hl_cuda_cnn.cu 45.8 KB
Newer Older
1
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
C
chengduoZH 已提交
2

Z
zhangjinchao01 已提交
3 4 5
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
C
chengduoZH 已提交
6

Z
zhangjinchao01 已提交
7
    http://www.apache.org/licenses/LICENSE-2.0
C
chengduoZH 已提交
8

Z
zhangjinchao01 已提交
9 10 11 12 13 14 15 16 17
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License. */

#include <float.h>
#include "hl_base.h"
#include "hl_cnn.h"
L
liaogang 已提交
18
#include "hl_device_functions.cuh"
Z
zhangjinchao01 已提交
19

L
liaogang 已提交
20 21 22 23
__global__ void KeMaxPoolForward(const int nthreads,
                                 const real* inputData,
                                 const int channels,
                                 const int height,
24
                                 const int width,
L
liaogang 已提交
25 26 27 28 29 30 31 32 33 34 35
                                 const int pooledH,
                                 const int pooledW,
                                 const int ksizeW,
                                 const int ksizeH,
                                 const int strideH,
                                 const int strideW,
                                 const int offsetH,
                                 const int offsetW,
                                 real* tgtData,
                                 const int tgtStride) {
  int index = blockIdx.x * blockDim.x + threadIdx.x;
Z
zhangjinchao01 已提交
36 37 38 39
  if (index < nthreads) {
    int pw = index % pooledW;
    int ph = (index / pooledW) % pooledH;
    int c = (index / pooledW / pooledH) % channels;
40 41 42 43 44 45 46
    int frameNum = index / pooledW / pooledH / channels;
    int hstart = ph * strideH - offsetH;
    int wstart = pw * strideW - offsetW;
    int hend = min(hstart + ksizeH, height);
    int wend = min(wstart + ksizeW, width);
    hstart = max(hstart, 0);
    wstart = max(wstart, 0);
Z
zhangjinchao01 已提交
47 48 49 50 51 52 53 54
    real maxval = -FLT_MAX;
    inputData += (frameNum * channels + c) * height * width;
    for (int h = hstart; h < hend; ++h) {
      for (int w = wstart; w < wend; ++w) {
        if (maxval < inputData[h * width + w])
          maxval = inputData[h * width + w];
      }
    }
L
liaogang 已提交
55 56
    int tgtIndex =
        index % (pooledW * pooledH * channels) + frameNum * tgtStride;
Q
qijun 已提交
57
    tgtData[tgtIndex] = maxval;
Z
zhangjinchao01 已提交
58 59 60
  }
}

L
liaogang 已提交
61 62
void hl_maxpool_forward(const int frameCnt,
                        const real* inputData,
63
                        const int channels,
L
liaogang 已提交
64 65 66 67 68 69 70 71 72 73 74 75
                        const int height,
                        const int width,
                        const int pooledH,
                        const int pooledW,
                        const int sizeX,
                        const int sizeY,
                        const int strideH,
                        const int strideW,
                        const int paddingH,
                        const int paddingW,
                        real* tgtData,
                        const int tgtStride) {
76 77
  int num_kernels = pooledH * pooledW * channels * frameCnt;
  int blocks = (num_kernels + 1024 - 1) / 1024;
Z
zhangjinchao01 已提交
78
  dim3 threads(1024, 1);
79 80
  dim3 grid(blocks, 1);

L
liaogang 已提交
81 82 83 84 85 86 87 88 89 90 91 92 93 94 95
  KeMaxPoolForward<<<grid, threads, 0, STREAM_DEFAULT>>>(num_kernels,
                                                         inputData,
                                                         channels,
                                                         height,
                                                         width,
                                                         pooledH,
                                                         pooledW,
                                                         sizeX,
                                                         sizeY,
                                                         strideH,
                                                         strideW,
                                                         paddingH,
                                                         paddingW,
                                                         tgtData,
                                                         tgtStride);
Z
zhangjinchao01 已提交
96 97 98
  CHECK_SYNC("hl_maxpool_forward failed");
}

L
liaogang 已提交
99 100 101 102 103 104
__global__ void KeMaxPoolBackward(const int nthreads,
                                  const real* inputData,
                                  const real* outData,
                                  const real* outGrad,
                                  const int channels,
                                  const int height,
105
                                  const int width,
L
liaogang 已提交
106 107 108 109 110 111 112 113 114 115 116 117 118
                                  const int pooledH,
                                  const int pooledW,
                                  const int sizeX,
                                  const int sizeY,
                                  const int strideH,
                                  const int strideW,
                                  const int padH,
                                  const int padW,
                                  real scaleA,
                                  real scaleB,
                                  real* targetGrad,
                                  const int outStride) {
  int index = blockIdx.x * blockDim.x + threadIdx.x;
Z
zhangjinchao01 已提交
119 120 121
  if (index < nthreads) {
    // find out the local index
    // find out the local offset
122 123
    int offsetW = index % width + padW;
    int offsetH = (index / width) % height + padH;
Z
zhangjinchao01 已提交
124
    int offsetC = (index / width / height) % channels;
125 126 127 128 129 130

