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

!10746 [cpu] Add type support for concat, reshape, transpose, tile, squeeze, expandims cpu op

Browse Source 
From: @yanglf1121
Reviewed-by: @wuxuejian
Signed-off-by: @wuxuejian
pull/10746/MERGE
mindspore-ci-bot 4 years ago committed by Gitee
parent 84303c539f 75aa938ee6
commit 25d5d43dea
10 changed files with 508 additions and 33 deletions
Show Stats Download Patch File Download Diff File

23
mindspore/ccsrc/backend/kernel_compiler/cpu/concat_cpu_kernel.h
Unescape View File

@ -1,5 +1,5 @@
/**
* Copyright 2020 Huawei Technologies Co., Ltd
* Copyright 2021 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.
@ -43,9 +43,30 @@ class ConcatCPUKernel : public CPUKernel {
MS_REG_CPU_KERNEL_T(
Concat, KernelAttr().SetAllSameAttr(true).AddInputAttr(kNumberTypeFloat32).AddOutputAttr(kNumberTypeFloat32),
ConcatCPUKernel, float);
MS_REG_CPU_KERNEL_T(Concat,
KernelAttr().SetAllSameAttr(true).AddInputAttr(kNumberTypeInt8).AddOutputAttr(kNumberTypeInt8),
ConcatCPUKernel, int8_t)
MS_REG_CPU_KERNEL_T(Concat,
KernelAttr().SetAllSameAttr(true).AddInputAttr(kNumberTypeInt16).AddOutputAttr(kNumberTypeInt16),
ConcatCPUKernel, int16_t)
MS_REG_CPU_KERNEL_T(Concat,
KernelAttr().SetAllSameAttr(true).AddInputAttr(kNumberTypeInt32).AddOutputAttr(kNumberTypeInt32),
ConcatCPUKernel, int)
MS_REG_CPU_KERNEL_T(Concat,
KernelAttr().SetAllSameAttr(true).AddInputAttr(kNumberTypeInt64).AddOutputAttr(kNumberTypeInt64),
ConcatCPUKernel, int64_t)
MS_REG_CPU_KERNEL_T(Concat,
KernelAttr().SetAllSameAttr(true).AddInputAttr(kNumberTypeUInt8).AddOutputAttr(kNumberTypeUInt8),
ConcatCPUKernel, uint8_t)
MS_REG_CPU_KERNEL_T(Concat,
KernelAttr().SetAllSameAttr(true).AddInputAttr(kNumberTypeUInt16).AddOutputAttr(kNumberTypeUInt16),
ConcatCPUKernel, uint16_t)
MS_REG_CPU_KERNEL_T(Concat,
KernelAttr().SetAllSameAttr(true).AddInputAttr(kNumberTypeUInt32).AddOutputAttr(kNumberTypeUInt32),
ConcatCPUKernel, uint32_t)
MS_REG_CPU_KERNEL_T(Concat,
KernelAttr().SetAllSameAttr(true).AddInputAttr(kNumberTypeUInt64).AddOutputAttr(kNumberTypeUInt64),
ConcatCPUKernel, uint64_t)
MS_REG_CPU_KERNEL_T(Concat,
KernelAttr().SetAllSameAttr(true).AddInputAttr(kNumberTypeBool).AddOutputAttr(kNumberTypeBool),
ConcatCPUKernel, bool)

56
mindspore/ccsrc/backend/kernel_compiler/cpu/reshape_cpu_kernel.h
Unescape View File

