Skip to content

P
Paddle

机器未来 / Paddle
与 Fork 源项目一致

Fork自 PaddlePaddle / Paddle

通知 1
Star 1
Fork 0
P
Paddle
提交 60d6348e 编写于 作者: Y Yu Yang
浏览文件
操作

Revert develop

上级 86af6bdc
隐藏空白更改
内联 并排
Showing with 30 addition and 52 deletion
paddle/fluid/operators/math/pooling.cu
浏览文件 @ 60d6348e
...@@ -12,8 +12,6 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. ...@@ -12,8 +12,6 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and See the License for the specific language governing permissions and
limitations under the License. */ limitations under the License. */
#include <algorithm>
#include <vector>
#include "paddle/fluid/operators/math/pooling.h" #include "paddle/fluid/operators/math/pooling.h"
#include "paddle/fluid/platform/cuda_primitives.h" #include "paddle/fluid/platform/cuda_primitives.h"
...@@ -22,7 +20,7 @@ namespace operators { ...@@ -22,7 +20,7 @@ namespace operators {
namespace math { namespace math {
template <typename PoolProcess, typename T> template <typename PoolProcess, typename T>
__global__ void KernelPool2D(const int nthreads, const T* input_data, // NOLINT __global__ void KernelPool2D(const int nthreads, const T* input_data,
const int channels, const int input_height, const int channels, const int input_height,
const int input_width, const int output_height, const int input_width, const int output_height,
const int output_width, const int ksize_height, const int output_width, const int ksize_height,
...@@ -60,8 +58,8 @@ __global__ void KernelPool2D(const int nthreads, const T* input_data, // NOLINT ...@@ -60,8 +58,8 @@ __global__ void KernelPool2D(const int nthreads, const T* input_data, // NOLINT
template <typename PoolProcess, typename T> template <typename PoolProcess, typename T>
__global__ void KernelPool2DGrad( __global__ void KernelPool2DGrad(
const int nthreads, const T* input_data, const T* output_data, // NOLINT const int nthreads, const T* input_data, const T* output_data,
const T* output_grad, const int channels, const int input_height, // NOLINT const T* output_grad, const int channels, const int input_height,
const int input_width, const int output_height, const int output_width, const int input_width, const int output_height, const int output_width,
const int ksize_height, const int ksize_width, const int stride_height, const int ksize_height, const int ksize_width, const int stride_height,
const int stride_width, const int padding_height, const int padding_width, const int stride_width, const int padding_height, const int padding_width,
...@@ -108,8 +106,8 @@ __global__ void KernelPool2DGrad( ...@@ -108,8 +106,8 @@ __global__ void KernelPool2DGrad(
template <typename T> template <typename T>
__global__ void KernelMaxPool2DGrad( __global__ void KernelMaxPool2DGrad(
const int nthreads, const T* input_data, const T* output_data, // NOLINT const int nthreads, const T* input_data, const T* output_data,
const T* output_grad, const int channels, const int input_height, // NOLINT const T* output_grad, const int channels, const int input_height,
const int input_width, const int output_height, const int output_width, const int input_width, const int output_height, const int output_width,
const int ksize_height, const int ksize_width, const int stride_height, const int ksize_height, const int ksize_width, const int stride_height,
const int stride_width, const int padding_height, const int padding_width, const int stride_width, const int padding_height, const int padding_width,
...@@ -160,10 +158,8 @@ template <typename PoolProcess, typename T> ...@@ -160,10 +158,8 @@ template <typename PoolProcess, typename T>
class Pool2dFunctor<platform::CUDADeviceContext, PoolProcess, T> { class Pool2dFunctor<platform::CUDADeviceContext, PoolProcess, T> {
public: public:
void operator()(const platform::CUDADeviceContext& context, void operator()(const platform::CUDADeviceContext& context,