    int frameNum = index / width / height / channels;
    int phstart = (offsetH < sizeY) ? 0 : (offsetH - sizeY) / strideH + 1;
    int pwstart = (offsetW < sizeX) ? 0 : (offsetW - sizeX) / strideW + 1;
    int phend = offsetH >= 0 ? min(offsetH / strideH + 1, pooledH) : 0;
    int pwend = offsetW >= 0 ? min(offsetW / strideW + 1, pooledW) : 0;
Z
zhangjinchao01 已提交
131 132
    real gradient = 0;
    real input = inputData[index];
Q
qijun 已提交
133 134
    outData += (frameNum * outStride + offsetC * pooledH * pooledW);
    outGrad += (frameNum * outStride + offsetC * pooledH * pooledW);
Z
zhangjinchao01 已提交
135 136 137 138 139 140 141
    for (int ph = phstart; ph < phend; ++ph) {
      for (int pw = pwstart; pw < pwend; ++pw) {
        if (input == outData[ph * pooledW + pw]) {
          gradient += outGrad[ph * pooledW + pw];
        }
      }
    }
L
liaogang 已提交
142
    targetGrad[index] = scaleB * targetGrad[index] + scaleA * gradient;
Z
zhangjinchao01 已提交
143 144 145
  }
}

L
liaogang 已提交
146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164
void hl_maxpool_backward(const int frameCnt,
                         const real* inputData,
                         const real* outData,
                         const real* outGrad,
                         const int channels,
                         const int height,
                         const int width,
                         const int pooledH,
                         const int pooledW,
                         const int sizeX,
                         const int sizeY,
                         const int strideH,
                         const int strideW,
                         const int paddingH,
                         const int paddingW,
                         real scaleA,
                         real scaleB,
                         real* targetGrad,
                         const int outStride) {
165 166
  int num_kernels = height * width * channels * frameCnt;
  int blocks = (num_kernels + 1024 - 1) / 1024;
Z
zhangjinchao01 已提交
167

L
liaogang 已提交
168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186
  KeMaxPoolBackward<<<blocks, 1024, 0, STREAM_DEFAULT>>>(num_kernels,
                                                         inputData,
                                                         outData,
                                                         outGrad,
                                                         channels,
                                                         height,
                                                         width,
                                                         pooledH,
                                                         pooledW,
                                                         sizeX,
                                                         sizeY,
                                                         strideH,
                                                         strideW,
                                                         paddingH,
                                                         paddingW,
                                                         scaleA,
                                                         scaleB,
                                                         targetGrad,
                                                         outStride);
Z
zhangjinchao01 已提交
187 188 189
  CHECK_SYNC("hl_maxpool_backward");
}

L
liaogang 已提交
190 191
__global__ void KeAvgPoolForward(const int nthreads,
                                 const real* inputData,
192
                                 const int channels,
L
liaogang 已提交
193 194 195 196 197 198 199 200 201 202 203 204
                                 const int height,
                                 const int width,
                                 const int pooledH,
                                 const int pooledW,
                                 const int sizeX,
                                 const int sizeY,
                                 const int strideH,
                                 const int strideW,
                                 const int padH,
                                 const int padW,
                                 real* tgtData,
                                 const int tgtStride) {
205
  int index = blockIdx.x * blockDim.x + threadIdx.x;
Z
zhangjinchao01 已提交
206 207 208 209
  if (index < nthreads) {
    int pw = index % pooledW;
    int ph = (index / pooledW) % pooledH;
    int c = (index / pooledW / pooledH) % channels;
210 211 212 213 214 215 216 217 218 219 220 221
    int frameNum = index / pooledW / pooledH / channels;

    int hstart = ph * strideH - padH;
    int wstart = pw * strideW - padW;
    int hend = min(hstart + sizeY, height + padH);
    int wend = min(wstart + sizeX, width + padW);
    int pool_size = (hend - hstart) * (wend - wstart);
    hstart = max(hstart, 0);
    wstart = max(wstart, 0);
    hend = min(hend, height);
    wend = min(wend, width);

Z
zhangjinchao01 已提交
222 223 224 225 226 227 228
    real aveval = 0;
    inputData += (frameNum * channels + c) * height * width;
    for (int h = hstart; h < hend; ++h) {
      for (int w = wstart; w < wend; ++w) {
        aveval += inputData[h * width + w];
      }
    }
L
liaogang 已提交
229 230
    int tgtIndex =
        index % (pooledW * pooledH * channels) + frameNum * tgtStride;
Q
qijun 已提交
231
    tgtData[tgtIndex] = aveval / pool_size;
Z
zhangjinchao01 已提交
232 233 234
  }
}

L
liaogang 已提交
235 236
void hl_avgpool_forward(const int frameCnt,
                        const real* inputData,
237
                        const int channels,
L
liaogang 已提交
238 239 240 241 242 243 244 245 246 247 248 249
                        const int height,
                        const int width,
                        const int pooledH,
                        const int pooledW,
                        const int sizeX,
                        const int sizeY,
                        const int strideH,
                        const int strideW,
                        const int paddingH,
                        const int paddingW,
                        real* tgtData,
                        const int tgtStride) {
250 251
  int num_kernels = pooledH * pooledW * channels * frameCnt;
  int blocks = (num_kernels + 1024 - 1) / 1024;
L
liaogang 已提交
252 253 254 255 256 257 258 259 260 261 262 263 264 265 266
  KeAvgPoolForward<<<blocks, 1024, 0, STREAM_DEFAULT>>>(num_kernels,
                                                        inputData,
                                                        channels,
                                                        height,
                                                        width,
                                                        pooledH,
                                                        pooledW,
                                                        sizeX,
                                                        sizeY,
                                                        strideH,
                                                        strideW,
                                                        paddingH,
                                                        paddingW,
                                                        tgtData,
                                                        tgtStride);
Z
zhangjinchao01 已提交
267 268 269
  CHECK_SYNC("hl_avgpool_forward failed");
}

