adjust main dir (#37916)

paddle/fluid/operators/elementwise/elementwise_op_function.h

@@ -31,8 +31,8 @@ limitations under the License. */
 
 // only can include the headers in paddle/pten/include dirs
 #include "paddle/pten/api/lib/utils/tensor_utils.h"
-#include "paddle/pten/kernels/functions/cpu/elementwise.h"
-#include "paddle/pten/kernels/functions/general/elementwise_base.h"
+#include "paddle/pten/kernels/hybird/cpu/elementwise.h"
+#include "paddle/pten/kernels/hybird/general/elementwise_base.h"
 
 #if defined(__NVCC__) || defined(__HIPCC__)
 #ifdef __NVCC__

paddle/fluid/operators/elementwise/elementwise_op_impl.cu.h

@@ -23,7 +23,7 @@ limitations under the License. */
 // only can include the headers in paddle/top/api dirs
 #include "paddle/pten/api/lib/utils/tensor_utils.h"
 #include "paddle/pten/include/core.h"
-#include "paddle/pten/kernels/functions/cuda/elementwise/elementwise.h"
+#include "paddle/pten/kernels/hybird/cuda/elementwise/elementwise.h"
 
 #ifdef __HIPCC__
 #define ELEMENTWISE_BLOCK_SIZE 256

paddle/fluid/operators/math/math_function.cc

@@ -29,7 +29,7 @@ limitations under the License. */
 #include "paddle/fluid/operators/math/math_function_impl.h"
 #include "paddle/fluid/platform/bfloat16.h"
 #include "paddle/fluid/platform/float16.h"
-#include "paddle/pten/kernels/functions/eigen/common.h"
+#include "paddle/pten/kernels/hybird/eigen/common.h"
 #include "unsupported/Eigen/CXX11/Tensor"
 
 namespace paddle {

paddle/fluid/operators/math/math_function.cu

@@ -21,7 +21,7 @@ limitations under the License. */
 #include "paddle/fluid/operators/math/math_function_impl.h"
 #include "paddle/fluid/platform/bfloat16.h"
 #include "paddle/fluid/platform/float16.h"
-#include "paddle/pten/kernels/functions/eigen/common.h"
+#include "paddle/pten/kernels/hybird/eigen/common.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/reduce_ops/reduce_op.h

@@ -28,7 +28,7 @@ limitations under the License. */
 #include "paddle/pten/api/lib/utils/tensor_utils.h"
 #include "paddle/pten/include/core.h"
 #include "paddle/pten/include/math.h"
-#include "paddle/pten/kernels/functions/general/reduce_impl.h"
+#include "paddle/pten/kernels/hybird/general/reduce_impl.h"
 
 #if defined(__HIPCC__) || defined(__NVCC__)
 #include "paddle/fluid/operators/reduce_ops/reduce_op.cu.h"

paddle/pten/CMakeLists.txt

@@ -11,10 +11,14 @@ add_subdirectory(common)
 add_subdirectory(api)
 # pten core components
 add_subdirectory(core)
+# pten components of specific backends
+add_subdirectory(backends)
 # pten kernels for diff device
 add_subdirectory(kernels)
 # pten infermeta
 add_subdirectory(infermeta)
+# pten operator definitions
+add_subdirectory(ops)
 # pten tests
 add_subdirectory(tests)
 

paddle/pten/infermeta/binary.cc

@@ -14,7 +14,7 @@ limitations under the License. */
 
 // See Note [ Why still include the fluid headers? ]
 #include "paddle/pten/infermeta/binary.h"
-#include "paddle/pten/kernels/functions/general/elementwise_base.h"
+#include "paddle/pten/kernels/hybird/general/elementwise_base.h"
 
