184 lines
6.3 KiB
C++
184 lines
6.3 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/slice_grad_kernel.h"
|
|
|
|
#include "paddle/phi/backends/xpu/enforce_xpu.h"
|
|
#include "paddle/phi/core/kernel_registry.h"
|
|
#include "paddle/phi/kernels/complex_kernel.h"
|
|
#include "paddle/phi/kernels/full_kernel.h"
|
|
#include "paddle/phi/kernels/funcs/slice_utils.h"
|
|
|
|
namespace phi {
|
|
|
|
template <typename T, typename Context>
|
|
void SliceGradKernel(const Context& dev_ctx,
|
|
const DenseTensor& input,
|
|
const DenseTensor& out_grad,
|
|
const std::vector<int64_t>& axes,
|
|
const IntArray& starts_t,
|
|
const IntArray& ends_t,
|
|
const std::vector<int64_t>& infer_flags,
|
|
const std::vector<int64_t>& decrease_axis,
|
|
DenseTensor* input_grad) {
|
|
using XPUType = typename XPUTypeTrait<T>::Type;
|
|
dev_ctx.template Alloc<T>(input_grad);
|
|
if (input_grad->numel() == 0) {
|
|
return;
|
|
}
|
|
if (out_grad.numel() == 0) {
|
|
Full<T, XPUContext>(dev_ctx, input_grad->dims(), T(0), input_grad);
|
|
return;
|
|
}
|
|
// Get the accurate attribute value of starts and ends
|
|
std::vector<int64_t> starts = starts_t.GetData();
|
|
std::vector<int64_t> ends = ends_t.GetData();
|
|
|
|
const auto& in_dims = input.dims();
|
|
int rank = in_dims.size();
|
|
|
|
std::vector<int64_t> pad_left(rank);
|
|
std::vector<int64_t> out_dims(rank);
|
|
std::vector<int64_t> pad_right(rank);
|
|
int64_t cnt = 0;
|
|
for (int i = 0; i < in_dims.size(); ++i) {
|
|
int64_t start = 0;
|
|
int64_t end = in_dims[i];
|
|
int64_t axis = cnt < static_cast<int64_t>(axes.size()) ? axes[cnt] : -1;
|
|
if (axis == i) {
|
|
bool zero_dim = false;
|
|
funcs::normalize_interval(starts[cnt],
|
|
ends[cnt],
|
|
static_cast<int64_t>(1),
|
|
in_dims[i],
|
|
&start,
|
|
&end,
|
|
&zero_dim);
|
|
cnt++;
|
|
}
|
|
|
|
pad_left[i] = start;
|
|
out_dims[i] = end - start;
|
|
pad_right[i] = in_dims[i] - out_dims[i] - pad_left[i];
|
|
}
|
|
|
|
int r =
|
|
xpu::pad<XPUType>(dev_ctx.x_context(),
|
|
reinterpret_cast<const XPUType*>(out_grad.data<T>()),
|
|
reinterpret_cast<XPUType*>(input_grad->data<T>()),
|
|
out_dims,
|
|
pad_left,
|
|
pad_right,
|
|
XPUType(0));
|
|
PADDLE_ENFORCE_XDNN_SUCCESS(r, "pad");
|
|
}
|
|
|
|
#ifdef PADDLE_WITH_XPU_FFT
|
|
template <>
|
|
void SliceGradKernel<phi::complex64, XPUContext>(
|
|
const XPUContext& dev_ctx,
|
|
const DenseTensor& input,
|
|
const DenseTensor& out_grad,
|
|
const std::vector<int64_t>& axes,
|
|
const IntArray& starts_t,
|
|
const IntArray& ends_t,
|
|
const std::vector<int64_t>& infer_flags,
|
|
const std::vector<int64_t>& decrease_axis,
|
|
DenseTensor* input_grad) {
|
|
using T = phi::complex64;
|
|
dev_ctx.template Alloc<T>(input_grad);
|
|
if (input_grad->numel() == 0) {
|
|
return;
|
|
}
|
|
if (out_grad.numel() == 0) {
|
|
Full<T, XPUContext>(dev_ctx, input_grad->dims(), T(0), input_grad);
|
|
return;
|
|
}
|
|
|
|
// Get the accurate attribute value of starts and ends
|
|
std::vector<int64_t> starts = starts_t.GetData();
|
|
std::vector<int64_t> ends = ends_t.GetData();
|
|
|
|
const auto& in_dims = input.dims();
|
|
int rank = in_dims.size();
|
|
|
|
std::vector<int64_t> pad_left(rank);
|
|
std::vector<int64_t> out_dims(rank);
|
|
std::vector<int64_t> pad_right(rank);
|
|
int64_t cnt = 0;
|
|
for (int i = 0; i < in_dims.size(); ++i) {
|
|
int64_t start = 0;
|
|
int64_t end = in_dims[i];
|
|
int64_t axis = cnt < static_cast<int64_t>(axes.size()) ? axes[cnt] : -1;
|
|
if (axis == i) {
|
|
bool zero_dim = false;
|
|
funcs::normalize_interval(starts[cnt],
|
|
ends[cnt],
|
|
static_cast<int64_t>(1),
|
|
in_dims[i],
|
|
&start,
|
|
&end,
|
|
&zero_dim);
|
|
cnt++;
|
|
}
|
|
|
|
pad_left[i] = start;
|
|
out_dims[i] = end - start;
|
|
pad_right[i] = in_dims[i] - out_dims[i] - pad_left[i];
|
|
}
|
|
|
|
// The current complex number implementation uses separate real/imaginary
|
|
// parts,resulting in redundant operations and performance
|
|
// penalties.Optimization should address this in future iterations.
|
|
const DenseTensor real = Real<T, XPUContext>(dev_ctx, out_grad);
|
|
const DenseTensor imag = Imag<T, XPUContext>(dev_ctx, out_grad);
|
|
DenseTensor real_out, imag_out;
|
|
real_out.Resize(input_grad->dims());
|
|
imag_out.Resize(input_grad->dims());
|
|
dev_ctx.template Alloc<float>(&real_out);
|
|
dev_ctx.template Alloc<float>(&imag_out);
|
|
int r = xpu::pad<float>(dev_ctx.x_context(),
|
|
real.data<float>(),
|
|
real_out.data<float>(),
|
|
out_dims,
|
|
pad_left,
|
|
pad_right,
|
|
static_cast<float>(0));
|
|
PADDLE_ENFORCE_XDNN_SUCCESS(r, "pad");
|
|
r = xpu::pad<float>(dev_ctx.x_context(),
|
|
imag.data<float>(),
|
|
imag_out.data<float>(),
|
|
out_dims,
|
|
pad_left,
|
|
pad_right,
|
|
static_cast<float>(0));
|
|
PADDLE_ENFORCE_XDNN_SUCCESS(r, "pad");
|
|
phi::ComplexKernel<float>(dev_ctx, real_out, imag_out, input_grad);
|
|
}
|
|
#endif
|
|
} // namespace phi
|
|
|
|
PD_REGISTER_KERNEL(slice_grad,
|
|
XPU,
|
|
ALL_LAYOUT,
|
|
phi::SliceGradKernel,
|
|
float,
|
|
int,
|
|
#ifdef PADDLE_WITH_XPU_FFT
|
|
phi::complex64,
|
|
#endif
|
|
phi::float16,
|
|
phi::bfloat16) {
|
|
}
|