[pten]Move dot, conj, sign dev_api into kernel.h (#38862)

* move dot_dev api into dot_kernel.h

* add infermate header

* modify to dotkerel in dot_op.h

* mvoe conj dev api into complex_kernel.h

* move sign dev api into  sign_kernel.h

paddle/fluid/operators/conj_op.h

@@ -39,7 +39,7 @@ class ConjKernel : public framework::OpKernel<T> {
     auto pt_out = paddle::experimental::MakePtenDenseTensor(*out);
 
     // call new kernel
-    pten::ConjKernel<T>(dev_ctx, *pt_x.get(), pt_out.get());
+    pten::ConjKernel<T, DeviceContext>(dev_ctx, *pt_x.get(), pt_out.get());
   }
 };
 

paddle/fluid/operators/dot_op.h

@@ -46,7 +46,8 @@ class DotKernel : public framework::OpKernel<T> {
     auto pt_out = paddle::experimental::MakePtenDenseTensor(*out);
 
     // call new kernel
-    pten::DotKernel<T>(dev_ctx, *pt_x.get(), *pt_y.get(), pt_out.get());
+    pten::DotKernel<T, DeviceContext>(dev_ctx, *pt_x.get(), *pt_y.get(),
+                                      pt_out.get());
   }
 };
 

paddle/fluid/operators/sign_op.h

@@ -39,7 +39,7 @@ class SignKernel : public framework::OpKernel<T> {
     auto pt_out = paddle::experimental::MakePtenDenseTensor(*out);
 
     // call new kernel
-    pten::Sign<T>(dev_ctx, *pt_x.get(), pt_out.get());
+    pten::SignKernel<T, DeviceContext>(dev_ctx, *pt_x.get(), pt_out.get());
   }
 };
 

paddle/pten/all.h

@@ -17,5 +17,4 @@ limitations under the License. */
 // developer apis
 #include "paddle/pten/include/core.h"
 #include "paddle/pten/include/infermeta.h"
-#include "paddle/pten/include/linalg.h"
 #include "paddle/pten/include/math.h"

paddle/pten/include/linalg.h

-// 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.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-
-#pragma once
-
-// See Note: [ How do we organize the kernel directory ]
-#include "paddle/pten/api/lib/utils/storage.h"
-#include "paddle/pten/include/infermeta.h"
-#include "paddle/pten/kernels/dot_kernel.h"
-
-namespace pten {
-
-template <typename T, typename ContextT>
-DenseTensor Dot(const ContextT& dev_ctx,
-                const DenseTensor& x,
-                const DenseTensor& y) {
-  auto out_meta = DotInferMeta(x.meta(), y.meta());
-  pten::DenseTensor dense_out(
-      pten::make_intrusive<paddle::experimental::SharedStorage>(
-          dev_ctx.GetPlace()),
-      std::move(out_meta));
-  DotKernel<T, ContextT>(dev_ctx, x, y, &dense_out);
-  return dense_out;
-}
-
-}  // namespace pten

paddle/pten/include/math.h

@@ -17,22 +17,10 @@ limitations under the License. */
 // See Note: [ How do we organize the kernel directory ]
 #include "paddle/pten/api/lib/utils/storage.h"
 #include "paddle/pten/include/infermeta.h"
-#include "paddle/pten/kernels/complex_kernel.h"
 #include "paddle/pten/kernels/scale_kernel.h"
 
 namespace pten {
 
-template <typename T, typename ContextT>
-DenseTensor Sign(const ContextT& dev_ctx, const DenseTensor& x) {
-  auto out_meta = UnchangedInferMeta(x.meta());
-  pten::DenseTensor dense_out(
-      pten::make_intrusive<paddle::experimental::SharedStorage>(
-          dev_ctx.GetPlace()),
-      std::move(out_meta));
-  Sign<T>(dev_ctx, x, &dense_out);
-  return dense_out;
-}
-
 template <typename T, typename ContextT>
 DenseTensor Scale(const ContextT& dev_ctx,
                   const DenseTensor& x,

paddle/pten/kernels/complex_kernel.h

@@ -15,6 +15,8 @@ limitations under the License. */
 #pragma once
 
 #include "paddle/pten/core/dense_tensor.h"
+#include "paddle/pten/include/infermeta.h"
+#include "paddle/pten/kernels/empty_kernel.h"
 
 #include "paddle/pten/infermeta/unary.h"
 #include "paddle/pten/kernels/empty_kernel.h"
@@ -27,7 +29,7 @@ void ConjKernel(const Context& dev_ctx, const DenseTensor& x, DenseTensor* out);
 template <typename T, typename Context>
 DenseTensor Conj(const Context& dev_ctx, const DenseTensor& x) {
   auto out_meta = UnchangedInferMeta(x.meta());
-  auto dense_out = Empty<T, Context>(dev_ctx, std::move(out_meta));
+  auto dense_out = pten::Empty<T, Context>(dev_ctx, std::move(out_meta));
   ConjKernel<T>(dev_ctx, x, &dense_out);
   return dense_out;
 }

paddle/pten/kernels/cpu/sign_kernel.cc

@@ -21,4 +21,5 @@ limitations under the License. */
 // See Note [ Why still include the fluid headers? ]
 #include "paddle/fluid/platform/bfloat16.h"
 
