m53297601
/
mindspore
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Add CumSum Kerner for CPU

Browse Source
pull/10066/head
shaoxiangdong 4 years ago
parent efa7d78b57
commit accf7c4671
4 changed files with 615 additions and 1 deletions
Show Stats Download Patch File Download Diff File

248
mindspore/ccsrc/backend/kernel_compiler/cpu/cumsum_cpu_kernel.cc
Unescape View File

@ -0,0 +1,248 @@
/**
* 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 <thread>
#include "backend/kernel_compiler/cpu/cumsum_cpu_kernel.h"
#include "runtime/device/cpu/cpu_device_address.h"
namespace mindspore {
namespace kernel {
void CumSumCPUKernel::InitKernel(const CNodePtr &kernel_node) {
CheckParam(kernel_node);
shape_ = AnfAlgo::GetPrevNodeOutputInferShape(kernel_node, 0);
dtype_ = AnfAlgo::GetPrevNodeOutputInferDataType(kernel_node, 0);
axis_ = static_cast<int>(AnfAlgo::GetNodeAttr<int64_t>(kernel_node, "axis"));
dst_shape = AnfAlgo::GetOutputInferShape(kernel_node, 0);
exclusive_ = AnfAlgo::GetNodeAttr<bool>(kernel_node, "exclusive");
reverse_ = AnfAlgo::GetNodeAttr<bool>(kernel_node, "reverse");
int input_dim_length = SizeToInt(shape_.size());
if (axis_ >= input_dim_length) {
MS_LOG(EXCEPTION) << "Axis out of bounds.";
}
while (axis_ < 0) {
axis_ += input_dim_length;
}
}
template <typename T>
void CumSumCPUKernel::InitWorkspaceSize() {
input_size_0_ = sizeof(T);
for (size_t i = 0; i < shape_.size(); i++) {
input_size_0_ *= shape_[i];
}
workspace_size_list_.emplace_back(input_size_0_);
}
void CumSumCPUKernel::InitInputOutputSize(const CNodePtr &kernel_node) {
CPUKernel::InitInputOutputSize(kernel_node);
if (dtype_ == kNumberTypeFloat32) {
InitWorkspaceSize<float_t>();
} else if (dtype_ == kNumberTypeFloat16) {
InitWorkspaceSize<float16>();
} else if (dtype_ == kNumberTypeInt32) {
InitWorkspaceSize<int32_t>();
} else if (dtype_ == kNumberTypeInt8) {
InitWorkspaceSize<int8_t>();
} else if (dtype_ == kNumberTypeUInt8) {
InitWorkspaceSize<uint8_t>();
}
}
bool CumSumCPUKernel::Launch(const std::vector<kernel::AddressPtr> &inputs,
const std::vector<kernel::AddressPtr> &workspace,
const std::vector<kernel::AddressPtr> &outputs) {
Reshape();
if (dtype_ == kNumberTypeFloat32) {
LaunchKernel<float_t>(inputs, workspace, outputs);
} else if (dtype_ == kNumberTypeFloat16) {
LaunchKernel<float16>(inputs, workspace, outputs);
} else if (dtype_ == kNumberTypeInt32) {
LaunchKernel<int32_t>(inputs, workspace, outputs);
} else if (dtype_ == kNumberTypeInt8) {
LaunchKernel<int8_t>(inputs, workspace, outputs);
} else if (dtype_ == kNumberTypeUInt8) {
LaunchKernel<uint8_t>(inputs, workspace, outputs);
}
return true;
}
void CumSumCPUKernel::Reshape() {
dims_[0] = 1;
dims_[1] = shape_[IntToSize(axis_)];
dims_[2] = 1;
for (size_t i = 0; i < IntToSize(axis_); i++) {
dims_[0] *= shape_[i];
}
for (size_t i = IntToSize(axis_) + 1; i < shape_.size(); i++) {
dims_[2] *= shape_[i];
}
stride_ = dims_[1] * dims_[2];
stride2_ = dims_[2];
return;
}
template <typename T>
void CumSumCPUKernel::LeftMove(const T *input, T *output, size_t dim0, size_t dim1, size_t dim2, size_t stride,
size_t stride2, size_t start, size_t end) {
for (size_t i = start; i < end; i++) {
size_t k1 = i / dim2 % dim0;
size_t k2 = i % dim2;
size_t offset = k1 * stride + k2;
for (size_t j = 0; j < dim1; ++j) {
size_t read_index = j * stride2 + offset;
if (j == 0) {
output[read_index] = (T)0;
} else {
size_t read_index2 = (j - 1) * stride2 + offset;
output[read_index] = input[read_index2];
}
}
}
}
template <typename T>
void CumSumCPUKernel::RightMove(const T *input, T *output, size_t dim0, size_t dim1, size_t dim2, size_t stride,
size_t stride2, size_t start, size_t end) {
for (size_t i = start; i < end; i++) {
size_t k1 = i / dim2 % dim0;
size_t k2 = i % dim2;
size_t offset = k1 * stride + k2;
for (int j = SizeToInt(dim1 - 1); j >= 0; --j) {
size_t read_index = j * stride2 + offset;
if (j == SizeToInt(dim1 - 1)) {
output[read_index] = (T)0;
} else {
size_t read_index2 = (j + 1) * stride2 + offset;
output[read_index] = input[read_index2];
}
}
}
}
template <typename T>
void CumSumCPUKernel::Copy(T *input, T *output, size_t dim0, size_t dim1, size_t dim2, size_t stride, size_t stride2,