@ -1,5 +1,5 @@
/**
* Copyright 2019 Huawei Technologies Co., Ltd
* Copyright 2021 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.
@ -38,23 +38,47 @@ class ReshapeCPUKernel : public CPUKernel {
size_t type_size_ = 4;
};
MS_REG_CPU_KERNEL(Reshape, KernelAttr().AddInputAttr(kNumberTypeFloat32).AddOutputAttr(kNumberTypeFloat32),
MS_REG_CPU_KERNEL(Reshape, KernelAttr().AddInputAttr(kNumberTypeInt8).AddOutputAttr(kNumberTypeInt8), ReshapeCPUKernel);
MS_REG_CPU_KERNEL(Reshape, KernelAttr().AddInputAttr(kNumberTypeInt16).AddOutputAttr(kNumberTypeInt16),
ReshapeCPUKernel);
MS_REG_CPU_KERNEL(Reshape, KernelAttr().AddInputAttr(kNumberTypeInt32).AddOutputAttr(kNumberTypeInt32),
ReshapeCPUKernel);
MS_REG_CPU_KERNEL(Reshape, KernelAttr().AddInputAttr(kNumberTypeInt64).AddOutputAttr(kNumberTypeInt64),
ReshapeCPUKernel);
MS_REG_CPU_KERNEL(Reshape, KernelAttr().AddInputAttr(kNumberTypeBool).AddOutputAttr(kNumberTypeBool), ReshapeCPUKernel);
MS_REG_CPU_KERNEL(Reshape, KernelAttr().AddInputAttr(kNumberTypeFloat32).AddOutputAttr(kNumberTypeFloat32),
ReshapeCPUKernel);
MS_REG_CPU_KERNEL(Reshape, KernelAttr().AddInputAttr(kNumberTypeUInt8).AddOutputAttr(kNumberTypeUInt8),
ReshapeCPUKernel);
MS_REG_CPU_KERNEL(Reshape, KernelAttr().AddInputAttr(kNumberTypeUInt16).AddOutputAttr(kNumberTypeUInt16),
ReshapeCPUKernel);
MS_REG_CPU_KERNEL(Reshape, KernelAttr().AddInputAttr(kNumberTypeUInt32).AddOutputAttr(kNumberTypeUInt32),
ReshapeCPUKernel);
MS_REG_CPU_KERNEL(Reshape, KernelAttr().AddInputAttr(kNumberTypeUInt64).AddOutputAttr(kNumberTypeUInt64),
ReshapeCPUKernel);
MS_REG_CPU_KERNEL(Flatten, KernelAttr().AddInputAttr(kNumberTypeFloat32).AddOutputAttr(kNumberTypeFloat32),
MS_REG_CPU_KERNEL(Flatten, KernelAttr().AddInputAttr(kNumberTypeInt8).AddOutputAttr(kNumberTypeInt8), ReshapeCPUKernel);
MS_REG_CPU_KERNEL(Flatten, KernelAttr().AddInputAttr(kNumberTypeInt16).AddOutputAttr(kNumberTypeInt16),
ReshapeCPUKernel);
MS_REG_CPU_KERNEL(Flatten, KernelAttr().AddInputAttr(kNumberTypeInt32).AddOutputAttr(kNumberTypeInt32),
ReshapeCPUKernel);
MS_REG_CPU_KERNEL(Flatten, KernelAttr().AddInputAttr(kNumberTypeInt64).AddOutputAttr(kNumberTypeInt64),
ReshapeCPUKernel);
MS_REG_CPU_KERNEL(Flatten, KernelAttr().AddInputAttr(kNumberTypeBool).AddOutputAttr(kNumberTypeBool), ReshapeCPUKernel);
MS_REG_CPU_KERNEL(Flatten, KernelAttr().AddInputAttr(kNumberTypeFloat32).AddOutputAttr(kNumberTypeFloat32),
ReshapeCPUKernel);
MS_REG_CPU_KERNEL(Flatten, KernelAttr().AddInputAttr(kNumberTypeUInt8).AddOutputAttr(kNumberTypeUInt8),
ReshapeCPUKernel);
MS_REG_CPU_KERNEL(Flatten, KernelAttr().AddInputAttr(kNumberTypeUInt16).AddOutputAttr(kNumberTypeUInt16),
ReshapeCPUKernel);
MS_REG_CPU_KERNEL(Flatten, KernelAttr().AddInputAttr(kNumberTypeUInt32).AddOutputAttr(kNumberTypeUInt32),
ReshapeCPUKernel);
