im2col.cu 14.9 KB
Newer Older
H
hedaoyuan 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.

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

    http://www.apache.org/licenses/LICENSE-2.0

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. */

H
hedaoyuan 已提交
15 16
#include "paddle/operators/math/im2col.h"
#include "paddle/platform/cuda_helper.h"
H
hedaoyuan 已提交
17 18

namespace paddle {
19
namespace operators {
20
namespace math {
H
hedaoyuan 已提交
21 22

template <class T>
H
hedaoyuan 已提交
23 24 25 26
__global__ void im2col(const T* data_im, int num_outs, int height, int width,
                       int filter_height, int filter_width, int stride_height,
                       int stride_width, int padding_height, int padding_width,
                       int output_height, int output_width, T* data_col) {
H
hedaoyuan 已提交
27
  int index = (blockIdx.x * gridDim.y + blockIdx.y) * blockDim.x + threadIdx.x;
H
hedaoyuan 已提交
28 29 30 31 32 33 34 35
  if (index < num_outs) {
    int w_out = index % output_width;
    index /= output_width;
    int h_out = index % output_height;
    int channel_in = index / output_height;
    int channel_out = channel_in * filter_height * filter_width;
    int h_in = h_out * stride_height;
    int w_in = w_out * stride_width;
H
hedaoyuan 已提交
36

H
hedaoyuan 已提交
37 38 39
    data_col += (channel_out * output_height + h_out) * output_width + w_out;
    for (int i = 0; i < filter_height; ++i) {
      for (int j = 0; j < filter_width; ++j) {
H
hedaoyuan 已提交
40 41
        int rIdx = int(h_in + i);
        int cIdx = int(w_in + j);
H
hedaoyuan 已提交
42 43 44 45
        if ((rIdx - (int)padding_height) >= (int)height ||
            (rIdx - (int)padding_height) < 0 ||
            (cIdx - (int)padding_width) >= (int)width ||
            (cIdx - (int)padding_width) < 0) {
H
hedaoyuan 已提交
46 47
          *data_col = 0;
        } else {
H
hedaoyuan 已提交
48 49
          rIdx = rIdx + channel_in * height - padding_height;
          cIdx = cIdx - padding_width;
H
hedaoyuan 已提交
50 51
          *data_col = data_im[rIdx * width + cIdx];
        }
H
hedaoyuan 已提交
52
        data_col += output_height * output_width;
H
hedaoyuan 已提交
53 54 55 56 57 58
      }
    }
  }
}

/*
H
hedaoyuan 已提交
59 60 61
 * im = [input_channels, input_height, input_width]
 * col =
 *   [input_channels, filter_height, filter_width, output_height, output_width]
H
hedaoyuan 已提交
62 63
 */
template <class T>
H
hedaoyuan 已提交
64 65
class Im2ColFunctor<paddle::operators::math::ColFormat::kCFO,
                    platform::GPUPlace, T> {
H
hedaoyuan 已提交
66
 public:
H
hedaoyuan 已提交
67 68
  void operator()(const framework::Tensor& im, framework::Tensor& col,
                  int stride_height, int stride_width, int padding_height,
69
                  int padding_width, platform::DeviceContext* context) {
H
hedaoyuan 已提交
70 71
    PADDLE_ENFORCE(im.dims().size() == 3);
    PADDLE_ENFORCE(col.dims().size() == 5);
H
hedaoyuan 已提交
72

