You cannot select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
72 lines
2.6 KiB
Plaintext
72 lines
2.6 KiB
Plaintext
8 months ago
|
// Copyright (c) 2021 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.
|
||
|
|
|
||
|
|
// reference
|
||
|
|
// https://github.com/PaddlePaddle/Paddle-Inference-Demo/blob/master/python/custom-operator/custom_relu_op.cu
|
||
|
|
|
||
|
|
#include "paddle/extension.h"
|
||
|
|
|
||
|
|
template <typename data_t>
|
||
|
|
__global__ void relu_cuda_forward_kernel(const data_t *x, data_t *y,
|
||
|
|
const int num) {
|
||
|
|
int gid = blockIdx.x * blockDim.x + threadIdx.x;
|
||
|
|
for (int i = gid; i < num; i += blockDim.x * gridDim.x) {
|
||
|
|
y[i] = max(x[i], static_cast<data_t>(0.));
|
||
|
|
}
|
||
|
|
}
|
||
|
|
|
||
|
|
template <typename data_t>
|
||
|
|
__global__ void relu_cuda_backward_kernel(const data_t *dy, const data_t *y,
|
||
|
|
data_t *dx, const int num) {
|
||
|
|
int gid = blockIdx.x * blockDim.x + threadIdx.x;
|
||
|
|
for (int i = gid; i < num; i += blockDim.x * gridDim.x) {
|
||
|
|
dx[i] = dy[i] * (y[i] > 0 ? 1. : 0.);
|
||
|
|
}
|
||
|
|
}
|
||
|
|
|
||
|
|
std::vector<paddle::Tensor> relu_cuda_forward(const paddle::Tensor &x) {
|
||
|
|
auto out = paddle::Tensor(paddle::PlaceType::kGPU);
|
||
|
|
|
||
|
|
out.reshape(x.shape());
|
||
|
|
int numel = x.size();
|
||
|
|
int block = 512;
|
||
|
|
int grid = (numel + block - 1) / block;
|
||
|
|
PD_DISPATCH_FLOATING_TYPES(
|
||
|
|
x.type(), "relu_cuda_forward_kernel", ([&] {
|
||
|
|
relu_cuda_forward_kernel<data_t><<<grid, block, 0, x.stream()>>>(
|
||
|
|
x.data<data_t>(), out.mutable_data<data_t>(x.place()), numel);
|
||
|
|
}));
|
||
|
|
|
||
|
|
return {out};
|
||
|
|
}
|
||
|
|
|
||
|
|
std::vector<paddle::Tensor> relu_cuda_backward(const paddle::Tensor &x,
|
||
|
|
const paddle::Tensor &out,
|
||
|
|
const paddle::Tensor &grad_out) {
|
||
|
|
auto grad_x = paddle::Tensor(paddle::PlaceType::kGPU);
|
||
|
|
grad_x.reshape(x.shape());
|
||
|
|
|
||
|
|
int numel = out.size();
|
||
|
|
int block = 512;
|
||
|
|
int grid = (numel + block - 1) / block;
|
||
|
|
PD_DISPATCH_FLOATING_TYPES(
|
||
|
|
out.type(), "relu_cuda_backward_kernel", ([&] {
|
||
|
|
relu_cuda_backward_kernel<data_t><<<grid, block, 0, x.stream()>>>(
|
||
|
|
grad_out.data<data_t>(), out.data<data_t>(),
|
||
|
|
grad_x.mutable_data<data_t>(x.place()), numel);
|
||
|
|
}));
|
||
|
|
|
||
|
|
return {grad_x};
|
||
|
|
}
|