[clang] [llvm] [OpenMP][amdgpu] Use max teams for amdgpu-max-num-workgroups (PR #195203)

[Kevin Sala Penades via cfe-commits]

================
@@ -8349,11 +8074,12 @@ OpenMPIRBuilder::readTeamBoundsForKernel(const Triple &, Function &Kernel) {
 
 void OpenMPIRBuilder::writeTeamsForKernel(const Triple &T, Function &Kernel,
                                           int32_t LB, int32_t UB) {
-  if (T.isNVPTX())
-    if (UB > 0)
+  if (UB > 0) {
+    if (T.isNVPTX())
       Kernel.addFnAttr(NVVMAttr::MaxClusterRank, llvm::utostr(UB));
-  if (T.isAMDGPU())
-    Kernel.addFnAttr("amdgpu-max-num-workgroups", llvm::utostr(LB) + ",1,1");
+    if (T.isAMDGPU())
+      Kernel.addFnAttr("amdgpu-max-num-workgroups", llvm::utostr(LB) + ",1,1");
+  }
 
   Kernel.addFnAttr("omp_target_num_teams", std::to_string(LB));
----------------
kevinsala wrote:

The `num_teams` clause accepts a _lower-bound_ since OpenMP 5.1 (doesn't seem supported yet). Also, the `bounds_launch` attribute for `ompx_attribute` can change the minimum number of teams.

https://github.com/llvm/llvm-project/pull/195203