 namespace pten {
 

paddle/pten/kernels/CMakeLists.txt

-# pten math functions called by kernels
-add_subdirectory(math)
-# pten basic functions called by kernels
-add_subdirectory(functions)
-# pten kernels for diff device
+# kernel primitive api
+add_subdirectory(primitive)
+# pten hybird functors and functions called by kernels
+add_subdirectory(hybird)
+
+# pten kernels for different backends
+# NOTE(chenweihang): We need to increase the compilation option of WITH_EIGEN,
+# which will support splitting eigen at compile time on demand in the future
+add_subdirectory(eigen)
+# NOTE(chenweihang): We need to increase the compilation option of WITH_BLAS,
+# which will support splitting eigen at compile time on demand in the future,
+# and if necessary, blas can be split into openblas and cublas
+add_subdirectory(blas)
 add_subdirectory(cpu)
 if(WITH_GPU OR WITH_ROCM)
-  # TODO(chenweihang): if hip can split from cuda impl, we should add hip dir
+  # NOTE(chenweihang): if hip can split from cuda impl, we should add hip dir
   add_subdirectory(cuda)
 endif()
-# TODO(chenweihang): migrate MKLDNN Kernel in the second phase of the project
 if(WITH_MKLDNN)
-  add_subdirectory(mkldnn)
+  # mkldnn will be deprecated and use the new name dnnl
+  add_subdirectory(dnnl)
 endif()
-# TODO(chenweihang): migrate NPU Kernel in the second phase of the project
 if(WITH_ASCEND_CL)
   add_subdirectory(npu)
 endif()
-# TODO(chenweihang): migrate XPU Kernel in the second phase of the project
 if(WITH_XPU)
   add_subdirectory(xpu)
 endif()

paddle/pten/kernels/cpu/creation.cc

@@ -15,7 +15,7 @@
 #include "paddle/pten/kernels/cpu/creation.h"
 
 #include "paddle/pten/core/kernel_registry.h"
-#include "paddle/pten/kernels/functions/eigen/fill.h"
+#include "paddle/pten/kernels/hybird/eigen/fill.h"
 
 namespace pten {
 

paddle/pten/kernels/cpu/linalg.cc

@@ -21,7 +21,7 @@
 #include "paddle/fluid/operators/math/blas.h"
 #include "paddle/fluid/platform/complex.h"
 
-#include "paddle/pten/kernels/functions/math/matmul_func.h"
+#include "paddle/pten/kernels/hybird/math/matmul_func.h"
 
 namespace pten {
 

paddle/pten/kernels/cpu/manipulation.cc

@@ -16,8 +16,8 @@
 #include "paddle/pten/api/ext/dispatch.h"
 #include "paddle/pten/infermeta/unary.h"
 #include "paddle/pten/kernels/cpu/utils.h"
-#include "paddle/pten/kernels/functions/general/manipulation.h"
-#include "paddle/pten/kernels/functions/math/cast_func.h"
+#include "paddle/pten/kernels/hybird/general/manipulation.h"
+#include "paddle/pten/kernels/hybird/math/cast_func.h"
 
 namespace pten {
 

paddle/pten/kernels/cpu/math.cc

@@ -15,12 +15,12 @@
 #include "paddle/pten/kernels/cpu/math.h"
 
 #include "paddle/pten/api/ext/dispatch.h"
-#include "paddle/pten/kernels/functions/cpu/elementwise.h"
-#include "paddle/pten/kernels/functions/eigen/reduce.h"
-#include "paddle/pten/kernels/functions/eigen/scale.h"
-#include "paddle/pten/kernels/functions/eigen/sign.h"
-#include "paddle/pten/kernels/functions/general/elementwise_functor.h"
-#include "paddle/pten/kernels/functions/general/reduce_impl.h"
+#include "paddle/pten/kernels/hybird/cpu/elementwise.h"
+#include "paddle/pten/kernels/hybird/eigen/reduce.h"
+#include "paddle/pten/kernels/hybird/eigen/scale.h"
+#include "paddle/pten/kernels/hybird/eigen/sign.h"
+#include "paddle/pten/kernels/hybird/general/elementwise_functor.h"
+#include "paddle/pten/kernels/hybird/general/reduce_impl.h"
 
 // See Note [ Why still include the fluid headers? ]
 #include "paddle/fluid/framework/eigen.h"

paddle/pten/kernels/cuda/creation.cu

@@ -15,7 +15,7 @@
 #include "paddle/pten/kernels/cuda/creation.h"
 
 #include "paddle/pten/core/kernel_registry.h"
-#include "paddle/pten/kernels/functions/eigen/fill.h"
+#include "paddle/pten/kernels/hybird/eigen/fill.h"
 
 namespace pten {
 

paddle/pten/kernels/cuda/linalg.cu

@@ -15,8 +15,8 @@
 #include "paddle/pten/kernels/cuda/linalg.h"
 
 #include "paddle/pten/core/kernel_registry.h"
-#include "paddle/pten/kernels/functions/eigen/dot.h"
-#include "paddle/pten/kernels/functions/math/matmul_func.h"
+#include "paddle/pten/kernels/hybird/eigen/dot.h"
+#include "paddle/pten/kernels/hybird/math/matmul_func.h"
 
