[Openmp-commits] [clang] [llvm] [mlir] [openmp] [OpenMP][offload] Cross-team reductions with variable number of teams (PR #195102)

[Robert Imschweiler via Openmp-commits]

================
@@ -6862,8 +6862,17 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
         CmdArgs.push_back("-fno-openmp-extensions");
       Args.AddAllArgs(CmdArgs, options::OPT_fopenmp_cuda_number_of_sm_EQ);
       Args.AddAllArgs(CmdArgs, options::OPT_fopenmp_cuda_blocks_per_sm_EQ);
-      Args.AddAllArgs(CmdArgs,
-                      options::OPT_fopenmp_cuda_teams_reduction_recs_num_EQ);
+      // '-fopenmp-cuda-teams-reduction-recs-num=' is deprecated and has no
+      // effect: the teams reduction buffer is sized at kernel launch by the
+      // offload plugin to match the actual number of teams. Honoring a
+      // smaller user-supplied value would silently truncate the buffer for
+      // larger launches.
+      if (Arg *A = Args.getLastArg(
+              options::OPT_fopenmp_cuda_teams_reduction_recs_num_EQ))
+        D.Diag(diag::warn_drv_deprecated_custom)
+            << A->getAsString(Args)
+            << "the value is ignored; the teams reduction buffer is sized "
+               "automatically at kernel launch";
----------------
ro-i wrote:

I don't think so?

https://github.com/llvm/llvm-project/pull/195102