OSSInnovation
/
mindspore
1
0
Fork 26
Code Issues Pull Requests 7 Releases Wiki Activity

!3495 Gpu StridedSlice performance optimize 

Merge pull request !3495 from chenweifeng/strided_slice
This commit is contained in:
mindspore-ci-bot 2020-07-27 11:04:50 +08:00 committed by Gitee
parent 295038d346 d3167da8af
commit c22792aab1
12 changed files with 867 additions and 190 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
mindspore/ccsrc/backend/kernel_compiler/gpu/arrays/slice_gpu_kernel.cc
View File

@ -24,11 +24,5 @@ MS_REG_GPU_KERNEL_ONE(Slice, KernelAttr().AddInputAttr(kNumberTypeInt32).AddOutp
SliceGpuFwdKernel, int)
MS_REG_GPU_KERNEL_ONE(Slice, KernelAttr().AddInputAttr(kNumberTypeFloat16).AddOutputAttr(kNumberTypeFloat16),
SliceGpuFwdKernel, half)
MS_REG_GPU_KERNEL_ONE(StridedSlice, KernelAttr().AddInputAttr(kNumberTypeFloat32).AddOutputAttr(kNumberTypeFloat32),
SliceGpuFwdKernel, float)
MS_REG_GPU_KERNEL_ONE(StridedSlice, KernelAttr().AddInputAttr(kNumberTypeFloat16).AddOutputAttr(kNumberTypeFloat16),
SliceGpuFwdKernel, half)
MS_REG_GPU_KERNEL_ONE(StridedSlice, KernelAttr().AddInputAttr(kNumberTypeInt32).AddOutputAttr(kNumberTypeInt32),
SliceGpuFwdKernel, int)
} // namespace kernel
} // namespace mindspore

11
mindspore/ccsrc/backend/kernel_compiler/gpu/arrays/slice_gpu_kernel.h
View File

@ -41,14 +41,9 @@ class SliceGpuFwdKernel : public GpuKernel {
}
T *input = GetDeviceAddress<T>(inputs, 0);
T *output = GetDeviceAddress<T>(outputs, 0);
if (is_strided_slice_) {
CalStridedSlice(output_size_ / sizeof(T), input, input_shape_, begin_, size_, strides_, output,
reinterpret_cast<cudaStream_t>(stream_ptr));
} else {
Slice4DKernel(begin_[0], begin_[1], begin_[2], begin_[3], size_[0], size_[1], size_[2], size_[3], input_shape_[0],
input_shape_[1], input_shape_[2], input_shape_[3], input, output,
reinterpret_cast<cudaStream_t>(stream_ptr));
}
Slice4DKernel(begin_[0], begin_[1], begin_[2], begin_[3], size_[0], size_[1], size_[2], size_[3], input_shape_[0],
input_shape_[1], input_shape_[2], input_shape_[3], input, output,
reinterpret_cast<cudaStream_t>(stream_ptr));
return true;
}
bool Init(const CNodePtr &kernel_node) override {

6
mindspore/ccsrc/backend/kernel_compiler/gpu/arrays/slice_grad_gpu_kernel.cc
View File

@ -29,11 +29,5 @@ MS_REG_GPU_KERNEL_ONE(
SliceGrad,
KernelAttr().AddInputAttr(kNumberTypeFloat16).AddInputAttr(kNumberTypeFloat16).AddOutputAttr(kNumberTypeFloat16),
SliceGradGpuKernel, half)
MS_REG_GPU_KERNEL_ONE(StridedSliceGrad, KernelAttr().AddInputAttr(kNumberTypeFloat32).AddOutputAttr(kNumberTypeFloat32),
SliceGradGpuKernel, float)
MS_REG_GPU_KERNEL_ONE(StridedSliceGrad, KernelAttr().AddInputAttr(kNumberTypeInt32).AddOutputAttr(kNumberTypeInt32),
SliceGradGpuKernel, int)
MS_REG_GPU_KERNEL_ONE(StridedSliceGrad, KernelAttr().AddInputAttr(kNumberTypeFloat16).AddOutputAttr(kNumberTypeFloat16),
SliceGradGpuKernel, half)
} // namespace kernel
} // namespace mindspore

11
mindspore/ccsrc/backend/kernel_compiler/gpu/arrays/slice_grad_gpu_kernel.h
View File

@ -38,13 +38,8 @@ class SliceGradGpuKernel : public GpuKernel {
T *dy = GetDeviceAddress<T>(inputs, 0);
T *dx = GetDeviceAddress<T>(outputs, 0);
FillDeviceArray(outputs[0]->size / sizeof(T), dx, 0.f, reinterpret_cast<cudaStream_t>(stream_ptr));
if (is_strided_slice_) {
CalStridedSliceGrad(output_size_ / sizeof(T), dy, input_shape_, begin_, size_, strides_, dx,
reinterpret_cast<cudaStream_t>(stream_ptr));
} else {
CalSliceGrad(output_size_ / sizeof(T), dy, input_shape_, begin_, size_, dx,
reinterpret_cast<cudaStream_t>(stream_ptr));
}
CalSliceGrad(output_size_ / sizeof(T), dy, input_shape_, begin_, size_, dx,
reinterpret_cast<cudaStream_t>(stream_ptr));
return true;
}
@ -140,7 +135,7 @@ class SliceGradGpuKernel : public GpuKernel {
size_t input_size_;
size_t output_size_;
size_t workspace_size_;
};
}; // namespace kernel
} // namespace kernel
} // namespace mindspore

28
mindspore/ccsrc/backend/kernel_compiler/gpu/arrays/strided_slice_gpu_kernel.cc Normal file
View File

@ -0,0 +1,28 @@
/**
* Copyright 2020 Huawei Technologies Co., Ltd
*
* 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 "backend/kernel_compiler/gpu/arrays/strided_slice_gpu_kernel.h"
namespace mindspore {
namespace kernel {
MS_REG_GPU_KERNEL_ONE(StridedSlice, KernelAttr().AddInputAttr(kNumberTypeFloat32).AddOutputAttr(kNumberTypeFloat32),
StridedSliceGpuKernel, float)
MS_REG_GPU_KERNEL_ONE(StridedSlice, KernelAttr().AddInputAttr(kNumberTypeFloat16).AddOutputAttr(kNumberTypeFloat16),
StridedSliceGpuKernel, half)
MS_REG_GPU_KERNEL_ONE(StridedSlice, KernelAttr().AddInputAttr(kNumberTypeInt32).AddOutputAttr(kNumberTypeInt32),
StridedSliceGpuKernel, int)
} // namespace kernel
} // namespace mindspore