 // See Note [ Why still include the fluid headers? ]
 #include "paddle/fluid/platform/complex.h"

paddle/pten/kernels/cuda/manipulation.cu

@@ -16,8 +16,8 @@
 #include "paddle/pten/infermeta/unary.h"
 #include "paddle/pten/kernels/cuda/manipulation.h"
 #include "paddle/pten/kernels/cuda/utils.h"
-#include "paddle/pten/kernels/functions/cuda/cast_kernel_impl.h"
-#include "paddle/pten/kernels/functions/general/manipulation.h"
+#include "paddle/pten/kernels/hybird/cuda/cast_kernel_impl.h"
+#include "paddle/pten/kernels/hybird/general/manipulation.h"
 
 namespace pten {
 

paddle/pten/kernels/cuda/math.cu

@@ -15,12 +15,12 @@ limitations under the License. */
 #include "paddle/pten/kernels/cuda/math.h"
 
 #include "paddle/fluid/operators/reduce_ops/reduce_functor_op.h"
-#include "paddle/pten/kernels/functions/cuda/elementwise/elementwise.h"
-#include "paddle/pten/kernels/functions/cuda/reduce/reduce.h"
-#include "paddle/pten/kernels/functions/eigen/scale.h"
-#include "paddle/pten/kernels/functions/eigen/sign.h"
-#include "paddle/pten/kernels/functions/general/elementwise_functor.h"
-#include "paddle/pten/kernels/functions/general/reduce_impl.h"
+#include "paddle/pten/kernels/hybird/cuda/elementwise/elementwise.h"
+#include "paddle/pten/kernels/hybird/cuda/reduce/reduce.h"
+#include "paddle/pten/kernels/hybird/eigen/scale.h"
+#include "paddle/pten/kernels/hybird/eigen/sign.h"
+#include "paddle/pten/kernels/hybird/general/elementwise_functor.h"
+#include "paddle/pten/kernels/hybird/general/reduce_impl.h"
 
 #ifdef __NVCC__
 #include "cub/cub.cuh"

paddle/pten/kernels/functions/CMakeLists.txt

-add_subdirectory(eigen)
-add_subdirectory(blas)
-add_subdirectory(general)

paddle/pten/kernels/math/cuda/CMakeLists.txt → paddle/pten/kernels/hybird/CMakeLists.txt

+add_subdirectory(eigen)
+add_subdirectory(blas)
+add_subdirectory(general)
+
+cc_library(pten_transpose_cpu SRCS transpose.cc DEPS dense_tensor device_context)
 if(WITH_GPU)
   nv_library(pten_transpose_cuda SRCS transpose.cu DEPS dense_tensor malloc device_context)
 elseif(WITH_ROCM)

paddle/pten/kernels/functions/cpu/elementwise.h → paddle/pten/kernels/hybird/cpu/elementwise.h

@@ -15,7 +15,7 @@ limitations under the License. */
 #pragma once
 
 #include "paddle/pten/core/dense_tensor.h"
-#include "paddle/pten/kernels/functions/general/elementwise_base.h"
+#include "paddle/pten/kernels/hybird/general/elementwise_base.h"
 
 namespace pten {
 

paddle/pten/kernels/functions/cuda/elementwise/elementwise.h → paddle/pten/kernels/hybird/cuda/elementwise/elementwise.h

@@ -14,8 +14,8 @@ limitations under the License. */
 
 #pragma once
 
-#include "paddle/pten/kernels/functions/cuda/elementwise/elementwise_broadcast.cu.h"
-#include "paddle/pten/kernels/functions/cuda/elementwise/elementwise_no_broadcast.cu.h"
+#include "paddle/pten/kernels/hybird/cuda/elementwise/elementwise_broadcast.cu.h"
+#include "paddle/pten/kernels/hybird/cuda/elementwise/elementwise_no_broadcast.cu.h"
 
 namespace pten {
 

paddle/pten/kernels/functions/cuda/elementwise/elementwise_broadcast.cu.h → paddle/pten/kernels/hybird/cuda/elementwise/elementwise_broadcast.cu.h

@@ -15,7 +15,7 @@ limitations under the License. */
 #pragma once
 
 #include "paddle/pten/core/dense_tensor.h"
-#include "paddle/pten/kernels/functions/cuda/elementwise/elementwise_common.cu.h"
+#include "paddle/pten/kernels/hybird/cuda/elementwise/elementwise_common.cu.h"
 
 namespace pten {
 

paddle/pten/kernels/functions/cuda/elementwise/elementwise_common.cu.h → paddle/pten/kernels/hybird/cuda/elementwise/elementwise_common.cu.h

