wehub-resource-sync
92 lines
3.6 KiB
C++
92 lines
3.6 KiB
C++
// Copyright (c) 2023 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/index_select_grad_kernel.h"
|
|
|
|
#include "paddle/phi/backends/xpu/enforce_xpu.h"
|
|
#include "paddle/phi/core/kernel_registry.h"
|
|
#include "paddle/phi/core/utils/data_type.h"
|
|
#include "paddle/phi/kernels/full_kernel.h"
|
|
namespace phi {
|
|
|
|
template <typename T, typename Context>
|
|
void IndexSelectGradKernel(const Context& dev_ctx,
|
|
const DenseTensor& x,
|
|
const DenseTensor& index,
|
|
const DenseTensor& out_grad,
|
|
int dim,
|
|
DenseTensor* x_grad) {
|
|
using XPUType = typename XPUTypeTrait<T>::Type;
|
|
if (out_grad.numel() == 0) {
|
|
Full<T, Context>(dev_ctx, x.dims(), 0, x_grad);
|
|
return;
|
|
}
|
|
if (dim < 0) {
|
|
dim += out_grad.dims().size();
|
|
}
|
|
const auto& index_type = index.dtype();
|
|
bool index_type_match =
|
|
index_type == DataType::INT32 || index_type == DataType::INT64;
|
|
PADDLE_ENFORCE_EQ(index_type_match,
|
|
true,
|
|
common::errors::InvalidArgument(
|
|
"Input(Index) holds the wrong type, it holds %s, but "
|
|
"desires to be %s or %s",
|
|
index_type,
|
|
DataType::INT32,
|
|
DataType::INT64));
|
|
|
|
XPUType* x_grad_data =
|
|
reinterpret_cast<XPUType*>((dev_ctx.template Alloc<T>(x_grad)));
|
|
const XPUType* out_grad_data =
|
|
reinterpret_cast<const XPUType*>(out_grad.data<T>());
|
|
|
|
auto out_grad_shape = vectorize<int64_t>(out_grad.dims());
|
|
auto x_grad_shape = vectorize<int64_t>(x_grad->dims());
|
|
|
|
int r = 0;
|
|
if (index_type == DataType::INT32) {
|
|
const int* index_data = index.data<int>();
|
|
r = xpu::index_select_grad<XPUType, int>(dev_ctx.x_context(),
|
|
nullptr,
|
|
index_data,
|
|
out_grad_data,
|
|
dim,
|
|
x_grad_data,
|
|
out_grad_shape,
|
|
x_grad_shape);
|
|
} else if (index_type == DataType::INT64) {
|
|
const int64_t* index_data = index.data<int64_t>();
|
|
r = xpu::index_select_grad<XPUType, int64_t>(dev_ctx.x_context(),
|
|
nullptr,
|
|
index_data,
|
|
out_grad_data,
|
|
dim,
|
|
x_grad_data,
|
|
out_grad_shape,
|
|
x_grad_shape);
|
|
}
|
|
PADDLE_ENFORCE_XDNN_SUCCESS(r, "index_select_grad");
|
|
}
|
|
|
|
} // namespace phi
|
|
|
|
PD_REGISTER_KERNEL(index_select_grad,
|
|
XPU,
|
|
ALL_LAYOUT,
|
|
phi::IndexSelectGradKernel,
|
|
float,
|
|
phi::float16,
|
|
phi::bfloat16) {}
|