[DCU] fix

transformer_engine/common/comm_gemm_overlap/comm_gemm_overlap.cpp

@@ -117,13 +117,17 @@ CommOverlapCore::CommOverlapCore(int myrank, int numranks, int mylocal, int numl
     NVTE_CHECK(false, "comm_cu_nums must be 4,8,16,32");
   }
 
+  const char *NVTE_UB_COMM_CU_NUMS = std::getenv("NVTE_UB_COMM_CU_NUMS");
   static cudaStream_t compute_streams[NVTE_COMM_OVERLAP_MAX_STREAMS];
   for (int i = 0; i < std::min(num_max_streams, num_splits); i++) {
     if (compute_streams[i] == nullptr) {
-      NVTE_CHECK_CUDA(cudaStreamCreateWithPriority(&compute_streams[i], cudaStreamNonBlocking, _gemm_priority));
+      if (NVTE_UB_COMM_CU_NUMS != nullptr && NVTE_UB_COMM_CU_NUMS[0] != '\0') {
 #ifdef __HIP_PLATFORM_AMD__
-      NVTE_CHECK_CUDA(hipExtStreamCreateWithCUMask(&compute_streams[i], cuMaskSize, cuMask));
+        NVTE_CHECK_CUDA(hipExtStreamCreateWithCUMask(&compute_streams[i], cuMaskSize, cuMask));
 #endif
+      } else {
+        NVTE_CHECK_CUDA(cudaStreamCreateWithPriority(&compute_streams[i], cudaStreamNonBlocking, _gemm_priority));
+      }
     }
     _stream_compute.push_back(compute_streams[i]);
   }
@@ -359,14 +363,18 @@ CommOverlapBase::CommOverlapBase(const std::vector<size_t> &buffer_shape, DType
     NVTE_CHECK(false, "comm_cu_nums must be 4,8,16,32");
   }
   
+  const char *NVTE_UB_COMM_CU_NUMS = std::getenv("NVTE_UB_COMM_CU_NUMS");
   static cudaStream_t comm_stream;
   if (comm_stream == nullptr) {
-    NVTE_CHECK_CUDA(
-      cudaStreamCreateWithPriority(&comm_stream, cudaStreamNonBlocking, _comm_priority));
+    if (NVTE_UB_COMM_CU_NUMS != nullptr && NVTE_UB_COMM_CU_NUMS[0] != '\0') {
 #ifdef __HIP_PLATFORM_AMD__
-    NVTE_CHECK_CUDA(
-      hipExtStreamCreateWithCUMask(&comm_stream, cuMaskSize, cuMask));
+      NVTE_CHECK_CUDA(
+        hipExtStreamCreateWithCUMask(&comm_stream, cuMaskSize, cuMask));
 #endif
+    } else {
+      NVTE_CHECK_CUDA(
+        cudaStreamCreateWithPriority(&comm_stream, cudaStreamNonBlocking, _comm_priority));
+    }
   }
   _stream_comm = comm_stream;
   NVTE_CHECK_CUDA(cudaEventCreateWithFlags(&_start_d2dcopy, 0));