[Fix] USE_NCCL -> ORT_USE_NCCL (microsoft#21136)

### Description Correct the macro used when NCCL enabled.
titaiwangms · Jun 24, 2024 · adaf0e8 · adaf0e8
1 parent 3a917e4
commit adaf0e8
Show file tree

Hide file tree

Showing 5 changed files with 6 additions and 6 deletions.
diff --git a/onnxruntime/contrib_ops/cuda/collective/custom_reduce_impl.cu b/onnxruntime/contrib_ops/cuda/collective/custom_reduce_impl.cu
@@ -29,7 +29,7 @@ namespace onnxruntime {
 namespace cuda {
 namespace collective {
 
-#if defined(USE_MPI) || defined(USE_NCCL)
+#if defined(USE_MPI) || defined(ORT_USE_NCCL)
 
 using namespace onnxruntime;
 using namespace onnxruntime::cuda;

diff --git a/onnxruntime/contrib_ops/cuda/collective/custom_reduce_impl.h b/onnxruntime/contrib_ops/cuda/collective/custom_reduce_impl.h
@@ -25,7 +25,7 @@ namespace onnxruntime {
 namespace cuda {
 namespace collective {
 
-#if defined(USE_MPI) || defined(USE_NCCL)
+#if defined(USE_MPI) || defined(ORT_USE_NCCL)
 
 constexpr size_t WARP_SIZE = 32;
 constexpr size_t MAX_ALL_REDUCE_BLOCKS = 24;

diff --git a/onnxruntime/contrib_ops/cuda/collective/ipc_utils.cc b/onnxruntime/contrib_ops/cuda/collective/ipc_utils.cc
@@ -23,7 +23,7 @@ namespace onnxruntime {
 namespace cuda {
 namespace collective {
 
-#if defined(USE_MPI) || defined(USE_NCCL)
+#if defined(USE_MPI) || defined(ORT_USE_NCCL)
 
 using namespace onnxruntime;
 

diff --git a/onnxruntime/contrib_ops/cuda/collective/ipc_utils.h b/onnxruntime/contrib_ops/cuda/collective/ipc_utils.h
@@ -24,7 +24,7 @@ namespace onnxruntime {
 namespace cuda {
 namespace collective {
 
-#if defined(USE_MPI) || defined(USE_NCCL)
+#if defined(USE_MPI) || defined(ORT_USE_NCCL)
 
 struct CudaDeleter {
   void operator()(void* ptr) const noexcept {
@@ -86,4 +86,4 @@ GetCustomAllReduceWorkspace(int rank, int world_size, size_t input_size, IPCMemo
 
 }  // namespace collective
 }  // namespace cuda
-}  // namespace onnxruntime
+}  // namespace onnxruntime
diff --git a/onnxruntime/contrib_ops/cuda/collective/nccl_kernels.h b/onnxruntime/contrib_ops/cuda/collective/nccl_kernels.h
@@ -5,7 +5,7 @@
 
 #include "core/providers/cuda/cuda_kernel.h"
 
-#if defined(ORT_USE_NCCL) || defined(ORT_USE_MPI)
+#if defined(ORT_USE_NCCL) || defined(USE_MPI)
 #ifndef USE_ROCM
 #include "custom_reduce_impl.h"
 #include "ipc_utils.h"