const framework::Tensor& input, const framework::Tensor& input, std::vector<int>& ksize,
std::vector<int>& ksize, // NOLINT std::vector<int>& strides, std::vector<int>& paddings,
std::vector<int>& strides, // NOLINT
std::vector<int>& paddings, // NOLINT
PoolProcess pool_process, framework::Tensor* output) { PoolProcess pool_process, framework::Tensor* output) {
const int batch_size = input.dims()[0]; const int batch_size = input.dims()[0];
const int input_channels = input.dims()[1]; const int input_channels = input.dims()[1];
...@@ -205,10 +201,8 @@ class Pool2dGradFunctor<platform::CUDADeviceContext, PoolProcess, T> { ...@@ -205,10 +201,8 @@ class Pool2dGradFunctor<platform::CUDADeviceContext, PoolProcess, T> {
void operator()(const platform::CUDADeviceContext& context, void operator()(const platform::CUDADeviceContext& context,
const framework::Tensor& input, const framework::Tensor& input,
const framework::Tensor& output, const framework::Tensor& output,
const framework::Tensor& output_grad, const framework::Tensor& output_grad, std::vector<int>& ksize,
std::vector<int>& ksize, // NOLINT std::vector<int>& strides, std::vector<int>& paddings,
std::vector<int>& strides, // NOLINT
std::vector<int>& paddings, // NOLINT
PoolProcess pool_process, framework::Tensor* input_grad) { PoolProcess pool_process, framework::Tensor* input_grad) {
const int batch_size = input.dims()[0]; const int batch_size = input.dims()[0];
const int input_channels = input.dims()[1]; const int input_channels = input.dims()[1];
...@@ -252,10 +246,8 @@ class MaxPool2dGradFunctor<platform::CUDADeviceContext, T> { ...@@ -252,10 +246,8 @@ class MaxPool2dGradFunctor<platform::CUDADeviceContext, T> {
void operator()(const platform::CUDADeviceContext& context, void operator()(const platform::CUDADeviceContext& context,
const framework::Tensor& input, const framework::Tensor& input,
const framework::Tensor& output, const framework::Tensor& output,
const framework::Tensor& output_grad, const framework::Tensor& output_grad, std::vector<int>& ksize,
std::vector<int>& ksize, // NOLINT std::vector<int>& strides, std::vector<int>& paddings,
std::vector<int>& strides, // NOLINT
std::vector<int>& paddings, // NOLINT
framework::Tensor* input_grad) { framework::Tensor* input_grad) {
const int batch_size = input.dims()[0]; const int batch_size = input.dims()[0];
const int input_channels = input.dims()[1]; const int input_channels = input.dims()[1];
...@@ -314,7 +306,7 @@ template class Pool2dGradFunctor<platform::CUDADeviceContext, ...@@ -314,7 +306,7 @@ template class Pool2dGradFunctor<platform::CUDADeviceContext,
double>; double>;
template <typename PoolProcess, typename T> template <typename PoolProcess, typename T>
__global__ void KernelPool3D(const int nthreads, const T* input_data, // NOLINT __global__ void KernelPool3D(const int nthreads, const T* input_data,
const int channels, const int input_depth, const int channels, const int input_depth,
const int input_height, const int input_width, const int input_height, const int input_width,
const int output_depth, const int output_height, const int output_depth, const int output_height,
...@@ -360,8 +352,8 @@ __global__ void KernelPool3D(const int nthreads, const T* input_data, // NOLINT ...@@ -360,8 +352,8 @@ __global__ void KernelPool3D(const int nthreads, const T* input_data, // NOLINT
template <typename PoolProcess, typename T> template <typename PoolProcess, typename T>
__global__ void KernelPool3DGrad( __global__ void KernelPool3DGrad(
const int nthreads, const T* input_data, const T* output_data, // NOLINT const int nthreads, const T* input_data, const T* output_data,
const T* output_grad, const int channels, const int input_depth, // NOLINT const T* output_grad, const int channels, const int input_depth,
const int input_height, const int input_width, const int output_depth, const int input_height, const int input_width, const int output_depth,