H
hedaoyuan 已提交
73 74 75 76 77 78 79 80 81 82 83 84
    int input_channels = im.dims()[0];
    int input_height = im.dims()[1];
    int input_width = im.dims()[2];
    int filter_height = col.dims()[1];
    int filter_width = col.dims()[2];
    int output_height = col.dims()[3];
    int output_width = col.dims()[4];

    int num_outputs = input_channels * output_height * output_width;
    int blocks = (num_outputs + 1024 - 1) / 1024;
    int block_x = 512;
    int block_y = (blocks + 512 - 1) / 512;
H
hedaoyuan 已提交
85
    dim3 threads(1024, 1);
H
hedaoyuan 已提交
86
    dim3 grid(block_x, block_y);
H
hedaoyuan 已提交
87 88 89
    im2col<T><<<
        grid, threads, 0,
        reinterpret_cast<platform::CUDADeviceContext*>(context)->stream()>>>(
H
hedaoyuan 已提交
90 91 92
        im.data<T>(), num_outputs, input_height, input_width, filter_height,
        filter_width, stride_height, stride_width, padding_height,
        padding_width, output_height, output_width, col.data<T>());
H
hedaoyuan 已提交
93 94 95 96 97
  }
};

template <class T>
__global__ void col2im(size_t n, const T* data_col, size_t height, size_t width,
H
hedaoyuan 已提交
98 99 100 101 102
                       size_t channels, size_t filter_height,
                       size_t filter_width, size_t stride_height,
                       size_t stride_width, size_t padding_height,
                       size_t padding_width, size_t output_height,
                       size_t output_width, T* data_im) {
H
hedaoyuan 已提交
103 104 105 106 107 108 109
  size_t index =
      (blockIdx.x * gridDim.y + blockIdx.y) * blockDim.x + threadIdx.x;
  if (index < n) {
    T val = 0;
    int w = int(index % width);
    int h = int((index / width) % height);
    int c = int(index / (width * height));
H
hedaoyuan 已提交
110 111 112 113
    if ((w - (int)padding_width) >= 0 &&
        (w - (int)padding_width) < (width - 2 * padding_width) &&
        (h - (int)padding_height) >= 0 &&
        (h - padding_height) < (height - 2 * padding_height)) {
H
hedaoyuan 已提交
114
      // compute the start and end of the output
H
hedaoyuan 已提交
115 116 117 118 119 120 121 122 123
      int w_col_start = (w < (int)filter_width)
                            ? 0
                            : (w - int(filter_width)) / (int)stride_width + 1;
      int w_col_end =
          min((int)(w / (int)stride_width + 1), (int)(output_width));
      int h_col_start = (h < (int)filter_height)
                            ? 0
                            : (h - (int)filter_height) / (int)stride_height + 1;
      int h_col_end = min(int(h / stride_height + 1), int(output_height));
H
hedaoyuan 已提交
124 125 126
      for (int h_col = h_col_start; h_col < h_col_end; ++h_col) {
        for (int w_col = w_col_start; w_col < w_col_end; ++w_col) {
          // the col location: [c * width * height + h_out, w_out]
H
hedaoyuan 已提交
127 128 129 130 131
          int c_col = int(c * filter_height * filter_width) +
                      (h - h_col * (int)stride_height) * (int)filter_width +
                      (w - w_col * (int)stride_width);
          val +=
              data_col[(c_col * output_height + h_col) * output_width + w_col];
H
hedaoyuan 已提交
132 133
        }
      }
H
hedaoyuan 已提交
134 135 136 137 138
      h -= padding_height;
      w -= padding_width;
      data_im[c * ((width - 2 * padding_width) *
                   (height - 2 * padding_height)) +
              h * (width - 2 * padding_width) + w] += val;
H
hedaoyuan 已提交
139 140 141 142 143
    }
  }
}

/*
H
hedaoyuan 已提交
144 145 146
 * im = [input_channels, input_height, input_width]
 * col =
 *   [input_channels, filter_height, filter_width, output_height, output_width]
H
hedaoyuan 已提交
147 148
 */
template <class T>
H
hedaoyuan 已提交
149 150
class Col2ImFunctor<paddle::operators::math::ColFormat::kCFO,
                    platform::GPUPlace, T> {
H
hedaoyuan 已提交
151
 public:
H
hedaoyuan 已提交
152 153
  void operator()(framework::Tensor& im, const framework::Tensor& col,
                  int stride_height, int stride_width, int padding_height,
154
                  int padding_width, platform::DeviceContext* context) {
H
hedaoyuan 已提交
155 156 157 158 159 160 161 162 163 164
    PADDLE_ENFORCE(im.dims().size() == 3);
    PADDLE_ENFORCE(col.dims().size() == 5);

    int input_channels = im.dims()[0];
    int input_height = im.dims()[1];
    int input_width = im.dims()[2];
    int filter_height = col.dims()[1];
    int filter_width = col.dims()[2];
    int output_height = col.dims()[3];
    int output_width = col.dims()[4];
H
hedaoyuan 已提交
165