190
mindspore/ccsrc/backend/kernel_compiler/gpu/arrays/strided_slice_gpu_kernel.h Normal file
View File

@ -0,0 +1,190 @@
/**
* Copyright 2020 Huawei Technologies Co., Ltd
*
* 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.
*/
#ifndef MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_GPU_STRIDED_SLICE_GPU_KERNEL_H
#define MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_GPU_STRIDED_SLICE_GPU_KERNEL_H
#include <vector>
#include <bitset>
#include <algorithm>
#include "backend/kernel_compiler/gpu/gpu_kernel.h"
#include "backend/kernel_compiler/gpu/gpu_kernel_factory.h"
#include "backend/kernel_compiler/gpu/cuda_impl/slice_impl.cuh"
namespace mindspore {
namespace kernel {
constexpr int MAX_DIMS = 4;
template <typename T>
class StridedSliceGpuKernel : public GpuKernel {
public:
StridedSliceGpuKernel() : null_output_(false) {}
~StridedSliceGpuKernel() override = default;
const std::vector<size_t> &GetInputSizeList() const override { return input_size_list_; }
const std::vector<size_t> &GetOutputSizeList() const override { return output_size_list_; }
const std::vector<size_t> &GetWorkspaceSizeList() const override { return workspace_size_list_; }
bool Launch(const std::vector<AddressPtr> &inputs, const std::vector<AddressPtr> &,
const std::vector<AddressPtr> &outputs, void *stream_ptr) override {
if (null_output_) {
return true;
}
T *input = GetDeviceAddress<T>(inputs, 0);
T *output = GetDeviceAddress<T>(outputs, 0);
StridedSlice(input_shape_, begin_, strides_, output_shape_, input, output,
reinterpret_cast<cudaStream_t>(stream_ptr));
return true;
}
bool Init(const CNodePtr &kernel_node) override {
input_shape_ = AnfAlgo::GetPrevNodeOutputInferShape(kernel_node, 0);
if (input_shape_.size() > MAX_DIMS) {
MS_LOG(ERROR) << "StridedSlice support support dims less than " << input_shape_.size();
return false;
}
FillEmptyDims(kernel_node);
ParseMasks(kernel_node);
FillOutputDim();
null_output_ = IsNullOutput();
InitSizeLists();
return true;
}
protected:
void InitSizeLists() override {
input_size_list_.push_back(input_shape_[0] * input_shape_[1] * input_shape_[2] * input_shape_[3] * sizeof(T));
output_size_list_.push_back(output_shape_[0] * output_shape_[1] * output_shape_[2] * output_shape_[3] * sizeof(T));
}
private:
void FillEmptyDims(const CNodePtr &kernel_node) {
begin_ = GetAttr<std::vector<int>>(kernel_node, "begin");
end_ = GetAttr<std::vector<int>>(kernel_node, "end");
strides_ = GetAttr<std::vector<int>>(kernel_node, "strides");
for (size_t i = 0; i < MAX_DIMS; i++) {
if (i < begin_.size()) {
begin_[i] =
std::min(begin_[i] < 0 ? SizeToInt(begin_[i] + input_shape_[i]) : begin_[i], SizeToInt(input_shape_[i] - 1));
} else {
begin_.push_back(0);
}
if (i < end_.size()) {
end_[i] = std::max(end_[i] < 0 ? end_[i] + SizeToInt(input_shape_[i]) : end_[i], -1);
} else {
end_.push_back(i < input_shape_.size() ? input_shape_[i] : 1);
}
if (i >= strides_.size()) {
strides_.push_back(1);
}
if (i >= input_shape_.size()) {
input_shape_.push_back(1);
}
}
}
void ParseMasks(const CNodePtr &kernel_node) {
auto begin_mask_int = GetAttr<int>(kernel_node, "begin_mask");
auto begin_mask = Dec2Bin(begin_mask_int);
for (size_t i = 0; i < begin_mask.size(); i++) {
if (begin_mask[i]) {
begin_[i] = 0;
}
}
auto end_mask_int = GetAttr<int>(kernel_node, "end_mask");
auto end_mask = Dec2Bin(end_mask_int);
for (size_t j = 0; j < end_mask.size(); j++) {
if (end_mask[j]) {
end_[j] = input_shape_[j];
}
}
auto ellipsis_mask_int = GetAttr<int>(kernel_node, "ellipsis_mask");
auto ellipsis_mask = Dec2Bin(ellipsis_mask_int);
for (size_t k = 0; k < ellipsis_mask.size(); k++) {
if (ellipsis_mask[k]) {
begin_[k] = 0;
end_[k] = input_shape_[k];
strides_[k] = 1;
}
}
auto shrink_axis_mask_str = GetAttr<int>(kernel_node, "shrink_axis_mask");
auto shrink_axis_mask = Dec2Bin(shrink_axis_mask_str);
for (size_t l = 0; l < shrink_axis_mask.size(); l++) {
if (shrink_axis_mask[l]) {
end_[l] = end_[l] > begin_[l] ? begin_[l] + 1 : begin_[l] - 1;
strides_[l] = end_[l] > begin_[l] ? 1 : -1;
}
}
}
std::vector<bool> Dec2Bin(const int &mask) {
auto mask_str = std::bitset<MAX_DIMS>(mask).to_string();
int dim_idx = 0;
std::vector<bool> result = {false, false, false, false};
for (int i = mask_str.size() - 1; i >= 0; i--) {
if (mask_str[i] == '1') {
result[dim_idx] = true;
}
dim_idx++;
}
return result;
}
void FillOutputDim() {
for (int i = 0; i < MAX_DIMS; i++) {
if (begin_[i] <= end_[i] && strides_[i] > 0) {
output_shape_.push_back((end_[i] - 1 - begin_[i]) / strides_[i] + 1);
} else if (begin_[i] > end_[i] && strides_[i] < 0) {
output_shape_.push_back((end_[i] - begin_[i] + 1) / strides_[i] + 1);
} else {
output_shape_.push_back(0);
}
}
}
bool IsNullOutput() {
for (int i = 0; i < MAX_DIMS; i++) {
if (begin_[i] >= end_[i] && strides_[i] > 0) {
return true;
}
if (begin_[i] < end_[i] && strides_[i] < 0) {
return true;
}
}
return false;
}
std::vector<int> begin_;
std::vector<int> end_;
std::vector<int> strides_;
std::vector<size_t> input_shape_;
std::vector<int> output_shape_;
int null_output_;
std::vector<size_t> input_size_list_;
std::vector<size_t> output_size_list_;
std::vector<size_t> workspace_size_list_;
};
} // namespace kernel
} // namespace mindspore
#endif // MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_GPU_STRIDED_SLICE_GPU_KERNEL_H

