Add flag comment

Author: trevor-m

xla/xla.proto

@@ -462,8 +462,9 @@ message DebugOptions {
 
   bool xla_gpu_enable_pipelined_reduce_scatter = 231;
 
-  // Enable NVSHMEM.
-  bool xla_gpu_experimental_enable_nvshmem = 379;
+  // Enable NVSHMEM. Must be set via XLA_FLAGS variable before XLA client is
+  // initialized and can't be set just through HLO Config->ExecutionOptions.
+  bool xla_gpu_experimental_enable_nvshmem = 382;
 
   // Enable allreduce reassociation on allreduces that are converted to a wider
   // type. The resulting allreduce will be promoted to a wider-typed allreduce.
@@ -1199,7 +1200,7 @@ message DebugOptions {
   // Note: when adding a new flag, please add it to one of the hardware-specific
   // or hardware-agnostic sections at the top of this proto message.
 
-  // Next id: 380
+  // Next id: 383
 
   // Extra options to pass to the compilation backend (e.g. LLVM); specific
   // interpretation of these values is left to the backend.