elementwise_ops.cc 5.0 KB
Newer Older
C
cen.li 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14
// Copyright (c) 2019 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.

C
cen.li 已提交
15
#include "lite/kernels/bm/bridges/registry.h"
C
cen.li 已提交
16 17
#include "bmcompiler_if.h"
#include "bmcompiler_if_lite.h"
C
cen.li 已提交
18 19 20 21 22

namespace paddle {
namespace lite {
namespace kernels {
namespace bm {
C
cen.li 已提交
23
namespace bridges {
C
cen.li 已提交
24

C
cen.li 已提交
25 26
node_map_type ElementwiseConverter(const std::shared_ptr<lite::OpLite> elementwise_op,
                            graph_ctx_type* graph_ctx,
C
cen.li 已提交
27
                            const node_map_type& input_nodes) {
C
cen.li 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 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136
    // output converted nodes
    node_map_type output_nodes;
    auto scope = elementwise_op->scope();
    auto op_info = elementwise_op->op_info();
    auto op_type = op_info->Type();
    
    // input
    const int input_num = 2;
    int **shape = new int *[input_num];
    int *dim = new int[input_num];
    const char **name = new const char *[input_num];
    
    auto x_var_name = op_info->Input("X").front();
    auto x = scope->FindVar(x_var_name)->GetMutable<lite::Tensor>();
    auto x_dims = x->dims();
    name[0] = static_cast<const char*>(x_var_name.c_str());
    dim[0] = x_dims.size();
    const long int* x_shape_data = const_cast<const long int*>(&x_dims.data()[0]);
    int i_x_shape_data[x_dims.size()];
    for (size_t i = 0; i < x_dims.size(); i++) {
        i_x_shape_data[i] = static_cast<int>(x_shape_data[i]);
    }
    shape[0] = i_x_shape_data;
    
    auto y_var_name = op_info->Input("Y").front();
    
    auto y = scope->FindVar(y_var_name)->GetMutable<lite::Tensor>();
    auto y_dims = y->dims();
    name[1] = static_cast<const char*>(y_var_name.c_str());
    dim[1] = y_dims.size();
    const long int* y_shape_data = const_cast<const long int*>(&y_dims.data()[0]);
    int i_y_shape_data[y_dims.size()];
    for (size_t i = 0; i < y_dims.size(); i++) {
        i_y_shape_data[i] = static_cast<int>(y_shape_data[i]);
    }
    shape[1] = i_y_shape_data;
    
    bool y_is_const = input_nodes.find(y_var_name) == input_nodes.end();
   
    // output
    auto output_var_name = op_info->Output("Out").front();
    auto output = scope->FindVar(output_var_name)->GetMutable<lite::Tensor>();
    auto output_dims = output->dims();
    const long int* output_shape_data = const_cast<const long int*>(&output_dims.data()[0]);
    int i_output_shape_data[output_dims.size()];
    for (size_t i = 0; i < output_dims.size(); i++) {
        i_output_shape_data[i] = static_cast<int>(output_shape_data[i]);
    }
    
    if (y_is_const) {
        CHECK(op_type == "elementwise_add");
    }
    
    int op_code{-1};
    float coeff[2] = {1.f, 1.f};

    if (op_type == "elementwise_mul") {
        op_code = 0;
    } else if (op_type == "elementwise_add") {
        op_code = 1;
    } else if(op_type == "elementwise_sub") {
        op_code = 1;
        coeff[1] = -1.f;
    } else {
        LOG(FATAL) << "UNSUPPORTED ELTWISE OPERATION: " << op_type;
    }
    
    if (!y_is_const) {
        add_eltwise_layer(graph_ctx->bm_compiler_handle,
                      input_num,
                      shape,
                      dim,
                      name,
                      const_cast<const int*>(i_output_shape_data),
                      output_dims.size(),
                      static_cast<const char*>(output_var_name.c_str()),
                      op_code,
                      coeff);
    } else {
        const float* y_data = const_cast<const float*>(y->mutable_data<float>());
        bm_add_const_tensor(graph_ctx->bm_compiler_handle,
                            name[0],
                            shape[0],
                            dim[0],
                            static_cast<bm_data_type_t>(0),
                            static_cast<const void*>(y_data));
                            
        
        add_binary_layer_v2(graph_ctx->bm_compiler_handle,
                          name[0],
                          shape[0],
                          dim[0],
                          0,
                          nullptr,
                          name[0],
                          shape[0],
                          dim[0],
                          0,
                          nullptr,
                          static_cast<const char*>(output_var_name.c_str()),
                          0);
    }

    delete [] shape;
    delete [] name;
    delete [] dim;
    
    output_nodes[output_var_name] = output_var_name;
    return output_nodes;
C
cen.li 已提交
137
}
C
cen.li 已提交
138

C
cen.li 已提交
139
}  // namespace bridges
C
cen.li 已提交
140 141 142 143
}  // namespace bm
}  // namespace kernels
}  // namespace lite
}  // namespace paddle
C
cen.li 已提交
144

C
cen.li 已提交
145
REGISTER_BM_BRIDGE(elementwise_add, paddle::lite::kernels::bm::bridges::ElementwiseConverter);