math_function.cpp 2.8 KB
Newer Older
Z
zhaojiaying01 已提交
1
/* Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
朔-望's avatar
朔-望 已提交
2 3 4 5 6 7 8 9 10 11 12 13 14

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. */

Z
zhaojiaying01 已提交
15
#include "operators/math/math_function.h"
Z
zhaojiaying01 已提交
16
#include "operators/math/gemm.h"
朔-望's avatar
朔-望 已提交
17 18

namespace paddle_mobile {
朔-望's avatar
朔-望 已提交
19 20 21
namespace operators {
namespace math {

朔-望's avatar
朔-望 已提交
22
template <>
朔-望's avatar
朔-望 已提交
23
void matmul<float>(const framework::Tensor &matrix_a, bool trans_a,
朔-望's avatar
朔-望 已提交
24
                   const framework::Tensor &matrix_b, bool trans_b, float alpha,
L
liuruilong 已提交
25
                   framework::Tensor *matrix_out, float beta, bool relu) {
26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43
  auto dim_a = matrix_a.dims();
  auto dim_b = matrix_b.dims();
  auto dim_out = matrix_out->dims();
  //  PADDLE_ENFORCE(dim_a.size() == 2 && dim_b.size() == 2 &&
  //  dim_out.size() ==
  //  2,
  //                 "The input and output of matmul be matrix");
  //
  //  PADDLE_ENFORCE(platform::is_cpu_place(matrix_a.place()) &&
  //                     platform::is_cpu_place(matrix_b.place())
  //                     &&
  //                     platform::is_cpu_place(matrix_out->place()),
  //                 "Matrix must all be in CPUPlace");

  int M = dim_out[0];
  int N = dim_out[1];
  int K = (trans_a == false) ? dim_a[1] : dim_a[0];

L
liuruilong 已提交
44 45 46 47 48 49 50 51
  if (relu) {
    sgemm_relu(M, N, K, alpha, matrix_a.data<float>(), K, matrix_b.data<float>(), N,
               beta, matrix_out->data<float>(), N);
  } else {
    sgemm(M, N, K, alpha, matrix_a.data<float>(), K, matrix_b.data<float>(), N,
          beta, matrix_out->data<float>(), N);
  }

朔-望's avatar
朔-望 已提交
52 53
}

朔-望's avatar
朔-望 已提交
54
template <>
朔-望's avatar
朔-望 已提交
55 56
void matmul<double>(const framework::Tensor &matrix_a, bool trans_a,
                    const framework::Tensor &matrix_b, bool trans_b,
L
liuruilong 已提交
57
                    double alpha, framework::Tensor *matrix_out, double beta, bool relu) {
58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74
  auto dim_a = matrix_a.dims();
  auto dim_b = matrix_b.dims();
  auto dim_out = matrix_out->dims();
  //  PADDLE_ENFORCE(dim_a.size() == 2 && dim_b.size() == 2 &&
  //  dim_out.size() ==
  //  2,
  //                 "The input and output of matmul be matrix");
  //
  //  PADDLE_ENFORCE(platform::is_cpu_place(matrix_a.place()) &&
  //                     platform::is_cpu_place(matrix_b.place())
  //                     &&
  //                     platform::is_cpu_place(matrix_out->place()),
  //                 "Matrix must all be in CPUPlace");

  int M = dim_out[0];
  int N = dim_out[1];
  int K = (trans_a == false) ? dim_a[1] : dim_a[0];
朔-望's avatar
朔-望 已提交
75 76
}

L
liuruilong 已提交
77 78


朔-望's avatar
朔-望 已提交
79 80 81
}  // namespace math
}  // namespace operators
}  // namespace paddle_mobile