MKLDNNMatrix.h 5.8 KB
Newer Older
T
tensor-tang 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
/* Copyright (c) 2017 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. */

#pragma once

T
tensor-tang 已提交
17 18
#include <vector>
#include "Matrix.h"
T
tensor-tang 已提交
19 20 21 22 23 24 25 26 27 28 29 30
#include "mkldnn.hpp"
#include "paddle/parameter/Parameter.h"

namespace paddle {

class MKLDNNMatrix;
typedef std::shared_ptr<MKLDNNMatrix> MKLDNNMatrixPtr;

/**
 * @brief MKLDNN Matrix.
 *
 */
T
tensor-tang 已提交
31
class MKLDNNMatrix : public CpuMatrix, public mkldnn::memory {
T
tensor-tang 已提交
32
public:
33 34 35 36
  MKLDNNMatrix(CpuMatrixPtr m, mkldnn::memory::primitive_desc pd)
      : CpuMatrix(m->getData(), m->getHeight(), m->getWidth(), false),
        mkldnn::memory(pd, m->getData()),
        m_(m) {}
T
tensor-tang 已提交
37

T
tensor-tang 已提交
38 39
  ~MKLDNNMatrix() {}

40 41 42
  /**
   * Create MKLDNNMatrix from a MatrixPtr and memory primitive_desc
   */
43 44
  static MKLDNNMatrixPtr create(mkldnn::memory::primitive_desc pd,
                                MatrixPtr m = nullptr);
45 46 47 48

  /**
   * Create MKLDNNMatrix from a MatrixPtr and memory details info
   */
T
tensor-tang 已提交
49 50 51 52
  static MKLDNNMatrixPtr create(
      mkldnn::memory::dims dims,
      mkldnn::memory::format fmt,
      mkldnn::engine& eg,
53
      MatrixPtr m = nullptr,
T
tensor-tang 已提交
54 55
      mkldnn::memory::data_type dtype = mkldnn::memory::data_type::f32);

56 57 58 59 60 61 62 63 64 65 66 67
  /**
   * Create primitive descriptor.
   * default with f32 dtype
   */
  static mkldnn::memory::primitive_desc createPrimitiveDesc(
      const mkldnn::memory::dims dims,
      const mkldnn::memory::format& fmt,
      const mkldnn::engine& eg,
      const mkldnn::memory::data_type& dtype = mkldnn::memory::data_type::f32) {
    return mkldnn::memory::primitive_desc(memory::desc(dims, dtype, fmt), eg);
  }

68 69 70 71 72
  /**
   * Create Memory descriptor.
   * default with any format and f32 dtype
   */
  static mkldnn::memory::desc createMemoryDesc(
73
      const mkldnn::memory::dims dims,
74 75 76 77 78 79 80
      const mkldnn::memory::format& fmt = mkldnn::memory::format::any,
      const mkldnn::memory::data_type& dtype = mkldnn::memory::data_type::f32) {
    return mkldnn::memory::desc(dims, dtype, fmt);
  }

  /**
   * Create reorder primitive.
81
   * Create a mkldnn::reorder handle for converting src MKLDNNMatrix to dst.
82 83 84 85 86 87
   * checkData: whether to check the data handle of src and dst.
   *            if true, it will check the data and do not allow them equal;
   *            otherwise, it will not check them, then the reorder created
   *            may have inplace buffer.
   *            Do not set false, if you can not guarantee the inplace logical
   *            would work with your reorder.
88 89 90 91 92 93
   */
  static std::shared_ptr<mkldnn::reorder> createReorder(
      const MKLDNNMatrixPtr& src,
      const MKLDNNMatrixPtr& dst,
      bool checkData = true);

T
tensor-tang 已提交
94 95 96 97 98
  void copyFrom(const Matrix& src) {
    // TODO(TJ): reorder data if this format is not nchw or x
    m_->copyFrom(src);
  }

99
public:
T
tensor-tang 已提交
100 101
  /**
   * Reorder this MKLDNNMatrix from other format.
T
refine  
tensor-tang 已提交
102 103 104
   * Support inplace reorder.
   * @note: this function would only reorder the data layout.
   *        will NOT change this original dim or format info
T
tensor-tang 已提交
105 106 107 108 109 110 111
   */
  void reorderDataFrom(const MKLDNNMatrixPtr& m,
                       memory::format srcFmt,
                       memory::dims targetDim);

