im2col.cu 14.0 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 math {
H
hedaoyuan 已提交
20 21

template <class T>
H
hedaoyuan 已提交
22 23 24 25
__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
26
  int index = (blockIdx.x * gridDim.y + blockIdx.y) * blockDim.x + threadIdx.x;
H
hedaoyuan 已提交
27 28 29 30 31 32 33 34
  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
35

H
hedaoyuan 已提交
36 37 38
    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
39 40
        int rIdx = int(h_in + i);
        int cIdx = int(w_in + j);
H
hedaoyuan 已提交
41 42 43 44
        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
45 46
          *data_col = 0;
        } else {
H
hedaoyuan 已提交
47 48
          rIdx = rIdx + channel_in * height - padding_height;
          cIdx = cIdx - padding_width;
H
hedaoyuan 已提交
49 50
          *data_col = data_im[rIdx * width + cIdx];
        }
H
hedaoyuan 已提交
51
        data_col += output_height * output_width;
H
hedaoyuan 已提交
52 53 54 55 56 57
      }
    }
  }
}

/*
H
hedaoyuan 已提交
58 59 60
 * im = [input_channels, input_height, input_width]
 * col =
 *   [input_channels, filter_height, filter_width, output_height, output_width]
H
hedaoyuan 已提交
61 62
 */
template <class T>
H
hedaoyuan 已提交
63
class Im2ColFunctor<kCFO, platform::GPUPlace, T> {
H
hedaoyuan 已提交
64
 public:
H
hedaoyuan 已提交
65 66
  void operator()(const framework::Tensor& im, framework::Tensor& col,
                  int stride_height, int stride_width, int padding_height,
67
                  int padding_width, platform::DeviceContext* context) {
H
hedaoyuan 已提交
68 69
    PADDLE_ENFORCE(im.dims().size() == 3);
    PADDLE_ENFORCE(col.dims().size() == 5);
H
hedaoyuan 已提交
70

H
hedaoyuan 已提交
71 72 73 74 75 76 77 78 79 80 81 82
    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
83
    dim3 threads(1024, 1);
H
hedaoyuan 已提交
84
    dim3 grid(block_x, block_y);
85
    // TODO(hedaoyuan): launch kernel on specified stream
H
hedaoyuan 已提交
86 87 88 89
    im2col<T><<<grid, threads>>>(
        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
90 91 92 93 94
  }
};

template <class T>
__global__ void col2im(size_t n, const T* data_col, size_t height, size_t width,
H
hedaoyuan 已提交
95 96 97 98 99
                       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
100 101 102 103 104 105 106
  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
107 108 109 110
    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
111
      // compute the start and end of the output
H
hedaoyuan 已提交
112 113 114 115 116 117 118 119 120
      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
121 122 123
      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
124 125 126 127 128
          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
129 130
        }
      }
H
hedaoyuan 已提交
131 132 133 134 135
      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
136 137 138 139 140
    }
  }
}

/*
H
hedaoyuan 已提交
141 142 143
 * im = [input_channels, input_height, input_width]
 * col =
 *   [input_channels, filter_height, filter_width, output_height, output_width]
H
hedaoyuan 已提交
144 145
 */
template <class T>
H
hedaoyuan 已提交
146
class Col2ImFunctor<kCFO, platform::GPUPlace, T> {
H
hedaoyuan 已提交
147
 public:
H
hedaoyuan 已提交
148 149
  void operator()(framework::Tensor& im, const framework::Tensor& col,
                  int stride_height, int stride_width, int padding_height,
150
                  int padding_width, platform::DeviceContext* context) {
H
hedaoyuan 已提交
151 152 153 154 155 156 157 158 159 160
    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
161

H
hedaoyuan 已提交
162 163
    size_t num_kernels = input_channels * (input_height + 2 * padding_height) *
                         (input_width + 2 * padding_width);
H
hedaoyuan 已提交
164

H
hedaoyuan 已提交
165 166 167
    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
168
    dim3 threads(1024, 1);
H
hedaoyuan 已提交
169
    dim3 grid(block_x, block_y);
H
hedaoyuan 已提交
170 171 172

    // To avoid involving atomic operations, we will launch one kernel per
    // bottom dimension, and then in the kernel add up the top dimensions.
173
    // TODO(hedaoyuan): launch kernel on specified stream
H
hedaoyuan 已提交
174 175 176 177 178
    col2im<T><<<grid, threads>>>(
        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
179 180 181
  }
};

