fix the bug for processgroup_hccl compiling (#40437)

paddle/fluid/distributed/collective/ProcessGroupHCCL.cc

@@ -139,11 +139,9 @@ bool ProcessGroupHCCL::HCCLTask::IsCompleted() {
 // TODO(sandyhouse): Add timeout for wait, now timeout unused
 bool ProcessGroupHCCL::HCCLTask::Wait(std::chrono::milliseconds timeout) {
   SynchronizeStreams();
-  if (FLAGS_hccl_blocking_wait) {
-    // NOTE(sandyhouse): It will block host for sync
-    while (!IsCompleted()) {
-      std::this_thread::sleep_for(std::chrono::milliseconds(kWaitBlockTImeout));
-    }
+  // NOTE(sandyhouse): It will block host for sync
+  while (!IsCompleted()) {
+    std::this_thread::sleep_for(std::chrono::milliseconds(kWaitBlockTImeout));
   }
   return true;
 }

paddle/fluid/distributed/collective/ProcessGroupHCCL.h

@@ -84,29 +84,6 @@ class ProcessGroupHCCL : public ProcessGroup {
       std::vector<Tensor>& tensors,
       const BroadcastOptions& = BroadcastOptions()) override;
 
-  std::shared_ptr<ProcessGroup::Task> Barrier(
-      const BarrierOptions& = BarrierOptions()) override;
-
-  std::shared_ptr<ProcessGroup::Task> Send(std::vector<Tensor>& tensors,
-                                           int dst_rank) override;
-
-  std::shared_ptr<ProcessGroup::Task> Recv(std::vector<Tensor>& tensors,
-                                           int src_rank) override;
-
-  std::shared_ptr<ProcessGroup::Task> AllGather(
-      std::vector<Tensor>& in_tensors,
-      std::vector<Tensor>& out_tensors) override;
-
-  std::shared_ptr<ProcessGroup::Task> AllToAll(
-      std::vector<Tensor>& in, std::vector<Tensor>& out) override;
-
-  std::shared_ptr<ProcessGroup::Task> Reduce(
-      std::vector<Tensor>& tensors, const ReduceOptions& opts) override;
-
-  std::shared_ptr<ProcessGroup::Task> Scatter(std::vector<Tensor>& in_tensors,
-                                              std::vector<Tensor>& out_tensors,
-                                              const ScatterOptions&) override;
-
  protected:
   virtual std::shared_ptr<ProcessGroupHCCL::HCCLTask> CreateTask(
       std::vector<Place> places, int rank, CommType opType,

paddle/fluid/pybind/CMakeLists.txt

@@ -92,7 +92,7 @@ if(NOT ON_INFER)
   if (WITH_GLOO)
     set (PYBIND_DEPS ${PYBIND_DEPS} processgroup_gloo)
   endif()
-  if(WITH_ASCEND)
+  if(WITH_ASCEND_CL)
     set (PYBIND_DEPS ${PYBIND_DEPS} processgroup_hccl)
   endif()
   set(PYBIND_SRCS ${PYBIND_SRCS} distributed_py.cc)