[OpenMPOpt] Move dedup runtime calls after init for target regions

[ggeorgakoudis]

Summary

Deduplication in OpenMPOpt finds redundant OpenMP runtime calls and replaces them with a single call placed in the earliest safe location in the IR. When deduplication happens in a target region this patch makes sure replacement calls are put after target_init.

[jdoerfert]

LG, one comment.

llvm/lib/Transforms/IPO/OpenMPOpt.cpp

1586

@jhuber6 suggested to not emit a remark here and removed it earlier.

llvm/test/Transforms/OpenMP/deduplication_target_remarks.ll

1 ↗

(On Diff #360823)

Probably don't need this anymore then.

[jhuber6]

llvm/lib/Transforms/IPO/OpenMPOpt.cpp

1586

Yeah, I decided that it wasn't really important for the user to know that the call was moved anywhere and just cluttered up the existing remarks. Mostly because the user has no clue where the beginning of the OpenMP region truly exists in their code and it's not really an optimization per-se. If you want to add it in you'd also need to give it an OMPxxx number and write some documentation for it.

[jdoerfert]

llvm/lib/Transforms/IPO/OpenMPOpt.cpp

1593

It could happen that this is empty, in the CGSCC pass for example.
In that case just continue and do not move anything.

[ggeorgakoudis]

Fix for comments

[ggeorgakoudis]

llvm/lib/Transforms/IPO/OpenMPOpt.cpp

1586

Makes sense, removed.

llvm/lib/Transforms/IPO/OpenMPOpt.cpp

bool deduplicateRuntimeCalls(Function &F,
     };
 
     if (!ReplVal) {
       for (Use *U : *UV)
         if (CallInst *CI = getCallIfRegularCall(*U, &RFI)) {
           if (!CanBeMoved(*CI))
             continue;
 
+          // If the function is a kernel, dedup will move
+          // the runtime call right after the kernel init callsite. Otherwise,
+          // it will move it to the beginning of the caller function.
+          if (isKernel(F)) {
+            auto &KernelInitRFI = OMPInfoCache.RFIs[OMPRTL___kmpc_target_init];
+            auto *KernelInitUV = KernelInitRFI.getUseVector(F);
+
+            if (KernelInitUV->empty())
+              continue;
+
+            assert(KernelInitUV->size() == 1 &&
+                   "Expected a single __kmpc_target_init in kernel\n");
+
+            CallInst *KernelInitCI =
+                getCallIfRegularCall(*KernelInitUV->front(), &KernelInitRFI);
+            assert(KernelInitCI &&
+                   "Expected a call to __kmpc_target_init in kernel\n");
+
+            CI->moveAfter(KernelInitCI);
+          } else
             CI->moveBefore(&*F.getEntryBlock().getFirstInsertionPt());
           ReplVal = CI;
           break;
         }
       if (!ReplVal)
         return false;
     }
 
     // If we use a call as a replacement value we need to make sure the ident is

llvm/test/Transforms/OpenMP/deduplication_target.ll

+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --function-signature --scrub-attributes
+; RUN: opt -openmp-opt-cgscc -S < %s | FileCheck %s
+; RUN: opt -passes=openmp-opt-cgscc -S < %s | FileCheck %s
+
+target datalayout = "e-i64:64-i128:128-v16:16-v32:32-n16:32:64"
+target triple = "nvptx64"
+
+%struct.ident_t = type { i32, i32, i32, i32, i8* }
+
+@0 = private unnamed_addr constant [23 x i8] c";unknown;unknown;0;0;;\00", align 1
+@1 = private unnamed_addr constant %struct.ident_t { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* @0, i32 0, i32 0) }, align 8
+@2 = private unnamed_addr constant %struct.ident_t { i32 0, i32 2, i32 1, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* @0, i32 0, i32 0) }, align 8
+@__omp_offloading_50_a3e09bf8_foo_l2_exec_mode = weak constant i8 0
[e-mail] = appending global [1 x i8*] [i8* @__omp_offloading_50_a3e09bf8_foo_l2_exec_mode], section "llvm.metadata"
+
+declare void @use(i32)
+
+define weak void @__omp_offloading_50_a3e09bf8_foo_l2() #0 {
+; CHECK-LABEL: define {{[^@]+}}@__omp_offloading_50_a3e09bf8_foo_l2
+; CHECK-SAME: () #[[ATTR0:[0-9]+]] {
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1:[0-9]+]], i1 true, i1 false, i1 true)
+; CHECK-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB2:[0-9]+]])
+; CHECK-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; CHECK-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; CHECK:       user_code.entry:
+; CHECK-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 true, i1 true)
+; CHECK-NEXT:    ret void
+; CHECK:       worker.exit:
+; CHECK-NEXT:    ret void
+;
+entry:
+  %captured_vars_addrs = alloca [0 x i8*], align 8
+  %0 = call i32 @__kmpc_target_init(%struct.ident_t* @1, i1 true, i1 false, i1 true)
+  %exec_user_code = icmp eq i32 %0, -1
+  br i1 %exec_user_code, label %user_code.entry, label %worker.exit
+
+user_code.entry:                                  ; preds = %entry
+  %1 = call i32 @__kmpc_global_thread_num(%struct.ident_t* @2)
+  %2 = call i32 @__kmpc_global_thread_num(%struct.ident_t* @2)
+  call void @__kmpc_target_deinit(%struct.ident_t* @1, i1 true, i1 true)
+  ret void
+
+worker.exit:                                      ; preds = %entry
+  ret void
+}
+
+declare i32 @__kmpc_target_init(%struct.ident_t*, i1, i1, i1)
+
+declare i32 @__kmpc_global_thread_num(%struct.ident_t*) #1
+
+declare void @__kmpc_target_deinit(%struct.ident_t*, i1, i1)
+
+attributes #0 = { convergent noinline norecurse nounwind "frame-pointer"="none" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+ptx32,+sm_20" }
+attributes #1 = { nounwind }
+
+!omp_offload.info = !{!0}
+!nvvm.annotations = !{!1}
+!llvm.module.flags = !{!2, !3, !4}
+
+!0 = !{i32 0, i32 80, i32 -1545561096, !"foo", i32 2, i32 0}
+!1 = !{void ()* @__omp_offloading_50_a3e09bf8_foo_l2, !"kernel", i32 1}
+!2 = !{i32 1, !"wchar_size", i32 4}
+!3 = !{i32 7, !"openmp", i32 50}
+!4 = !{i32 7, !"openmp-device", i32 50}