Skip to content

Logical Ops support more data types #34141

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 10 commits into from
Jul 23, 2021
Merged

Logical Ops support more data types #34141

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
10 commits
Select commit Hold shift + click to select a range
6d88046
logical ops support int8, int16, int32, int64, float, double
will-jl944 Jul 13, 2021
fda2abf
update docs of logical ops
will-jl944 Jul 14, 2021
ac8b193
fix npu and xpu logical ops
will-jl944 Jul 14, 2021
c130cf2
fix npu and xpu logical ops
will-jl944 Jul 15, 2021
af4943f
merge develop
will-jl944 Jul 15, 2021
78716a6
Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into…
will-jl944 Jul 15, 2021
611cb66
fix bug in xpu logical op code
will-jl944 Jul 15, 2021
22a4d68
Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into…
will-jl944 Jul 15, 2021
d6c1cb2
update test_logical_op_npu and test_logical_op_xpu
will-jl944 Jul 15, 2021
acdaaf8
correct error type
will-jl944 Jul 15, 2021
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
23 changes: 11 additions & 12 deletions paddle/fluid/operators/controlflow/logical_op.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,11 +1,8 @@
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved.

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
Expand All @@ -26,15 +23,16 @@ class BinaryLogicalOpProtoMaker : public framework::OpProtoAndCheckerMaker {
void Make() override {
OpComment comment;
AddInput("X", string::Sprintf("Left hand operand of %s operator. Must be "
"a Variable of type bool.",
"a Variable of type being one of bool, int8, "
"int16, int32, int64, float32, float64.",
comment.type));
AddInput("Y", string::Sprintf("Right hand operand of %s operator. Must be "
"a Variable of type bool.",
"a Variable of type being one of bool, int8, "
"int16, int32, int64, float32, float64.",
comment.type));
AddOutput("Out", string::Sprintf("n-dim bool Variable"));
AddComment(string::Sprintf(R"DOC(%s Operator

It operates element-wise on X and Y, and returns the Out. X, Y and Out are N-dim boolean LoDTensor or Tensor.
It operates element-wise on X and Y, and returns the Out. X, Y and Out are N-dim LoDTensor or Tensor.
Each element of Out is calculated by %s
)DOC",
comment.type, comment.equation));
Expand All @@ -46,13 +44,14 @@ class UnaryLogicalOpProtoMaker : public framework::OpProtoAndCheckerMaker {
public:
void Make() override {
OpComment comment;
AddInput("X", string::Sprintf("Operand of %s operator. Must be "
"a LoDTensor or Tensor of type bool.",
comment.type));
AddInput("X",
string::Sprintf("Operand of %s operator. Must be "
"a LoDTensor or Tensor of type being one of bool, "
"int8, int16, int32, int64, float32, float64.",
comment.type));
AddOutput("Out", string::Sprintf("n-dim bool LoDTensor or Tensor."));
AddComment(string::Sprintf(R"DOC(%s Operator

It operates element-wise on X, and returns the Out. X and Out are N-dim boolean LoDTensor or Tensor.
It operates element-wise on X, and returns the Out. X and Out are N-dim LoDTensor or Tensor.
Each element of Out is calculated by %s
)DOC",
comment.type, comment.equation));
Expand Down
31 changes: 17 additions & 14 deletions paddle/fluid/operators/controlflow/logical_op.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,11 +1,8 @@
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved.

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
Expand All @@ -21,13 +18,13 @@ namespace plat = paddle::platform;
namespace paddle {
namespace operators {

#define LOGICAL_BINARY_FUNCTOR(func_name, op) \
template <typename T> \
struct func_name { \
using ELEMENT_TYPE = T; \
HOSTDEVICE bool operator()(const T* args) const { \
return args[0] op args[1]; \
} \
#define LOGICAL_BINARY_FUNCTOR(func_name, op) \
template <typename T> \
struct func_name { \
using ELEMENT_TYPE = T; \
HOSTDEVICE bool operator()(const T* args) const { \
return static_cast<bool>(args[0]) op static_cast<bool>(args[1]); \
} \
};