H
hedaoyuan 已提交
166 167
    size_t num_kernels = input_channels * (input_height + 2 * padding_height) *
                         (input_width + 2 * padding_width);
H
hedaoyuan 已提交
168

H
hedaoyuan 已提交
169 170 171
    size_t blocks = (num_kernels + 1024 - 1) / 1024;
    size_t block_x = 512;
    size_t block_y = (blocks + 512 - 1) / 512;
H
hedaoyuan 已提交
172
    dim3 threads(1024, 1);
H
hedaoyuan 已提交
173
    dim3 grid(block_x, block_y);
H
hedaoyuan 已提交
174 175 176

    // To avoid involving atomic operations, we will launch one kernel per
    // bottom dimension, and then in the kernel add up the top dimensions.
H
hedaoyuan 已提交
177 178 179
    col2im<T><<<
        grid, threads, 0,
        reinterpret_cast<platform::CUDADeviceContext*>(context)->stream()>>>(
H
hedaoyuan 已提交
180 181 182 183
        num_kernels, col.data<T>(), input_height + 2 * padding_height,
        input_width + 2 * padding_width, input_channels, filter_height,
        filter_width, stride_height, stride_width, padding_height,
        padding_width, output_height, output_width, im.data<T>());
H
hedaoyuan 已提交
184 185 186
  }
};

H
hedaoyuan 已提交
187 188 189 190 191 192 193 194
template class Im2ColFunctor<paddle::operators::math::ColFormat::kCFO,
                             platform::GPUPlace, float>;
template class Im2ColFunctor<paddle::operators::math::ColFormat::kCFO,
                             platform::GPUPlace, double>;
template class Col2ImFunctor<paddle::operators::math::ColFormat::kCFO,
                             platform::GPUPlace, float>;
template class Col2ImFunctor<paddle::operators::math::ColFormat::kCFO,
                             platform::GPUPlace, double>;
H
hedaoyuan 已提交
195 196

template <class T>
H
hedaoyuan 已提交
197 198 199 200 201 202 203 204 205 206 207 208 209 210 211
__global__ void im2colOCF(const T* im_data, T* col_data, int input_channels,
                          int input_height, int input_width, int filter_height,
                          int filter_width, int stride_height, int stride_width,
                          int padding_height, int padding_width,
                          int output_height, int output_width) {
  int swid = blockIdx.x;
  int shid = blockIdx.y;
  for (int channelid = threadIdx.z; channelid < input_channels;
       channelid += blockDim.z) {
    for (int idy = threadIdx.y; idy < filter_height; idy += blockDim.y) {
      for (int idx = threadIdx.x; idx < filter_width; idx += blockDim.x) {
        int width_offset = idx + swid * stride_width - padding_width;
        int height_offset = idy + shid * stride_height - padding_height;
        int im_offset = width_offset + height_offset * input_width +
                        channelid * input_height * input_width;
H
hedaoyuan 已提交
212

H
hedaoyuan 已提交
213 214 215 216
        int col_offset = idx + idy * filter_width +
                         channelid * filter_height * filter_width +
                         (shid * output_width + swid) *
                             (input_channels * filter_height * filter_width);
H
hedaoyuan 已提交
217

H
hedaoyuan 已提交
218 219 220
        if (height_offset >= input_height || height_offset < 0 ||
            width_offset >= input_width || width_offset < 0) {
          col_data[col_offset] = T(0);
H
hedaoyuan 已提交
221
        } else {
H
hedaoyuan 已提交
222
          col_data[col_offset] = im_data[im_offset];
H
hedaoyuan 已提交
223 224 225 226 227 228 229
        }
      }
    }
  }
}

