[OPENMP][NVPTX]Use faster teams reduction algorithm.

A faster way to reduce the values in teams reductions was found, the
codegen is updated to use this faster algorithm and new runtime functions.

llvm-svn: 354479
diff --git a/clang/lib/Frontend/CompilerInvocation.cpp b/clang/lib/Frontend/CompilerInvocation.cpp
index d4094ac..1a33a00 100644
--- a/clang/lib/Frontend/CompilerInvocation.cpp
+++ b/clang/lib/Frontend/CompilerInvocation.cpp
@@ -2840,6 +2840,9 @@
     Opts.OpenMPCUDABlocksPerSM =
         getLastArgIntValue(Args, options::OPT_fopenmp_cuda_blocks_per_sm_EQ,
                            Opts.OpenMPCUDABlocksPerSM, Diags);
+    Opts.OpenMPCUDAReductionBufNum = getLastArgIntValue(
+        Args, options::OPT_fopenmp_cuda_teams_reduction_recs_num_EQ,
+        Opts.OpenMPCUDAReductionBufNum, Diags);
   }
 
   // Prevent auto-widening the representation of loop counters during an