LOGICAL_BINARY_FUNCTOR(CudaOrFunctor, ||)
Expand Down Expand Up @@ -68,10 +65,16 @@ class BinaryLogicalOpKernel<platform::CUDADeviceContext, Functor>
} // namespace operators
} // namespace paddle

#define REGISTER_LOGICAL_CUDA_KERNEL(op_name, func) \
REGISTER_OP_CUDA_KERNEL( \
op_name, \
ops::BinaryLogicalOpKernel<plat::CUDADeviceContext, ops::func<bool>>);
#define REGISTER_LOGICAL_CUDA_KERNEL(op_name, func) \
REGISTER_OP_CUDA_KERNEL( \
op_name, \
ops::BinaryLogicalOpKernel<plat::CUDADeviceContext, ops::func<bool>>, \
ops::BinaryLogicalOpKernel<plat::CUDADeviceContext, ops::func<int8_t>>, \
ops::BinaryLogicalOpKernel<plat::CUDADeviceContext, ops::func<int16_t>>, \
ops::BinaryLogicalOpKernel<plat::CUDADeviceContext, ops::func<int>>, \
ops::BinaryLogicalOpKernel<plat::CUDADeviceContext, ops::func<int64_t>>, \
ops::BinaryLogicalOpKernel<plat::CUDADeviceContext, ops::func<float>>, \
ops::BinaryLogicalOpKernel<plat::CUDADeviceContext, ops::func<double>>);

REGISTER_LOGICAL_CUDA_KERNEL(logical_or, CudaOrFunctor)
REGISTER_LOGICAL_CUDA_KERNEL(logical_and, CudaAndFunctor)
Expand Down
43 changes: 32 additions & 11 deletions paddle/fluid/operators/controlflow/logical_op.h
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,11 +1,8 @@
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved.

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
Expand Down Expand Up @@ -82,12 +79,36 @@ class UnaryLogicalOpKernel
} // namespace operators
} // namespace paddle

#define REGISTER_BINARY_LOGICAL_KERNEL(op_type, dev, functor) \
REGISTER_OP_##dev##_KERNEL( \
op_type, ::paddle::operators::BinaryLogicalOpKernel< \
::paddle::platform::dev##DeviceContext, functor<bool>>);
#define REGISTER_BINARY_LOGICAL_KERNEL(op_type, dev, functor) \
REGISTER_OP_##dev##_KERNEL( \
op_type, ::paddle::operators::BinaryLogicalOpKernel< \
::paddle::platform::dev##DeviceContext, functor<bool>>, \
::paddle::operators::BinaryLogicalOpKernel< \
::paddle::platform::dev##DeviceContext, functor<int8_t>>, \
::paddle::operators::BinaryLogicalOpKernel< \
::paddle::platform::dev##DeviceContext, functor<int16_t>>, \
::paddle::operators::BinaryLogicalOpKernel< \
::paddle::platform::dev##DeviceContext, functor<int>>, \
::paddle::operators::BinaryLogicalOpKernel< \
::paddle::platform::dev##DeviceContext, functor<int64_t>>, \
::paddle::operators::BinaryLogicalOpKernel< \
::paddle::platform::dev##DeviceContext, functor<float>>, \
::paddle::operators::BinaryLogicalOpKernel< \
::paddle::platform::dev##DeviceContext, functor<double>>);

