wehub-resource-sync
115 lines
3.3 KiB
C++
115 lines
3.3 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/stack_grad_kernel.h"
|
|
#include "paddle/phi/backends/xpu/enforce_xpu.h"
|
|
#include "paddle/phi/common/memory_utils.h"
|
|
#include "paddle/phi/core/ddim.h"
|
|
#include "paddle/phi/core/kernel_registry.h"
|
|
#include "paddle/phi/core/tensor_utils.h"
|
|
|
|
namespace phi {
|
|
|
|
template <typename T, typename Context>
|
|
void StackGradKernel(const Context& dev_ctx,
|
|
const DenseTensor& out_grad,
|
|
int axis,
|
|
std::vector<DenseTensor*> x_grad) {
|
|
using XPUType = typename XPUTypeTrait<T>::Type;
|
|
|
|
auto og_dims = out_grad.dims();
|
|
int rank = og_dims.size();
|
|
if (axis < 0) {
|
|
axis += rank;
|
|
}
|
|
int64_t n_slices = og_dims[axis];
|
|
|
|
struct ValidSlice {
|
|
DenseTensor* dx;
|
|
DDim final_dims;
|
|
};
|
|
std::vector<ValidSlice> valid_slices;
|
|
valid_slices.reserve(x_grad.size());
|
|
|
|
for (size_t i = 0; i < x_grad.size(); ++i) {
|
|
DenseTensor* dx_i = x_grad[i];
|
|
if (dx_i == nullptr || dx_i->numel() == 0) {
|
|
continue;
|
|
}
|
|
ValidSlice vs;
|
|
vs.dx = dx_i;
|
|
vs.final_dims = dx_i->dims();
|
|
valid_slices.push_back(vs);
|
|
}
|
|
|
|
if (valid_slices.empty()) {
|
|
return;
|
|
}
|
|
|
|
int64_t needed_slices = static_cast<int64_t>(valid_slices.size());
|
|
|
|
PADDLE_ENFORCE_LE(
|
|
needed_slices,
|
|
n_slices,
|
|
common::errors::InvalidArgument(
|
|
"Number of valid slices (%ld) exceeds out_grad's dimension (%ld) "
|
|
"along axis %d in stack_grad kernel. Mismatch between forward and "
|
|
"backward shapes.",
|
|
needed_slices,
|
|
n_slices,
|
|
axis));
|
|
|
|
std::vector<int64_t> partial_shape = vectorize<int64_t>(og_dims);
|
|
partial_shape[axis] = needed_slices;
|
|
|
|
std::vector<XPUType*> dx_ptrs;
|
|
dx_ptrs.reserve(needed_slices);
|
|
|
|
std::vector<int64_t> dx_dims_list;
|
|
dx_dims_list.reserve(needed_slices);
|
|
|
|
for (auto& vs : valid_slices) {
|
|
dev_ctx.template Alloc<T>(vs.dx);
|
|
dx_ptrs.push_back(reinterpret_cast<XPUType*>(vs.dx->template data<T>()));
|
|
dx_dims_list.push_back(1);
|
|
}
|
|
|
|
const XPUType* og_data =
|
|
reinterpret_cast<const XPUType*>(out_grad.template data<T>());
|
|
|
|
int r = xpu::split<XPUType>(
|
|
dev_ctx.x_context(), og_data, dx_ptrs, partial_shape, dx_dims_list, axis);
|
|
|
|
PADDLE_ENFORCE_XDNN_SUCCESS(r, "split in stack_grad op");
|
|
|
|
for (auto& vs : valid_slices) {
|
|
vs.dx->Resize(vs.final_dims);
|
|
}
|
|
}
|
|
|
|
} // namespace phi
|
|
|
|
PD_REGISTER_KERNEL(stack_grad,
|
|
XPU,
|
|
ALL_LAYOUT,
|
|
phi::StackGradKernel,
|
|
float,
|
|
phi::float16,
|
|
phi::bfloat16,
|
|
int64_t,
|
|
int,
|
|
int16_t,
|
|
int8_t,
|
|
uint8_t) {}
|