// Copyright (c) 2022 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. #pragma once #include #include "paddle/phi/common/amp_type_traits.h" #include "paddle/phi/core/dense_tensor.h" #include "paddle/phi/kernels/funcs/for_range.h" namespace phi { template struct DigammaGradFunctor { DigammaGradFunctor(const T* dout, const T* x, T* output, int64_t numel) : dout_(dout), x_(x), output_(output), numel_(numel) {} HOSTDEVICE void operator()(int64_t idx) const { using MPType = typename MPTypeTrait::Type; const MPType mp_dout = static_cast(dout_[idx]); const MPType mp_x = static_cast(x_[idx]); output_[idx] = static_cast(mp_dout * Eigen::numext::polygamma(MPType(1), mp_x)); } private: const T* dout_; const T* x_; T* output_; int64_t numel_; }; template void DigammaGradKernel(const Context& dev_ctx, const DenseTensor& x, const DenseTensor& out_grad, DenseTensor* x_grad) { dev_ctx.template Alloc(x_grad); if (x_grad && x_grad->numel() == 0) { return; } auto* dout_data = out_grad.data(); auto* x_data = x.data(); auto* dx_data = x_grad->data(); auto numel = out_grad.numel(); funcs::ForRange for_range(dev_ctx, numel); DigammaGradFunctor functor(dout_data, x_data, dx_data, numel); for_range(functor); } } // namespace phi