MS_REG_CPU_KERNEL(Flatten, KernelAttr().AddInputAttr(kNumberTypeUInt64).AddOutputAttr(kNumberTypeUInt64),
ReshapeCPUKernel);
MS_REG_CPU_KERNEL(ExpandDims, KernelAttr().AddInputAttr(kNumberTypeFloat32).AddOutputAttr(kNumberTypeFloat32),
MS_REG_CPU_KERNEL(ExpandDims, KernelAttr().AddInputAttr(kNumberTypeInt8).AddOutputAttr(kNumberTypeInt8),
ReshapeCPUKernel);
MS_REG_CPU_KERNEL(ExpandDims, KernelAttr().AddInputAttr(kNumberTypeInt16).AddOutputAttr(kNumberTypeInt16),
ReshapeCPUKernel);
MS_REG_CPU_KERNEL(ExpandDims, KernelAttr().AddInputAttr(kNumberTypeInt32).AddOutputAttr(kNumberTypeInt32),
ReshapeCPUKernel);
@ -62,14 +86,36 @@ MS_REG_CPU_KERNEL(ExpandDims, KernelAttr().AddInputAttr(kNumberTypeInt64).AddOut
ReshapeCPUKernel);
MS_REG_CPU_KERNEL(ExpandDims, KernelAttr().AddInputAttr(kNumberTypeBool).AddOutputAttr(kNumberTypeBool),
ReshapeCPUKernel);
MS_REG_CPU_KERNEL(ExpandDims, KernelAttr().AddInputAttr(kNumberTypeFloat32).AddOutputAttr(kNumberTypeFloat32),
ReshapeCPUKernel);
MS_REG_CPU_KERNEL(ExpandDims, KernelAttr().AddInputAttr(kNumberTypeUInt8).AddOutputAttr(kNumberTypeUInt8),
ReshapeCPUKernel);
MS_REG_CPU_KERNEL(ExpandDims, KernelAttr().AddInputAttr(kNumberTypeUInt16).AddOutputAttr(kNumberTypeUInt16),
ReshapeCPUKernel);
MS_REG_CPU_KERNEL(ExpandDims, KernelAttr().AddInputAttr(kNumberTypeUInt32).AddOutputAttr(kNumberTypeUInt32),
ReshapeCPUKernel);
MS_REG_CPU_KERNEL(ExpandDims, KernelAttr().AddInputAttr(kNumberTypeUInt64).AddOutputAttr(kNumberTypeUInt64),
ReshapeCPUKernel);
MS_REG_CPU_KERNEL(Squeeze, KernelAttr().AddInputAttr(kNumberTypeFloat32).AddOutputAttr(kNumberTypeFloat32),
MS_REG_CPU_KERNEL(Squeeze, KernelAttr().AddInputAttr(kNumberTypeInt8).AddOutputAttr(kNumberTypeInt8), ReshapeCPUKernel);
MS_REG_CPU_KERNEL(Squeeze, KernelAttr().AddInputAttr(kNumberTypeInt16).AddOutputAttr(kNumberTypeInt16),
ReshapeCPUKernel);
MS_REG_CPU_KERNEL(Squeeze, KernelAttr().AddInputAttr(kNumberTypeInt32).AddOutputAttr(kNumberTypeInt32),
ReshapeCPUKernel);
MS_REG_CPU_KERNEL(Squeeze, KernelAttr().AddInputAttr(kNumberTypeInt64).AddOutputAttr(kNumberTypeInt64),
ReshapeCPUKernel);
MS_REG_CPU_KERNEL(Squeeze, KernelAttr().AddInputAttr(kNumberTypeBool).AddOutputAttr(kNumberTypeBool), ReshapeCPUKernel);
MS_REG_CPU_KERNEL(Squeeze, KernelAttr().AddInputAttr(kNumberTypeFloat32).AddOutputAttr(kNumberTypeFloat32),
ReshapeCPUKernel);
MS_REG_CPU_KERNEL(Squeeze, KernelAttr().AddInputAttr(kNumberTypeUInt8).AddOutputAttr(kNumberTypeUInt8),
ReshapeCPUKernel);
MS_REG_CPU_KERNEL(Squeeze, KernelAttr().AddInputAttr(kNumberTypeUInt16).AddOutputAttr(kNumberTypeUInt16),
ReshapeCPUKernel);
MS_REG_CPU_KERNEL(Squeeze, KernelAttr().AddInputAttr(kNumberTypeUInt32).AddOutputAttr(kNumberTypeUInt32),
ReshapeCPUKernel);
MS_REG_CPU_KERNEL(Squeeze, KernelAttr().AddInputAttr(kNumberTypeUInt64).AddOutputAttr(kNumberTypeUInt64),
ReshapeCPUKernel);
} // namespace kernel
} // namespace mindspore