28
mindspore/ccsrc/backend/kernel_compiler/gpu/arrays/strided_slice_grad_gpu_kernel.cc Normal file
View File

@ -0,0 +1,28 @@
/**
* Copyright 2020 Huawei Technologies Co., Ltd
*
* 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 "backend/kernel_compiler/gpu/arrays/strided_slice_grad_gpu_kernel.h"
namespace mindspore {
namespace kernel {
MS_REG_GPU_KERNEL_ONE(StridedSliceGrad, KernelAttr().AddInputAttr(kNumberTypeFloat32).AddOutputAttr(kNumberTypeFloat32),
StridedSliceGradGpuKernel, float)
MS_REG_GPU_KERNEL_ONE(StridedSliceGrad, KernelAttr().AddInputAttr(kNumberTypeFloat16).AddOutputAttr(kNumberTypeFloat16),
StridedSliceGradGpuKernel, half)
MS_REG_GPU_KERNEL_ONE(StridedSliceGrad, KernelAttr().AddInputAttr(kNumberTypeInt32).AddOutputAttr(kNumberTypeInt32),
StridedSliceGradGpuKernel, int)
} // namespace kernel
} // namespace mindspore

191
mindspore/ccsrc/backend/kernel_compiler/gpu/arrays/strided_slice_grad_gpu_kernel.h Normal file
View File

@ -0,0 +1,191 @@
/**
* Copyright 2020 Huawei Technologies Co., Ltd
*
* 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.
*/
#ifndef MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_GPU_STRIDED_SLICE_GRAD_GPU_KERNEL_H
#define MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_GPU_STRIDED_SLICE_GRAD_GPU_KERNEL_H
#include <vector>
#include <bitset>
#include <algorithm>
#include "backend/kernel_compiler/gpu/gpu_kernel.h"
#include "backend/kernel_compiler/gpu/gpu_kernel_factory.h"
#include "backend/kernel_compiler/gpu/cuda_impl/slice_impl.cuh"
namespace mindspore {
namespace kernel {
constexpr int MAX_DIMS = 4;
template <typename T>
class StridedSliceGradGpuKernel : public GpuKernel {
public:
StridedSliceGradGpuKernel() : null_output_(false) {}
~StridedSliceGradGpuKernel() override = default;
const std::vector<size_t> &GetInputSizeList() const override { return input_size_list_; }
const std::vector<size_t> &GetOutputSizeList() const override { return output_size_list_; }
const std::vector<size_t> &GetWorkspaceSizeList() const override { return workspace_size_list_; }
bool Launch(const std::vector<AddressPtr> &inputs, const std::vector<AddressPtr> &,
const std::vector<AddressPtr> &outputs, void *stream_ptr) override {
T *dy = GetDeviceAddress<T>(inputs, 0);
T *dx = GetDeviceAddress<T>(outputs, 0);
FillDeviceArray(outputs[0]->size / sizeof(T), dx, 0.f, reinterpret_cast<cudaStream_t>(stream_ptr));
if (null_output_) {
return true;
}
StridedSliceGrad(output_shape_, begin_, strides_, input_shape_, dy, dx, reinterpret_cast<cudaStream_t>(stream_ptr));
return true;
}
bool Init(const CNodePtr &kernel_node) override {
input_shape_ = GetAttr<std::vector<int>>(kernel_node, "shapex");
if (input_shape_.size() > MAX_DIMS) {
MS_LOG(ERROR) << "StridedSliceGrad support support dims less than " << input_shape_.size();
return false;
}
FillEmptyDims(kernel_node);
ParseMasks(kernel_node);
FillOutputDim();
null_output_ = IsNullOutput();
InitSizeLists();
return true;
}
protected:
void InitSizeLists() override {
input_size_list_.push_back(output_shape_[0] * output_shape_[1] * output_shape_[2] * output_shape_[3] * sizeof(T));
output_size_list_.push_back(input_shape_[0] * input_shape_[1] * input_shape_[2] * input_shape_[3] * sizeof(T));
}
private:
void FillEmptyDims(const CNodePtr &kernel_node) {
begin_ = GetAttr<std::vector<int>>(kernel_node, "begin");
end_ = GetAttr<std::vector<int>>(kernel_node, "end");
strides_ = GetAttr<std::vector<int>>(kernel_node, "strides");
for (size_t i = 0; i < MAX_DIMS; i++) {
if (i < begin_.size()) {
begin_[i] =
std::min(begin_[i] < 0 ? SizeToInt(begin_[i] + input_shape_[i]) : begin_[i], SizeToInt(input_shape_[i] - 1));
} else {
begin_.push_back(0);
}
if (i < end_.size()) {
end_[i] = std::max(end_[i] < 0 ? end_[i] + SizeToInt(input_shape_[i]) : end_[i], -1);
} else {
end_.push_back(i < input_shape_.size() ? input_shape_[i] : 1);
}
if (i >= strides_.size()) {
strides_.push_back(1);
}
if (i >= input_shape_.size()) {
input_shape_.push_back(1);
}
}
}
void ParseMasks(const CNodePtr &kernel_node) {
auto begin_mask_int = GetAttr<int>(kernel_node, "begin_mask");
auto begin_mask = Dec2Bin(begin_mask_int);
for (size_t i = 0; i < begin_mask.size(); i++) {
if (begin_mask[i]) {
begin_[i] = 0;
}
}
auto end_mask_int = GetAttr<int>(kernel_node, "end_mask");
auto end_mask = Dec2Bin(end_mask_int);
for (size_t j = 0; j < end_mask.size(); j++) {
if (end_mask[j]) {
end_[j] = input_shape_[j];
}
}
auto ellipsis_mask_int = GetAttr<int>(kernel_node, "ellipsis_mask");
auto ellipsis_mask = Dec2Bin(ellipsis_mask_int);
for (size_t k = 0; k < ellipsis_mask.size(); k++) {
if (ellipsis_mask[k]) {
begin_[k] = 0;
end_[k] = input_shape_[k];
strides_[k] = 1;
}
}
auto shrink_axis_mask_str = GetAttr<int>(kernel_node, "shrink_axis_mask");
auto shrink_axis_mask = Dec2Bin(shrink_axis_mask_str);
for (size_t l = 0; l < shrink_axis_mask.size(); l++) {
if (shrink_axis_mask[l]) {
end_[l] = end_[l] > begin_[l] ? begin_[l] + 1 : begin_[l] - 1;
strides_[l] = end_[l] > begin_[l] ? 1 : -1;
}
}
}
std::vector<bool> Dec2Bin(const int &mask) {
auto mask_str = std::bitset<MAX_DIMS>(mask).to_string();
int dim_idx = 0;
std::vector<bool> result = {false, false, false, false};
for (int i = mask_str.size() - 1; i >= 0; i--) {
if (mask_str[i] == '1') {
result[dim_idx] = true;
}
dim_idx++;
}
return result;
}
void FillOutputDim() {
for (int i = 0; i < MAX_DIMS; i++) {
if (begin_[i] <= end_[i] && strides_[i] > 0) {
output_shape_.push_back((end_[i] - 1 - begin_[i]) / strides_[i] + 1);
} else if (begin_[i] > end_[i] && strides_[i] < 0) {
output_shape_.push_back((end_[i] - begin_[i] + 1) / strides_[i] + 1);
} else {
output_shape_.push_back(0);
}
}
}
bool IsNullOutput() {
for (int i = 0; i < MAX_DIMS; i++) {
if (begin_[i] >= end_[i] && strides_[i] > 0) {
return true;
}
if (begin_[i] < end_[i] && strides_[i] < 0) {
return true;
}
}
return false;
}
std::vector<int> begin_;
std::vector<int> end_;
std::vector<int> strides_;
std::vector<int> input_shape_;
std::vector<int> output_shape_;
int null_output_;
std::vector<size_t> input_size_list_;
std::vector<size_t> output_size_list_;
std::vector<size_t> workspace_size_list_;
};
} // namespace kernel
} // namespace mindspore
#endif // MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_GPU_STRIDED_SLICE_GRAD_GPU_KERNEL_H