const int output_height, const int output_width, const int ksize_depth, const int output_height, const int output_width, const int ksize_depth,
const int ksize_height, const int ksize_width, const int stride_depth, const int ksize_height, const int ksize_width, const int stride_depth,
...@@ -424,8 +416,8 @@ __global__ void KernelPool3DGrad( ...@@ -424,8 +416,8 @@ __global__ void KernelPool3DGrad(
template <typename T> template <typename T>
__global__ void KernelMaxPool3DGrad( __global__ void KernelMaxPool3DGrad(
const int nthreads, const T* input_data, const T* output_data, // NOLINT const int nthreads, const T* input_data, const T* output_data,
const T* output_grad, const int channels, const int input_depth, // NOLINT const T* output_grad, const int channels, const int input_depth,
const int input_height, const int input_width, const int output_depth, const int input_height, const int input_width, const int output_depth,
const int output_height, const int output_width, const int ksize_depth, const int output_height, const int output_width, const int ksize_depth,
const int ksize_height, const int ksize_width, const int stride_depth, const int ksize_height, const int ksize_width, const int stride_depth,
...@@ -482,10 +474,8 @@ template <typename PoolProcess, class T> ...@@ -482,10 +474,8 @@ template <typename PoolProcess, class T>
class Pool3dFunctor<platform::CUDADeviceContext, PoolProcess, T> { class Pool3dFunctor<platform::CUDADeviceContext, PoolProcess, T> {
public: public:
void operator()(const platform::CUDADeviceContext& context, void operator()(const platform::CUDADeviceContext& context,
const framework::Tensor& input, const framework::Tensor& input, std::vector<int>& ksize,
std::vector<int>& ksize, // NOLINT std::vector<int>& strides, std::vector<int>& paddings,
std::vector<int>& strides, // NOLINT
std::vector<int>& paddings, // NOLINT
PoolProcess pool_process, framework::Tensor* output) { PoolProcess pool_process, framework::Tensor* output) {
const int batch_size = input.dims()[0]; const int batch_size = input.dims()[0];
const int input_channels = input.dims()[1]; const int input_channels = input.dims()[1];
...@@ -535,10 +525,8 @@ class Pool3dGradFunctor<platform::CUDADeviceContext, PoolProcess, T> { ...@@ -535,10 +525,8 @@ class Pool3dGradFunctor<platform::CUDADeviceContext, PoolProcess, T> {
void operator()(const platform::CUDADeviceContext& context, void operator()(const platform::CUDADeviceContext& context,
const framework::Tensor& input, const framework::Tensor& input,
const framework::Tensor& output, const framework::Tensor& output,
const framework::Tensor& output_grad, const framework::Tensor& output_grad, std::vector<int>& ksize,
std::vector<int>& ksize, // NOLINT std::vector<int>& strides, std::vector<int>& paddings,
std::vector<int>& strides, // NOLINT
std::vector<int>& paddings, // NOLINT
PoolProcess pool_process, framework::Tensor* input_grad) { PoolProcess pool_process, framework::Tensor* input_grad) {
const int batch_size = input.dims()[0]; const int batch_size = input.dims()[0];
const int input_channels = input.dims()[1]; const int input_channels = input.dims()[1];
...@@ -590,10 +578,8 @@ class MaxPool3dGradFunctor<platform::CUDADeviceContext, T> { ...@@ -590,10 +578,8 @@ class MaxPool3dGradFunctor<platform::CUDADeviceContext, T> {
void operator()(const platform::CUDADeviceContext& context, void operator()(const platform::CUDADeviceContext& context,
const framework::Tensor& input, const framework::Tensor& input,
const framework::Tensor& output, const framework::Tensor& output,
const framework::Tensor& output_grad, const framework::Tensor& output_grad, std::vector<int>& ksize,
std::vector<int>& ksize, // NOLINT std::vector<int>& strides, std::vector<int>& paddings,
std::vector<int>& strides, // NOLINT
std::vector<int>& paddings, // NOLINT
framework::Tensor* input_grad) { framework::Tensor* input_grad) {
const int batch_size = input.dims()[0]; const int batch_size = input.dims()[0];