28
mindspore/ccsrc/backend/kernel_compiler/cpu/tile_cpu_kernel.cc
Unescape View File

@ -1,5 +1,5 @@
/**
* Copyright 2020 Huawei Technologies Co., Ltd
* Copyright 2021 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.
@ -31,18 +31,30 @@ void TileCPUKernel::InitKernel(const CNodePtr &kernel_node) {
if (dtype_ == kTypeUnknown) {
dtype_ = AnfAlgo::GetPrevNodeOutputInferDataType(kernel_node, 0);
}
launch_map_[kNumberTypeInt8] = &TileCPUKernel::LaunchKernel<int8_t>;
launch_map_[kNumberTypeInt16] = &TileCPUKernel::LaunchKernel<int16_t>;
launch_map_[kNumberTypeInt32] = &TileCPUKernel::LaunchKernel<int>;
launch_map_[kNumberTypeInt64] = &TileCPUKernel::LaunchKernel<int64_t>;
launch_map_[kNumberTypeUInt8] = &TileCPUKernel::LaunchKernel<uint8_t>;
launch_map_[kNumberTypeUInt16] = &TileCPUKernel::LaunchKernel<uint16_t>;
launch_map_[kNumberTypeUInt32] = &TileCPUKernel::LaunchKernel<uint32_t>;
launch_map_[kNumberTypeUInt64] = &TileCPUKernel::LaunchKernel<uint64_t>;
launch_map_[kNumberTypeFloat32] = &TileCPUKernel::LaunchKernel<float>;
launch_map_[kNumberTypeBool] = &TileCPUKernel::LaunchKernel<bool>;
auto iter = launch_map_.find(dtype_);
if (iter != launch_map_.end()) {
launch_func_ = iter->second;
} else {
MS_LOG(EXCEPTION) << "Input data type: " << dtype_ << "is not supported for Tile kernel on CPU.";
}
}
bool TileCPUKernel::Launch(const std::vector<kernel::AddressPtr> &inputs,
const std::vector<kernel::AddressPtr> & /*workspace*/,
const std::vector<kernel::AddressPtr> &outputs) {
if (dtype_ == kNumberTypeInt32) {
LaunchKernel<int>(inputs, outputs);
} else if (dtype_ == kNumberTypeFloat32) {
LaunchKernel<float>(inputs, outputs);
} else if (dtype_ == kNumberTypeInt64) {
LaunchKernel<int64_t>(inputs, outputs);
}
launch_func_(this, inputs, outputs);
return true;
}

18
mindspore/ccsrc/backend/kernel_compiler/cpu/tile_cpu_kernel.h
Unescape View File

@ -1,5 +1,5 @@
/**
* Copyright 2020 Huawei Technologies Co., Ltd
* Copyright 2021 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.
@ -43,14 +43,30 @@ class TileCPUKernel : public CPUKernel {
std::vector<size_t> y_shape_;
std::vector<int> multiples_;
TypeId dtype_{kTypeUnknown};
using TypeKernel =
std::function<void(TileCPUKernel *, const std::vector<AddressPtr> &inputs, const std::vector<AddressPtr> &outputs)>;
std::unordered_map<TypeId, TypeKernel> launch_map_;
TypeKernel launch_func_;
};
MS_REG_CPU_KERNEL(Tile, KernelAttr().AddInputAttr(kNumberTypeInt8).AddOutputAttr(kNumberTypeInt8), TileCPUKernel);
MS_REG_CPU_KERNEL(Tile, KernelAttr().AddInputAttr(kNumberTypeInt16).AddOutputAttr(kNumberTypeInt16), TileCPUKernel);
MS_REG_CPU_KERNEL(Tile, KernelAttr().AddInputAttr(kNumberTypeInt32).AddOutputAttr(kNumberTypeInt32), TileCPUKernel);
MS_REG_CPU_KERNEL(Tile, KernelAttr().AddInputAttr(kNumberTypeInt64).AddOutputAttr(kNumberTypeInt64), TileCPUKernel);
MS_REG_CPU_KERNEL(Tile, KernelAttr().AddInputAttr(kNumberTypeFloat32).AddOutputAttr(kNumberTypeFloat32), TileCPUKernel);
MS_REG_CPU_KERNEL(Tile, KernelAttr().AddInputAttr(kNumberTypeUInt8).AddOutputAttr(kNumberTypeUInt8), TileCPUKernel);
MS_REG_CPU_KERNEL(Tile, KernelAttr().AddInputAttr(kNumberTypeUInt16).AddOutputAttr(kNumberTypeUInt16), TileCPUKernel);
MS_REG_CPU_KERNEL(Tile, KernelAttr().AddInputAttr(kNumberTypeUInt32).AddOutputAttr(kNumberTypeUInt32), TileCPUKernel);
MS_REG_CPU_KERNEL(Tile, KernelAttr().AddInputAttr(kNumberTypeUInt64).AddOutputAttr(kNumberTypeUInt64), TileCPUKernel);
MS_REG_CPU_KERNEL(Tile, KernelAttr().AddInputAttr(kNumberTypeBool).AddOutputAttr(kNumberTypeBool), TileCPUKernel);
} // namespace kernel
} // namespace mindspore