  /**
   * Reorder this MKLDNNMatrix to other format.
T
refine  
tensor-tang 已提交
112 113 114
   * Support inplace reorder.
   * @note: this function would only reorder the data layout.
   *        will NOT change the dst dim or format info
T
tensor-tang 已提交
115 116 117 118 119
   */
  void reorderDataTo(const MKLDNNMatrixPtr& m,
                     memory::format dstFmt,
                     memory::dims targetDim);

120 121 122 123 124 125 126
  /**
   * Dimensionality reduction.
   * Change format "nchw --> nc" or "oihw --> oi" if the h and w are both 1
   */
  void downSpatial();

  /**
127
   * set the memory data handle.
128 129 130
   * Caution: This will not check the buffer size of the data,
   *          it should be coverd by user.
   */
131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149
  void setData(real* data) {
    set_data_handle(data);
    CpuMatrix::setData(data);
    m_.reset();
  }

  /**
   * override Matrix::getData
   * check data before return
   */
  real* getData() override {
    CHECK_EQ((void*)data_, get_data_handle());
    return data_;
  }

  const real* getData() const override {
    CHECK_EQ((void*)data_, get_data_handle());
    return data_;
  }
150

T
tensor-tang 已提交
151
  /**
T
tensor-tang 已提交
152
   * Get primitive descriptor.
T
tensor-tang 已提交
153
   */
T
refine  
tensor-tang 已提交
154 155 156
  mkldnn::memory::primitive_desc getPrimitiveDesc() {
    return this->get_primitive_desc();
  }
T
tensor-tang 已提交
157

T
tensor-tang 已提交
158
  /**
T
tensor-tang 已提交
159
   * Get memory descriptor.
T
tensor-tang 已提交
160
   */
T
refine  
tensor-tang 已提交
161
  mkldnn::memory::desc getMemoryDesc() { return getPrimitiveDesc().desc(); }
T
tensor-tang 已提交
162 163

  /**
164
   * Get dimensions.
T
tensor-tang 已提交
165
   */
T
tensor-tang 已提交
166
  mkldnn::memory::dims getDims() {
T
refine  
tensor-tang 已提交
167
    mkldnn::memory::desc md = getMemoryDesc();
168 169
    const int* src = md.data.dims;
    int ndims = md.data.ndims;
T
tensor-tang 已提交
170 171 172 173 174 175 176
    mkldnn::memory::dims dst;
    dst.resize(ndims);
    for (int i = 0; i < ndims; ++i) {
      dst[i] = src[i];
    }
    return dst;
  }
T
tensor-tang 已提交
177

T
tensor-tang 已提交
178 179 180 181
  /**
   * Get format.
   */
  mkldnn::memory::format getFormat() {
T
refine  
tensor-tang 已提交
182
    return (mkldnn::memory::format)(getMemoryDesc().data.format);
T
tensor-tang 已提交
183 184 185
  }

  /**
186
   * Get memory data type.
T
tensor-tang 已提交
187
   */
188
  mkldnn::memory::data_type getDtype() {
T
refine  
tensor-tang 已提交
189
    return (mkldnn::memory::data_type)(getMemoryDesc().data.data_type);
190 191 192 193 194
  }

  /**
   * Get engine.
   */
T
refine  
tensor-tang 已提交
195
  mkldnn::engine getEngine() { return getPrimitiveDesc().get_engine(); }
T
tensor-tang 已提交
196 197 198

protected:
  /**
T
refine  
tensor-tang 已提交
199 200
   * Do reorder once.
   * Can support inplace.
T
tensor-tang 已提交
201 202 203 204 205 206
   */
  void reorderOnce(void* srcData,
                   void* dstData,
                   memory::format srcFmt,
                   memory::format dstFmt,
                   memory::dims dm);
207 208 209 210

private:
  // save the CpuMatrixPtr in case the buffer released outside
  CpuMatrixPtr m_;
T
tensor-tang 已提交
211 212 213
};

}  // namespace paddle