wehub-resource-sync
88 lines
3.5 KiB
C++
88 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/temporal_shift_grad_kernel.h"
|
|
|
|
#include "paddle/common/layout.h"
|
|
#include "paddle/phi/backends/xpu/enforce_xpu.h"
|
|
#include "paddle/phi/core/kernel_registry.h"
|
|
#include "paddle/phi/kernels/funcs/axis_utils.h"
|
|
|
|
namespace phi {
|
|
|
|
template <typename T, typename Context>
|
|
void TemporalShiftGradKernel(const Context& dev_ctx,
|
|
const DenseTensor& out_grad,
|
|
int seg_num,
|
|
float shift_ratio,
|
|
const std::string& data_format_str,
|
|
DenseTensor* x_grad) {
|
|
if (x_grad && x_grad->numel() == 0) {
|
|
dev_ctx.template Alloc<T>(x_grad);
|
|
return;
|
|
}
|
|
auto* input_grad = x_grad;
|
|
auto* output_grad = &out_grad;
|
|
int64_t t = seg_num;
|
|
const DataLayout data_layout = StringToDataLayout(data_format_str);
|
|
|
|
const int64_t nt = output_grad->dims()[0];
|
|
const int64_t n = nt / t;
|
|
const int64_t c = (data_layout == DataLayout::NCHW ? output_grad->dims()[1]
|
|
: output_grad->dims()[3]);
|
|
const int64_t h = (data_layout == DataLayout::NCHW ? output_grad->dims()[2]
|
|
: output_grad->dims()[1]);
|
|
const int64_t w = (data_layout == DataLayout::NCHW ? output_grad->dims()[3]
|
|
: output_grad->dims()[2]);
|
|
|
|
DDim in_grad_dims =
|
|
(data_layout == DataLayout::NCHW ? make_ddim({nt, c, h, w})
|
|
: make_ddim({nt, h, w, c}));
|
|
const T* output_grad_data = output_grad->data<T>();
|
|
input_grad->Resize(in_grad_dims);
|
|
T* input_grad_data = dev_ctx.template Alloc<T>(input_grad);
|
|
|
|
if (data_layout == DataLayout::NCHW) {
|
|
int r = xpu::temporal_shift_grad(dev_ctx.x_context(),
|
|
output_grad_data,
|
|
input_grad_data,
|
|
n,
|
|
c,
|
|
h,
|
|
w,
|
|
t,
|
|
shift_ratio,
|
|
false);
|
|
PADDLE_ENFORCE_XDNN_SUCCESS(r, "temporal_shift_grad");
|
|
} else {
|
|
int r = xpu::temporal_shift_grad(dev_ctx.x_context(),
|
|
output_grad_data,
|
|
input_grad_data,
|
|
n,
|
|
c,
|
|
h,
|
|
w,
|
|
t,
|
|
shift_ratio,
|
|
true);
|
|
PADDLE_ENFORCE_XDNN_SUCCESS(r, "temporal_shift_grad");
|
|
}
|
|
}
|
|
|
|
} // namespace phi
|
|
|
|
PD_REGISTER_KERNEL(
|
|
temporal_shift_grad, XPU, ALL_LAYOUT, phi::TemporalShiftGradKernel, float) {
|
|
}
|