44
mindspore/ccsrc/backend/kernel_compiler/cpu/transpose_cpu_kernel.cc
Unescape View File

@ -1,5 +1,5 @@
/**
* Copyright 2020 Huawei Technologies Co., Ltd
* Copyright 2021 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.
@ -29,13 +29,36 @@ void TransposeCPUFwdKernel::InitKernel(const CNodePtr &kernel_node) {
if (shape_.size() != axis_.size()) {
MS_LOG(EXCEPTION) << "The size of input shape and transpose axis shape must be equal.";
}
dtype_ = AnfAlgo ::GetPrevNodeOutputDeviceDataType(kernel_node, 0);
if (dtype_ == kTypeUnknown) {
dtype_ = AnfAlgo::GetPrevNodeOutputInferDataType(kernel_node, 0);
}
launch_map_[kNumberTypeInt8] = &TransposeCPUFwdKernel::LaunchKernel<int8_t>;
launch_map_[kNumberTypeInt16] = &TransposeCPUFwdKernel::LaunchKernel<int16_t>;
launch_map_[kNumberTypeInt32] = &TransposeCPUFwdKernel::LaunchKernel<int>;
launch_map_[kNumberTypeInt64] = &TransposeCPUFwdKernel::LaunchKernel<int64_t>;
launch_map_[kNumberTypeUInt8] = &TransposeCPUFwdKernel::LaunchKernel<uint8_t>;
launch_map_[kNumberTypeUInt16] = &TransposeCPUFwdKernel::LaunchKernel<uint16_t>;
launch_map_[kNumberTypeUInt32] = &TransposeCPUFwdKernel::LaunchKernel<uint32_t>;
launch_map_[kNumberTypeUInt64] = &TransposeCPUFwdKernel::LaunchKernel<uint64_t>;
launch_map_[kNumberTypeFloat32] = &TransposeCPUFwdKernel::LaunchKernel<float>;
launch_map_[kNumberTypeBool] = &TransposeCPUFwdKernel::LaunchKernel<bool>;
auto iter = launch_map_.find(dtype_);
if (iter != launch_map_.end()) {
launch_func_ = iter->second;
} else {
MS_LOG(EXCEPTION) << "Input data type: " << dtype_ << "is not supported for Transpose kernel on CPU.";
}
}
bool TransposeCPUFwdKernel::Launch(const std::vector<kernel::AddressPtr> &inputs,
const std::vector<kernel::AddressPtr> & /*workspace*/,
const std::vector<kernel::AddressPtr> &outputs) {
auto input = reinterpret_cast<float *>(inputs[0]->addr);
auto output = reinterpret_cast<float *>(outputs[0]->addr);
size_t size = IntToSize(inputs[0]->size / sizeof(float));
template <typename T>
void TransposeCPUFwdKernel::LaunchKernel(const std::vector<AddressPtr> &inputs,
const std::vector<AddressPtr> &outputs) {
auto input = reinterpret_cast<T *>(inputs[0]->addr);
auto output = reinterpret_cast<T *>(outputs[0]->addr);
size_t size = IntToSize(inputs[0]->size / sizeof(T));
size_t shape_size = IntToSize(shape_.size());
if (shape_size > kMaxDim) {
MS_LOG(EXCEPTION) << "Input is " << shape_size << "-D, but transpose supports max " << kMaxDim << "-D inputs.";
@ -61,7 +84,14 @@ bool TransposeCPUFwdKernel::Launch(const std::vector<kernel::AddressPtr> &inputs
}
output[new_position] = input[position];
}
}
bool TransposeCPUFwdKernel::Launch(const std::vector<kernel::AddressPtr> &inputs,
const std::vector<kernel::AddressPtr> & /*workspace*/,
const std::vector<kernel::AddressPtr> &outputs) {
launch_func_(this, inputs, outputs);
return true;
}
} // namespace kernel
} // namespace mindspore

42
mindspore/ccsrc/backend/kernel_compiler/cpu/transpose_cpu_kernel.h
Unescape View File

