Skip to content
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

[X86] add elementwise_pow, sqrt #5928

Merged
merged 1 commit into from
Apr 21, 2021
Merged

[X86] add elementwise_pow, sqrt #5928

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
17 changes: 10 additions & 7 deletions lite/kernels/x86/activation_compute.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,7 +14,6 @@

#include "lite/kernels/x86/activation_compute.h"

// float
REGISTER_LITE_KERNEL(square,
kX86,
kFloat,
Expand All @@ -25,7 +24,6 @@ REGISTER_LITE_KERNEL(square,
.BindOutput("Out", {LiteType::GetTensorTy(TARGET(kX86))})
.Finalize();

// float
REGISTER_LITE_KERNEL(relu,
kX86,
kFloat,
Expand All @@ -36,7 +34,6 @@ REGISTER_LITE_KERNEL(relu,
.BindOutput("Out", {LiteType::GetTensorTy(TARGET(kX86))})
.Finalize();

// float
REGISTER_LITE_KERNEL(leaky_relu,
kX86,
kFloat,
Expand All @@ -48,7 +45,6 @@ REGISTER_LITE_KERNEL(leaky_relu,
.BindPaddleOpVersion("leaky_relu", 1)
.Finalize();

// float
REGISTER_LITE_KERNEL(tanh,
kX86,
kFloat,
Expand All @@ -59,7 +55,6 @@ REGISTER_LITE_KERNEL(tanh,
.BindOutput("Out", {LiteType::GetTensorTy(TARGET(kX86))})
.Finalize();

// float
REGISTER_LITE_KERNEL(gelu,
kX86,
kFloat,
Expand All @@ -80,7 +75,6 @@ REGISTER_LITE_KERNEL(softsign,
.BindOutput("Out", {LiteType::GetTensorTy(TARGET(kX86))})
.Finalize();

// float
REGISTER_LITE_KERNEL(sigmoid,
kX86,
kFloat,
Expand All @@ -91,7 +85,6 @@ REGISTER_LITE_KERNEL(sigmoid,
.BindOutput("Out", {LiteType::GetTensorTy(TARGET(kX86))})
.Finalize();

// float
REGISTER_LITE_KERNEL(relu6,
kX86,
kFloat,
Expand All @@ -101,3 +94,13 @@ REGISTER_LITE_KERNEL(relu6,
.BindInput("X", {LiteType::GetTensorTy(TARGET(kX86))})
.BindOutput("Out", {LiteType::GetTensorTy(TARGET(kX86))})
.Finalize();

REGISTER_LITE_KERNEL(sqrt,
kX86,
kFloat,
kNCHW,
paddle::lite::kernels::x86::SqrtCompute<float>,
def)
.BindInput("X", {LiteType::GetTensorTy(TARGET(kX86))})
.BindOutput("Out", {LiteType::GetTensorTy(TARGET(kX86))})
.Finalize();
22 changes: 22 additions & 0 deletions lite/kernels/x86/activation_compute.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -307,6 +307,28 @@ class Relu6Compute : public KernelLite<TARGET(kX86), PRECISION(kFloat)> {
virtual ~Relu6Compute() = default;
};

template <typename T>
struct SqrtFunctor : public BaseActivationFunctor<T> {
template <typename Device, typename X, typename Out>
void operator()(Device d, X x, Out out) const {
out.device(d) = x.sqrt();
}
};

template <typename T>
class SqrtCompute : public KernelLite<TARGET(kX86), PRECISION(kFloat)> {
public:
using param_t = operators::ActivationParam;

void Run() override {
auto& param = *param_.get_mutable<operators::ActivationParam>();
param.Out->template mutable_data<T>();
Activate<SqrtFunctor<T>>(param.X, param.Out);
}

virtual ~SqrtCompute() = default;
};

} // namespace x86
} // namespace kernels
} // namespace lite
Expand Down
11 changes: 11 additions & 0 deletions lite/kernels/x86/elementwise_compute.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -182,6 +182,17 @@ REGISTER_LITE_KERNEL(
.BindOutput("Out", {LiteType::GetTensorTy(TARGET(kX86), PRECISION(kInt64))})
.Finalize();

REGISTER_LITE_KERNEL(elementwise_pow,
kX86,
kFloat,
kNCHW,
paddle::lite::kernels::x86::ElementwisePowCompute<float>,
def)
.BindInput("X", {LiteType::GetTensorTy(TARGET(kX86), PRECISION(kFloat))})
.BindInput("Y", {LiteType::GetTensorTy(TARGET(kX86), PRECISION(kFloat))})
.BindOutput("Out", {LiteType::GetTensorTy(TARGET(kX86), PRECISION(kFloat))})
.Finalize();

REGISTER_LITE_KERNEL(elementwise_mod,
kX86,
kFloat,
Expand Down
40 changes: 27 additions & 13 deletions lite/kernels/x86/elementwise_compute.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -11,8 +11,9 @@
// 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.
#pragma once

