| @@ -50,6 +50,14 @@ int DoSplit(float *in_data, float **out_data, const int *input_shape, int offset | |||||
| split_which = i % num_split; | split_which = i % num_split; | ||||
| split_times = i / num_split; | split_times = i / num_split; | ||||
| int split_size = split_sizes[split_which]; | int split_size = split_sizes[split_which]; | ||||
| // support split size is -1 in the end. | |||||
| if (split_size == -1) { | |||||
| int split_dim_i = input_shape[split_dim]; | |||||
| for (int j = 0; j < num_split - 1; ++j) { | |||||
| split_dim_i -= split_sizes[j]; | |||||
| } | |||||
| split_size = split_dim_i; | |||||
| } | |||||
| float *dst = out_data[split_which] + split_times * in_stride * split_size; | float *dst = out_data[split_which] + split_times * in_stride * split_size; | ||||
| (void)memcpy(dst, src, split_size * in_stride_bytes); | (void)memcpy(dst, src, split_size * in_stride_bytes); | ||||
| src += split_size * in_stride; | src += split_size * in_stride; | ||||
| @@ -88,7 +88,7 @@ int Split::InferShape(std::vector<tensor::Tensor *> inputs_, std::vector<tensor: | |||||
| if (!GetInferFlag()) { | if (!GetInferFlag()) { | ||||
| return RET_OK; | return RET_OK; | ||||
| } | } | ||||
| int split_dim = GetSplitDim(); | |||||
| size_t split_dim = GetSplitDim() == -1 ? input->shape().size() - 1 : GetSplitDim(); | |||||
| std::vector<int> input_shape = input->shape(); | std::vector<int> input_shape = input->shape(); | ||||
| std::vector<int> size_split; | std::vector<int> size_split; | ||||
| for (size_t i = 0; i < GetSizeSplits().size(); ++i) { | for (size_t i = 0; i < GetSizeSplits().size(); ++i) { | ||||
| @@ -97,7 +97,15 @@ int Split::InferShape(std::vector<tensor::Tensor *> inputs_, std::vector<tensor: | |||||
| for (int i = 0; i < number_split; ++i) { | for (int i = 0; i < number_split; ++i) { | ||||
| std::vector<int> output_shape; | std::vector<int> output_shape; | ||||
| output_shape.insert(output_shape.begin(), input_shape.begin(), input_shape.end()); | output_shape.insert(output_shape.begin(), input_shape.begin(), input_shape.end()); | ||||
| auto split_dim_i = size_split.empty() ? input_shape[split_dim] / number_split : size_split[i]; | |||||
| int split_dim_i = input_shape[split_dim]; | |||||
| // support split size is -1 in the end. | |||||
| if (i == number_split - 1 && size_split[i] == -1) { | |||||
| for (size_t j = 0; j < size_split.size() - 1; ++j) { | |||||
| split_dim_i -= size_split[j]; | |||||
| } | |||||
| } else { | |||||
| split_dim_i = size_split.empty() ? input_shape[split_dim] / number_split : size_split[i]; | |||||
| } | |||||
| output_shape[split_dim] = split_dim_i; | output_shape[split_dim] = split_dim_i; | ||||
| outputs_[i]->set_shape(output_shape); | outputs_[i]->set_shape(output_shape); | ||||
| outputs_[i]->set_data_type(input->data_type()); | outputs_[i]->set_data_type(input->data_type()); | ||||
| @@ -29,4 +29,6 @@ add_library(caffe_parser_mid OBJECT | |||||
| ${CMAKE_CURRENT_SOURCE_DIR}/caffe_permute_parser.cc | ${CMAKE_CURRENT_SOURCE_DIR}/caffe_permute_parser.cc | ||||
| ${CMAKE_CURRENT_SOURCE_DIR}/caffe_tile_parser.cc | ${CMAKE_CURRENT_SOURCE_DIR}/caffe_tile_parser.cc | ||||
| ${CMAKE_CURRENT_SOURCE_DIR}/caffe_tanh_parser.cc | ${CMAKE_CURRENT_SOURCE_DIR}/caffe_tanh_parser.cc | ||||
| ${CMAKE_CURRENT_SOURCE_DIR}/caffe_exp_parser.cc) | |||||
| ${CMAKE_CURRENT_SOURCE_DIR}/caffe_exp_parser.cc | |||||
| ${CMAKE_CURRENT_SOURCE_DIR}/caffe_slice_parser.cc | |||||