@ -1,5 +1,5 @@
/**
* Copyright 2020 Huawei Technologies Co., Ltd
* Copyright 2021 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.
@ -16,6 +16,7 @@
#ifndef MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_CPU_TRANSPOSE_CPU_KERNEL_H_
#define MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_CPU_TRANSPOSE_CPU_KERNEL_H_
#include <vector>
#include <unordered_map>
#include <memory>
#include <string>
#include "backend/kernel_compiler/cpu/cpu_kernel.h"
@ -32,12 +33,47 @@ class TransposeCPUFwdKernel : public CPUKernel {
bool Launch(const std::vector<AddressPtr> &inputs, const std::vector<AddressPtr> &workspace,
const std::vector<AddressPtr> &outputs) override;
template <typename T>
void LaunchKernel(const std::vector<AddressPtr> &inputs, const std::vector<AddressPtr> &outputs);
private:
std::vector<size_t> shape_;
std::vector<int> axis_;
TypeId dtype_{kTypeUnknown};
using TypeKernel =
std::function<void(TransposeCPUFwdKernel *, const std::vector<AddressPtr> &, const std::vector<AddressPtr> &)>;
std::unordered_map<TypeId, TypeKernel> launch_map_;
TypeKernel launch_func_;
};
MS_REG_CPU_KERNEL(Transpose, KernelAttr().AddInputAttr(kNumberTypeFloat32).AddOutputAttr(kNumberTypeFloat32),
MS_REG_CPU_KERNEL(Transpose,
KernelAttr().SetAllSameAttr(true).AddInputAttr(kNumberTypeFloat32).AddOutputAttr(kNumberTypeFloat32),
TransposeCPUFwdKernel);
MS_REG_CPU_KERNEL(Transpose,
KernelAttr().SetAllSameAttr(true).AddInputAttr(kNumberTypeInt8).AddOutputAttr(kNumberTypeInt8),
TransposeCPUFwdKernel);
MS_REG_CPU_KERNEL(Transpose,
KernelAttr().SetAllSameAttr(true).AddInputAttr(kNumberTypeInt16).AddOutputAttr(kNumberTypeInt16),
TransposeCPUFwdKernel);
MS_REG_CPU_KERNEL(Transpose,
KernelAttr().SetAllSameAttr(true).AddInputAttr(kNumberTypeInt32).AddOutputAttr(kNumberTypeInt32),
TransposeCPUFwdKernel);
MS_REG_CPU_KERNEL(Transpose,
KernelAttr().SetAllSameAttr(true).AddInputAttr(kNumberTypeInt64).AddOutputAttr(kNumberTypeInt64),
TransposeCPUFwdKernel);
MS_REG_CPU_KERNEL(Transpose,
KernelAttr().SetAllSameAttr(true).AddInputAttr(kNumberTypeUInt8).AddOutputAttr(kNumberTypeUInt8),
TransposeCPUFwdKernel);
MS_REG_CPU_KERNEL(Transpose,
KernelAttr().SetAllSameAttr(true).AddInputAttr(kNumberTypeUInt16).AddOutputAttr(kNumberTypeUInt16),
TransposeCPUFwdKernel);
MS_REG_CPU_KERNEL(Transpose,
KernelAttr().SetAllSameAttr(true).AddInputAttr(kNumberTypeUInt32).AddOutputAttr(kNumberTypeUInt32),
TransposeCPUFwdKernel);
MS_REG_CPU_KERNEL(Transpose,
KernelAttr().SetAllSameAttr(true).AddInputAttr(kNumberTypeUInt64).AddOutputAttr(kNumberTypeUInt64),
TransposeCPUFwdKernel);
MS_REG_CPU_KERNEL(Transpose,
KernelAttr().SetAllSameAttr(true).AddInputAttr(kNumberTypeBool).AddOutputAttr(kNumberTypeBool),
TransposeCPUFwdKernel);
} // namespace kernel
} // namespace mindspore

