wehub-resource-sync
105 lines
3.5 KiB
C++
105 lines
3.5 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/reduce_sum_grad_kernel.h"
|
|
|
|
#include <set>
|
|
|
|
#include "paddle/phi/backends/xpu/enforce_xpu.h"
|
|
#include "paddle/phi/core/kernel_registry.h"
|
|
#include "paddle/phi/kernels/cast_kernel.h"
|
|
#include "paddle/phi/kernels/empty_kernel.h"
|
|
|
|
namespace phi {
|
|
|
|
template <typename T, typename Context>
|
|
void ReduceSumGradKernel(const Context& dev_ctx,
|
|
const DenseTensor& x,
|
|
const DenseTensor& out_grad,
|
|
const IntArray& dims_arr,
|
|
bool keep_dim,
|
|
bool reduce_all,
|
|
DenseTensor* x_grad) {
|
|
using XPUType = typename XPUTypeTrait<T>::Type;
|
|
if (x_grad && x_grad->numel() == 0) {
|
|
dev_ctx.template Alloc<T>(x_grad);
|
|
return;
|
|
}
|
|
reduce_all = recompute_reduce_all(x, dims_arr, reduce_all);
|
|
auto dims = dims_arr.GetData();
|
|
dev_ctx.Alloc(x_grad, x.dtype());
|
|
auto* out_data = reinterpret_cast<const XPUType*>(out_grad.data());
|
|
auto* x_grad_data = reinterpret_cast<XPUType*>(x_grad->data());
|
|
const auto& input_dim_size = x.dims().size();
|
|
std::vector<int64_t> true_dims;
|
|
for (size_t i = 0; i < dims.size(); ++i) {
|
|
if (dims[i] < 0) {
|
|
true_dims.push_back(dims[i] + input_dim_size);
|
|
} else {
|
|
true_dims.push_back(dims[i]);
|
|
}
|
|
}
|
|
|
|
std::vector<int64_t> ydims(input_dim_size);
|
|
std::vector<int64_t> xdims((input_dim_size));
|
|
std::set<int64_t> dims_set(true_dims.begin(), true_dims.end());
|
|
for (auto i = 0; i < input_dim_size; i++) {
|
|
xdims[i] = x.dims()[i];
|
|
if (dims_set.find(i) != dims_set.end() || reduce_all) {
|
|
ydims[i] = 1;
|
|
} else {
|
|
ydims[i] = x.dims()[i];
|
|
}
|
|
}
|
|
|
|
// use [1] to replace [], because xpu not support []
|
|
if (xdims.size() == 0) {
|
|
xdims = std::vector<int64_t>({1});
|
|
}
|
|
if (ydims.size() == 0) {
|
|
ydims = std::vector<int64_t>({1});
|
|
}
|
|
|
|
if (x.dtype() != out_grad.dtype()) {
|
|
DenseTensorMeta x_grad_meta(
|
|
out_grad.dtype(), x_grad->dims(), x_grad->layout());
|
|
DenseTensor x_grad_tmp = Empty<Context>(dev_ctx, std::move(x_grad_meta));
|
|
auto* x_grad_tmp_data = reinterpret_cast<XPUType*>(x_grad_tmp.data());
|
|
|
|
int r = xpu::broadcast<XPUType>(
|
|
dev_ctx.x_context(), out_data, x_grad_tmp_data, ydims, xdims);
|
|
PADDLE_ENFORCE_XDNN_SUCCESS(r, "broadcast");
|
|
|
|
CastKernel<T>(dev_ctx, x_grad_tmp, x.dtype(), x_grad);
|
|
} else {
|
|
int r = xpu::broadcast<XPUType>(
|
|
dev_ctx.x_context(), out_data, x_grad_data, ydims, xdims);
|
|
PADDLE_ENFORCE_XDNN_SUCCESS(r, "broadcast");
|
|
}
|
|
}
|
|
|
|
} // namespace phi
|
|
|
|
PD_REGISTER_KERNEL(sum_grad,
|
|
XPU,
|
|
ALL_LAYOUT,
|
|
phi::ReduceSumGradKernel,
|
|
float,
|
|
phi::float16,
|
|
phi::bfloat16,
|
|
int64_t,
|
|
int,
|
|
bool) {
|
|
kernel->OutputAt(0).SetDataType(phi::DataType::UNDEFINED);
|
|
}
|