wehub-resource-sync
90 lines
2.9 KiB
C++
90 lines
2.9 KiB
C++
// 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.
|
|
|
|
#include "paddle/phi/kernels/gather_grad_kernel.h"
|
|
|
|
#include "paddle/phi/core/kernel_registry.h"
|
|
#include "paddle/phi/kernels/full_kernel.h"
|
|
#include "paddle/phi/kernels/funcs/eigen/common.h"
|
|
#include "paddle/phi/kernels/funcs/gather.h"
|
|
#include "paddle/phi/kernels/funcs/scatter.h"
|
|
|
|
namespace phi {
|
|
|
|
template <typename T, typename Context>
|
|
void GatherGradKernel(const Context& dev_ctx,
|
|
const DenseTensor& x UNUSED,
|
|
const DenseTensor& index,
|
|
const DenseTensor& out_grad,
|
|
const Scalar& axis,
|
|
DenseTensor* x_grad) {
|
|
if (out_grad.numel() == 0) {
|
|
if (x_grad) {
|
|
Full<T, Context>(dev_ctx, x_grad->dims(), 0, x_grad);
|
|
}
|
|
return;
|
|
}
|
|
const auto& index_type = index.dtype();
|
|
auto axis_v = axis.to<int>();
|
|
if (axis_v < 0) {
|
|
axis_v += static_cast<int>(x.dims().size());
|
|
}
|
|
|
|
if (axis_v != 0) {
|
|
if (index_type == DataType::INT32) {
|
|
funcs::GatherV2GradFunction<T, int32_t>(
|
|
dev_ctx, &out_grad, &index, axis_v, x_grad);
|
|
} else if (index_type == DataType::INT64) {
|
|
funcs::GatherV2GradFunction<T, int64_t>(
|
|
dev_ctx, &out_grad, &index, axis_v, x_grad);
|
|
}
|
|
return;
|
|
}
|
|
|
|
dev_ctx.template Alloc<T>(x_grad);
|
|
|
|
auto dxt = EigenVector<T>::Flatten(*x_grad);
|
|
auto& place = *dev_ctx.eigen_device();
|
|
dxt.device(place) = dxt.constant(static_cast<T>(0));
|
|
if (x_grad->numel() == 0) return;
|
|
|
|
if (index_type == DataType::INT32) {
|
|
funcs::ScatterAssignAdd<T, int32_t>(dev_ctx, out_grad, index, x_grad);
|
|
} else if (index_type == DataType::INT64) {
|
|
funcs::ScatterAssignAdd<T, int64_t>(dev_ctx, out_grad, index, x_grad);
|
|
} else {
|
|
PADDLE_THROW(common::errors::InvalidArgument(
|
|
"The data type of Input(Index) of gather_grad must be int32 or int64 "
|
|
"on CPU."));
|
|
}
|
|
}
|
|
|
|
} // namespace phi
|
|
|
|
PD_REGISTER_KERNEL(gather_grad,
|
|
CPU,
|
|
ALL_LAYOUT,
|
|
phi::GatherGradKernel,
|
|
float,
|
|
double,
|
|
uint8_t,
|
|
int8_t,
|
|
int16_t,
|
|
int32_t,
|
|
int64_t,
|
|
bool,
|
|
phi::bfloat16,
|
|
phi::complex64,
|
|
phi::complex128) {}
|