16
tests/st/ops/cpu/test_concat_op.py
Unescape View File

@ -1,4 +1,4 @@
# Copyright 2020 Huawei Technologies Co., Ltd
# Copyright 2021 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.
@ -46,6 +46,7 @@ def axis10(nptype):
print(output)
assert (output.asnumpy() == expect).all()
@pytest.mark.level0
@pytest.mark.platform_x86_cpu
@pytest.mark.env_onecard
@ -171,6 +172,7 @@ def axis21(nptype):
assert (output.asnumpy() == expect).all()
print(output)
@pytest.mark.level0
@pytest.mark.platform_x86_cpu
@pytest.mark.env_onecard
@ -287,6 +289,18 @@ def test_concat_4i_float32():
def test_concat_4i_int32():
concat_4i(np.int32)
@pytest.mark.level0
@pytest.mark.platform_x86_cpu
@pytest.mark.env_onecard
def test_concat_4i_int8():
concat_4i(np.int8)
@pytest.mark.level0
@pytest.mark.platform_x86_cpu
@pytest.mark.env_onecard
def test_concat_4i_uint64():
concat_4i(np.uint64)
@pytest.mark.level0
@pytest.mark.platform_x86_cpu
@pytest.mark.env_onecard

37
tests/st/ops/cpu/test_squeeze_op.py
Unescape View File

@ -1,4 +1,4 @@
# Copyright 2020 Huawei Technologies Co., Ltd
# Copyright 2021 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.
@ -40,7 +40,8 @@ def test_squeeze_shape_float32():
expect = np.ones(shape=[2, 8, 3]).astype(np.float32)
net = SqueezeNet()
result = net(Tensor(x))
assert np.allclose(result.asnumpy(), expect, rtol=1.e-4, atol=1.e-8, equal_nan=True)
assert np.allclose(result.asnumpy(), expect, rtol=1.e-4,
atol=1.e-8, equal_nan=True)
@pytest.mark.level0
@ -51,7 +52,8 @@ def test_squeeze_shape_int32():
expect = np.array([7, 11]).astype(np.int32)
net = SqueezeNet()
result = net(Tensor(x))
assert np.allclose(result.asnumpy(), expect, rtol=1.e-4, atol=1.e-8, equal_nan=True)
assert np.allclose(result.asnumpy(), expect, rtol=1.e-4,
atol=1.e-8, equal_nan=True)
@pytest.mark.level0
@ -62,4 +64,31 @@ def test_squeeze_shape_bool():
expect = np.array([True, False]).astype(np.bool_)
net = SqueezeNet()
result = net(Tensor(x))
assert np.allclose(result.asnumpy(), expect, rtol=1.e-4, atol=1.e-8, equal_nan=True)
assert np.allclose(result.asnumpy(), expect, rtol=1.e-4,
atol=1.e-8, equal_nan=True)
@pytest.mark.level0
@pytest.mark.platform_x86_cpu
@pytest.mark.env_onecard
def test_squeeze_shape_float64():
x = np.random.random([1, 2, 1, 1, 8, 3, 1]).astype(np.float64)
expect = np.squeeze(x)
net = SqueezeNet()
result = net(Tensor(x))
print(result.asnumpy()[0][0], expect[0][0])
assert np.allclose(result.asnumpy(), expect, rtol=1.e-4,
atol=1.e-8, equal_nan=True)
@pytest.mark.level0
@pytest.mark.platform_x86_cpu
@pytest.mark.env_onecard
def test_squeeze_shape_uint16():
x = np.random.random([1, 2, 1, 1, 8, 3, 1]).astype(np.uint16)
expect = np.squeeze(x)
net = SqueezeNet()
result = net(Tensor(x))
print(result.asnumpy()[0][0], expect[0][0])
assert np.allclose(result.asnumpy(), expect, rtol=1.e-4,
atol=1.e-8, equal_nan=True)

28
tests/st/ops/cpu/test_tile_op.py
Unescape View File

@ -1,4 +1,4 @@
# Copyright 2020 Huawei Technologies Co., Ltd
# Copyright 2021 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.
@ -43,3 +43,29 @@ def test_net():
print(arr_x)
output = tile(Tensor(arr_x))
print(output.asnumpy())
arr_x = np.array([[0], [1], [2], [3]]).astype(np.float64)
@pytest.mark.level0
@pytest.mark.platform_x86_cpu
@pytest.mark.env_onecard
def test_net_float64():
tile = Net()
print(arr_x)
output = tile(Tensor(arr_x))
print(output.asnumpy())
arr_x = np.array([[0], [1], [2], [3]]).astype(np.bool_)
@pytest.mark.level0
@pytest.mark.platform_x86_cpu
@pytest.mark.env_onecard
def test_net_bool():
tile = Net()
print(arr_x)
output = tile(Tensor(arr_x))
print(output.asnumpy())

249
tests/st/ops/cpu/test_transpose_op.py
View File

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