size_t start, size_t end) {
for (size_t i = start; i < end; i++) {
size_t k1 = i / dim2 % dim0;
size_t k2 = i % dim2;
size_t offset = k1 * stride + k2;
for (size_t j = 0; j < dim1; ++j) {
size_t read_index = j * stride2 + offset;
input[read_index] = output[read_index];
}
}
}
template <typename T>
void CumSumCPUKernel::CumSumKernelReverse(const T *input, T *output, size_t dim0, size_t dim1, size_t dim2,
size_t stride, size_t stride2, size_t start, size_t end) {
for (size_t i = start; i < end; i++) {
size_t k1 = i / dim2 % dim0;
size_t k2 = i % dim2;
size_t offset = k1 * stride + k2;
for (int j = SizeToInt(dim1 - 1); j >= 0; --j) {
size_t read_index = j * stride2 + offset;
if (j == SizeToInt(dim1 - 1)) {
output[read_index] = input[read_index];
} else {
size_t read_index2 = (j + 1) * stride2 + offset;
output[read_index] = output[read_index2] + input[read_index];
}
}
}
}
template <typename T>
void CumSumCPUKernel::CumSumKernel(const T *input, T *output, size_t dim0, size_t dim1, size_t dim2, size_t stride,
size_t stride2, size_t start, size_t end) {
for (size_t i = start; i < end; i++) {
size_t k1 = i / dim2 % dim0;
size_t k2 = i % dim2;
size_t offset = k1 * stride + k2;
for (size_t j = 0; j < dim1; ++j) {
size_t read_index = j * stride2 + offset;
if (j == 0) {
output[read_index] = input[read_index];
} else {
size_t read_index2 = (j - 1) * stride2 + offset;
output[read_index] = output[read_index2] + input[read_index];
}
}
}
}
template <typename T>
void CumSumCPUKernel::LaunchCumSum(const T *input, T *output, T *workspace, size_t start, size_t end) {
start = start / dims_[1];
end = end / dims_[1];
if (exclusive_) {
if (reverse_) {
RightMove(input, output, dims_[0], dims_[1], dims_[2], stride_, stride2_, start, end);
Copy(workspace, output, dims_[0], dims_[1], dims_[2], stride_, stride2_, start, end);
CumSumKernelReverse(workspace, output, dims_[0], dims_[1], dims_[2], stride_, stride2_, start, end);
} else {
LeftMove(input, output, dims_[0], dims_[1], dims_[2], stride_, stride2_, start, end);
Copy(workspace, output, dims_[0], dims_[1], dims_[2], stride_, stride2_, start, end);
CumSumKernel(workspace, output, dims_[0], dims_[1], dims_[2], stride_, stride2_, start, end);
}
} else {
if (reverse_) {
CumSumKernelReverse(input, output, dims_[0], dims_[1], dims_[2], stride_, stride2_, start, end);
} else {
CumSumKernel(input, output, dims_[0], dims_[1], dims_[2], stride_, stride2_, start, end);
}
}
return;
}
template <typename T>
void CumSumCPUKernel::LaunchKernel(const std::vector<kernel::AddressPtr> &inputs,
const std::vector<kernel::AddressPtr> &workspace,
const std::vector<kernel::AddressPtr> &outputs) {
auto input = reinterpret_cast<T *>(inputs[0]->addr);
auto ws = reinterpret_cast<T *>(workspace[0]->addr);
auto output = reinterpret_cast<T *>(outputs[0]->addr);
// multithreading
size_t lens = inputs[0]->size > 0 ? static_cast<size_t>(inputs[0]->size / sizeof(T)) : 1;
auto max_thread_num = std::thread::hardware_concurrency();
size_t thread_num = lens < 128 * max_thread_num ? std::ceil(lens / 128.0) : max_thread_num;
MS_LOG(INFO) << "Lens=" << lens << "; use thread_num=" << thread_num << "; max_thread_num: " << max_thread_num;
std::vector<std::thread> threads;
threads.reserve(thread_num);
size_t start = 0;
size_t once_compute_size = (lens + thread_num - 1) / thread_num;
if (thread_num < 1 || once_compute_size < 1) {
MS_LOG(ERROR) << "Invalid value: thread_num " << thread_num << "; once_compute_size " << once_compute_size;
return;
}
while (start < lens) {
size_t end = (start + once_compute_size) > lens ? lens : (start + once_compute_size);
threads.emplace_back(std::thread(&CumSumCPUKernel::LaunchCumSum<T>, this, input, output, ws, start, end));
start += once_compute_size;
}
for (size_t i = 0; i < threads.size(); ++i) {
threads[i].join();
}
return;
}
void CumSumCPUKernel::CheckParam(const CNodePtr &kernel_node) {
size_t input_num = AnfAlgo::GetInputTensorNum(kernel_node);
if (input_num != 1) {
MS_LOG(EXCEPTION) << "Argument number is " << input_num << ", but CumSumGpuKernel needs 1.";
}
}
} // namespace kernel
} // namespace mindspore