#define REGISTER_UNARY_LOGICAL_KERNEL(op_type, dev, functor) \
REGISTER_OP_##dev##_KERNEL( \
op_type, ::paddle::operators::UnaryLogicalOpKernel< \
::paddle::platform::dev##DeviceContext, functor<bool>>);
#define REGISTER_UNARY_LOGICAL_KERNEL(op_type, dev, functor) \
REGISTER_OP_##dev##_KERNEL( \
op_type, ::paddle::operators::UnaryLogicalOpKernel< \
::paddle::platform::dev##DeviceContext, functor<bool>>, \
::paddle::operators::UnaryLogicalOpKernel< \
::paddle::platform::dev##DeviceContext, functor<int8_t>>, \
::paddle::operators::UnaryLogicalOpKernel< \
::paddle::platform::dev##DeviceContext, functor<int16_t>>, \
::paddle::operators::UnaryLogicalOpKernel< \
::paddle::platform::dev##DeviceContext, functor<int>>, \
::paddle::operators::UnaryLogicalOpKernel< \
::paddle::platform::dev##DeviceContext, functor<int64_t>>, \
::paddle::operators::UnaryLogicalOpKernel< \
::paddle::platform::dev##DeviceContext, functor<float>>, \
::paddle::operators::UnaryLogicalOpKernel< \
::paddle::platform::dev##DeviceContext, functor<double>>);
29 changes: 22 additions & 7 deletions paddle/fluid/operators/controlflow/logical_op_npu.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,11 +1,8 @@
/* Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
Expand Down Expand Up @@ -82,11 +79,29 @@ class LogicalAndPUKernel : public framework::OpKernel<T> {
namespace ops = paddle::operators;
namespace plat = paddle::platform;

REGISTER_OP_NPU_KERNEL(logical_not,
ops::LogicalNotNPUKernel<plat::NPUDeviceContext, bool>);
REGISTER_OP_NPU_KERNEL(
logical_not, ops::LogicalNotNPUKernel<plat::NPUDeviceContext, bool>,
ops::LogicalNotNPUKernel<plat::NPUDeviceContext, int8_t>,
ops::LogicalNotNPUKernel<plat::NPUDeviceContext, int16_t>,
ops::LogicalNotNPUKernel<plat::NPUDeviceContext, int>,
ops::LogicalNotNPUKernel<plat::NPUDeviceContext, int64_t>,
ops::LogicalNotNPUKernel<plat::NPUDeviceContext, float>,
ops::LogicalNotNPUKernel<plat::NPUDeviceContext, double>);

REGISTER_OP_NPU_KERNEL(logical_or,
ops::LogicalOrNPUKernel<plat::NPUDeviceContext, bool>);
ops::LogicalOrNPUKernel<plat::NPUDeviceContext, bool>,
ops::LogicalOrNPUKernel<plat::NPUDeviceContext, int8_t>,
ops::LogicalOrNPUKernel<plat::NPUDeviceContext, int16_t>,
ops::LogicalOrNPUKernel<plat::NPUDeviceContext, int>,
ops::LogicalOrNPUKernel<plat::NPUDeviceContext, int64_t>,
ops::LogicalOrNPUKernel<plat::NPUDeviceContext, float>,
ops::LogicalOrNPUKernel<plat::NPUDeviceContext, double>);

REGISTER_OP_NPU_KERNEL(logical_and,
ops::LogicalAndPUKernel<plat::NPUDeviceContext, bool>);
ops::LogicalAndPUKernel<plat::NPUDeviceContext, bool>,
Copy link
Contributor

@Shixiaowei02 Shixiaowei02 Jul 19, 2021

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LogicalAndPUKernel 请确认下这个类型名称；推理认为增加这些实例带来的体积增量 9 MB 有些大了，请给出一些原因吧，谢谢

Copy link
Contributor Author

@will-jl944 will-jl944 Jul 19, 2021

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LogicalAndPuKernel这个名称是在#34159 和#34170 这两个pr中定义的。这个pr当中只增加了对更多输入数据类型的支持。