L
liaogang 已提交
270 271 272 273
__global__ void KeAvgPoolBackward(const int nthreads,
                                  const real* outGrad,
                                  const int channels,
                                  const int height,
274
                                  const int width,
L
liaogang 已提交
275 276 277 278 279 280 281 282 283 284 285 286
                                  const int pooledH,
                                  const int pooledW,
                                  const int sizeX,
                                  const int sizeY,
                                  const int strideH,
                                  const int strideW,
                                  const int padH,
                                  const int padW,
                                  real scaleA,
                                  real scaleB,
                                  real* tgtGrad,
                                  const int outStride) {
287
  int index = blockIdx.x * blockDim.x + threadIdx.x;
Z
zhangjinchao01 已提交
288
  if (index < nthreads) {
289 290
    int offsetW = index % width + padW;
    int offsetH = (index / width) % height + padH;
Z
zhangjinchao01 已提交
291
    int offsetC = (index / width / height) % channels;
292 293 294 295 296 297
    int frameNum = index / width / height / channels;

    int phstart = (offsetH < sizeY) ? 0 : (offsetH - sizeY) / strideH + 1;
    int pwstart = (offsetW < sizeX) ? 0 : (offsetW - sizeX) / strideW + 1;
    int phend = offsetH >= 0 ? min(offsetH / strideH + 1, pooledH) : 0;
    int pwend = offsetW >= 0 ? min(offsetW / strideW + 1, pooledW) : 0;
Z
zhangjinchao01 已提交
298
    real gradient = 0;
Q
qijun 已提交
299 300
    outGrad += (frameNum * outStride + offsetC * pooledH * pooledW);

Z
zhangjinchao01 已提交
301 302 303
    for (int ph = phstart; ph < phend; ++ph) {
      for (int pw = pwstart; pw < pwend; ++pw) {
        // figure out the pooling size
304 305 306 307 308
        int hstart = ph * strideH - padH;
        int wstart = pw * strideW - padW;
        int hend = min(hstart + sizeY, height + padH);
        int wend = min(wstart + sizeX, width + padW);
        int poolsize = (hend - hstart) * (wend - wstart);
L
liaogang 已提交
309
        gradient += outGrad[ph * pooledW + pw] / poolsize;
Z
zhangjinchao01 已提交
310 311 312 313 314 315
      }
    }
    tgtGrad[index] = scaleB * tgtGrad[index] + scaleA * gradient;
  }
}

L
liaogang 已提交
316 317
void hl_avgpool_backward(const int frameCnt,
                         const real* outGrad,
318
                         const int channels,
L
liaogang 已提交
319 320 321 322 323 324 325 326 327 328 329 330 331 332
                         const int height,
                         const int width,
                         const int pooledH,
                         const int pooledW,
                         const int sizeX,
                         const int sizeY,
                         const int strideH,
                         const int strideW,
                         const int paddingH,
                         const int paddingW,
                         real scaleA,
                         real scaleB,
                         real* backGrad,
                         const int outStride) {
333 334
  int num_kernels = height * width * channels * frameCnt;
  int blocks = (num_kernels + 1024 - 1) / 1024;
Z
zhangjinchao01 已提交
335

L
liaogang 已提交
336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352
  KeAvgPoolBackward<<<blocks, 1024, 0, STREAM_DEFAULT>>>(num_kernels,
                                                         outGrad,
                                                         channels,
                                                         height,
                                                         width,
                                                         pooledH,
                                                         pooledW,
                                                         sizeX,
                                                         sizeY,
                                                         strideH,
                                                         strideW,
                                                         paddingH,
                                                         paddingW,
                                                         scaleA,
                                                         scaleB,
                                                         backGrad,
                                                         outStride);
Z
zhangjinchao01 已提交
353 354 355
  CHECK_SYNC("hl_avgpool_backward failed");
}

