[OpenMP] Add Missing Runtime Call for Globalization Remarks

[jhuber6]

Summary

Add a missing runtime call to perform data globalization checks.

[jdoerfert]

LGTM

llvm/lib/Transforms/IPO/OpenMPOpt.cpp

auto SplitMemTransfers = [&](Use &U, Function &Decl) {
       return WasSplit;
     };
     RFI.foreachUse(SCC, SplitMemTransfers);
 
     return Changed;
   }
 
   void analysisGlobalization() {
-    auto &RFI =
-        OMPInfoCache.RFIs[OMPRTL___kmpc_data_sharing_coalesced_push_stack];
+    RuntimeFunction GlobalizationRuntimeIDs[] = {
+        OMPRTL___kmpc_data_sharing_coalesced_push_stack,
+        OMPRTL___kmpc_data_sharing_push_stack};
 
-    auto checkGlobalization = [&](Use &U, Function &Decl) {
+    for (const auto GlobalizationCallID : GlobalizationRuntimeIDs) {
+      auto &RFI = OMPInfoCache.RFIs[GlobalizationCallID];
+
+      auto CheckGlobalization = [&](Use &U, Function &Decl) {
         if (CallInst *CI = getCallIfRegularCall(U, &RFI)) {
           auto Remark = [&](OptimizationRemarkAnalysis ORA) {
             return ORA
                    << "Found thread data sharing on the GPU. "
                    << "Expect degraded performance due to data globalization.";
           };
           emitRemark<OptimizationRemarkAnalysis>(CI, "OpenMPGlobalization",
                                                  Remark);
         }
 
         return false;
       };
 
-    RFI.foreachUse(SCC, checkGlobalization);
+      RFI.foreachUse(SCC, CheckGlobalization);
+    }
     return;
   }
 
   /// Maps the values stored in the offload arrays passed as arguments to
   /// \p RuntimeCall into the offload arrays in \p OAs.
   bool getValuesInOffloadArrays(CallInst &RuntimeCall,
                                 MutableArrayRef<OffloadArray> OAs) {
     assert(OAs.size() == 3 && "Need space for three offload arrays!");

llvm/test/Transforms/OpenMP/globalization_remarks.ll

 define hidden i32 @_Z3barv() local_unnamed_addr #3 !dbg !15 {
 entry:
   %a1 = alloca i32, align 4
   %0 = tail call i8 @__kmpc_is_spmd_exec_mode() #4
   %.not = icmp eq i8 %0, 0
   br i1 %.not, label %.non-spmd, label %.exit
 
 .non-spmd:                                        ; preds = %entry
-  %1 = tail call i8* @__kmpc_data_sharing_coalesced_push_stack(i64 128, i16 0) #4, !dbg !31
+  %1 = tail call i8* @__kmpc_data_sharing_push_stack(i64 128, i16 0) #4, !dbg !31
   %2 = bitcast i8* %1 to %struct._globalized_locals_ty*
   br label %.exit
 
 .exit:                                            ; preds = %entry, %.non-spmd
   %_select_stack = phi %struct._globalized_locals_ty* [ %2, %.non-spmd ], [ null, %entry ]
   %nvptx_tid = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !range !28
   %nvptx_lane_id = and i32 %nvptx_tid, 31
   %3 = zext i32 %nvptx_lane_id to i64

 .exit3:                                           ; preds = %.non-spmd2, %.exit
   ret i32 %6, !dbg !31
 }
 
 declare i8 @__kmpc_is_spmd_exec_mode() local_unnamed_addr
 
 declare i8* @__kmpc_data_sharing_coalesced_push_stack(i64, i16) local_unnamed_addr
 
+declare i8* @__kmpc_data_sharing_push_stack(i64, i16) local_unnamed_addr
+
 ; Function Attrs: nounwind readnone
 declare i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1
 
 declare void @__kmpc_data_sharing_pop_stack(i8*) local_unnamed_addr
 
 ; Function Attrs: nounwind
 declare i32 @__kmpc_global_thread_num(%struct.ident_t*) local_unnamed_addr #4