@@ -18,7 +18,7 @@ limitations under the License. */
 #include "paddle/fluid/platform/aligned_vector.h"
 #include "paddle/fluid/platform/function_traits.h"
 #include "paddle/pten/core/dense_tensor.h"
-#include "paddle/pten/kernels/functions/general/elementwise_base.h"
+#include "paddle/pten/kernels/hybird/general/elementwise_base.h"
 
 namespace pten {
 namespace kps = paddle::operators::kernel_primitives;

paddle/pten/kernels/functions/cuda/elementwise/elementwise_no_broadcast.cu.h → paddle/pten/kernels/hybird/cuda/elementwise/elementwise_no_broadcast.cu.h

@@ -14,7 +14,7 @@ limitations under the License. */
 
 #pragma once
 
-#include "paddle/pten/kernels/functions/cuda/elementwise/elementwise_common.cu.h"
+#include "paddle/pten/kernels/hybird/cuda/elementwise/elementwise_common.cu.h"
 
 #ifdef __HIPCC__
 #define ELEMENTWISE_BLOCK_SIZE 256

paddle/pten/kernels/functions/cuda/reduce/reduce.h → paddle/pten/kernels/hybird/cuda/reduce/reduce.h

@@ -21,7 +21,7 @@
 #include "paddle/pten/core/dense_tensor.h"
 
 #include "paddle/fluid/platform/device_context.h"
-#include "paddle/pten/kernels/functions/cuda/reduce/reduce_cuda_impl.h"
+#include "paddle/pten/kernels/hybird/cuda/reduce/reduce_cuda_impl.h"
 
 namespace pten {
 

paddle/pten/kernels/functions/cuda/reduce/reduce_cuda_impl.h → paddle/pten/kernels/hybird/cuda/reduce/reduce_cuda_impl.h

@@ -43,7 +43,7 @@ namespace cub = hipcub;
 #include "paddle/pten/api/ext/dispatch.h"
 #include "paddle/pten/api/include/tensor.h"
 #include "paddle/pten/kernels/cuda/utils.h"
-#include "paddle/pten/kernels/functions/math/cast_func.h"
+#include "paddle/pten/kernels/hybird/math/cast_func.h"
 
 // Reduce split or not, Whether to use ReduceHigherDim
 #define REDUCE_SPLIT_BOUNDARY 512

paddle/pten/kernels/functions/eigen/dot.h → paddle/pten/kernels/hybird/eigen/dot.h

@@ -15,7 +15,7 @@ limitations under the License. */
 #pragma once
 
 #include "paddle/pten/core/dense_tensor.h"
-#include "paddle/pten/kernels/functions/eigen/common.h"
+#include "paddle/pten/kernels/hybird/eigen/common.h"
 
 // See Note [ Why still include the fluid headers? ]
 #include "paddle/fluid/operators/eigen/eigen_function.h"

paddle/pten/kernels/functions/eigen/elementwise.h → paddle/pten/kernels/hybird/eigen/elementwise.h

@@ -15,7 +15,7 @@ limitations under the License. */
 #pragma once
 
 #include "paddle/pten/core/dense_tensor.h"
-#include "paddle/pten/kernels/functions/eigen/common.h"
+#include "paddle/pten/kernels/hybird/eigen/common.h"
 
 namespace pten {
 namespace eigen {

paddle/pten/kernels/functions/eigen/fill.h → paddle/pten/kernels/hybird/eigen/fill.h

@@ -15,7 +15,7 @@ limitations under the License. */
 #pragma once
 
 #include "paddle/pten/core/dense_tensor.h"
-#include "paddle/pten/kernels/functions/eigen/common.h"
+#include "paddle/pten/kernels/hybird/eigen/common.h"
 
 // See Note [ Why still include the fluid headers? ]
 #include "paddle/fluid/operators/eigen/eigen_function.h"

paddle/pten/kernels/functions/eigen/reduce.h → paddle/pten/kernels/hybird/eigen/reduce.h

@@ -16,8 +16,8 @@
 
 #include "paddle/pten/api/lib/utils/allocator.h"
 #include "paddle/pten/core/dense_tensor.h"
-#include "paddle/pten/kernels/functions/eigen/common.h"
-#include "paddle/pten/kernels/math/transpose.h"
+#include "paddle/pten/kernels/hybird/eigen/common.h"
+#include "paddle/pten/kernels/hybird/transpose.h"
 
