math_function.h 3.1 KB
Newer Older
Q
qijun 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
/* 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. */

#pragma once
Q
qijun 已提交
16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53
#ifdef PADDLE_USE_MKLML
#include <mkl_cblas.h>
#include <mkl_lapacke.h>
#include <mkl_vml_functions.h>
#endif

#ifdef PADDLE_USE_MKL
#include <mkl.h>
#include <mkl_lapacke.h>
#endif

#ifdef PADDLE_USE_ATLAS
extern "C" {
#include <cblas.h>
#include <clapack.h>
}
#endif

#ifdef PADDLE_USE_OPENBLAS
#include <cblas.h>
#include <lapacke.h>
#endif

#ifndef LAPACK_FOUND
extern "C" {
#include <cblas.h>
int LAPACKE_sgetrf(int matrix_layout, int m, int n, float* a, int lda,
                   int* ipiv);
int LAPACKE_dgetrf(int matrix_layout, int m, int n, double* a, int lda,
                   int* ipiv);
int LAPACKE_sgetri(int matrix_layout, int n, float* a, int lda,
                   const int* ipiv);
int LAPACKE_dgetri(int matrix_layout, int n, double* a, int lda,
                   const int* ipiv);
}
#endif

#include <cmath>
Q
qijun 已提交
54

55
#include "paddle/framework/eigen.h"
Q
qijun 已提交
56
#include "paddle/framework/tensor.h"
Q
qijun 已提交
57
#include "paddle/platform/device_context.h"
Q
qijun 已提交
58
#include "paddle/platform/enforce.h"
Q
qijun 已提交
59 60 61 62 63

namespace paddle {
namespace operators {
namespace math {

Q
qijun 已提交
64 65 66 67 68
// Support continuous memory now
// If transA = N, and transB = N
// Then matrixA: M * K, matrixB: K * N matrixC : M * N
// For more detailed info, please refer to
// http://www.netlib.org/lapack/explore-html/d4/de2/sgemm_8f.html
Q
qijun 已提交
69
template <typename Place, typename T>
70 71 72
void gemm(const platform::DeviceContext& context, const CBLAS_TRANSPOSE transA,
          const CBLAS_TRANSPOSE transB, const int M, const int N, const int K,
          const T alpha, const T* A, const T* B, const T beta, T* C);
Q
qijun 已提交
73

G
guosheng 已提交
74 75 76 77 78 79 80
// gemm wrapper with stride args for matrix uncontinuous in memory
template <typename Place, typename T>
void gemm(const platform::DeviceContext& context, const bool transA,
          const bool transB, const int M, const int N, const int K,
          const T alpha, const T* A, const int lda, const T* B, const int ldb,
          const T beta, T* C, const int ldc);

Q
qijun 已提交
81
// matrix multiply with continuous memory
Q
qijun 已提交
82
template <typename Place, typename T>
83 84
void matmul(const platform::DeviceContext& context,
            const framework::Tensor& matrix_a, bool trans_a,
Q
qijun 已提交
85
            const framework::Tensor& matrix_b, bool trans_b, T alpha,
86
            framework::Tensor* matrix_out, T beta);
Q
qijun 已提交
87

88 89 90 91 92 93 94
template <typename Place, typename T>
void SetConstant(const platform::DeviceContext& context,
                 framework::Tensor* tensor, T num) {
  auto t = framework::EigenVector<T>::Flatten(*tensor);
  t.device(*context.GetEigenDevice<Place>()) = t.constant(static_cast<T>(num));
}

Q
qijun 已提交
95 96 97
}  // namespace math
}  // namespace operators
}  // namespace paddle