216
mindspore/ccsrc/backend/kernel_compiler/gpu/cuda_impl/slice_impl.cu Executable file → Normal file
View File

@ -21,48 +21,29 @@
#include "backend/kernel_compiler/gpu/cuda_impl/slice_impl.cuh"
template <typename T>
__global__ void Slice4D(const int s1, const int s2, const int s3, const int s4,
const int l1, const int l2, const int l3, const int l4,
const int d1, const int d2, const int d3, const int d4,
const T *input, T *output) {
__global__ void Slice4D(const int s1, const int s2, const int s3, const int s4, const int l1, const int l2,
const int l3, const int l4, const int d1, const int d2, const int d3, const int d4,
const T *input, T *output) {
for (size_t pos = blockIdx.x * blockDim.x + threadIdx.x; pos < (l1 * l2 * l3 * l4); pos += blockDim.x * gridDim.x) {
int i = pos / (l2 * l3 * l4) % l1;
int j = pos / (l3 * l4) % l2;
int k = pos / l4 % l3;
int o = pos % l4;
int offset = (i + s1) * (d2 * d3 * d4) +
(j + s2) * (d3 * d4) +
(k + s3) * d4 +
(o + s4);
int offset = (i + s1) * (d2 * d3 * d4) + (j + s2) * (d3 * d4) + (k + s3) * d4 + (o + s4);
output[pos] = input[offset];
}
}
template <typename T>
__global__ void SliceGrad(const T* dy, int p, int start, int length, T* output) {
__global__ void SliceGrad(const T *dy, int p, int start, int length, T *output) {
for (size_t pos = blockIdx.x * blockDim.x + threadIdx.x; pos < (length); pos += blockDim.x * gridDim.x) {
output[start + pos] = dy[p + pos];
}
return;
}
template <typename T>
__global__ void StridedSlice(const T* input, int p, int start, int begin, int stride, int ended, T* output) {
for (size_t pos = blockIdx.x * blockDim.x + threadIdx.x; pos < std::ceil(static_cast<float>(ended - begin) / stride);
pos += blockDim.x * gridDim.x) {
output[p + pos] = input[start + pos * stride];
}
return;
}
template <typename T>
__global__ void StridedSliceGrad(const T* dy, int p, int start, int begin, int stride, int ended, T* dx) {
for (size_t pos = blockIdx.x * blockDim.x + threadIdx.x; pos < std::ceil(static_cast<float>(ended - begin) / stride);
pos += blockDim.x * gridDim.x) {
dx[start + pos * stride] = dy[p + pos];
}
return;
}
template <typename T>
__global__ void FillArray(T* addr, const size_t len, const float value) {
__global__ void FillArray(T *addr, const size_t len, const float value) {
T value_ = static_cast<T>(value);
for (size_t pos = blockIdx.x * blockDim.x + threadIdx.x; pos < len; pos += blockDim.x * gridDim.x) {
addr[pos] = value_;
@ -70,21 +51,20 @@ __global__ void FillArray(T* addr, const size_t len, const float value) {
return;
}
template <typename T>
void FillDeviceArray(const size_t input_size, T* addr, const float value, cudaStream_t cuda_stream) {
void FillDeviceArray(const size_t input_size, T *addr, const float value, cudaStream_t cuda_stream) {
FillArray<<<GET_BLOCKS(input_size), GET_THREADS, 0, cuda_stream>>>(addr, input_size, value);
return;
}
template <typename T>
void Slice4DKernel(const int s1, const int s2, const int s3, const int s4,
const int l1, const int l2, const int l3, const int l4,
const int d1, const int d2, const int d3, const int d4,
const T *input, T *output, cudaStream_t stream) {
Slice4D<<<GET_BLOCKS(l1 * l2 * l3 * l4), GET_THREADS, 0, stream>>>(s1, s2, s3, s4, l1, l2, l3, l4,
d1, d2, d3, d4, input, output);
void Slice4DKernel(const int s1, const int s2, const int s3, const int s4, const int l1, const int l2, const int l3,
const int l4, const int d1, const int d2, const int d3, const int d4, const T *input, T *output,
cudaStream_t stream) {
Slice4D<<<GET_BLOCKS(l1 * l2 * l3 * l4), GET_THREADS, 0, stream>>>(s1, s2, s3, s4, l1, l2, l3, l4, d1, d2, d3, d4,
input, output);
}
template <typename T>
void CalSliceGrad(const size_t input_size, const T* dy, const std::vector<int> in_shape, const std::vector<int> begin,
const std::vector<int> size, T* output, cudaStream_t cuda_stream) {
void CalSliceGrad(const size_t input_size, const T *dy, const std::vector<int> in_shape, const std::vector<int> begin,
const std::vector<int> size, T *output, cudaStream_t cuda_stream) {
int block = in_shape[1] * in_shape[2] * in_shape[3];
int map = in_shape[2] * in_shape[3];
int w = in_shape[3];
@ -100,92 +80,100 @@ void CalSliceGrad(const size_t input_size, const T* dy, const std::vector<int> i
}
}
}
template <typename T>
void CalStridedSlice(const size_t input_size, const T* input, const std::vector<int> in_shape,
const std::vector<int> begin, const std::vector<int> end, const std::vector<int> strides,
T* output, cudaStream_t cuda_stream) {
int block = in_shape[1] * in_shape[2] * in_shape[3];
int map = in_shape[2] * in_shape[3];
int w = in_shape[3];
int ended = end[3];
int p = 0;
int start = 0;
for (int i = begin[0]; i < ((end[0] > begin[0]) ? end[0] : (2 * begin[0] - end[0])); i += std::abs(strides[0])) {
for (int j = begin[1]; j < ((end[1] > begin[1]) ? end[1] : (2 * begin[1] - end[1])); j += std::abs(strides[1])) {
for (int k = begin[2]; k < ((end[2] > begin[2]) ? end[2] : (2 * begin[2] - end[2])); k += std::abs(strides[2])) {
