[OpenMP] Add RTL function to externalization RAII

[jhuber6]

Summary

This patch adds the '_kmpc_get_hardware_num_threads_in_block'
OpenMP RTL function to the externalization RAII struct. This was getting
optimized out and then being replaced with an undefined value once added
back in, causing bugs for complex reductions.

Fixes #53909.

[jdoerfert]

LG. FTR, same principle as D114802.

llvm/lib/Transforms/IPO/OpenMPOpt.cpp

bool runAttributor(bool IsModulePass) {
     ExternalizationRAII EndParallel(OMPInfoCache,
                                     OMPRTL___kmpc_kernel_end_parallel);
     ExternalizationRAII BarrierSPMD(OMPInfoCache,
                                     OMPRTL___kmpc_barrier_simple_spmd);
     ExternalizationRAII BarrierGeneric(OMPInfoCache,
                                        OMPRTL___kmpc_barrier_simple_generic);
     ExternalizationRAII ThreadId(OMPInfoCache,
                                  OMPRTL___kmpc_get_hardware_thread_id_in_block);
+    ExternalizationRAII NumThreads(
+        OMPInfoCache, OMPRTL___kmpc_get_hardware_num_threads_in_block);
     ExternalizationRAII WarpSize(OMPInfoCache, OMPRTL___kmpc_get_warp_size);
 
     registerAAs(IsModulePass);
 
     ChangeStatus Changed = A.run();
 
     LLVM_DEBUG(dbgs() << "[Attributor] Done with " << SCC.size()
                       << " functions, result: " << Changed << ".\n");

llvm/test/Transforms/OpenMP/get_hardware_num_threads_in_block_fold.ll

 ; CHECK-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) {
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    ret void
 ;
 entry:
   ret void
 }
 
-declare i32 @__kmpc_get_hardware_num_threads_in_block()
+define internal i32 @__kmpc_get_hardware_num_threads_in_block() {
+; CHECK-LABEL: define {{[^@]+}}@__kmpc_get_hardware_num_threads_in_block
+; CHECK-SAME: () #[[ATTR1]] {
+; CHECK-NEXT:    [[RET:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block_dummy()
+; CHECK-NEXT:    ret i32 [[RET]]
+;
+  %ret = call i32 @__kmpc_get_hardware_num_threads_in_block_dummy()
+  ret i32 %ret
+}
+declare i32 @__kmpc_get_hardware_num_threads_in_block_dummy()
 declare i32 @__kmpc_target_init(%struct.ident_t*, i8, i1 zeroext, i1 zeroext) #1
 declare void @__kmpc_target_deinit(%struct.ident_t* nocapture readnone, i8, i1 zeroext) #1
 declare void @__kmpc_parallel_51(%struct.ident_t*, i32, i32, i32, i32, i8*, i8*, i8**, i64)
 declare i32 @__kmpc_global_thread_num(%struct.ident_t*)
 
 
 !llvm.module.flags = !{!0, !1}
 !nvvm.annotations = !{!2, !3, !4}