C
chengduoZH 已提交
356 357 358 359 360 361 362 363 364 365 366 367 368 369 370
__global__ void KeMaxPool3DForward(const int nthreads,
                                   const real* inputData,
                                   const int channels,
                                   const int depth,
                                   const int height,
                                   const int width,
                                   const int pooledD,
                                   const int pooledH,
                                   const int pooledW,
                                   const int ksizeD,
                                   const int ksizeH,
                                   const int ksizeW,
                                   const int strideD,
                                   const int strideH,
                                   const int strideW,
C
chengduoZH 已提交
371 372 373
                                   const int padD,
                                   const int padH,
                                   const int padW,
C
chengduoZH 已提交
374
                                   real* tgtData,
C
chengduoZH 已提交
375
                                   real* maxPoolIdxData,
C
chengduoZH 已提交
376 377 378 379 380 381 382 383
                                   const int tgtStride) {
  for (int index = blockIdx.x * blockDim.x + threadIdx.x; index < (nthreads);
       index += blockDim.x * gridDim.x) {
    int pw = index % pooledW;
    int ph = (index / pooledW) % pooledH;
    int pd = (index / pooledW / pooledH) % pooledD;
    int c = (index / pooledW / pooledH / pooledD) % channels;
    int frameNum = index / pooledW / pooledH / pooledD / channels;
C
chengduoZH 已提交
384 385 386
    int dstart = pd * strideD - padD;
    int hstart = ph * strideH - padH;
    int wstart = pw * strideW - padW;
C
chengduoZH 已提交
387 388 389 390 391 392 393
    int dend = min(dstart + ksizeD, depth);
    int hend = min(hstart + ksizeH, height);
    int wend = min(wstart + ksizeW, width);
    dstart = max(dstart, 0);
    hstart = max(hstart, 0);
    wstart = max(wstart, 0);
    real maxval = -FLT_MAX;
C
chengduoZH 已提交
394
    int maxIdx = -1;
C
chengduoZH 已提交
395 396 397 398
    inputData += (frameNum * channels + c) * depth * height * width;
    for (int d = dstart; d < dend; ++d) {
      for (int h = hstart; h < hend; ++h) {
        for (int w = wstart; w < wend; ++w) {
C
chengduoZH 已提交
399
          if (maxval < inputData[(d * height + h) * width + w]) {
C
chengduoZH 已提交
400
            maxval = inputData[(d * height + h) * width + w];
C
chengduoZH 已提交
401 402
            maxIdx = (d * height + h) * width + w;
          }
C
chengduoZH 已提交
403 404 405 406 407 408
        }
      }
    }
    int tgtIndex =
        index % (pooledW * pooledH * pooledD * channels) + frameNum * tgtStride;
    tgtData[tgtIndex] = maxval;
C
chengduoZH 已提交
409
    maxPoolIdxData[tgtIndex] = maxIdx;
C
chengduoZH 已提交
410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427
  }
}

void hl_maxpool3D_forward(const int frameCnt,
                          const real* inputData,
                          const int channels,
                          const int depth,
                          const int height,
                          const int width,
                          const int pooledD,
                          const int pooledH,
                          const int pooledW,
                          const int sizeZ,
                          const int sizeY,
                          const int sizeX,
                          const int strideD,
                          const int strideH,
                          const int strideW,
C
chengduoZH 已提交
428 429 430
                          const int padD,
                          const int padH,
                          const int padW,
C
chengduoZH 已提交
431
                          real* tgtData,
C
chengduoZH 已提交
432
                          real* maxPoolIdxData,
C
chengduoZH 已提交
433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453
                          const int tgtStride) {
  int num_kernels = pooledD * pooledH * pooledW * channels * frameCnt;
  int blocks = (num_kernels + 1024 - 1) / 1024;
  dim3 threads(1024, 1);
  dim3 grid(blocks, 1);

  KeMaxPool3DForward<<<grid, threads, 0, STREAM_DEFAULT>>>(num_kernels,
                                                           inputData,
                                                           channels,
                                                           depth,
                                                           height,
                                                           width,
                                                           pooledD,
                                                           pooledH,
                                                           pooledW,
                                                           sizeZ,
                                                           sizeY,
                                                           sizeX,
                                                           strideD,
                                                           strideH,
                                                           strideW,
C
chengduoZH 已提交
454 455 456
                                                           padD,
                                                           padH,
                                                           padW,
C
chengduoZH 已提交
457
                                                           tgtData,
C
chengduoZH 已提交
458
                                                           maxPoolIdxData,
C
chengduoZH 已提交
459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483
                                                           tgtStride);
  CHECK_SYNC("hl_maxpool3D_forward failed");
}

__global__ void KeMaxPool3DBackward(const int nthreads,
                                    const real* outGrad,
                                    const int channels,
                                    const int depth,
                                    const int height,
                                    const int width,
                                    const int pooledD,
                                    const int pooledH,
                                    const int pooledW,
                                    const int sizeZ,
                                    const int sizeY,
                                    const int sizeX,
                                    const int strideD,
                                    const int strideH,
                                    const int strideW,
                                    const int padD,
                                    const int padH,
                                    const int padW,
                                    real scaleA,
                                    real scaleB,
                                    real* targetGrad,
C
chengduoZH 已提交
484
                                    real* maxPoolIdxData,
C
chengduoZH 已提交
485 486 487
                                    const int outStride) {
  for (int index = blockIdx.x * blockDim.x + threadIdx.x; index < (nthreads);
       index += blockDim.x * gridDim.x) {
C
chengduoZH 已提交
488 489 490
    int offsetW = index % width;
    int offsetH = (index / width) % height;
    int offsetD = (index / width / height) % depth;
C
chengduoZH 已提交
491 492 493
    int offsetC = (index / width / height / depth) % channels;
    int frameNum = index / width / height / depth / channels;