start = (strides[0] > 0 ? i : 2 * begin[0] - i) * block + (strides[1] > 0 ? j : 2 * begin[1] - j) * map +
(strides[2] > 0 ? k : 2 * begin[2] - k) * w + begin[3];
StridedSlice<<<GET_BLOCKS(input_size), GET_THREADS, 0, cuda_stream>>>(input, p, start, begin[3], strides[3],
ended, output);
p = p + std::ceil(static_cast<float>(end[3] - begin[3]) / strides[3]);
}
}
}
}
template <typename T>
void CalStridedSliceGrad(const size_t input_size, const T* dy, const std::vector<int> in_shape,
const std::vector<int> begin, const std::vector<int> end, const std::vector<int> strides,
T* dx, cudaStream_t cuda_stream) {
int block = in_shape[1] * in_shape[2] * in_shape[3];
int map = in_shape[2] * in_shape[3];
int w = in_shape[3];
int ended = end[3];
int p = 0;
int start = 0;
for (int i = begin[0]; i < ((end[0] > begin[0]) ? end[0] : (2 * begin[0] - end[0] + 1)); i += std::abs(strides[0])) {
for (int j = begin[1]; j < ((end[1] > begin[1]) ? end[1] : (2 * begin[1] - end[1] + 1));
j += std::abs(strides[1])) {
for (int k = begin[2]; k < ((end[2] > begin[2]) ? end[2] : (2 * begin[2] - end[2] + 1));
k += std::abs(strides[2])) {
start = (strides[0] > 0 ? i : 2 * begin[0] - i) * block + (strides[1] > 0 ? j : 2 * begin[1] - j) * map +
(strides[2] > 0 ? k : 2 * begin[2] - k) * w + begin[3];
StridedSliceGrad<<<GET_BLOCKS(input_size), GET_THREADS, 0, cuda_stream>>>(dy, p, start, begin[3], strides[3],
ended, dx);
p = p + std::ceil(static_cast<float>(end[3] - begin[3]) / strides[3]);
}
}
__global__ void StridedSliceKernel(const int b0, const int b1, const int b2, const int b3, const int s0, const int s1,
const int s2, const int s3, const int i0, const int i1, const int i2, const int i3,
const int o0, const int o1, const int o2, const int o3, const T *input_addr,
T *output_addr) {
int output_num = o0 * o1 * o2 * o3;
for (size_t pos = blockIdx.x * blockDim.x + threadIdx.x; pos < output_num; pos += blockDim.x * gridDim.x) {
int i = pos / (o1 * o2 * o3) % o0;
int j = pos / (o2 * o3) % o1;
int k = pos / o3 % o2;
int l = pos % o3;
int input_idx = (i * s0 + b0) * i1 * i2 * i3 + (j * s1 + b1) * i2 * i3 + (k * s2 + b2) * i3 + (l * s3 + b3);
output_addr[pos] = input_addr[input_idx];
}
}
template void FillDeviceArray<float>(const size_t input_size, float* addr, const float value, cudaStream_t cuda_stream);
template void Slice4DKernel(const int s1, const int s2, const int s3, const int s4,
const int l1, const int l2, const int l3, const int l4,
const int d1, const int d2, const int d3, const int d4,
template <typename T>
void StridedSlice(const std::vector<size_t> &input_shape, const std::vector<int> &begin,
const std::vector<int> &strides, const std::vector<int> &output_shape, const T *input, T *output,
cudaStream_t cuda_stream) {
int size = output_shape[0] * output_shape[1] * output_shape[2] * output_shape[3];
StridedSliceKernel<<<GET_BLOCKS(size), GET_THREADS, 0, cuda_stream>>>(
begin[0], begin[1], begin[2], begin[3], strides[0], strides[1], strides[2], strides[3], input_shape[0],
input_shape[1], input_shape[2], input_shape[3], output_shape[0], output_shape[1], output_shape[2], output_shape[3],
input, output);
}
template <typename T>
__global__ void StridedSliceGradKernel(const int b0, const int b1, const int b2, const int b3, const int s0,
const int s1, const int s2, const int s3, const int i0, const int i1,
const int i2, const int i3, const int o0, const int o1, const int o2,
const int o3, const T *dy, T *dx) {
int output_num = o0 * o1 * o2 * o3;
for (size_t pos = blockIdx.x * blockDim.x + threadIdx.x; pos < output_num; pos += blockDim.x * gridDim.x) {
int i = pos / (o1 * o2 * o3) % o0;
int j = pos / (o2 * o3) % o1;
int k = pos / o3 % o2;
int l = pos % o3;
int input_idx = (i * s0 + b0) * i1 * i2 * i3 + (j * s1 + b1) * i2 * i3 + (k * s2 + b2) * i3 + (l * s3 + b3);
dx[input_idx] = dy[pos];
}
return;
}
template <typename T>
void StridedSliceGrad(const std::vector<int> &dy_shape, const std::vector<int> &begin, const std::vector<int> &strides,
const std::vector<int> &dx_shape, const T *dy, T *dx, cudaStream_t cuda_stream) {
int size = dy_shape[0] * dy_shape[1] * dy_shape[2] * dy_shape[3];
StridedSliceGradKernel<<<GET_BLOCKS(size), GET_THREADS, 0, cuda_stream>>>(
begin[0], begin[1], begin[2], begin[3], strides[0], strides[1], strides[2], strides[3], dx_shape[0], dx_shape[1],
dx_shape[2], dx_shape[3], dy_shape[0], dy_shape[1], dy_shape[2], dy_shape[3], dy, dx);
}
template void FillDeviceArray<float>(const size_t input_size, float *addr, const float value, cudaStream_t cuda_stream);
template void Slice4DKernel(const int s1, const int s2, const int s3, const int s4, const int l1, const int l2,
const int l3, const int l4, const int d1, const int d2, const int d3, const int d4,
const float *input, float *output, cudaStream_t stream);
template void CalSliceGrad<float>(const size_t input_size, const float* dy, const std::vector<int> in_shape,
const std::vector<int> begin, const std::vector<int> size, float* output,
template void CalSliceGrad<float>(const size_t input_size, const float *dy, const std::vector<int> in_shape,
