Fix compilation problem (#31100)

Fix compilation problem (#31100)

cmake/external/ascend.cmake

@@ -57,28 +57,24 @@ if(WITH_ASCEND)
   SET_PROPERTY(TARGET atlas_acl PROPERTY IMPORTED_LOCATION ${atlas_acl_lib})
 
   add_custom_target(extern_ascend DEPENDS ascend_ge ascend_graph atlas_acl)
+endif()
 
-elseif(WITH_ASCEND_CL)
-  set(ASCEND_ATC_DIR ${ASCEND_DIR}/atc/lib64)
-  set(ASCEND_ACL_DIR ${ASCEND_DIR}/acllib/lib64)
-  set(STATIC_ACL_LIB ${ASCEND_ACL_DIR})
+if(WITH_ASCEND_CL)
+  set(ASCEND_CL_DIR ${ASCEND_DIR}/ascend-toolkit/latest/fwkacllib/lib64)
 
-  set(ATLAS_ACL_DIR ${ASCEND_DIR}/ascend-toolkit/latest/fwkacllib/lib64)
-  set(ATLAS_ATC_DIR ${ASCEND_DIR}/ascend-toolkit/latest/fwkacllib/lib64)
-  
-  set(atlas_acl_lib ${ATLAS_ACL_DIR}/libascendcl.so)
-  set(atlas_acl_op_compiler_lib ${ATLAS_ACL_DIR}/libacl_op_compiler.so)
-  set(ATLAS_ACL_INC_DIR ${ASCEND_DIR}/ascend-toolkit/latest/fwkacllib/include)
+  set(ascendcl_lib ${ASCEND_CL_DIR}/libascendcl.so)
+  set(acl_op_compiler_lib ${ASCEND_CL_DIR}/libacl_op_compiler.so)
+  set(ASCEND_CL_INC_DIR ${ASCEND_DIR}/ascend-toolkit/latest/fwkacllib/include)
 
-  message(STATUS "ATLAS_ACL_INC_DIR ${ATLAS_ACL_INC_DIR}")
-  message(STATUS "ATLAS_ACL_LIB_DIR ${ATLAS_ACL_DIR}")
-  INCLUDE_DIRECTORIES(${ATLAS_ACL_INC_DIR})
+  message(STATUS "ASCEND_CL_INC_DIR ${ASCEND_CL_INC_DIR}")
+  message(STATUS "ASCEND_CL_DIR ${ASCEND_CL_DIR}")
+  INCLUDE_DIRECTORIES(${ASCEND_CL_INC_DIR})
 
-  ADD_LIBRARY(atlas_acl SHARED IMPORTED GLOBAL)
-  SET_PROPERTY(TARGET atlas_acl PROPERTY IMPORTED_LOCATION ${atlas_acl_lib})
+  ADD_LIBRARY(ascendcl SHARED IMPORTED GLOBAL)
+  SET_PROPERTY(TARGET ascendcl PROPERTY IMPORTED_LOCATION ${ascendcl_lib})
 
-  ADD_LIBRARY(atlas_acl_op_compiler SHARED IMPORTED GLOBAL)
-  SET_PROPERTY(TARGET atlas_acl_op_compiler PROPERTY IMPORTED_LOCATION ${atlas_acl_op_compiler_lib})
-  add_custom_target(extern_ascend DEPENDS atlas_acl atlas_acl_op_compiler)
+  ADD_LIBRARY(acl_op_compiler SHARED IMPORTED GLOBAL)
+  SET_PROPERTY(TARGET acl_op_compiler PROPERTY IMPORTED_LOCATION ${acl_op_compiler_lib})
+  add_custom_target(extern_ascend_cl DEPENDS ascendcl acl_op_compiler)
 
 endif()

cmake/external/protobuf.cmake

@@ -201,6 +201,9 @@ FUNCTION(build_protobuf TARGET_NAME BUILD_FOR_HOST)
 if(WITH_ASCEND AND NOT WITH_ASCEND_CXX11)
     SET(PROTOBUF_REPOSITORY  https://gitee.com/tianjianhe/protobuf.git)
     SET(PROTOBUF_TAG         v3.8.0)
+elseif(WITH_ASCEND_CL AND NOT WITH_ASCEND_CXX11)
+    SET(PROTOBUF_REPOSITORY  https://gitee.com/tianjianhe/protobuf.git)
+    SET(PROTOBUF_TAG         v3.8.0)
 else()
     SET(PROTOBUF_REPOSITORY  ${GIT_URL}/protocolbuffers/protobuf.git)
     SET(PROTOBUF_TAG         9f75c5aa851cd877fb0d93ccc31b8567a6706546)