C
chengduoZH 已提交
494 495 496 497 498 499 500 501 502
    int pdstart =
        (offsetD + padD < sizeZ) ? 0 : (offsetD + padD - sizeZ) / strideD + 1;
    int phstart =
        (offsetH + padH < sizeY) ? 0 : (offsetH + padH - sizeY) / strideH + 1;
    int pwstart =
        (offsetW + padW < sizeX) ? 0 : (offsetW + padW - sizeX) / strideW + 1;
    int pdend = min((offsetD + padD) / strideD + 1, pooledD);
    int phend = min((offsetH + padH) / strideH + 1, pooledH);
    int pwend = min((offsetW + padW) / strideW + 1, pooledW);
C
chengduoZH 已提交
503 504 505

    real gradient = 0;
    outGrad += ((frameNum * channels + offsetC) * pooledD * pooledH * pooledW);
C
chengduoZH 已提交
506 507
    maxPoolIdxData +=
        ((frameNum * channels + offsetC) * pooledD * pooledH * pooledW);
C
chengduoZH 已提交
508 509 510
    for (int pd = pdstart; pd < pdend; ++pd) {
      for (int ph = phstart; ph < phend; ++ph) {
        for (int pw = pwstart; pw < pwend; ++pw) {
C
chengduoZH 已提交
511 512
          if (((offsetD * height + offsetH) * width + offsetW) ==
              maxPoolIdxData[(pd * pooledH + ph) * pooledW + pw])
C
chengduoZH 已提交
513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541
            gradient += outGrad[(pd * pooledH + ph) * pooledW + pw];
        }
      }
    }
    targetGrad[index] = scaleA * gradient + scaleB * targetGrad[index];
  }
}

void hl_maxpool3D_backward(const int frameCnt,
                           const real* outGrad,
                           const int channels,
                           const int depth,
                           const int height,
                           const int width,
                           const int outputD,
                           const int outputH,
                           const int outputW,
                           const int sizeZ,
                           const int sizeY,
                           const int sizeX,
                           const int strideD,
                           const int strideH,
                           const int strideW,
                           const int paddingD,
                           const int paddingH,
                           const int paddingW,
                           real scaleA,
                           real scaleB,
                           real* targetGrad,
C
chengduoZH 已提交
542
                           real* maxPoolIdxData,
C
chengduoZH 已提交
543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567
                           const int outStride) {
  int num_kernels = depth * height * width * channels * frameCnt;
  int blocks = (num_kernels + 1024 - 1) / 1024;

  KeMaxPool3DBackward<<<blocks, 1024, 0, STREAM_DEFAULT>>>(num_kernels,
                                                           outGrad,
                                                           channels,
                                                           depth,
                                                           height,
                                                           width,
                                                           outputD,
                                                           outputH,
                                                           outputW,
                                                           sizeZ,
                                                           sizeY,
                                                           sizeX,
                                                           strideD,
                                                           strideH,
                                                           strideW,
                                                           paddingD,
                                                           paddingH,
                                                           paddingW,
                                                           scaleA,
                                                           scaleB,
                                                           targetGrad,
C
chengduoZH 已提交
568
                                                           maxPoolIdxData,
C
chengduoZH 已提交
569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782
                                                           outStride);
  CHECK_SYNC("hl_maxpool3D_backward");
}

__global__ void KeAvgPool3DForward(const int nthreads,
                                   const real* inputData,
                                   const int channels,
                                   const int depth,
                                   const int height,
                                   const int width,
                                   const int pooledD,
                                   const int pooledH,
                                   const int pooledW,
                                   const int sizeZ,
                                   const int sizeY,
                                   const int sizeX,
                                   const int strideD,
                                   const int strideH,
                                   const int strideW,
                                   const int padD,
                                   const int padH,
                                   const int padW,
                                   real* tgtData,
                                   const int tgtStride) {
  for (int index = blockIdx.x * blockDim.x + threadIdx.x; index < (nthreads);
       index += blockDim.x * gridDim.x) {
    int pw = index % pooledW;
    int ph = (index / pooledW) % pooledH;
    int pd = (index / pooledW / pooledH) % pooledD;
    int c = (index / pooledW / pooledH / pooledD) % channels;
    int frameNum = index / pooledW / pooledH / pooledD / channels;
    int dstart = pd * strideD - padD;
    int hstart = ph * strideH - padH;
    int wstart = pw * strideW - padW;
    int dend = min(dstart + sizeZ, depth + padD);
    int hend = min(hstart + sizeY, height + padH);
    int wend = min(wstart + sizeX, width + padW);
    int pool_size = (dend - dstart) * (hend - hstart) * (wend - wstart);
    dstart = max(dstart, 0);
    hstart = max(hstart, 0);
    wstart = max(wstart, 0);
    dend = min(dend, depth);
    hend = min(hend, height);
    wend = min(wend, width);

    real aveval = 0;
    inputData += (frameNum * channels + c) * depth * height * width;
    for (int d = dstart; d < dend; ++d) {
      for (int h = hstart; h < hend; ++h) {
        for (int w = wstart; w < wend; ++w) {
          aveval += inputData[(d * height + h) * width + w];
        }
      }
    }
    int tgtIndex =
        index % (pooledW * pooledH * pooledD * channels) + frameNum * tgtStride;
    tgtData[tgtIndex] = aveval / pool_size;
  }
}