const std::vector<int> begin, const std::vector<int> size, float *output,
cudaStream_t cuda_stream);
template void CalStridedSlice<float>(const size_t input_size, const float* input, const std::vector<int> in_shape,
const std::vector<int> begin, const std::vector<int> end,
const std::vector<int> strides, float* output, cudaStream_t cuda_stream);
template void CalStridedSliceGrad<float>(const size_t input_size, const float* dy, const std::vector<int> in_shape,
const std::vector<int> begin, const std::vector<int> end,
const std::vector<int> strides, float* dx, cudaStream_t cuda_stream);
template void FillDeviceArray<half>(const size_t input_size, half* addr, const float value, cudaStream_t cuda_stream);
template void Slice4DKernel(const int s1, const int s2, const int s3, const int s4,
const int l1, const int l2, const int l3, const int l4,
const int d1, const int d2, const int d3, const int d4,
template void FillDeviceArray<half>(const size_t input_size, half *addr, const float value, cudaStream_t cuda_stream);
template void Slice4DKernel(const int s1, const int s2, const int s3, const int s4, const int l1, const int l2,
const int l3, const int l4, const int d1, const int d2, const int d3, const int d4,
const half *input, half *output, cudaStream_t stream);
template void CalSliceGrad<half>(const size_t input_size, const half* dy, const std::vector<int> in_shape,
const std::vector<int> begin, const std::vector<int> size, half* output,
template void CalSliceGrad<half>(const size_t input_size, const half *dy, const std::vector<int> in_shape,
const std::vector<int> begin, const std::vector<int> size, half *output,
cudaStream_t cuda_stream);
template void CalStridedSlice<half>(const size_t input_size, const half* input, const std::vector<int> in_shape,
const std::vector<int> begin, const std::vector<int> end,
const std::vector<int> strides, half* output, cudaStream_t cuda_stream);
template void CalStridedSliceGrad<half>(const size_t input_size, const half* dy, const std::vector<int> in_shape,
const std::vector<int> begin, const std::vector<int> end,
const std::vector<int> strides, half* dx, cudaStream_t cuda_stream);
template void FillDeviceArray<int>(const size_t input_size, int* addr, const float value, cudaStream_t cuda_stream);
template void Slice4DKernel(const int s1, const int s2, const int s3, const int s4,
const int l1, const int l2, const int l3, const int l4,
const int d1, const int d2, const int d3, const int d4,
template void FillDeviceArray<int>(const size_t input_size, int *addr, const float value, cudaStream_t cuda_stream);
template void Slice4DKernel(const int s1, const int s2, const int s3, const int s4, const int l1, const int l2,
const int l3, const int l4, const int d1, const int d2, const int d3, const int d4,
const int *input, int *output, cudaStream_t stream);
template void CalSliceGrad<int>(const size_t input_size, const int* dy, const std::vector<int> in_shape,
const std::vector<int> begin, const std::vector<int> size, int* output,
template void CalSliceGrad<int>(const size_t input_size, const int *dy, const std::vector<int> in_shape,
const std::vector<int> begin, const std::vector<int> size, int *output,
cudaStream_t cuda_stream);
template void CalStridedSlice<int>(const size_t input_size, const int* input, const std::vector<int> in_shape,
const std::vector<int> begin, const std::vector<int> end,
const std::vector<int> strides, int* output, cudaStream_t cuda_stream);
template void CalStridedSliceGrad<int>(const size_t input_size, const int* dy, const std::vector<int> in_shape,
const std::vector<int> begin, const std::vector<int> end,
const std::vector<int> strides, int* dx, cudaStream_t cuda_stream);
template void StridedSlice(const std::vector<size_t> &input_shape, const std::vector<int> &begin,
const std::vector<int> &strides, const std::vector<int> &output_shape, const float *input,
float *output, cudaStream_t cuda_stream);
template void StridedSlice(const std::vector<size_t> &input_shape, const std::vector<int> &begin,
const std::vector<int> &strides, const std::vector<int> &output_shape, const half *input,
half *output, cudaStream_t cuda_stream);
template void StridedSlice(const std::vector<size_t> &input_shape, const std::vector<int> &begin,
const std::vector<int> &strides, const std::vector<int> &output_shape, const int *input,
int *output, cudaStream_t cuda_stream);
template void StridedSliceGrad(const std::vector<int> &dy_shape, const std::vector<int> &begin,
const std::vector<int> &strides, const std::vector<int> &dx_shape, const float *dy,
float *dx, cudaStream_t cuda_stream);
template void StridedSliceGrad(const std::vector<int> &dy_shape, const std::vector<int> &begin,
const std::vector<int> &strides, const std::vector<int> &dx_shape, const half *dy,
half *dx, cudaStream_t cuda_stream);
template void StridedSliceGrad(const std::vector<int> &dy_shape, const std::vector<int> &begin,
const std::vector<int> &strides, const std::vector<int> &dx_shape, const int *dy,
int *dx, cudaStream_t cuda_stream);