/*
H
hedaoyuan 已提交
230 231 232
 * im = [input_channels, input_height, input_width]
 * col =
 *   [output_height, output_width, input_channels, filter_height, filter_width]
H
hedaoyuan 已提交
233 234
 */
template <class T>
H
hedaoyuan 已提交
235 236
class Im2ColFunctor<paddle::operators::math::ColFormat::kOCF,
                    platform::GPUPlace, T> {
H
hedaoyuan 已提交
237
 public:
H
hedaoyuan 已提交
238 239
  void operator()(const framework::Tensor& im, framework::Tensor& col,
                  int stride_height, int stride_width, int padding_height,
240
                  int padding_width, platform::DeviceContext* context) {
H
hedaoyuan 已提交
241 242 243 244 245 246 247 248 249
    PADDLE_ENFORCE(im.dims().size() == 3);
    PADDLE_ENFORCE(col.dims().size() == 5);
    int input_channels = im.dims()[0];
    int input_height = im.dims()[1];
    int input_width = im.dims()[2];
    int filter_height = col.dims()[3];
    int filter_width = col.dims()[4];
    int output_height = col.dims()[0];
    int output_width = col.dims()[1];
H
hedaoyuan 已提交
250

H
hedaoyuan 已提交
251 252 253 254 255 256 257 258 259 260 261
    int block_dim_x = 0;
    int block_dim_y = 0;
    if (filter_height <= 4 && filter_width <= 4) {
      block_dim_x = 4;
      block_dim_y = 4;
    } else if (filter_height <= 8 && filter_width <= 8) {
      block_dim_x = 8;
      block_dim_y = 8;
    } else if (filter_height <= 16 && filter_width <= 16) {
      block_dim_x = 16;
      block_dim_y = 16;
H
hedaoyuan 已提交
262
    } else {
H
hedaoyuan 已提交
263 264
      block_dim_x = 32;
      block_dim_y = 32;
H
hedaoyuan 已提交
265 266
    }

H
hedaoyuan 已提交
267 268 269 270
    int block_dim_z = 1024 / block_dim_x / block_dim_y;
    dim3 threads(block_dim_x, block_dim_y,
                 std::min(block_dim_z, input_channels));
    dim3 grid(output_width, output_height);
H
hedaoyuan 已提交
271 272 273
    im2colOCF<T><<<
        grid, threads, 0,
        reinterpret_cast<platform::CUDADeviceContext*>(context)->stream()>>>(
H
hedaoyuan 已提交
274 275 276
        im.data<T>(), col.data<T>(), input_channels, input_height, input_width,
        filter_height, filter_width, stride_height, stride_width,
        padding_height, padding_width, output_height, output_width);
H
hedaoyuan 已提交
277 278 279 280
  }
};

template <class T>
H
hedaoyuan 已提交
281 282 283 284 285 286 287 288 289 290 291 292 293 294 295
__global__ void col2imOCF(T* im_data, const T* col_data, int input_channels,
                          int input_height, int input_width, int filter_height,
                          int filter_width, int stride_height, int stride_width,
                          int padding_height, int padding_width,
                          int output_height, int output_width) {
  int swid = blockIdx.x;
  int shid = blockIdx.y;
  for (int channelid = threadIdx.z; channelid < input_channels;
       channelid += blockDim.z) {
    for (int idy = threadIdx.y; idy < filter_height; idy += blockDim.y) {
      for (int idx = threadIdx.x; idx < filter_width; idx += blockDim.x) {
        int width_offset = idx + swid * stride_width - padding_width;
        int height_offset = idy + shid * stride_height - padding_height;
        int im_offset = width_offset + height_offset * input_width +
                        channelid * input_height * input_width;
H
hedaoyuan 已提交
296