void hl_avgpool3D_forward(const int frameCnt,
                          const real* inputData,
                          const int channels,
                          const int depth,
                          const int height,
                          const int width,
                          const int pooledD,
                          const int pooledH,
                          const int pooledW,
                          const int sizeZ,
                          const int sizeY,
                          const int sizeX,
                          const int strideD,
                          const int strideH,
                          const int strideW,
                          const int paddingD,
                          const int paddingH,
                          const int paddingW,
                          real* tgtData,
                          const int tgtStride) {
  int num_kernels = pooledD * pooledH * pooledW * channels * frameCnt;
  int blocks = (num_kernels + 1024 - 1) / 1024;
  KeAvgPool3DForward<<<blocks, 1024, 0, STREAM_DEFAULT>>>(num_kernels,
                                                          inputData,
                                                          channels,
                                                          depth,
                                                          height,
                                                          width,
                                                          pooledD,
                                                          pooledH,
                                                          pooledW,
                                                          sizeZ,
                                                          sizeY,
                                                          sizeX,
                                                          strideD,
                                                          strideH,
                                                          strideW,
                                                          paddingD,
                                                          paddingH,
                                                          paddingW,
                                                          tgtData,
                                                          tgtStride);
  CHECK_SYNC("hl_avgpool3D_forward failed");
}

__global__ void KeAvgPool3DBackward(const int nthreads,
                                    const real* outGrad,
                                    const int channels,
                                    const int depth,
                                    const int height,
                                    const int width,
                                    const int pooledD,
                                    const int pooledH,
                                    const int pooledW,
                                    const int sizeZ,
                                    const int sizeY,
                                    const int sizeX,
                                    const int strideD,
                                    const int strideH,
                                    const int strideW,
                                    const int padD,
                                    const int padH,
                                    const int padW,
                                    real scaleA,
                                    real scaleB,
                                    real* tgtGrad,
                                    const int outStride) {
  for (int index = blockIdx.x * blockDim.x + threadIdx.x; index < (nthreads);
       index += blockDim.x * gridDim.x) {
    int offsetW = index % width + padW;
    int offsetH = (index / width) % height + padH;
    int offsetD = (index / width / height) % depth + padD;
    int offsetC = (index / width / height / depth) % channels;
    int frameNum = index / width / height / depth / channels;

    int pdstart = (offsetD < sizeZ) ? 0 : (offsetD - sizeZ) / strideD + 1;
    int phstart = (offsetH < sizeY) ? 0 : (offsetH - sizeY) / strideH + 1;
    int pwstart = (offsetW < sizeX) ? 0 : (offsetW - sizeX) / strideW + 1;
    int pdend = min(offsetD / strideD + 1, pooledD);
    int phend = min(offsetH / strideH + 1, pooledH);
    int pwend = min(offsetW / strideW + 1, pooledW);

    real gradient = 0;
    outGrad += (frameNum * channels + offsetC) * pooledD * pooledH * pooledW;

    for (int pd = pdstart; pd < pdend; ++pd) {
      for (int ph = phstart; ph < phend; ++ph) {
        for (int pw = pwstart; pw < pwend; ++pw) {
          // figure out the pooling size
          int dstart = pd * strideD - padD;
          int hstart = ph * strideH - padH;
          int wstart = pw * strideW - padW;
          int dend = min(dstart + sizeZ, depth + padD);
          int hend = min(hstart + sizeY, height + padH);
          int wend = min(wstart + sizeX, width + padW);
          int poolsize = (dend - dstart) * (hend - hstart) * (wend - wstart);
          gradient += outGrad[(pd * pooledH + ph) * pooledW + pw] / poolsize;
        }
      }
    }
    tgtGrad[index] = scaleA * gradient + scaleB * tgtGrad[index];
  }
}

void hl_avgpool3D_backward(const int frameCnt,
                           const real* outGrad,
                           const int channels,
                           const int depth,
                           const int height,
                           const int width,
                           const int outputD,
                           const int outputH,
                           const int outputW,
                           const int sizeZ,
                           const int sizeY,
                           const int sizeX,
                           const int strideD,
                           const int strideH,
                           const int strideW,
                           int paddingD,
                           int paddingH,
                           int paddingW,
                           real scaleA,
                           real scaleB,
                           real* backGrad,
                           const int outStride) {
  int num_kernels = depth * height * width * channels * frameCnt;
  int blocks = (num_kernels + 1024 - 1) / 1024;

  KeAvgPool3DBackward<<<blocks, 1024, 0, STREAM_DEFAULT>>>(num_kernels,
                                                           outGrad,
                                                           channels,
                                                           depth,
                                                           height,
                                                           width,
                                                           outputD,
                                                           outputH,
                                                           outputW,
                                                           sizeZ,
                                                           sizeY,
                                                           sizeX,
                                                           strideD,
                                                           strideH,
                                                           strideW,
                                                           paddingD,
                                                           paddingH,
                                                           paddingW,
                                                           scaleA,
                                                           scaleB,
                                                           backGrad,
                                                           outStride);
  CHECK_SYNC("hl_avgpool3D_backward failed");
}

