refine reshard (#58734)

ForFishes · web-flow · commit 7c70a8c8052f · 2023-11-07T13:12:15.000+08:00
diff --git a/paddle/fluid/eager/grad_node_info.h b/paddle/fluid/eager/grad_node_info.h
@@ -21,7 +21,7 @@
 #include "paddle/fluid/eager/hooks.h"
 #include "paddle/phi/api/all.h"
 #include "paddle/phi/core/distributed/auto_parallel/dist_attr.h"
-#include "paddle/phi/core/distributed/auto_parallel/reshard_utils.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/reshard_utils.h"
 #include "paddle/utils/test_macros.h"
 
 namespace egr {
diff --git a/paddle/fluid/operators/collective/c_comm_init_op.cc b/paddle/fluid/operators/collective/c_comm_init_op.cc
@@ -34,7 +34,7 @@ limitations under the License. */
 PHI_DECLARE_bool(dynamic_static_unified_comm);
 #endif
 
-#include "paddle/phi/core/distributed/auto_parallel/reshard_utils.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/reshard_utils.h"
 #include "paddle/phi/core/distributed/comm_context_manager.h"
 #include "paddle/phi/core/distributed/store/store_utils.h"
 #include "paddle/phi/core/distributed/store/tcp_store.h"
diff --git a/paddle/fluid/pybind/auto_parallel_py.cc b/paddle/fluid/pybind/auto_parallel_py.cc
@@ -40,13 +40,13 @@
 #include "paddle/phi/backends/context_pool.h"
 #include "paddle/phi/core/dense_tensor.h"
 #include "paddle/phi/core/distributed/auto_parallel/dist_tensor.h"
-#include "paddle/phi/core/distributed/auto_parallel/nd_mesh_reshard_function.h"
-#include "paddle/phi/core/distributed/auto_parallel/p_to_r_reshard_function.h"
-#include "paddle/phi/core/distributed/auto_parallel/r_to_p_reshard_function.h"
-#include "paddle/phi/core/distributed/auto_parallel/r_to_s_reshard_function.h"
-#include "paddle/phi/core/distributed/auto_parallel/s_to_r_reshard_function.h"
-#include "paddle/phi/core/distributed/auto_parallel/s_to_s_reshard_function.h"
-#include "paddle/phi/core/distributed/auto_parallel/same_status_reshard_function.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/nd_mesh_reshard_function.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/p_to_r_reshard_function.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/r_to_p_reshard_function.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/r_to_s_reshard_function.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/s_to_r_reshard_function.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/s_to_s_reshard_function.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/same_status_reshard_function.h"
 #include "paddle/phi/core/enforce.h"
 
 #ifdef PADDLE_WITH_DISTRIBUTE
diff --git a/paddle/fluid/pybind/eager_method.cc b/paddle/fluid/pybind/eager_method.cc
@@ -61,8 +61,8 @@ typedef SSIZE_T ssize_t;
 #include "paddle/phi/api/lib/data_transform.h"
 #include "paddle/phi/core/ddim.h"
 #include "paddle/phi/core/distributed/auto_parallel/dist_tensor.h"
-#include "paddle/phi/core/distributed/auto_parallel/reshard_function.h"
-#include "paddle/phi/core/distributed/auto_parallel/reshard_utils.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/reshard_function.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/reshard_utils.h"
 #include "paddle/phi/core/flags.h"
 #include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
diff --git a/paddle/phi/api/include/tensor_utils.h b/paddle/phi/api/include/tensor_utils.h
@@ -18,8 +18,8 @@ limitations under the License. */
 
 #include "paddle/phi/api/include/tensor.h"
 #ifdef PADDLE_WITH_DISTRIBUTE
-#include "paddle/phi/core/distributed/auto_parallel/reshard_function.h"
-#include "paddle/phi/core/distributed/auto_parallel/reshard_utils.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/reshard_function.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/reshard_utils.h"
 #endif
 