-PT_REGISTER_CTX_KERNEL(sign, CPU, ALL_LAYOUT, pten::Sign, float, double) {}
+PT_REGISTER_CTX_KERNEL(sign, CPU, ALL_LAYOUT, pten::SignKernel, float, double) {
+}

paddle/pten/kernels/dot_kernel.h

@@ -15,7 +15,8 @@
 #pragma once
 
 #include "paddle/pten/core/dense_tensor.h"
-
+#include "paddle/pten/infermeta/binary.h"
+#include "paddle/pten/kernels/empty_kernel.h"
 namespace pten {
 
 template <typename T, typename Context>
@@ -24,4 +25,13 @@ void DotKernel(const Context& dev_ctx,
                const DenseTensor& y,
                DenseTensor* out);
 
+template <typename T, typename Context>
+DenseTensor Dot(const Context& dev_ctx,
+                const DenseTensor& x,
+                const DenseTensor& y) {
+  auto out_meta = DotInferMeta(x.meta(), y.meta());
+  auto dense_out = pten::Empty<T, Context>(dev_ctx, std::move(out_meta));
+  DotKernel<T, Context>(dev_ctx, x, y, &dense_out);
+  return dense_out;
+}
 }  // namespace pten

paddle/pten/kernels/gpu/sign_kernel.cu

@@ -24,4 +24,4 @@ limitations under the License. */
 using float16 = paddle::platform::float16;
 
 PT_REGISTER_CTX_KERNEL(
-    sign, GPU, ALL_LAYOUT, pten::Sign, float, double, float16) {}
+    sign, GPU, ALL_LAYOUT, pten::SignKernel, float, double, float16) {}

paddle/pten/kernels/impl/complex_kernel_impl.h

@@ -21,14 +21,14 @@
 namespace pten {
 
 template <typename T, typename Context>
-void ConjKernel(const Context& context,
+void ConjKernel(const Context& dev_ctx,
                 const DenseTensor& x,
                 DenseTensor* out) {
   auto numel = x.numel();
   auto* x_data = x.data<T>();
   auto* out_data = out->mutable_data<T>();
 
-  paddle::platform::ForRange<Context> for_range(context, numel);
+  paddle::platform::ForRange<Context> for_range(dev_ctx, numel);
   paddle::operators::math::ConjFunctor<T> functor(x_data, numel, out_data);
   for_range(functor);
 }

paddle/pten/kernels/impl/sign_kernel_impl.h

@@ -23,7 +23,9 @@ limitations under the License. */
 namespace pten {
 
 template <typename T, typename Context>
-void Sign(const Context& dev_ctx, const DenseTensor& x, DenseTensor* out) {
+void SignKernel(const Context& dev_ctx,
+                const DenseTensor& x,
+                DenseTensor* out) {
   out->mutable_data<T>();
   auto eigen_out = pten::EigenVector<T>::Flatten(*out);
   auto eigen_x = pten::EigenVector<T>::Flatten(x);

paddle/pten/kernels/sign_kernel.h

@@ -15,10 +15,20 @@ limitations under the License. */
 #pragma once
 
 #include "paddle/pten/core/dense_tensor.h"
+#include "paddle/pten/include/infermeta.h"
+#include "paddle/pten/kernels/empty_kernel.h"
 
 namespace pten {
 
 template <typename T, typename Context>
-void Sign(const Context& dev_ctx, const DenseTensor& x, DenseTensor* out);
+void SignKernel(const Context& dev_ctx, const DenseTensor& x, DenseTensor* out);
+
+template <typename T, typename Context>
+DenseTensor Sign(const Context& dev_ctx, const DenseTensor& x) {
+  auto out_meta = UnchangedInferMeta(x.meta());
+  auto dense_out = pten::Empty<T, Context>(dev_ctx, std::move(out_meta));
+  SignKernel<T, Context>(dev_ctx, x, &dense_out);
+  return dense_out;
+}
 
 }  // namespace pten

paddle/pten/tests/kernels/test_conj_dev_api.cc

@@ -15,7 +15,7 @@ limitations under the License. */
 #include <gtest/gtest.h>
 #include <memory>
 
-#include "paddle/pten/include/math.h"
+#include "paddle/pten/kernels/complex_kernel.h"
 
 #include "paddle/pten/api/lib/utils/allocator.h"
 #include "paddle/pten/core/dense_tensor.h"

paddle/pten/tests/kernels/test_dot_dev_api.cc

@@ -15,7 +15,7 @@ limitations under the License. */
 #include <gtest/gtest.h>
 #include <memory>
 
-#include "paddle/pten/include/linalg.h"
+#include "paddle/pten/kernels/dot_kernel.h"
 
 #include "paddle/pten/api/lib/utils/allocator.h"
 #include "paddle/pten/core/dense_tensor.h"