“24ea602e183ca20a7577ebe253323d0e5d0f9847”上不存在“drivers/net/ethernet/broadcom/b44.c”
math_function_int8.cpp 2.6 KB
Newer Older
Z
Zhen Wang 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22
/* Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.

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. */

#include <cstring>
#include <string>
#include "operators/math/gemm.h"
#include "operators/math/math_function.h"

namespace paddle_mobile {
namespace operators {
namespace math {
H
hjchen2 已提交
23 24 25 26 27
template <>
void matmul<int8_t>(const framework::Tensor &matrix_a, bool trans_a,
                    const framework::Tensor &matrix_b, bool trans_b,
                    int8_t alpha, framework::Tensor *matrix_out, int8_t beta,
                    bool relu, int8_t *bias) {
Z
Zhen Wang 已提交
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
  auto dim_a = matrix_a.dims();
  auto dim_b = matrix_b.dims();
  auto dim_out = matrix_out->dims();
  PADDLE_MOBILE_ENFORCE(
      dim_a.size() == 2 && dim_b.size() == 2 && dim_out.size() == 2,
      "The input and output of matmul be matrix");

  int32_t M = dim_out[0];
  int32_t N = dim_out[1];
  int32_t K = (!trans_a) ? dim_a[1] : dim_a[0];
  Gemm gemm;

  if (trans_a) {
    int32_t numel = matrix_a.numel();
    int32_t m = matrix_a.dims()[0];
    int32_t n = matrix_a.dims()[1];
    int8_t *tmp = (int8_t *)(matrix_a.data<int8_t>());  // NOLINT
    int8_t *a = static_cast<int8_t *>(
        paddle_mobile::memory::Alloc(sizeof(int8_t) * numel));
    int32_t index = 0;
    for (int32_t j = 0; j < n; j++) {
      for (int32_t i = 0; i < m; i++) {
        a[index++] = tmp[i * n + j];
      }
    }

54
#ifdef _OPENMP
H
hjchen2 已提交
55 56
    gemm.Sgemm_omp(M, N, K, alpha, a, K, matrix_b.data<int8_t>(), N, beta,
                   matrix_out->data<int32_t>(), N, relu, bias);
57
#else
H
hjchen2 已提交
58 59
    gemm.Sgemm(M, N, K, alpha, a, K, matrix_b.data<int8_t>(), N, beta,
               matrix_out->data<int32_t>(), N, relu, bias);
60
#endif
Z
Zhen Wang 已提交
61
  } else {
62
#ifdef _OPENMP
H
hjchen2 已提交
63 64 65
    gemm.Sgemm_omp(M, N, K, alpha, matrix_a.data<int8_t>(), K,
                   matrix_b.data<int8_t>(), N, beta,
                   matrix_out->data<int32_t>(), N, relu, bias);
66
#else
H
hjchen2 已提交
67 68 69
    gemm.Sgemm(M, N, K, alpha, matrix_a.data<int8_t>(), K,
               matrix_b.data<int8_t>(), N, beta, matrix_out->data<int32_t>(), N,
               relu, bias);
70
#endif
Z
Zhen Wang 已提交
71 72 73 74 75
  }
}
}  // namespace math
}  // namespace operators
}  // namespace paddle_mobile
反馈
建议
客服 返回
顶部