95
mindspore/ccsrc/backend/kernel_compiler/cpu/cumsum_cpu_kernel.h
Unescape View File

@ -0,0 +1,95 @@
/**
* 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_CPU_CUMSUM_CPU_KERNEL_H_
#define MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_CPU_CUMSUM_CPU_KERNEL_H_
#include <memory>
#include <vector>
#include "backend/kernel_compiler/cpu/cpu_kernel.h"
#include "backend/kernel_compiler/cpu/cpu_kernel_factory.h"
namespace mindspore {
namespace kernel {
class CumSumCPUKernel : public CPUKernel {
public:
CumSumCPUKernel() = default;
~CumSumCPUKernel() override = default;
void InitKernel(const CNodePtr &kernel_node) override;
void InitInputOutputSize(const CNodePtr &kernel_node) override;
bool Launch(const std::vector<AddressPtr> &inputs, const std::vector<AddressPtr> &workspace,
const std::vector<AddressPtr> &outputs) override;
template <typename T>
void InitWorkspaceSize();
template <typename T>
void LaunchKernel(const std::vector<AddressPtr> &inputs, const std::vector<AddressPtr> &workspace,
const std::vector<AddressPtr> &outputs);
template <typename T>
void LaunchCumSum(const T *input_addr, T *output_addr, T *ws_addr, size_t start, size_t end);
private:
void CheckParam(const CNodePtr &kernel_node);
void Reshape();
template <typename T>
void LeftMove(const T *input, T *output, size_t dim0, size_t dim1, size_t dim2, size_t stride, size_t stride2,
size_t start, size_t end);
template <typename T>
void RightMove(const T *input, T *output, size_t dim0, size_t dim1, size_t dim2, size_t stride, size_t stride2,
size_t start, size_t end);
template <typename T>
void Copy(T *input, T *output, size_t dim0, size_t dim1, size_t dim2, size_t stride, size_t stride2, size_t start,
size_t end);
template <typename T>
void CumSumKernelReverse(const T *input, T *output, size_t dim0, size_t dim1, size_t dim2, size_t stride,
size_t stride2, size_t start, size_t end);
template <typename T>
void CumSumKernel(const T *input, T *output, size_t dim0, size_t dim1, size_t dim2, size_t stride, size_t stride2,
size_t start, size_t end);
std::vector<size_t> shape_;
std::vector<size_t> dst_shape;
size_t input_size_0_;
size_t stride_;
size_t stride2_;
size_t dims_[3] = {};
int exclusive_;
int reverse_;
int axis_;
TypeId dtype_{kTypeUnknown};
};
MS_REG_CPU_KERNEL(CumSum, KernelAttr().AddInputAttr(kNumberTypeFloat32).AddOutputAttr(kNumberTypeFloat32),
CumSumCPUKernel);
MS_REG_CPU_KERNEL(CumSum, KernelAttr().AddInputAttr(kNumberTypeFloat16).AddOutputAttr(kNumberTypeFloat16),
CumSumCPUKernel);
MS_REG_CPU_KERNEL(CumSum, KernelAttr().AddInputAttr(kNumberTypeInt32).AddOutputAttr(kNumberTypeInt32), CumSumCPUKernel);
MS_REG_CPU_KERNEL(CumSum, KernelAttr().AddInputAttr(kNumberTypeInt8).AddOutputAttr(kNumberTypeInt8), CumSumCPUKernel);
MS_REG_CPU_KERNEL(CumSum, KernelAttr().AddInputAttr(kNumberTypeUInt8).AddOutputAttr(kNumberTypeUInt8), CumSumCPUKernel);
} // namespace kernel
} // namespace mindspore
#endif // MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_CPU_CUMSUM_CPU_KERNEL_H_

2
mindspore/ops/operations/math_ops.py
Unescape View File

@ -844,7 +844,7 @@ class CumSum(PrimitiveWithInfer):
Tensor, the shape of the output tensor is consistent with the input tensor's.
Supported Platforms:
``Ascend`` ``GPU``
``Ascend`` ``GPU`` ``CPU``
Examples:
>>> input = Tensor(np.array([[3, 4, 6, 10],[1, 6, 7, 9],[4, 3, 8, 7],[1, 3, 7, 9]]).astype(np.float32))

271
tests/st/ops/cpu/test_cumsum_op.py
View File

File diff suppressed because it is too large Load Diff
Write Preview
Loading…
Cancel
Save