 namespace paddle {
diff --git a/paddle/phi/api/lib/api_custom_impl.cc b/paddle/phi/api/lib/api_custom_impl.cc
@@ -29,7 +29,7 @@ limitations under the License. */
 #include "paddle/phi/infermeta/nullary.h"
 #include "paddle/phi/infermeta/unary.h"
 #ifdef PADDLE_WITH_DISTRIBUTE
-#include "paddle/phi/core/distributed/auto_parallel/reshard_utils.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/reshard_utils.h"
 #include "paddle/phi/infermeta/spmd_rules/rules.h"
 #endif
 namespace paddle {
diff --git a/paddle/phi/api/lib/data_transform.cc b/paddle/phi/api/lib/data_transform.cc
@@ -23,8 +23,8 @@ limitations under the License. */
 #include "paddle/phi/api/lib/utils/allocator.h"
 #include "paddle/phi/backends/context_pool.h"
 #include "paddle/phi/core/distributed/auto_parallel/dist_tensor.h"
-#include "paddle/phi/core/distributed/auto_parallel/reshard_function.h"
-#include "paddle/phi/core/distributed/auto_parallel/reshard_utils.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/reshard_function.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/reshard_utils.h"
 #include "paddle/phi/core/flags.h"
 #include "paddle/phi/core/kernel_registry.h"
 #include "paddle/phi/core/tensor_utils.h"
diff --git a/paddle/phi/api/lib/tensor_copy.cc b/paddle/phi/api/lib/tensor_copy.cc
@@ -26,7 +26,7 @@ limitations under the License. */
 #include "paddle/phi/infermeta/unary.h"
 #ifdef PADDLE_WITH_DISTRIBUTE
 #include "paddle/phi/api/lib/data_transform.h"
-#include "paddle/phi/core/distributed/auto_parallel/reshard_utils.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/reshard_utils.h"
 #include "paddle/phi/infermeta/spmd_rules/rules.h"
 #endif
 namespace paddle {
diff --git a/paddle/phi/api/lib/tensor_method.cc b/paddle/phi/api/lib/tensor_method.cc
@@ -29,7 +29,7 @@ limitations under the License. */
 // clang-format off
 #ifdef PADDLE_WITH_DISTRIBUTE
 #include "paddle/phi/infermeta/spmd_rules/rules.h"
-#include "paddle/phi/core/distributed/auto_parallel/reshard_utils.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/reshard_utils.h"
 #include "paddle/phi/api/lib/data_transform.h"
 #endif
 namespace paddle {
diff --git a/paddle/phi/api/yaml/generator/api_gen.py b/paddle/phi/api/yaml/generator/api_gen.py
@@ -382,7 +382,7 @@ def source_include(header_file_path):
 
 #ifdef PADDLE_WITH_DISTRIBUTE
 #include "paddle/phi/infermeta/spmd_rules/rules.h"
-#include "paddle/phi/core/distributed/auto_parallel/reshard_utils.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/reshard_utils.h"
 #endif
 
 PD_DECLARE_bool(conv2d_disable_cudnn);
diff --git a/paddle/phi/api/yaml/generator/dist_bw_api_gen.py b/paddle/phi/api/yaml/generator/dist_bw_api_gen.py
@@ -340,7 +340,7 @@ def source_include(header_file_path, fw_header_file_path):
 
 #ifdef PADDLE_WITH_DISTRIBUTE
 #include "paddle/phi/infermeta/spmd_rules/rules.h"
-#include "paddle/phi/core/distributed/auto_parallel/reshard_utils.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/reshard_utils.h"
 #endif
 
 PD_DECLARE_bool(conv2d_disable_cudnn);
diff --git a/paddle/phi/core/distributed/auto_parallel/CMakeLists.txt b/paddle/phi/core/distributed/auto_parallel/CMakeLists.txt
@@ -7,15 +7,8 @@ collect_srcs(
   process_mesh.cc
   dist_attr.cc
   dist_mapper.cc
-  reshard_utils.cc
   dist_tensor.cc
   dist_meta_tensor.cc
-  inferspmd_utils.cc
-  reshard_function.cc
-  r_to_s_reshard_function.cc
-  s_to_r_reshard_function.cc
-  r_to_p_reshard_function.cc
-  p_to_r_reshard_function.cc
-  s_to_s_reshard_function.cc
-  nd_mesh_reshard_function.cc
-  same_status_reshard_function.cc)
+  inferspmd_utils.cc)
+
+add_subdirectory(reshard)
diff --git a/paddle/phi/core/distributed/auto_parallel/dist_tensor.cc b/paddle/phi/core/distributed/auto_parallel/dist_tensor.cc
@@ -16,8 +16,8 @@
 