L
liaogang 已提交
783
__global__ void KeBilinearInterpFw(const real* in,
L
liaogang 已提交
784 785 786 787 788 789 790 791 792 793 794 795
                                   const size_t inImgH,
                                   const size_t inImgW,
                                   const size_t inputH,
                                   const size_t inputW,
                                   real* out,
                                   const size_t outImgH,
                                   const size_t outImgW,
                                   const size_t outputH,
                                   const size_t outputW,
                                   const size_t numChannels,
                                   const real ratioH,
                                   const real ratioW) {
L
liaogang 已提交
796
  int nthreads = outputH * outputW;
L
liaogang 已提交
797
  int tid = blockIdx.x * blockDim.x + threadIdx.x;
L
liaogang 已提交
798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816
  if (tid < nthreads) {
    int outIdH = tid / outputW;
    int outIdW = tid % outputW;
    int inImgSize = inputW / numChannels;
    int outImgSize = outputW / numChannels;
    int channelId = outIdW / outImgSize;

    int outImgIdy = (outIdW % outImgSize) / outImgW;
    int inImgIdy = ratioH * outImgIdy;
    int hId = (inImgIdy < inImgH - 1) ? 1 : 0;
    real h1lambda = ratioH * outImgIdy - inImgIdy;
    real h2lambda = 1.f - h1lambda;

    int outImgIdx = tid % outImgW;
    int inImgIdx = ratioW * outImgIdx;
    int wId = (inImgIdx < inImgW - 1) ? 1 : 0;
    real w1lambda = ratioW * outImgIdx - inImgIdx;
    real w2lambda = 1.f - w1lambda;

L
liaogang 已提交
817 818
    const real* inPos = &in[outIdH * inputW + channelId * inImgSize +
                            inImgIdy * inImgW + inImgIdx];
L
liaogang 已提交
819 820 821

    // bilinear interpolation
    out[outIdH * outputW + outIdW] =
L
liaogang 已提交
822 823 824
        h2lambda * (w2lambda * inPos[0] + w1lambda * inPos[wId]) +
        h1lambda * (w2lambda * inPos[hId * inImgW] +
                    w1lambda * inPos[hId * inImgW + wId]);
L
liaogang 已提交
825 826 827 828 829 830 831 832 833 834 835 836 837
  }
}

void hl_bilinear_forward(const real* inData,
                         const size_t inImgH,
                         const size_t inImgW,
                         const size_t inputH,
                         const size_t inputW,
                         real* outData,
                         const size_t outImgH,
                         const size_t outImgW,
                         const size_t outputH,
                         const size_t outputW,
L
liaogang 已提交
838 839 840 841
                         const size_t numChannels,
                         const real ratioH,
                         const real ratioW) {
  int threadNum = outputH * outputW;
L
liaogang 已提交
842 843
  int blocks = (threadNum + 1024 - 1) / 1024;

L
liaogang 已提交
844 845 846 847 848 849 850 851 852 853 854 855 856
  KeBilinearInterpFw<<<blocks, 1024, 0, STREAM_DEFAULT>>>(inData,
                                                          inImgH,
                                                          inImgW,
                                                          inputH,
                                                          inputW,
                                                          outData,
                                                          outImgH,
                                                          outImgW,
                                                          outputH,
                                                          outputW,
                                                          numChannels,
                                                          ratioH,
                                                          ratioW);
L
liaogang 已提交
857 858 859
  CHECK_SYNC("hl_bilinear_forward failed");
}

L
liaogang 已提交
860
__global__ void KeBilinearInterpBw(real* in,
L
liaogang 已提交
861 862 863 864 865 866 867 868 869 870 871 872
                                   const size_t inImgH,
                                   const size_t inImgW,
                                   const size_t inputH,
                                   const size_t inputW,
                                   const real* out,
                                   const size_t outImgH,
                                   const size_t outImgW,
                                   const size_t outputH,
                                   const size_t outputW,
                                   const size_t numChannels,
                                   const real ratioH,
                                   const real ratioW) {
L
liaogang 已提交
873
  int nthreads = outputH * outputW;
L
liaogang 已提交
874
  int tid = blockIdx.x * blockDim.x + threadIdx.x;
L
liaogang 已提交
875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893
  if (tid < nthreads) {
    int outIdH = tid / outputW;
    int outIdW = tid % outputW;
    int inImgSize = inputW / numChannels;
    int outImgSize = outputW / numChannels;
    int channelId = outIdW / outImgSize;

    int outImgIdy = (outIdW % outImgSize) / outImgW;
    int inImgIdy = ratioH * outImgIdy;
    int hId = (inImgIdy < inImgH - 1) ? 1 : 0;
    real h1lambda = ratioH * outImgIdy - inImgIdy;
    real h2lambda = 1.f - h1lambda;

    int outImgIdx = tid % outImgW;
    int inImgIdx = ratioW * outImgIdx;
    int wId = (inImgIdx < inImgW - 1) ? 1 : 0;
    real w1lambda = ratioW * outImgIdx - inImgIdx;
    real w2lambda = 1.f - w1lambda;

