wehub-resource-sync
87 lines
2.9 KiB
Plaintext
87 lines
2.9 KiB
Plaintext
// 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/scatter_grad_kernel.h"
|
|
|
|
#include "paddle/phi/backends/gpu/gpu_context.h"
|
|
#include "paddle/phi/core/kernel_registry.h"
|
|
#include "paddle/phi/core/tensor_utils.h"
|
|
#include "paddle/phi/kernels/full_kernel.h"
|
|
#include "paddle/phi/kernels/funcs/gather.cu.h"
|
|
#include "paddle/phi/kernels/funcs/scatter.cu.h"
|
|
|
|
namespace phi {
|
|
|
|
template <typename T, typename Context>
|
|
void ScatterGradKernel(const Context &dev_ctx,
|
|
const DenseTensor &index,
|
|
const DenseTensor &updates,
|
|
const DenseTensor &out_grad,
|
|
bool overwrite,
|
|
DenseTensor *x_grad,
|
|
DenseTensor *updates_grad) {
|
|
if (out_grad.numel() == 0) {
|
|
if (x_grad) {
|
|
dev_ctx.template Alloc<T>(x_grad);
|
|
}
|
|
if (updates_grad) {
|
|
Full<T, Context>(dev_ctx, updates_grad->dims(), 0, updates_grad);
|
|
}
|
|
return;
|
|
}
|
|
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(
|
|
"scatter_op index holds the wrong type, it holds [%s],"
|
|
"but desires to be [%s] or [%s]",
|
|
index_type,
|
|
DataType::INT32,
|
|
DataType::INT64));
|
|
|
|
if (x_grad) {
|
|
Copy(dev_ctx, out_grad, dev_ctx.GetPlace(), false, x_grad);
|
|
if (index_type == DataType::INT32) {
|
|
funcs::GPUScatterGradForX<T, int32_t>(dev_ctx, index, x_grad);
|
|
} else {
|
|
funcs::GPUScatterGradForX<T, int64_t>(dev_ctx, index, x_grad);
|
|
}
|
|
}
|
|
|
|
if (updates_grad) {
|
|
dev_ctx.template Alloc<T>(updates_grad);
|
|
// Gradient by Gather: dUpdates = dO[Ids]
|
|
if (index_type == DataType::INT32) {
|
|
funcs::GPUGather<T, int32_t>(dev_ctx, out_grad, index, updates_grad);
|
|
} else {
|
|
funcs::GPUGather<T, int64_t>(dev_ctx, out_grad, index, updates_grad);
|
|
}
|
|
}
|
|
}
|
|
|
|
} // namespace phi
|
|
|
|
PD_REGISTER_KERNEL(scatter_grad,
|
|
GPU,
|
|
ALL_LAYOUT,
|
|
phi::ScatterGradKernel,
|
|
float,
|
|
double,
|
|
int,
|
|
int64_t,
|
|
phi::float16,
|
|
phi::bfloat16) {}
|