ops::LogicalAndPUKernel<plat::NPUDeviceContext, int8_t>,
ops::LogicalAndPUKernel<plat::NPUDeviceContext, int16_t>,
ops::LogicalAndPUKernel<plat::NPUDeviceContext, int>,
ops::LogicalAndPUKernel<plat::NPUDeviceContext, int64_t>,
ops::LogicalAndPUKernel<plat::NPUDeviceContext, float>,
ops::LogicalAndPUKernel<plat::NPUDeviceContext, double>);
4 changes: 2 additions & 2 deletions paddle/fluid/operators/controlflow/logical_op_xpu.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -45,7 +45,7 @@ class BinaryLogicalOpXPUKernel : public framework::OpKernel<T> {
auto* x = context.Input<framework::Tensor>("X");
auto* y = context.Input<framework::Tensor>("Y");
auto* out = context.Output<framework::Tensor>("Out");
T* out_ptr = out->mutable_data<T>(context.GetPlace());
bool* out_ptr = out->mutable_data<bool>(context.GetPlace());
const T* x_ptr = x->data<T>();
const T* y_ptr = y->data<T>();
auto& dev_ctx =
Expand Down Expand Up @@ -153,7 +153,7 @@ class UnaryLogicalOpXPUKernel : public framework::OpKernel<T> {
if (x->numel() == 0) {
return;
}
out->mutable_data<T>(context.GetPlace());
out->mutable_data<bool>(context.GetPlace());
auto& dev_ctx =
context.template device_context<paddle::platform::XPUDeviceContext>();
int ret = xpu::logical_not<bool>(dev_ctx.x_context(), x->data<T>(),
Expand Down
8 changes: 7 additions & 1 deletion paddle/fluid/operators/controlflow/logicaland_op_xpu.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,5 +17,11 @@ limitations under the License. */
namespace ops = paddle::operators;
REGISTER_OP_XPU_KERNEL(
logical_and,
ops::BinaryLogicalOpXPUKernel<ops::XpuLogicalType::XPU_AND, bool>);
ops::BinaryLogicalOpXPUKernel<ops::XpuLogicalType::XPU_AND, bool>,
ops::BinaryLogicalOpXPUKernel<ops::XpuLogicalType::XPU_AND, int8_t>,
ops::BinaryLogicalOpXPUKernel<ops::XpuLogicalType::XPU_AND, int16_t>,
ops::BinaryLogicalOpXPUKernel<ops::XpuLogicalType::XPU_AND, int>,
ops::BinaryLogicalOpXPUKernel<ops::XpuLogicalType::XPU_AND, int64_t>,
ops::BinaryLogicalOpXPUKernel<ops::XpuLogicalType::XPU_AND, float>,
ops::BinaryLogicalOpXPUKernel<ops::XpuLogicalType::XPU_AND, double>);
#endif
8 changes: 7 additions & 1 deletion paddle/fluid/operators/controlflow/logicalnot_op_xpu.cc
100755 → 100644
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,5 +15,11 @@ limitations under the License. */
#ifdef PADDLE_WITH_XPU
#include "paddle/fluid/operators/controlflow/logical_op_xpu.h"
namespace ops = paddle::operators;
REGISTER_OP_XPU_KERNEL(logicalnot, ops::UnaryLogicalOpXPUKernel<bool>);
REGISTER_OP_XPU_KERNEL(logicalnot, ops::UnaryLogicalOpXPUKernel<bool>,
ops::UnaryLogicalOpXPUKernel<int8_t>,
ops::UnaryLogicalOpXPUKernel<int16_t>,
ops::UnaryLogicalOpXPUKernel<int>,
ops::UnaryLogicalOpXPUKernel<int64_t>,
ops::UnaryLogicalOpXPUKernel<float>,
ops::UnaryLogicalOpXPUKernel<double>);
#endif
8 changes: 7 additions & 1 deletion paddle/fluid/operators/controlflow/logicalor_op_xpu.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -18,5 +18,11 @@ limitations under the License. */
namespace ops = paddle::operators;
REGISTER_OP_XPU_KERNEL(
logical_or,
ops::BinaryLogicalOpXPUKernel<ops::XpuLogicalType::XPU_OR, bool>);
ops::BinaryLogicalOpXPUKernel<ops::XpuLogicalType::XPU_OR, bool>,
ops::BinaryLogicalOpXPUKernel<ops::XpuLogicalType::XPU_OR, int8_t>,
ops::BinaryLogicalOpXPUKernel<ops::XpuLogicalType::XPU_OR, int16_t>,
ops::BinaryLogicalOpXPUKernel<ops::XpuLogicalType::XPU_OR, int>,
ops::BinaryLogicalOpXPUKernel<ops::XpuLogicalType::XPU_OR, int64_t>,
ops::BinaryLogicalOpXPUKernel<ops::XpuLogicalType::XPU_OR, float>,
ops::BinaryLogicalOpXPUKernel<ops::XpuLogicalType::XPU_OR, double>);
#endif
Loading