/* 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 #include #include namespace paddle { namespace operators { template class SoftmaxKernel : public framework::OpKernel { public: void Compute(const framework::KernelContext& context) const override { auto input = context.Input(0)->Get(); auto* output = context.Output(0)->GetMutable(); output->mutable_data(context.GetPlace()); auto logits = input.matrix(); auto softmax = output->matrix(); const int kBatchDim = 0; const int kClassDim = 1; const int batch_size = logits.dimension(kBatchDim); const int num_classes = logits.dimension(kClassDim); Eigen::DSizes along_class(kClassDim); Eigen::DSizes batch_by_one(batch_size, 1); Eigen::DSizes one_by_class(1, num_classes); auto shifted_logits = (logits - logits.maximum(along_class) .eval() .reshape(batch_by_one) .broadcast(one_by_class)); softmax.device(*(context.GetEigenDevice())) = shifted_logits.exp(); softmax.device(*(context.GetEigenDevice())) = (softmax * softmax.sum(along_class) .inverse() .eval() .reshape(batch_by_one) .broadcast(one_by_class)); } }; } // namespace operators } // namespace paddle