| ) | |||||
| @@ -0,0 +1,72 @@ | |||||
| /** | |||||
| * Copyright 2020 Huawei Technologies Co., Ltd | |||||
| * | |||||
| * 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 "tools/converter/parser/caffe/caffe_slice_parser.h" | |||||
| #include <memory> | |||||
| namespace mindspore { | |||||
| namespace lite { | |||||
| STATUS CaffeSliceParser::Parse(const caffe::LayerParameter &proto, const caffe::LayerParameter &weight, | |||||
| schema::CNodeT *op, std::vector<schema::TensorT *> *weightVec) { | |||||
| MS_LOG(DEBUG) << "parse CaffeSliceParser"; | |||||
| if (op == nullptr) { | |||||
| MS_LOG(ERROR) << "op is null"; | |||||
| return RET_NULL_PTR; | |||||
| } | |||||
| op->primitive = std::make_unique<schema::PrimitiveT>(); | |||||
| if (op->primitive == nullptr) { | |||||
| MS_LOG(ERROR) << "op->primitive is null"; | |||||
| return RET_NULL_PTR; | |||||
| } | |||||
| std::unique_ptr<schema::SplitT> attr = std::make_unique<schema::SplitT>(); | |||||
| if (attr == nullptr) { | |||||
| MS_LOG(ERROR) << "new op failed"; | |||||
| return RET_NULL_PTR; | |||||
| } | |||||
| const caffe::SliceParameter &slice_param = proto.slice_param(); | |||||
| if (!slice_param.slice_point().empty()) { | |||||
| attr->numberSplit = slice_param.slice_point_size() + 1; | |||||
| std::vector<int32_t> size_splits; | |||||
| for (int i = 0; i < slice_param.slice_point_size(); ++i) { | |||||
| if (i == 0) { | |||||
| size_splits.push_back(slice_param.slice_point(i)); | |||||
| } else { | |||||
| size_splits.push_back(slice_param.slice_point(i) - slice_param.slice_point(i - 1)); | |||||
| } | |||||
| } | |||||
| size_splits.push_back(-1); | |||||
| attr->sizeSplits = size_splits; | |||||
| } | |||||
| // The axis along which to slice -- may be negative to index from the end (e.g., -1 for the last axis). | |||||
| if (slice_param.has_axis()) { | |||||
| attr->splitDim = slice_param.axis(); | |||||
| } else if (slice_param.has_slice_dim()) { | |||||
| attr->splitDim = slice_param.slice_dim(); | |||||
| } | |||||
| op->name = proto.name(); | |||||
| op->primitive->value.type = schema::PrimitiveType_Split; | |||||
| op->primitive->value.value = attr.release(); | |||||
| return RET_OK; | |||||
| } | |||||
| CaffeNodeRegistrar g_caffeSliceParser("Slice", new CaffeSliceParser()); | |||||
| } // namespace lite | |||||
| } // namespace mindspore | |||||
| @@ -0,0 +1,36 @@ | |||||
| /** | |||||
| * Copyright 2020 Huawei Technologies Co., Ltd | |||||
| * | |||||
| * 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. | |||||
| */ | |||||
| #ifndef MMINDSPORE_LITE_TOOLS_CONVERTER_PARSER_CAFFE_CAFFE_SLICE_PARSER_H_ | |||||
| #define MINDSPORE_LITE_TOOLS_CONVERTER_PARSER_CAFFE_CAFFE_SLICE_PARSER_H_ | |||||
| #include <vector> | |||||
| #include "mindspore/lite/tools/converter/parser/caffe/caffe_node_parser.h" | |||||
| #include "mindspore/lite/tools/converter/parser/caffe/caffe_node_parser_registry.h" | |||||
| namespace mindspore { | |||||
| namespace lite { | |||||
| class CaffeSliceParser : public CaffeNodeParser { | |||||
| public: | |||||
| CaffeSliceParser() : CaffeNodeParser("slice") {} | |||||
| STATUS Parse(const caffe::LayerParameter &proto, const caffe::LayerParameter &weight, schema::CNodeT *op, | |||||
| std::vector<schema::TensorT *> *weightVec) override; | |||||
| }; | |||||
| } // namespace lite | |||||
| } // namespace mindspore | |||||
| #endif // MINDSPORE_LITE_TOOLS_CONVERTER_PARSER_CAFFE_CAFFE_SLICE_PARSER_H_ | |||||