25
mindspore/ccsrc/backend/kernel_compiler/gpu/cuda_impl/slice_impl.cuh Executable file → Normal file
View File

@ -21,23 +21,20 @@
#include <vector>
#include "runtime/device/gpu/cuda_common.h"
template <typename T>
void Slice4DKernel(const int s1, const int s2, const int s3, const int s4,
const int l1, const int l2, const int l3, const int l4,
const int d1, const int d2, const int d3, const int d4,
const T *input, T *output, cudaStream_t stream);
void Slice4DKernel(const int s1, const int s2, const int s3, const int s4, const int l1, const int l2, const int l3,
const int l4, const int d1, const int d2, const int d3, const int d4, const T *input, T *output,
cudaStream_t stream);
template <typename T>
void CalSliceGrad(const size_t input_size, const T* input, const std::vector<int> in_shape,
const std::vector<int> begin, const std::vector<int> size, T* output, cudaStream_t cuda_stream);
void CalSliceGrad(const size_t input_size, const T *input, const std::vector<int> in_shape,
const std::vector<int> begin, const std::vector<int> size, T *output, cudaStream_t cuda_stream);
template <typename T>
void CalStridedSlice(const size_t input_size, const T* input, const std::vector<int> in_shape,
const std::vector<int> begin, const std::vector<int> end, const std::vector<int> strides,
T* output, cudaStream_t cuda_stream);
void StridedSlice(const std::vector<size_t> &input_shape, const std::vector<int> &begin,
const std::vector<int> &strides, const std::vector<int> &output_shape, const T *input, T *output,
cudaStream_t cuda_stream);
template <typename T>
void CalStridedSliceGrad(const size_t input_size, const T* dy, const std::vector<int> in_shape,
const std::vector<int> begin, const std::vector<int> end, const std::vector<int> strides,
T* dx, cudaStream_t cuda_stream);
void StridedSliceGrad(const std::vector<int> &dy_shape, const std::vector<int> &begin, const std::vector<int> &strides,
const std::vector<int> &dx_shape, const T *dy, T *dx, cudaStream_t cuda_stream);
template <typename T>
void FillDeviceArray(const size_t input_size, T* addr, const float value, cudaStream_t cuda_stream);
void FillDeviceArray(const size_t input_size, T *addr, const float value, cudaStream_t cuda_stream);
#endif // MINDSPORE_CCSRC_KERNEL_GPU_CUDA_IMPL_SLICEIMPL_H_

261
tests/st/ops/gpu/test_stridedslice_grad_op.py
View File

@ -19,31 +19,258 @@ import pytest
import mindspore.context as context
import mindspore.nn as nn
from mindspore import Tensor
from mindspore.common.api import ms_function
from mindspore.ops import operations as P
from mindspore.ops.operations import _grad_ops as G
from mindspore.ops import composite as C
context.set_context(mode=context.GRAPH_MODE, device_target='GPU')
class StridedSliceGrad(nn.Cell):
def __init__(self):
super(StridedSliceGrad, self).__init__()
self.ssg = G.StridedSliceGrad()
self.shape = P.Shape()
class StridedSliceNet(nn.Cell):
def __init__(self, begin, end, stride, begin_mask=0, end_mask=0, ellipsis_mask=0):
super(StridedSliceNet, self).__init__()
self.begin = begin
self.end = end
self.strides = stride
self.slice = P.StridedSlice(begin_mask, end_mask, ellipsis_mask)
@ms_function
def construct(self, dy, x):
return self.ssg(dy, self.shape(x), (2, 0, 0), (3, 2, 3), (1, 1, 1))
def construct(self, x):
return self.slice(x, self.begin, self.end, self.strides)
class GradData(nn.Cell):
def __init__(self, network):
super(GradData, self).__init__()
self.grad = C.GradOperation(name="get_all", get_all=True, sens_param=False)
self.network = network
def construct(self, x):
return self.grad(self.network)(x)
@pytest.mark.level0
@pytest.mark.platform_x86_gpu_training
@pytest.mark.env_onecard
def test_slice():
x = Tensor(np.array([[[1, 1, 1], [2, 2, 2]], [[3, 3, 3], [4, 4, 4]], [[5, 5, 5], [6, 7, 8]]]).astype(np.float32))
dy = Tensor(np.array([[[5., 1., 5.], [6., 1., 8.]]]).astype(np.float32))
ssg = StridedSliceGrad()
output = ssg(dy, x)
expect = [[[0, 0, 0], [0, 0, 0]], [[0, 0, 0], [0, 0, 0]], [[5, 1, 5], [6, 1, 8]]]
assert (output.asnumpy() == expect).all()
def test_strided_slice_grad():
x = Tensor(np.arange(0, 2*3*4*5).reshape(2, 3, 4, 5).astype(np.float32))
net = StridedSliceNet((1, 0, 0, 2), (2, 2, 2, 4), (1, 1, 1, 1))
dx = GradData(net)(x)
expect = np.array([[[[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.]],
[[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.]],
[[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.]]],
[[[0., 0., 1., 1., 0.],
[0., 0., 1., 1., 0.],
[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.]],
[[0., 0., 1., 1., 0.],
[0., 0., 1., 1., 0.],
[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.]],
[[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.]]]])
assert np.allclose(dx[0].asnumpy(), expect)
net = StridedSliceNet((1, 0, 0, 5), (2, 2, 2, 1), (1, 1, 1, -2))
dx = GradData(net)(x)
expect = np.array([[[[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.]],
[[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.]],
[[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.]]],
[[[0., 0., 1., 0., 1.],
[0., 0., 1., 0., 1.],
[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.]],
[[0., 0., 1., 0., 1.],
[0., 0., 1., 0., 1.],
[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.]],
[[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.]]]])
assert np.allclose(dx[0].asnumpy(), expect)
net = StridedSliceNet((1, 0, 0, -1), (2, 2, 2, 1), (1, 1, 1, -1))
dx = GradData(net)(x)
expect = np.array([[[[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.]],
[[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.]],
[[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.]]],
[[[0., 0., 1., 1., 1.],
[0., 0., 1., 1., 1.],
[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.]],
[[0., 0., 1., 1., 1.],
[0., 0., 1., 1., 1.],
[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.]],
[[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.]]]])
assert np.allclose(dx[0].asnumpy(), expect)
# ME infer fault
# y = GradData()(x, (1, 0, -1, -2), (2, 2, 0, -5), (1, 1, -1, -2))
# expect = np.array([[[[0., 0., 0., 0., 0.],
# [0., 0., 0., 0., 0.],
# [0., 0., 0., 0., 0.],
# [0., 0., 0., 0., 0.]],
# [[0., 0., 0., 0., 0.],
# [0., 0., 0., 0., 0.],
# [0., 0., 0., 0., 0.],
# [0., 0., 0., 0., 0.]],
# [[0., 0., 0., 0., 0.],
# [0., 0., 0., 0., 0.],
# [0., 0., 0., 0., 0.],
# [0., 0., 0., 0., 0.]]],
# [[[0., 0., 0., 0., 0.],
# [0., 1., 0., 1., 0.],
# [0., 1., 0., 1., 0.],
# [0., 1., 0., 1., 0.]],
# [[0., 0., 0., 0., 0.],
# [0., 1., 0., 1., 0.],
# [0., 1., 0., 1., 0.],
# [0., 1., 0., 1., 0.]],begin_mask=0b1000, end_mask=0b0010, ellipsis_mask=0b0100
# [[0., 0., 0., 0., 0.],
# [0., 0., 0., 0., 0.],
# [0., 0., 0., 0., 0.],
# [0., 0., 0., 0., 0.]]]])
# assert np.allclose(y.asnumpy(), expect)
# y = Grad(begin_mask=0b1000, end_mask=0b0010)(x, (1, 0, 0, 2), (2, 2, 2, 4), (1, 1, 1, 1))
# expect = np.array([[[[0., 0., 0., 0., 0.],
# [0., 0., 0., 0., 0.],
# [0., 0., 0., 0., 0.],
# [0., 0., 0., 0., 0.]],
# [[0., 0., 0., 0., 0.],
# [0., 0., 0., 0., 0.],
# [0., 0., 0., 0., 0.],
# [0., 0., 0., 0., 0.]],
# [[0., 0., 0., 0., 0.],
# [0., 0., 0., 0., 0.],
# [0., 0., 0., 0., 0.],
# [0., 0., 0., 0., 0.]]],
# [[[0., 0., 1., 1., 0.],
# [0., 0., 1., 1., 0.],
# [0., 0., 0., 0., 0.],
# [0., 0., 0., 0., 0.]],
# [[0., 0., 1., 1., 0.],
# [0., 0., 1., 1., 0.],
# [0., 0., 0., 0., 0.],
# [0., 0., 0., 0., 0.]],
# [[0., 0., 0., 0., 0.],
# [0., 0., 0., 0., 0.],
# [0., 0., 0., 0., 0.],
# [0., 0., 0., 0., 0.]]]])
# assert np.allclose(y.asnumpy(), expect)
net = StridedSliceNet((1, 0, 0, 2), (2, 2, 2, 4), (1, 1, 1, 1),
begin_mask=0b1000, end_mask=0b0010, ellipsis_mask=0b0100)
dx = GradData(net)(x)
expect = np.array([[[[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.]],
[[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.]],
[[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.]]],
[[[1., 1., 1., 1., 0.],
[1., 1., 1., 1., 0.],
[1., 1., 1., 1., 0.],
[1., 1., 1., 1., 0.]],
[[1., 1., 1., 1., 0.],
[1., 1., 1., 1., 0.],
[1., 1., 1., 1., 0.],
[1., 1., 1., 1., 0.]],
[[1., 1., 1., 1., 0.],
[1., 1., 1., 1., 0.],
[1., 1., 1., 1., 0.],
[1., 1., 1., 1., 0.]]]])
assert np.allclose(dx[0].asnumpy(), expect)
x = Tensor(np.arange(0, 3*4*5).reshape(3, 4, 5).astype(np.float32))
net = StridedSliceNet((1, 0, 0), (2, -3, 3), (1, 1, 3))
dx = GradData(net)(x)
expect = np.array([[[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.]],
[[1., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.]],
[[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.]]])
assert np.allclose(dx[0].asnumpy(), expect)