 // See Note [ Why still include the fluid headers? ]
 #include "paddle/fluid/operators/eigen/eigen_function.h"

paddle/pten/kernels/functions/eigen/scale.h → paddle/pten/kernels/hybird/eigen/scale.h

@@ -15,7 +15,7 @@ limitations under the License. */
 #pragma once
 
 #include "paddle/pten/core/dense_tensor.h"
-#include "paddle/pten/kernels/functions/eigen/common.h"
+#include "paddle/pten/kernels/hybird/eigen/common.h"
 
 // See Note [ Why still include the fluid headers? ]
 #include "paddle/fluid/operators/eigen/eigen_function.h"

paddle/pten/kernels/functions/eigen/sign.h → paddle/pten/kernels/hybird/eigen/sign.h

@@ -15,7 +15,7 @@ limitations under the License. */
 #pragma once
 
 #include "paddle/pten/core/dense_tensor.h"
-#include "paddle/pten/kernels/functions/eigen/common.h"
+#include "paddle/pten/kernels/hybird/eigen/common.h"
 
 // See Note [ Why still include the fluid headers? ]
 #include "paddle/fluid/operators/eigen/eigen_function.h"

paddle/pten/kernels/functions/general/elementwise_functor.h → paddle/pten/kernels/hybird/general/elementwise_functor.h

@@ -18,8 +18,8 @@ limitations under the License. */
 #include "paddle/fluid/platform/float16.h"
 #include "paddle/fluid/platform/hostdevice.h"
 #include "paddle/pten/core/dense_tensor.h"
-#include "paddle/pten/kernels/functions/blas/elementwise.h"
-#include "paddle/pten/kernels/functions/eigen/elementwise.h"
+#include "paddle/pten/kernels/hybird/blas/elementwise.h"
+#include "paddle/pten/kernels/hybird/eigen/elementwise.h"
 
 namespace pten {
 namespace general {

paddle/pten/kernels/functions/general/reduce_impl.h → paddle/pten/kernels/hybird/general/reduce_impl.h

@@ -16,8 +16,8 @@
 #include "paddle/fluid/platform/transform.h"
 #include "paddle/pten/api/ext/dispatch.h"
 #include "paddle/pten/core/dense_tensor.h"
-#include "paddle/pten/kernels/functions/eigen/reduce.h"
-#include "paddle/pten/kernels/functions/math/cast_func.h"
+#include "paddle/pten/kernels/hybird/eigen/reduce.h"
+#include "paddle/pten/kernels/hybird/math/cast_func.h"
 namespace pten {
 namespace general {
 

paddle/pten/kernels/math/cpu/transpose.cc → paddle/pten/kernels/hybird/transpose.cc

@@ -12,7 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-#include "paddle/pten/kernels/math/transpose.h"
+#include "paddle/pten/kernels/hybird/transpose.h"
 #include "paddle/fluid/framework/ddim.h"
 #include "paddle/pten/core/dense_tensor.h"
 

paddle/pten/kernels/math/cuda/transpose.cu → paddle/pten/kernels/hybird/transpose.cu

@@ -15,8 +15,8 @@
 #include "paddle/fluid/framework/ddim.h"
 #include "paddle/fluid/memory/memcpy.h"
 #include "paddle/pten/core/dense_tensor.h"
-#include "paddle/pten/kernels/functions/math/cast_func.h"
-#include "paddle/pten/kernels/math/transpose.h"
+#include "paddle/pten/kernels/hybird/math/cast_func.h"
+#include "paddle/pten/kernels/hybird/transpose.h"
 
 // See Note [ Why still include the fluid headers? ]
 #include "paddle/fluid/platform/bfloat16.h"

paddle/pten/kernels/math/CMakeLists.txt

-add_subdirectory(cpu)
-if(WITH_GPU OR WITH_ROCM)
-  add_subdirectory(cuda)
-endif()

paddle/pten/kernels/math/cpu/CMakeLists.txt

-cc_library(pten_transpose_cpu SRCS transpose.cc DEPS dense_tensor device_context)

paddle/pten/kernels/xpu/manipulation.cc

@@ -14,7 +14,7 @@
 
 #include "paddle/pten/kernels/xpu/manipulation.h"
 #include "paddle/pten/infermeta/unary.h"
-#include "paddle/pten/kernels/functions/general/manipulation.h"
+#include "paddle/pten/kernels/hybird/general/manipulation.h"
 #include "paddle/pten/kernels/xpu/utils.h"
 
 namespace pten {