113 lines
4.5 KiB
C++
113 lines
4.5 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/dropout_grad_kernel.h"
|
|
|
|
#include <memory>
|
|
#include <string>
|
|
|
|
#include "paddle/phi/backends/xpu/enforce_xpu.h"
|
|
#include "paddle/phi/core/kernel_registry.h"
|
|
|
|
namespace phi {
|
|
|
|
template <typename T, typename Context>
|
|
void DropoutGradRawKernel(const Context& dev_ctx,
|
|
const DenseTensor& mask,
|
|
const DenseTensor& out_grad,
|
|
const Scalar& p,
|
|
bool is_test,
|
|
const std::string& mode,
|
|
DenseTensor* x_grad) {
|
|
using XPUType = typename XPUTypeTrait<T>::Type;
|
|
PADDLE_ENFORCE_EQ(!is_test,
|
|
true,
|
|
common::errors::InvalidArgument(
|
|
"GradOp is only callable when is_test is false"));
|
|
auto* grad_x = x_grad;
|
|
auto* grad_y = &out_grad;
|
|
dev_ctx.template Alloc<T>(grad_x);
|
|
float dropout_prob = p.to<float>();
|
|
const uint8_t* mask_data = mask.data<uint8_t>();
|
|
|
|
xpu::ctx_guard RAII_GUARD(dev_ctx.x_context());
|
|
XPUType* mask_tmp_data = nullptr;
|
|
auto dev_version =
|
|
backends::xpu::get_xpu_version(dev_ctx.GetPlace().GetDeviceId());
|
|
if (mode != "upscale_in_train") {
|
|
mask_tmp_data = RAII_GUARD.alloc_l3_or_gm<XPUType>(mask.numel());
|
|
int r = xpu::cast<uint8_t, XPUType>(
|
|
dev_ctx.x_context(), mask_data, mask_tmp_data, mask.numel());
|
|
PADDLE_ENFORCE_XDNN_SUCCESS(r, "cast");
|
|
r = xpu::mul(dev_ctx.x_context(),
|
|
reinterpret_cast<const XPUType*>(grad_y->data<T>()),
|
|
reinterpret_cast<const XPUType*>(mask_tmp_data),
|
|
reinterpret_cast<XPUType*>(grad_x->data<T>()),
|
|
grad_y->numel());
|
|
PADDLE_ENFORCE_XDNN_SUCCESS(r, "mul");
|
|
return;
|
|
}
|
|
if (dev_version == backends::xpu::XPUVersion::XPU3) {
|
|
int r = xpu::dropout_grad_v2(
|
|
dev_ctx.x_context(),
|
|
mask_data,
|
|
reinterpret_cast<const XPUType*>(grad_y->data<T>()),
|
|
reinterpret_cast<XPUType*>(grad_x->data<T>()),
|
|
dropout_prob,
|
|
grad_y->numel());
|
|
PADDLE_ENFORCE_XDNN_SUCCESS(r, "dropout_grad_v2");
|
|
} else if (dev_version == backends::xpu::XPUVersion::XPU1) {
|
|
mask_tmp_data = RAII_GUARD.alloc_l3_or_gm<XPUType>(mask.numel());
|
|
int r = xpu::cast<uint8_t, XPUType>(
|
|
dev_ctx.x_context(), mask_data, mask_tmp_data, mask.numel());
|
|
float scale =
|
|
(dropout_prob == 1.0f) ? (1.0f) : (1.0f / (1.0f - dropout_prob));
|
|
r = xpu::scale(dev_ctx.x_context(),
|
|
reinterpret_cast<const XPUType*>(mask_tmp_data),
|
|
reinterpret_cast<XPUType*>(mask_tmp_data),
|
|
mask.numel(),
|
|
false,
|
|
scale,
|
|
0.0f);
|
|
PADDLE_ENFORCE_XDNN_SUCCESS(r, "scale");
|
|
r = xpu::mul(dev_ctx.x_context(),
|
|
reinterpret_cast<const XPUType*>(grad_y->data<T>()),
|
|
reinterpret_cast<const XPUType*>(mask_tmp_data),
|
|
reinterpret_cast<XPUType*>(grad_x->data<T>()),
|
|
grad_y->numel());
|
|
PADDLE_ENFORCE_XDNN_SUCCESS(r, "mul");
|
|
} else {
|
|
mask_tmp_data = RAII_GUARD.alloc_l3_or_gm<XPUType>(mask.numel());
|
|
int r = xpu::cast<uint8_t, XPUType>(
|
|
dev_ctx.x_context(), mask_data, mask_tmp_data, mask.numel());
|
|
r = xpu::dropout_grad(dev_ctx.x_context(),
|
|
reinterpret_cast<const XPUType*>(mask_tmp_data),
|
|
reinterpret_cast<const XPUType*>(grad_y->data<T>()),
|
|
reinterpret_cast<XPUType*>(grad_x->data<T>()),
|
|
dropout_prob,
|
|
grad_y->numel());
|
|
PADDLE_ENFORCE_XDNN_SUCCESS(r, "dropout_grad");
|
|
}
|
|
}
|
|
|
|
} // namespace phi
|
|
|
|
PD_REGISTER_KERNEL(dropout_grad,
|
|
XPU,
|
|
ALL_LAYOUT,
|
|
phi::DropoutGradRawKernel,
|
|
float,
|
|
phi::float16,
|
|
phi::bfloat16) {}
|