84
tests/st/ops/gpu/test_stridedslice_op.py
View File

@ -17,29 +17,79 @@ import numpy as np
import pytest
import mindspore.context as context
import mindspore.nn as nn
from mindspore import Tensor
from mindspore.ops import operations as P
context.set_context(mode=context.GRAPH_MODE, device_target='GPU')
class StridedSlice(nn.Cell):
def __init__(self):
super(StridedSlice, self).__init__()
self.stridedslice = P.StridedSlice()
def construct(self, x):
return self.stridedslice(x, (2, 0, 0), (3, 2, 3), (1, 1, 1))
@pytest.mark.level0
@pytest.mark.platform_x86_gpu_training
@pytest.mark.env_onecard
def test_slice():
x = Tensor(np.array([[[1, 1, 1], [2, 2, 2]], [[3, 3, 3], [4, 4, 4]], [[5, 5, 5], [6, 7, 8]]]).astype(np.int32))
stridedslice = StridedSlice()
output = stridedslice(x)
expect = [[[5., 5., 5.],
[6., 7., 8.]]]
assert (output.asnumpy() == expect).all()
def test_stridedslice():
x = Tensor(np.arange(0, 2*3*4*5).reshape(2, 3, 4, 5).astype(np.float32))
y = P.StridedSlice()(x, (1, 0, 0, 2), (2, 2, 2, 4), (1, 1, 1, 1))
expect = np.array([[[[62, 63],
[67, 68]],
[[82, 83],
[87, 88]]]])
assert np.allclose(y.asnumpy(), expect)
y = P.StridedSlice()(x, (1, 0, 0, 5), (2, 2, 2, 1), (1, 1, 1, -2))
expect = np.array([[[[64, 62],
[69, 67]],
[[84, 82],
[89, 87]]]])
assert np.allclose(y.asnumpy(), expect)
y = P.StridedSlice()(x, (1, 0, 0, -1), (2, 2, 2, 1), (1, 1, 1, -1))
expect = np.array([[[[64, 63, 62],
[69, 68, 67]],
[[84, 83, 82],
[89, 88, 87]]]])
assert np.allclose(y.asnumpy(), expect)
# ME infer fault
# y = P.StridedSlice()(x, (1, 0, -1, -2), (2, 2, 0, -5), (1, 1, -1, -2))
# expect = np.array([[[[78, 76],
# [73, 71],
# [68, 66]],
# [[98, 96],
# [93, 91],
# [88, 86]]]])
# assert np.allclose(y.asnumpy(), expect)
# y = P.StridedSlice(begin_mask=0b1000, end_mask=0b0010)(x, (1, 0, 0, 2), (2, 2, 2, 4), (1, 1, 1, 1))
# expect = np.array([[[[ 62, 63],
# [ 67, 68]],
# [[ 82, 83],
# [ 87, 88]],
# [[102, 103],
# [107, 108]]]])
# assert np.allclose(y.asnumpy(), expect)
op = P.StridedSlice(begin_mask=0b1000, end_mask=0b0010, ellipsis_mask=0b0100)
y = op(x, (1, 0, 0, 2), (2, 2, 2, 4), (1, 1, 1, 1))
expect = np.array([[[[60, 61, 62, 63],
[65, 66, 67, 68],
[70, 71, 72, 73],
[75, 76, 77, 78]],
[[80, 81, 82, 83],
[85, 86, 87, 88],
[90, 91, 92, 93],
[95, 96, 97, 98]],
[[100, 101, 102, 103],
[105, 106, 107, 108],
[110, 111, 112, 113],
[115, 116, 117, 118]]]])
assert np.allclose(y.asnumpy(), expect)
x = Tensor(np.arange(0, 3*4*5).reshape(3, 4, 5).astype(np.float32))
y = P.StridedSlice()(x, (1, 0, 0), (2, -3, 3), (1, 1, 3))
expect = np.array([[[20]]])
assert np.allclose(y.asnumpy(), expect)
x_np = np.arange(0, 4*5).reshape(4, 5).astype(np.float32)
y = Tensor(x_np)[:, ::-1]
expect = x_np[:, ::-1]
assert np.allclose(y.asnumpy(), expect)