H
hedaoyuan 已提交
182 183 184 185
template class Im2ColFunctor<kCFO, platform::GPUPlace, float>;
template class Im2ColFunctor<kCFO, platform::GPUPlace, double>;
template class Col2ImFunctor<kCFO, platform::GPUPlace, float>;
template class Col2ImFunctor<kCFO, platform::GPUPlace, double>;
H
hedaoyuan 已提交
186 187

template <class T>
H
hedaoyuan 已提交
188 189 190 191 192 193 194 195 196 197 198 199 200 201 202
__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
203

H
hedaoyuan 已提交
204 205 206 207
        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
208

H
hedaoyuan 已提交
209 210 211
        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
212
        } else {
H
hedaoyuan 已提交
213
          col_data[col_offset] = im_data[im_offset];
H
hedaoyuan 已提交
214 215 216 217 218 219 220
        }
      }
    }
  }
}

/*
H
hedaoyuan 已提交
221 222 223
 * im = [input_channels, input_height, input_width]
 * col =
 *   [output_height, output_width, input_channels, filter_height, filter_width]
H
hedaoyuan 已提交
224 225
 */
template <class T>
H
hedaoyuan 已提交
226
class Im2ColFunctor<kOCF, platform::GPUPlace, T> {
H
hedaoyuan 已提交
227
 public:
H
hedaoyuan 已提交
228 229
  void operator()(const framework::Tensor& im, framework::Tensor& col,
                  int stride_height, int stride_width, int padding_height,
230
                  int padding_width, platform::DeviceContext* context) {
H
hedaoyuan 已提交
231 232 233 234 235 236 237 238 239
    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
240

H
hedaoyuan 已提交
241 242 243 244 245 246 247 248 249 250 251
    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
252
    } else {
H
hedaoyuan 已提交
253 254
      block_dim_x = 32;
      block_dim_y = 32;
H
hedaoyuan 已提交
255 256
    }

H
hedaoyuan 已提交
257 258 259 260
    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);
261
    // TODO(hedaoyuan): launch kernel on specified stream
H
hedaoyuan 已提交
262 263 264 265
    im2colOCF<T><<<grid, threads>>>(
        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
266 267 268 269
  }
};

template <class T>
H
hedaoyuan 已提交
270 271 272 273 274 275 276 277 278 279 280 281 282 283 284
__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
285

H
hedaoyuan 已提交
286 287 288 289
        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
290

H
hedaoyuan 已提交
291 292 293 294
        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
295 296 297 298 299 300 301
        }
      }
    }
  }
}

/*
H
hedaoyuan 已提交
302 303 304
 * im = [input_channels, input_height, input_width]
 * col =
 *   [output_height, output_width, input_channels, filter_height, filter_width]
H
hedaoyuan 已提交
305 306
 */
template <class T>
H
hedaoyuan 已提交
307
class Col2ImFunctor<kOCF, platform::GPUPlace, T> {
H
hedaoyuan 已提交
308
 public:
H
hedaoyuan 已提交
309 310
  void operator()(framework::Tensor& im, const framework::Tensor& col,
                  int stride_height, int stride_width, int padding_height,
311
                  int padding_width, platform::DeviceContext* context) {
H
hedaoyuan 已提交
312 313 314 315 316 317 318 319 320
    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
321

H
hedaoyuan 已提交
322 323 324 325 326 327 328 329 330 331 332
    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
333
    } else {
H
hedaoyuan 已提交
334 335
      block_dim_x = 32;
      block_dim_y = 32;
H
hedaoyuan 已提交
336 337
    }

H
hedaoyuan 已提交
338 339 340 341
    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);
342 343
    // TODO(hedaoyuan): launch kernel on specified stream
    col2imOCF<T><<<grid, threads>>>(
H
hedaoyuan 已提交
344 345 346
        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
347 348 349
  }
};

H
hedaoyuan 已提交
350 351 352 353
template class Im2ColFunctor<kOCF, platform::GPUPlace, float>;
template class Im2ColFunctor<kOCF, platform::GPUPlace, double>;
template class Col2ImFunctor<kOCF, platform::GPUPlace, float>;
template class Col2ImFunctor<kOCF, platform::GPUPlace, double>;
H
hedaoyuan 已提交
354

355
}  // namespace math
H
hedaoyuan 已提交
356
}  // namespace paddle