H
hedaoyuan 已提交
297 298 299 300
        int col_offset = idx + idy * filter_width +
                         channelid * filter_height * filter_width +
                         (shid * output_width + swid) *
                             (input_channels * filter_height * filter_width);
H
hedaoyuan 已提交
301

H
hedaoyuan 已提交
302 303 304 305
        if (height_offset >= 0 && height_offset < input_height &&
            width_offset >= 0 && width_offset < input_width) {
          paddle::platform::CudaAtomicAdd(im_data + im_offset,
                                          col_data[col_offset]);
H
hedaoyuan 已提交
306 307 308 309 310 311 312
        }
      }
    }
  }
}

/*
H
hedaoyuan 已提交
313 314 315
 * im = [input_channels, input_height, input_width]
 * col =
 *   [output_height, output_width, input_channels, filter_height, filter_width]
H
hedaoyuan 已提交
316 317
 */
template <class T>
H
hedaoyuan 已提交
318 319
class Col2ImFunctor<paddle::operators::math::ColFormat::kOCF,
                    platform::GPUPlace, T> {
H
hedaoyuan 已提交
320
 public:
H
hedaoyuan 已提交
321 322
  void operator()(framework::Tensor& im, const framework::Tensor& col,
                  int stride_height, int stride_width, int padding_height,
323
                  int padding_width, platform::DeviceContext* context) {
H
hedaoyuan 已提交
324 325 326 327 328 329 330 331 332
    PADDLE_ENFORCE(im.dims().size() == 3);
    PADDLE_ENFORCE(col.dims().size() == 5);
    int input_channels = im.dims()[0];
    int input_height = im.dims()[1];
    int input_width = im.dims()[2];
    int filter_height = col.dims()[3];
    int filter_width = col.dims()[4];
    int output_height = col.dims()[0];
    int output_width = col.dims()[1];
H
hedaoyuan 已提交
333

H
hedaoyuan 已提交
334 335 336 337 338 339 340 341 342 343 344
    int block_dim_x = 0;
    int block_dim_y = 0;
    if (filter_height <= 4 && filter_width <= 4) {
      block_dim_x = 4;
      block_dim_y = 4;
    } else if (filter_height <= 8 && filter_width <= 8) {
      block_dim_x = 8;
      block_dim_y = 8;
    } else if (filter_height <= 16 && filter_width <= 16) {
      block_dim_x = 16;
      block_dim_y = 16;
H
hedaoyuan 已提交
345
    } else {
H
hedaoyuan 已提交
346 347
      block_dim_x = 32;
      block_dim_y = 32;
H
hedaoyuan 已提交
348 349
    }

H
hedaoyuan 已提交
350 351 352 353
    int block_dim_z = 1024 / block_dim_x / block_dim_y;
    dim3 threads(block_dim_x, block_dim_y,
                 std::min(block_dim_z, input_channels));
    dim3 grid(output_width, output_height);
H
hedaoyuan 已提交
354 355 356
    col2imOCF<T><<<
        grid, threads, 0,
        reinterpret_cast<platform::CUDADeviceContext*>(context)->stream()>>>(
H
hedaoyuan 已提交
357 358 359
        im.data<T>(), col.data<T>(), input_channels, input_height, input_width,
        filter_height, filter_width, stride_height, stride_width,
        padding_height, padding_width, output_height, output_width);
H
hedaoyuan 已提交
360 361 362
  }
};

H
hedaoyuan 已提交
363 364 365 366 367 368 369 370
template class Im2ColFunctor<paddle::operators::math::ColFormat::kOCF,
                             platform::GPUPlace, float>;
template class Im2ColFunctor<paddle::operators::math::ColFormat::kOCF,
                             platform::GPUPlace, double>;
template class Col2ImFunctor<paddle::operators::math::ColFormat::kOCF,
                             platform::GPUPlace, float>;
template class Col2ImFunctor<paddle::operators::math::ColFormat::kOCF,
                             platform::GPUPlace, double>;
H
hedaoyuan 已提交
371

372
}  // namespace math
373
}  // namespace operators
H
hedaoyuan 已提交
374
}  // namespace paddle