 #include "glog/logging.h"
 #include "paddle/phi/backends/context_pool.h"
-#include "paddle/phi/core/distributed/auto_parallel/reshard_function.h"
-#include "paddle/phi/core/distributed/auto_parallel/reshard_utils.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/reshard_function.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/reshard_utils.h"
 #include "paddle/phi/core/distributed/store/store_utils.h"
 
 namespace phi {
diff --git a/paddle/phi/core/distributed/auto_parallel/reshard/CMakeLists.txt b/paddle/phi/core/distributed/auto_parallel/reshard/CMakeLists.txt
@@ -0,0 +1,12 @@
+collect_srcs(
+  core_srcs
+  SRCS
+  reshard_utils.cc
+  reshard_function.cc
+  r_to_s_reshard_function.cc
+  s_to_r_reshard_function.cc
+  r_to_p_reshard_function.cc
+  p_to_r_reshard_function.cc
+  s_to_s_reshard_function.cc
+  nd_mesh_reshard_function.cc
+  same_status_reshard_function.cc)
diff --git a/paddle/phi/core/distributed/auto_parallel/reshard/nd_mesh_reshard_function.cc b/paddle/phi/core/distributed/auto_parallel/reshard/nd_mesh_reshard_function.cc
@@ -12,17 +12,17 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-#include "paddle/phi/core/distributed/auto_parallel/nd_mesh_reshard_function.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/nd_mesh_reshard_function.h"
 
 #include "glog/logging.h"
 #include "paddle/phi/common/int_array.h"
 #include "paddle/phi/core/distributed/auto_parallel/dist_attr.h"
 #include "paddle/phi/core/distributed/auto_parallel/dist_tensor.h"
-#include "paddle/phi/core/distributed/auto_parallel/p_to_r_reshard_function.h"
-#include "paddle/phi/core/distributed/auto_parallel/r_to_p_reshard_function.h"
-#include "paddle/phi/core/distributed/auto_parallel/r_to_s_reshard_function.h"
-#include "paddle/phi/core/distributed/auto_parallel/reshard_utils.h"
-#include "paddle/phi/core/distributed/auto_parallel/s_to_r_reshard_function.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/p_to_r_reshard_function.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/r_to_p_reshard_function.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/r_to_s_reshard_function.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/reshard_utils.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/s_to_r_reshard_function.h"
 
 namespace phi {
 namespace distributed {
diff --git a/paddle/phi/core/distributed/auto_parallel/reshard/nd_mesh_reshard_function.h b/paddle/phi/core/distributed/auto_parallel/reshard/nd_mesh_reshard_function.h
@@ -14,7 +14,7 @@
 
 #pragma once
 
-#include "paddle/phi/core/distributed/auto_parallel/reshard_function.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/reshard_function.h"
 
 namespace phi {
 namespace distributed {
diff --git a/paddle/phi/core/distributed/auto_parallel/reshard/p_to_r_reshard_function.cc b/paddle/phi/core/distributed/auto_parallel/reshard/p_to_r_reshard_function.cc
@@ -12,13 +12,13 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-#include "paddle/phi/core/distributed/auto_parallel/p_to_r_reshard_function.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/p_to_r_reshard_function.h"
 
 #include "glog/logging.h"
 
 #include "paddle/phi/core/distributed/auto_parallel/dist_attr.h"
 #include "paddle/phi/core/distributed/auto_parallel/dist_tensor.h"
-#include "paddle/phi/core/distributed/auto_parallel/reshard_utils.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/reshard_utils.h"
 #include "paddle/phi/kernels/all_reduce_kernel.h"
 #include "paddle/phi/kernels/elementwise_divide_kernel.h"
 #include "paddle/phi/kernels/full_kernel.h"
diff --git a/paddle/phi/core/distributed/auto_parallel/reshard/p_to_r_reshard_function.h b/paddle/phi/core/distributed/auto_parallel/reshard/p_to_r_reshard_function.h
@@ -14,7 +14,7 @@
 
