; l++) |
| 128 | { |
| 129 | horPred += rightColumn[k]; |
| 130 | topRow[l] += bottomRow[l]; |
| 131 | dst[k * dstStride + l] = (pixel)((horPred + topRow[l]) >> shift2D); |
| 132 | } |
| 133 | } |
| 134 | } |
| 135 | |
| 136 | template<int width> |
| 137 | void intra_pred_ang_c(pixel* dst, intptr_t dstStride, pixel *refLeft, pixel *refAbove, int dirMode, int bFilter) |
| 138 | { |
| 139 | // Map the mode index to main prediction direction and angle |
| 140 | int k, l; |
| 141 | bool modeHor = (dirMode < 18); |
| 142 | bool modeVer = !modeHor; |
| 143 | int intraPredAngle = modeVer ? (int)dirMode - VER_IDX : modeHor ? -((int)dirMode - HOR_IDX) : 0; |
| 144 | int absAng = abs(intraPredAngle); |
| 145 | int signAng = intraPredAngle < 0 ? -1 : 1; |
| 146 | |
| 147 | // Set bitshifts and scale the angle parameter to block size |
| 148 | static const int angTable[9] = { 0, 2, 5, 9, 13, 17, 21, 26, 32 }; |
| 149 | static const int invAngTable[9] = { 0, 4096, 1638, 910, 630, 482, 390, 315, 256 }; // (256 * 32) / Angle |
| 150 | int invAngle = invAngTable[absAng]; |
| 151 | |
| 152 | absAng = angTable[absAng]; |
| 153 | intraPredAngle = signAng * absAng; |
| 154 | |
| 155 | // Do angular predictions |
| 156 | { |
| 157 | pixel* refMain; |
| 158 | pixel* refSide; |
| 159 | |
| 160 | // Initialise the Main and Left reference array. |
| 161 | if (intraPredAngle < 0) |
| 162 | { |
| 163 | refMain = (modeVer ? refAbove : refLeft); // + (width - 1); |
| 164 | refSide = (modeVer ? refLeft : refAbove); // + (width - 1); |
| 165 | |
| 166 | // Extend the Main reference to the left. |
| 167 | int invAngleSum = 128; // rounding for (shift by 8) |
| 168 | for (k = -1; k > width * intraPredAngle >> 5; k--) |
| 169 | { |
| 170 | invAngleSum += invAngle; |
| 171 | refMain[k] = refSide[invAngleSum >> 8]; |
| 172 | } |
| 173 | } |
| 174 | else |
| 175 | { |
| 176 | refMain = modeVer ? refAbove : refLeft; |
| 177 | refSide = modeVer ? refLeft : refAbove; |
| 178 | } |
| 179 | |
| 180 | if (intraPredAngle == 0) |
| 181 | { |
| 182 | for (k = 0; k < width; k++) |
| 183 | { |
| 184 | for (l = 0; l < width; l++) |
| 185 | { |
| 186 | dst[k * dstStride + l] = refMain[l + 1]; |
| 187 | } |
| 188 | } |
| 189 | |
| 190 | if (bFilter) |
| 191 | { |
| 192 | for (k = 0; k < width; k++) |
| 193 | { |
| 194 | dst[k * dstStride] = (pixel)Clip3((int16_t)0, (int16_t)((1 << X265_DEPTH) - 1), static_cast<int16_t>((dst[k * dstStride]) + ((refSide[k + 1] - refSide[0]) >> 1))); |
| 195 | } |
| 196 | } |
| 197 | } |
| 198 | else |
| 199 | { |
| 200 | int deltaPos = 0; |
| 201 | int deltaInt; |
| 202 | int deltaFract; |
| 203 | int refMainIndex; |
| 204 | |
| 205 | for (k = 0; k < width; k++) |
| 206 | { |
| 207 | deltaPos += intraPredAngle; |
| 208 | deltaInt = deltaPos >> 5; |
| 209 | deltaFract = deltaPos & (32 - 1); |
| 210 | |
| 211 | if (deltaFract) |
| 212 | { |
| 213 | // Do linear filtering |
| 214 | for (l = 0; l < width; l++) |
| 215 | { |
| 216 | refMainIndex = l + deltaInt + 1; |
| 217 | dst[k * dstStride + l] = (pixel)(((32 - deltaFract) * refMain[refMainIndex] + deltaFract * refMain[refMainIndex + 1] + 16) >> 5); |