#pragma once
#include <cmath>
#include "lite/core/kernel.h"
#include "lite/core/op_registry.h"
#include "lite/fluid/eigen.h"
Expand Down Expand Up @@ -50,6 +51,11 @@ struct FloorDivFunctor {
}
};

template <typename T>
struct PowFunctor {
inline HOSTDEVICE T operator()(T a, T b) const { return std::pow(a, b); }
};

template <typename T>
struct ModFunctor {
inline HOSTDEVICE T operator()(T a, T b) const {
Expand Down Expand Up @@ -78,9 +84,7 @@ class ElementwiseAddCompute
auto& param = *param_.get_mutable<param_t>();
auto& context = ctx_->As<X86Context>();
param.Out->template mutable_data<T>();
paddle::lite::kernels::x86::ElementwiseComputeEx<AddFunctor<T>,
lite::TargetType::kX86,
T>(
ElementwiseComputeEx<AddFunctor<T>, lite::TargetType::kX86, T>(
context, param.X, param.Y, param.axis, AddFunctor<T>(), param.Out);
}

Expand All @@ -98,9 +102,7 @@ class ElementwiseSubCompute
auto& context = ctx_->As<X86Context>();

param.Out->template mutable_data<T>();
paddle::lite::kernels::x86::ElementwiseComputeEx<SubFunctor<T>,
lite::TargetType::kX86,
T>(
ElementwiseComputeEx<SubFunctor<T>, lite::TargetType::kX86, T>(
context, param.X, param.Y, param.axis, SubFunctor<T>(), param.Out);
}

Expand All @@ -116,9 +118,7 @@ class ElementwiseMulCompute
auto& param = *param_.get_mutable<param_t>();
auto& context = ctx_->As<X86Context>();
param.Out->template mutable_data<T>();
paddle::lite::kernels::x86::ElementwiseComputeEx<MulFunctor<T>,
lite::TargetType::kX86,
T>(
ElementwiseComputeEx<MulFunctor<T>, lite::TargetType::kX86, T>(
context, param.X, param.Y, param.axis, MulFunctor<T>(), param.Out);
}

Expand All @@ -134,9 +134,7 @@ class ElementwiseDivCompute
auto& param = *param_.get_mutable<param_t>();
auto& context = ctx_->As<X86Context>();
param.Out->template mutable_data<T>();
paddle::lite::kernels::x86::ElementwiseComputeEx<DivFunctor<T>,
lite::TargetType::kX86,
T>(
ElementwiseComputeEx<DivFunctor<T>, lite::TargetType::kX86, T>(
context, param.X, param.Y, param.axis, DivFunctor<T>(), param.Out);
}

Expand All @@ -159,6 +157,22 @@ class ElementwiseFloorDivCompute
virtual ~ElementwiseFloorDivCompute() = default;
};

template <typename T>
class ElementwisePowCompute
: public KernelLite<TARGET(kX86), PRECISION(kFloat)> {
public:
using param_t = operators::ElementwiseParam;
void Run() override {
auto& param = *param_.get_mutable<param_t>();
auto& context = ctx_->As<X86Context>();
param.Out->template mutable_data<T>();
ElementwiseComputeEx<PowFunctor<T>, lite::TargetType::kX86, T>(
context, param.X, param.Y, param.axis, PowFunctor<T>(), param.Out);
}

virtual ~ElementwisePowCompute() = default;
};

template <typename T>
class ElementwiseModCompute
: public KernelLite<TARGET(kX86), PRECISION(kFloat)> {
Expand Down
2 changes: 2 additions & 0 deletions lite/tests/kernels/activation_compute_test.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -687,6 +687,8 @@ TEST(Activation_sqrt, precision) {
place = Place(TARGET(kOpenCL), PRECISION(kFP16), DATALAYOUT(kImageDefault));
#elif defined(LITE_WITH_ARM)
place = TARGET(kARM);
#elif defined(LITE_WITH_X86)
place = TARGET(kX86);
#else
return;
#endif
Expand Down
11 changes: 6 additions & 5 deletions lite/tests/kernels/elementwise_compute_test.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -64,6 +64,11 @@ T floordiv(T a, T b) {
return static_cast<T>(std::trunc(a / b));
}

template <class T>
T pow(T a, T b) {
return std::pow(a, b);
}

template <class T>
T max(T a, T b) {
return std::max(a, b);
Expand All @@ -74,11 +79,6 @@ T min(T a, T b) {
return std::min(a, b);
}

template <class T>
T pow(T a, T b) {
return std::pow(a, b);
}

template <class T>
T mod(T a, T b) {
T res = a % b;
Expand Down Expand Up @@ -379,6 +379,7 @@ TEST(elementwise_x86, precison) {
TestEltX86<int64_t>(place, abs_error, op, "int64");
}

TestEltX86<float>(place, abs_error, "pow", "def");
TestEltX86<int>(place, abs_error, "mod", "int32");
TestEltX86<int64_t>(place, abs_error, "mod", "int64");
}
Expand Down