 #pragma once
 
-#include "paddle/phi/core/distributed/auto_parallel/reshard_function.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/reshard_function.h"
 
 namespace phi {
 namespace distributed {
diff --git a/paddle/phi/core/distributed/auto_parallel/reshard/r_to_p_reshard_function.cc b/paddle/phi/core/distributed/auto_parallel/reshard/r_to_p_reshard_function.cc
@@ -12,13 +12,13 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-#include "paddle/phi/core/distributed/auto_parallel/r_to_p_reshard_function.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/r_to_p_reshard_function.h"
 
 #include "glog/logging.h"
 
 #include "paddle/phi/core/distributed/auto_parallel/dist_attr.h"
 #include "paddle/phi/core/distributed/auto_parallel/dist_tensor.h"
-#include "paddle/phi/core/distributed/auto_parallel/reshard_utils.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/reshard_utils.h"
 #include "paddle/phi/kernels/assign_kernel.h"
 #include "paddle/phi/kernels/full_kernel.h"
 
diff --git a/paddle/phi/core/distributed/auto_parallel/reshard/r_to_p_reshard_function.h b/paddle/phi/core/distributed/auto_parallel/reshard/r_to_p_reshard_function.h
@@ -14,7 +14,7 @@
 
 #pragma once
 
-#include "paddle/phi/core/distributed/auto_parallel/reshard_function.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/reshard_function.h"
 
 namespace phi {
 namespace distributed {
diff --git a/paddle/phi/core/distributed/auto_parallel/reshard/r_to_s_reshard_function.cc b/paddle/phi/core/distributed/auto_parallel/reshard/r_to_s_reshard_function.cc
@@ -12,13 +12,13 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-#include "paddle/phi/core/distributed/auto_parallel/r_to_s_reshard_function.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/r_to_s_reshard_function.h"
 
 #include "glog/logging.h"
 #include "paddle/phi/core/distributed/auto_parallel/dist_attr.h"
 #include "paddle/phi/core/distributed/auto_parallel/dist_tensor.h"
-#include "paddle/phi/core/distributed/auto_parallel/reshard_utils.h"
-#include "paddle/phi/core/distributed/auto_parallel/same_status_reshard_function.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/reshard_utils.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/same_status_reshard_function.h"
 #include "paddle/phi/kernels/split_kernel.h"
 
 namespace phi {
diff --git a/paddle/phi/core/distributed/auto_parallel/reshard/r_to_s_reshard_function.h b/paddle/phi/core/distributed/auto_parallel/reshard/r_to_s_reshard_function.h
@@ -14,7 +14,7 @@
 
 #pragma once
 
-#include "paddle/phi/core/distributed/auto_parallel/reshard_function.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/reshard_function.h"
 
 namespace phi {
 namespace distributed {
diff --git a/paddle/phi/core/distributed/auto_parallel/reshard/reshard_function.cc b/paddle/phi/core/distributed/auto_parallel/reshard/reshard_function.cc
@@ -12,7 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-#include "paddle/phi/core/distributed/auto_parallel/reshard_function.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/reshard_function.h"
 
 #include "paddle/phi/core/distributed/auto_parallel/dist_attr.h"
 #include "paddle/phi/core/distributed/auto_parallel/dist_tensor.h"
diff --git a/paddle/phi/core/distributed/auto_parallel/reshard/reshard_function.h b/paddle/phi/core/distributed/auto_parallel/reshard/reshard_function.h
diff --git a/paddle/phi/core/distributed/auto_parallel/reshard/reshard_utils.cc b/paddle/phi/core/distributed/auto_parallel/reshard/reshard_utils.cc
@@ -12,13 +12,13 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-#include "paddle/phi/core/distributed/auto_parallel/reshard_utils.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/reshard_utils.h"
 
 #include "glog/logging.h"
 #include "paddle/phi/backends/context_pool.h"
 #include "paddle/phi/core/device_context.h"
 #include "paddle/phi/core/distributed/auto_parallel/process_mesh.h"
-#include "paddle/phi/core/distributed/auto_parallel/reshard_function.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/reshard_function.h"
 #include "paddle/phi/core/distributed/comm_context_manager.h"
 #include "paddle/phi/core/distributed/store/store_utils.h"
 #include "paddle/phi/core/enforce.h"
diff --git a/paddle/phi/core/distributed/auto_parallel/reshard/reshard_utils.h b/paddle/phi/core/distributed/auto_parallel/reshard/reshard_utils.h
diff --git a/paddle/phi/core/distributed/auto_parallel/reshard/s_to_r_reshard_function.cc b/paddle/phi/core/distributed/auto_parallel/reshard/s_to_r_reshard_function.cc
@@ -12,13 +12,13 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-#include "paddle/phi/core/distributed/auto_parallel/s_to_r_reshard_function.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/s_to_r_reshard_function.h"
 