| 218 | } |
| 219 | } |
| 220 | else |
| 221 | { |
| 222 | // Just copy the integer samples |
| 223 | for (l = 0; l < width; l++) |
| 224 | { |
| 225 | dst[k * dstStride + l] = refMain[l + deltaInt + 1]; |
| 226 | } |
| 227 | } |
| 228 | } |
| 229 | } |
| 230 | |
| 231 | // Flip the block if this is the horizontal mode |
| 232 | if (modeHor) |
| 233 | { |
| 234 | for (k = 0; k < width - 1; k++) |
| 235 | { |
| 236 | for (l = k + 1; l < width; l++) |
| 237 | { |
| 238 | pixel tmp = dst[k * dstStride + l]; |
| 239 | dst[k * dstStride + l] = dst[l * dstStride + k]; |
| 240 | dst[l * dstStride + k] = tmp; |
| 241 | } |
| 242 | } |
| 243 | } |
| 244 | } |
| 245 | } |
| 246 | |
| 247 | template<int log2Size> |
| 248 | void all_angs_pred_c(pixel *dest, pixel *above0, pixel *left0, pixel *above1, pixel *left1, int bLuma) |
| 249 | { |
| 250 | const int size = 1 << log2Size; |
| 251 | for (int mode = 2; mode <= 34; mode++) |
| 252 | { |
| 253 | pixel *left = (g_intraFilterFlags[mode] & size ? left1 : left0); |
| 254 | pixel *above = (g_intraFilterFlags[mode] & size ? above1 : above0); |
| 255 | pixel *out = dest + ((mode - 2) << (log2Size * 2)); |
| 256 | |
| 257 | intra_pred_ang_c<size>(out, size, left, above, mode, bLuma); |
| 258 | |
| 259 | // Optimize code don't flip buffer |
| 260 | bool modeHor = (mode < 18); |
| 261 | |
| 262 | // transpose the block if this is a horizontal mode |
| 263 | if (modeHor) |
| 264 | { |
| 265 | for (int k = 0; k < size - 1; k++) |
| 266 | { |
| 267 | for (int l = k + 1; l < size; l++) |
| 268 | { |
| 269 | pixel tmp = out[k * size + l]; |
| 270 | out[k * size + l] = out[l * size + k]; |
| 271 | out[l * size + k] = tmp; |
| 272 | } |
| 273 | } |
| 274 | } |
| 275 | } |
| 276 | } |
| 277 | } |
| 278 | |
| 279 | namespace x265 { |
| 280 | // x265 private namespace |
| 281 | |
| 282 | void Setup_C_IPredPrimitives(EncoderPrimitives& p) |
| 283 | { |
| 284 | p.intra_pred[0][BLOCK_4x4] = planar_pred_c<2>; |
| 285 | p.intra_pred[0][BLOCK_8x8] = planar_pred_c<3>; |
| 286 | p.intra_pred[0][BLOCK_16x16] = planar_pred_c<4>; |
| 287 | p.intra_pred[0][BLOCK_32x32] = planar_pred_c<5>; |
| 288 | |
| 289 | // Intra Prediction DC |
| 290 | p.intra_pred[1][BLOCK_4x4] = intra_pred_dc_c<4>; |
| 291 | p.intra_pred[1][BLOCK_8x8] = intra_pred_dc_c<8>; |
| 292 | p.intra_pred[1][BLOCK_16x16] = intra_pred_dc_c<16>; |
| 293 | p.intra_pred[1][BLOCK_32x32] = intra_pred_dc_c<32>; |
| 294 | for (int i = 2; i < NUM_INTRA_MODE; i++) |
| 295 | { |
| 296 | p.intra_pred[i][BLOCK_4x4] = intra_pred_ang_c<4>; |
| 297 | p.intra_pred[i][BLOCK_8x8] = intra_pred_ang_c<8>; |
| 298 | p.intra_pred[i][BLOCK_16x16] = intra_pred_ang_c<16>; |
| 299 | p.intra_pred[i][BLOCK_32x32] = intra_pred_ang_c<32>; |
| 300 | } |
| 301 | |
| 302 | p.intra_pred_allangs[BLOCK_4x4] = all_angs_pred_c<2>; |
| 303 | p.intra_pred_allangs[BLOCK_8x8] = all_angs_pred_c<3>; |
| 304 | p.intra_pred_allangs[BLOCK_16x16] = all_angs_pred_c<4>; |
| 305 | p.intra_pred_allangs[BLOCK_32x32] = all_angs_pred_c<5>; |
| 306 | } |
| 307 | } |