L
liaogang 已提交
894 895
    real* inPos = &in[outIdH * inputW + channelId * inImgSize +
                      inImgIdy * inImgW + inImgIdx];
L
liaogang 已提交
896
    const real* outPos = &out[outIdH * outputW + outIdW];
L
liaogang 已提交
897 898
    paddle::paddleAtomicAdd(&inPos[0], h2lambda * w2lambda * outPos[0]);
    paddle::paddleAtomicAdd(&inPos[wId], h2lambda * w1lambda * outPos[0]);
L
liaogang 已提交
899 900 901 902
    paddle::paddleAtomicAdd(&inPos[hId * inImgW],
                            h1lambda * w2lambda * outPos[0]);
    paddle::paddleAtomicAdd(&inPos[hId * inImgW + wId],
                            h1lambda * w1lambda * outPos[0]);
L
liaogang 已提交
903 904 905 906 907 908 909 910 911 912 913 914 915
  }
}

void hl_bilinear_backward(real* inGrad,
                          const size_t inImgH,
                          const size_t inImgW,
                          const size_t inputH,
                          const size_t inputW,
                          const real* outGrad,
                          const size_t outImgH,
                          const size_t outImgW,
                          const size_t outputH,
                          const size_t outputW,
L
liaogang 已提交
916 917 918 919
                          const size_t numChannels,
                          const real ratioH,
                          const real ratioW) {
  int threadNum = outputH * outputW;
L
liaogang 已提交
920 921
  int blocks = (threadNum + 1024 - 1) / 1024;

L
liaogang 已提交
922 923 924 925 926 927 928 929 930 931 932 933 934
  KeBilinearInterpBw<<<blocks, 1024, 0, STREAM_DEFAULT>>>(inGrad,
                                                          inImgH,
                                                          inImgW,
                                                          inputH,
                                                          inputW,
                                                          outGrad,
                                                          outImgH,
                                                          outImgW,
                                                          outputH,
                                                          outputW,
                                                          numChannels,
                                                          ratioH,
                                                          ratioW);
L
liaogang 已提交
935
  CHECK_SYNC("hl_bilinear_backward failed");
L
liaogang 已提交
936 937
}

L
liaogang 已提交
938 939 940 941 942 943 944
__global__ void maxoutFpCompute(size_t nthreads,
                                const real* inData,
                                real* outData,
                                int* idData,
                                size_t size,
                                size_t featLen,
                                size_t groups) {
945
  int index = blockIdx.x * blockDim.x + threadIdx.x;
L
liaogang 已提交
946
  if (index < nthreads) {
947 948 949 950
    size_t batch_idx = index / size;
    size_t i = index % size;
    size_t channel_idx = i / featLen;
    size_t feat_idx = i % featLen;
L
liaogang 已提交
951 952
    size_t data_idx =
        (batch_idx * size + channel_idx * featLen) * groups + feat_idx;
953 954 955 956 957 958 959 960 961 962 963 964 965 966
    real max = inData[data_idx];
    int maxId = 0;
    for (size_t g = 1; g < groups; ++g) {
      real tmp = inData[data_idx + g * featLen];
      if (tmp > max) {
        max = tmp;
        maxId = g;
      }
    }
    outData[index] = max;
    idData[index] = maxId;
  }
}

L
liaogang 已提交
967 968 969 970 971 972 973
void hl_maxout_forward(const real* inData,
                       real* outData,
                       int* idData,
                       size_t batchSize,
                       size_t size,
                       size_t featLen,
                       size_t groups) {
974 975
  int num_kernels = size * batchSize;
  int blocks = (num_kernels + 1024 - 1) / 1024;
L
liaogang 已提交
976 977
  maxoutFpCompute<<<blocks, 1024, 0, STREAM_DEFAULT>>>(
      num_kernels, inData, outData, idData, size, featLen, groups);
978 979 980
  CHECK_SYNC("hl_maxout_forward failed");
}

L
liaogang 已提交
981 982 983 984 985 986 987
__global__ void maxoutBpCompute(size_t nthreads,
                                real* inGrad,
                                const real* outGrad,
                                const int* idData,
                                size_t size,
                                size_t featLen,
                                size_t groups) {
988
  int index = blockIdx.x * blockDim.x + threadIdx.x;
L
liaogang 已提交
989
  if (index < nthreads) {
990 991 992 993 994
    size_t batch_idx = index / size;
    size_t i = index % size;
    size_t channel_idx = i / featLen;
    size_t feat_idx = i % featLen;
    size_t newIndex = batch_idx * size;
L
liaogang 已提交
995 996
    size_t gradIdx =
        (channel_idx * groups + (idData + newIndex)[i]) * featLen + feat_idx;
997 998 999 1000
    (inGrad + newIndex * groups)[gradIdx] += (outGrad + newIndex)[i];
  }
}

L
liaogang 已提交
1001 1002 1003 1004 1005 1006 1007
void hl_maxout_backward(real* inGrad,
                        const real* outGrad,
                        const int* idData,
                        size_t batchSize,
                        size_t size,
                        size_t featLen,
                        size_t groups) {
1008 1009
  int num_kernels = size * batchSize;
  int blocks = (num_kernels + 1024 - 1) / 1024;
L
liaogang 已提交
1010 1011
  maxoutBpCompute<<<blocks, 1024, 0, STREAM_DEFAULT>>>(
      num_kernels, inGrad, outGrad, idData, size, featLen, groups);
1012 1013
  CHECK_SYNC("hl_maxout_backward failed");
}