const int input_channels = input.dims()[1]; const int input_channels = input.dims()[1];
...@@ -750,10 +736,8 @@ template <typename T1, typename T2> ...@@ -750,10 +736,8 @@ template <typename T1, typename T2>
class MaxPool2dWithIndexFunctor<platform::CUDADeviceContext, T1, T2> { class MaxPool2dWithIndexFunctor<platform::CUDADeviceContext, T1, T2> {
public: public:
void operator()(const platform::CUDADeviceContext& context, void operator()(const platform::CUDADeviceContext& context,
const framework::Tensor& input, const framework::Tensor& input, std::vector<int>& ksize,
std::vector<int>& ksize, // NOLINT std::vector<int>& strides, std::vector<int>& paddings,
std::vector<int>& strides, // NOLINT
std::vector<int>& paddings, // NOLINT
framework::Tensor* output, framework::Tensor* mask) { framework::Tensor* output, framework::Tensor* mask) {
const int batch_size = input.dims()[0]; const int batch_size = input.dims()[0];
const int input_channels = input.dims()[1]; const int input_channels = input.dims()[1];
...@@ -795,10 +779,8 @@ class MaxPool2dWithIndexGradFunctor<platform::CUDADeviceContext, T1, T2> { ...@@ -795,10 +779,8 @@ class MaxPool2dWithIndexGradFunctor<platform::CUDADeviceContext, T1, T2> {
public: public:
void operator()(const platform::CUDADeviceContext& context, void operator()(const platform::CUDADeviceContext& context,
const framework::Tensor& output_grad, const framework::Tensor& output_grad,
const framework::Tensor& mask, const framework::Tensor& mask, std::vector<int>& ksize,
std::vector<int>& ksize, // NOLINT std::vector<int>& strides, std::vector<int>& paddings,
std::vector<int>& strides, // NOLINT
std::vector<int>& paddings, // NOLINT
framework::Tensor* input_grad) { framework::Tensor* input_grad) {
const int batch_size = input_grad->dims()[0]; const int batch_size = input_grad->dims()[0];
const int input_channels = input_grad->dims()[1]; const int input_channels = input_grad->dims()[1];
...@@ -955,10 +937,8 @@ template <typename T1, typename T2> ...@@ -955,10 +937,8 @@ template <typename T1, typename T2>
class MaxPool3dWithIndexFunctor<platform::CUDADeviceContext, T1, T2> { class MaxPool3dWithIndexFunctor<platform::CUDADeviceContext, T1, T2> {
public: public:
void operator()(const platform::CUDADeviceContext& context, void operator()(const platform::CUDADeviceContext& context,
const framework::Tensor& input, const framework::Tensor& input, std::vector<int>& ksize,
std::vector<int>& ksize, // NOLINT std::vector<int>& strides, std::vector<int>& paddings,
std::vector<int>& strides, // NOLINT
std::vector<int>& paddings, // NOLINT
framework::Tensor* output, framework::Tensor* mask) { framework::Tensor* output, framework::Tensor* mask) {
const int batch_size = input.dims()[0]; const int batch_size = input.dims()[0];
const int input_channels = input.dims()[1]; const int input_channels = input.dims()[1];
...@@ -1007,10 +987,8 @@ class MaxPool3dWithIndexGradFunctor<platform::CUDADeviceContext, T1, T2> { ...@@ -1007,10 +987,8 @@ class MaxPool3dWithIndexGradFunctor<platform::CUDADeviceContext, T1, T2> {
public: public:
void operator()(const platform::CUDADeviceContext& context, void operator()(const platform::CUDADeviceContext& context,
const framework::Tensor& output_grad, const framework::Tensor& output_grad,
const framework::Tensor& mask, const framework::Tensor& mask, std::vector<int>& ksize,
std::vector<int>& ksize, // NOLINT std::vector<int>& strides, std::vector<int>& paddings,
std::vector<int>& strides, // NOLINT
std::vector<int>& paddings, // NOLINT
framework::Tensor* input_grad) { framework::Tensor* input_grad) {
const int batch_size = input_grad->dims()[0]; const int batch_size = input_grad->dims()[0];
const int input_channels = input_grad->dims()[1]; const int input_channels = input_grad->dims()[1];
......
Markdown is supported
0% 或 .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑！
想要评论请 注册 或