 #include "glog/logging.h"
 #include "paddle/phi/common/int_array.h"
 #include "paddle/phi/core/distributed/auto_parallel/dist_attr.h"
 #include "paddle/phi/core/distributed/auto_parallel/dist_tensor.h"
-#include "paddle/phi/core/distributed/auto_parallel/reshard_utils.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/reshard_utils.h"
 #include "paddle/phi/kernels/all_gather_kernel.h"
 #include "paddle/phi/kernels/concat_kernel.h"
 #include "paddle/phi/kernels/split_kernel.h"
diff --git a/paddle/phi/core/distributed/auto_parallel/reshard/s_to_r_reshard_function.h b/paddle/phi/core/distributed/auto_parallel/reshard/s_to_r_reshard_function.h
@@ -13,8 +13,8 @@
 // limitations under the License.
 
 #pragma once
-#include "paddle/phi/core/distributed/auto_parallel/reshard_function.h"
-#include "paddle/phi/core/distributed/auto_parallel/same_status_reshard_function.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/reshard_function.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/same_status_reshard_function.h"
 
 namespace phi {
 namespace distributed {
diff --git a/paddle/phi/core/distributed/auto_parallel/reshard/s_to_s_reshard_function.cc b/paddle/phi/core/distributed/auto_parallel/reshard/s_to_s_reshard_function.cc
@@ -12,13 +12,13 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-#include "paddle/phi/core/distributed/auto_parallel/s_to_s_reshard_function.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/s_to_s_reshard_function.h"
 
 #include "glog/logging.h"
 
 #include "paddle/phi/core/distributed/auto_parallel/dist_attr.h"
 #include "paddle/phi/core/distributed/auto_parallel/dist_tensor.h"
-#include "paddle/phi/core/distributed/auto_parallel/reshard_utils.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/reshard_utils.h"
 #include "paddle/phi/kernels/all_to_all_kernel.h"
 #include "paddle/phi/kernels/reshape_kernel.h"
 #include "paddle/phi/kernels/transpose_kernel.h"
diff --git a/paddle/phi/core/distributed/auto_parallel/reshard/s_to_s_reshard_function.h b/paddle/phi/core/distributed/auto_parallel/reshard/s_to_s_reshard_function.h
@@ -14,7 +14,7 @@
 
 #pragma once
 
-#include "paddle/phi/core/distributed/auto_parallel/reshard_function.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/reshard_function.h"
 
 namespace phi {
 namespace distributed {
diff --git a/paddle/phi/core/distributed/auto_parallel/reshard/same_status_reshard_function.cc b/paddle/phi/core/distributed/auto_parallel/reshard/same_status_reshard_function.cc
@@ -12,14 +12,14 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-#include "paddle/phi/core/distributed/auto_parallel/same_status_reshard_function.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/same_status_reshard_function.h"
 
 #include <algorithm>
 
 #include "glog/logging.h"
 #include "paddle/phi/core/distributed/auto_parallel/dist_attr.h"
 #include "paddle/phi/core/distributed/auto_parallel/dist_tensor.h"
-#include "paddle/phi/core/distributed/auto_parallel/reshard_utils.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/reshard_utils.h"
 #include "paddle/phi/core/distributed/store/store_utils.h"
 #include "paddle/phi/kernels/p_recv_kernel.h"
 #include "paddle/phi/kernels/p_send_kernel.h"
diff --git a/paddle/phi/core/distributed/auto_parallel/reshard/same_status_reshard_function.h b/paddle/phi/core/distributed/auto_parallel/reshard/same_status_reshard_function.h
@@ -14,7 +14,7 @@
 
 #pragma once
 
-#include "paddle/phi/core/distributed/auto_parallel/reshard_function.h"
+#include "paddle/phi/core/distributed/auto_parallel/reshard/reshard_function.h"
 
 namespace phi {
 namespace distributed {