cmake/third_party.cmake

@@ -276,7 +276,12 @@ endif(WITH_BOX_PS)
 
 if(WITH_ASCEND OR WITH_ASCEND_CL)
     include(external/ascend)
-    list(APPEND third_party_deps extern_ascend)
+    if(WITH_ASCEND)
+        list(APPEND third_party_deps extern_ascend)
+    endif()
+    if(WITH_ASCEND_CL)
+        list(APPEND third_party_deps extern_ascend_cl)
+    endif()
 endif ()
 
 if (WITH_PSCORE)

paddle/fluid/framework/garbage_collector.cc

@@ -89,7 +89,8 @@ StreamGarbageCollector::StreamGarbageCollector(const platform::CUDAPlace &place,
     : GarbageCollector(place, max_memory_size) {
   platform::CUDADeviceGuard guard(place.device);
   PADDLE_ENFORCE_CUDA_SUCCESS(cudaStreamCreate(&stream_));
-  callback_manager_.reset(new platform::StreamCallbackManager(stream_));
+  callback_manager_.reset(
+      new platform::StreamCallbackManager<cudaStream_t>(stream_));
 }
 
 StreamGarbageCollector::~StreamGarbageCollector() {

paddle/fluid/framework/garbage_collector.h

@@ -117,7 +117,8 @@ class StreamGarbageCollector : public GarbageCollector {
 
  private:
   cudaStream_t stream_;
-  std::unique_ptr<platform::StreamCallbackManager> callback_manager_;
+  std::unique_ptr<platform::StreamCallbackManager<cudaStream_t>>
+      callback_manager_;
 };
 
 class CUDAPinnedGarbageCollector : public GarbageCollector {

paddle/fluid/operators/elementwise/CMakeLists.txt

@@ -8,4 +8,7 @@ register_operators(DEPS op_version_registry)
 cc_test(test_elementwise_add_op_inplace SRCS test_elementwise_add_op_inplace.cc DEPS op_registry elementwise_add_op scope device_context enforce executor)
 cc_test(test_elementwise_div_grad_grad SRCS test_elementwise_div_grad_grad.cc DEPS op_registry elementwise_div_op scope device_context enforce executor)
 cc_test(test_elementwise_add_grad_grad SRCS test_elementwise_add_grad_grad.cc DEPS op_registry elementwise_add_op scope device_context enforce executor)
+
+if(WITH_ASCEND_CL)
 cc_test(elementwise_op_npu_test SRCS elementwise_op_npu_test.cc DEPS op_registry elementwise_add_op elementwise_sub_op scope device_context enforce executor)
+endif()

paddle/fluid/platform/CMakeLists.txt

@@ -72,7 +72,7 @@ if(WITH_ASCEND)
 endif()
 
 if(WITH_ASCEND_CL)
-    cc_library(npu_info SRCS npu_info.cc DEPS gflags glog enforce monitor atlas_acl atlas_acl_op_compiler)
+    cc_library(npu_info SRCS npu_info.cc DEPS gflags glog enforce monitor ascendcl acl_op_compiler)
 endif()
 
 add_subdirectory(dynload)
@@ -105,7 +105,7 @@ nv_library(stream_callback_manager SRCS stream_callback_manager.cc DEPS simple_t
 ENDIF()
 
 IF(WITH_ASCEND_CL)
-cc_library(stream_callback_manager SRCS stream_callback_manager.cc DEPS simple_threadpool enforce atlas_acl) 
+cc_library(stream_callback_manager SRCS stream_callback_manager.cc DEPS simple_threadpool enforce) 
 ENDIF()
 
 IF(WITH_GPU)

paddle/fluid/platform/stream/cuda_stream.cc

@@ -35,7 +35,7 @@ bool CUDAStream::Init(const Place& place, const Priority& priority) {
     PADDLE_ENFORCE_CUDA_SUCCESS(
         cudaStreamCreateWithPriority(&stream_, kDefaultFlag, 0));
   }
-  callback_manager_.reset(new StreamCallbackManager(stream_));
+  callback_manager_.reset(new StreamCallbackManager<cudaStream_t>(stream_));
   VLOG(3) << "CUDAStream Init stream: " << stream_
           << ", priority: " << static_cast<int>(priority);
   return true;

paddle/fluid/platform/stream/cuda_stream.h

@@ -74,7 +74,7 @@ class CUDAStream final {
   Place place_;
   cudaStream_t stream_{nullptr};
   Priority priority_{Priority::kNormal};
-  std::unique_ptr<StreamCallbackManager> callback_manager_;
+  std::unique_ptr<StreamCallbackManager<cudaStream_t>> callback_manager_;
 
   DISABLE_COPY_AND_ASSIGN(CUDAStream);
 };