[OpenMP] Add additional remarks for OpenMPOpt

[jhuber6]

Summary

This patch adds additional remarks, suggesting the use of noescape for failed
globalization and indicating when internalization failed.

[jhuber6]

Forgot to format.

[jdoerfert]

LG, one nit

llvm/lib/Transforms/IPO/AttributorAttributes.cpp

5210

leftover

[jhuber6]

Removing leftover debug print.

llvm/lib/Transforms/IPO/AttributorAttributes.cpp

auto Pred = [&](const Use &U, bool &Follow) -> bool {
         // If a callsite argument use is nofree, we are fine.
         const auto &ArgNoFreeAA = A.getAAFor<AANoFree>(
             *this, IRPosition::callsite_argument(*CB, ArgNo),
             DepClassTy::OPTIONAL);
 
         if (!NoCaptureAA.isAssumedNoCapture() ||
             !ArgNoFreeAA.isAssumedNoFree()) {
 
+          llvm::errs() << *CB << "\n";
           // Emit a missed remark if this is missed OpenMP globalization.
           auto Remark = [&](OptimizationRemarkMissed ORM) {
             return ORM << "Could not move globalized variable to the stack. "
                        << "Variable is potentially "
-                       << ((!NoCaptureAA.isAssumedNoCapture()) ? "captured."
-                                                               : "freed.");
+                       << (!NoCaptureAA.isAssumedNoCapture() ? "captured. "
+                                                               : "freed. ")
+                       << "Mark as noescape to override.";
+                  
           };
 
           LibFunc IsAllocShared;
           if (auto *AllocShared = dyn_cast<CallBase>(&I)) {
             TLI->getLibFunc(*AllocShared, IsAllocShared);
             if (IsAllocShared == LibFunc___kmpc_alloc_shared)
               A.emitRemark<OptimizationRemarkMissed>(
                   AllocShared, "HeapToStackFailed", Remark);

llvm/lib/Transforms/IPO/OpenMPOpt.cpp

 }
 
 PreservedAnalyses OpenMPOptPass::run(Module &M, ModuleAnalysisManager &AM) {
   if (!containsOpenMP(M))
     return PreservedAnalyses::all();
   if (DisableOpenMPOptimizations)
     return PreservedAnalyses::all();
 
+  FunctionAnalysisManager &FAM =
+      AM.getResult<FunctionAnalysisManagerModuleProxy>(M).getManager();
   KernelSet Kernels = getDeviceKernels(M);
 
   auto IsCalled = [&](Function &F) {
     if (Kernels.contains(&F))
       return true;
     for (const User *U : F.users())
       if (!isa<BlockAddress>(U))
         return true;
     return false;
   };
 
+  auto EmitRemark = [&](Function &F) {
+    auto &ORE = FAM.getResult<OptimizationRemarkEmitterAnalysis>(F);
+    ORE.emit([&]() { 
+      OptimizationRemarkMissed ORM(DEBUG_TYPE, "InternalizationFailure", &F);
+        return ORM << "Could not internalize function. " 
+                   << "Some optimizations may not be possible.";
+    });
+  };
+
   // Create internal copies of each function if this is a kernel Module. This
   // allows iterprocedural passes to see every call edge.
   DenseSet<const Function *> InternalizedFuncs;
   if (isOpenMPDevice(M))
     for (Function &F : M)
       if (!F.isDeclaration() && !Kernels.contains(&F) && IsCalled(F))
         if (Attributor::internalizeFunction(F, /* Force */ true))
           InternalizedFuncs.insert(&F);
+        else if (!F.hasLocalLinkage())
+          EmitRemark(F);
 
   // Look at every function in the Module unless it was internalized.
   SmallVector<Function *, 16> SCC;
   for (Function &F : M)
     if (!F.isDeclaration() && !InternalizedFuncs.contains(&F))
       SCC.push_back(&F);
 
   if (SCC.empty())
     return PreservedAnalyses::all();
 
-  FunctionAnalysisManager &FAM =
-      AM.getResult<FunctionAnalysisManagerModuleProxy>(M).getManager();
-
   AnalysisGetter AG(FAM);
 
   auto OREGetter = [&FAM](Function *F) -> OptimizationRemarkEmitter & {
     return FAM.getResult<OptimizationRemarkEmitterAnalysis>(*F);
   };
 
   BumpPtrAllocator Allocator;
   CallGraphUpdater CGUpdater;

llvm/test/Transforms/OpenMP/remove_globalization.ll

 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --function-signature
 ; RUN: opt -S -passes=openmp-opt < %s | FileCheck %s
 ; RUN: opt -passes=openmp-opt -pass-remarks=openmp-opt -pass-remarks-missed=openmp-opt -disable-output < %s 2>&1 | FileCheck %s -check-prefix=CHECK-REMARKS
 target datalayout = "e-i64:64-i128:128-v16:16-v32:32-n16:32:64"
 target triple = "nvptx64"
 
-; CHECK-REMARKS: remark: remove_globalization.c:4:2: Could not move globalized variable to the stack. Variable is potentially captured.
+; CHECK-REMARKS: remark: remove_globalization.c:4:2: Could not move globalized variable to the stack. Variable is potentially captured. Mark as noescape to override.
 ; CHECK-REMARKS: remark: remove_globalization.c:2:2: Moving globalized variable to the stack.
 ; CHECK-REMARKS: remark: remove_globalization.c:6:2: Moving globalized variable to the stack.
 
 @S = external local_unnamed_addr global i8*
 
 define void @kernel() {
 ; CHECK-LABEL: define {{[^@]+}}@kernel() {
 ; CHECK-NEXT:  entry:

llvm/test/Transforms/OpenMP/single_threaded_execution.ll

 ; RUN: opt -passes=openmp-opt -debug-only=openmp-opt -disable-output < %s 2>&1 | FileCheck %s
+; RUN: opt -passes=openmp-opt -pass-remarks-missed=openmp-opt -disable-output < %s 2>&1 | FileCheck %s --check-prefix=REMARKS
 ; REQUIRES: asserts
 ; ModuleID = 'single_threaded_exeuction.c'
 
 define weak void @kernel() {
   call void @__kmpc_kernel_init(i32 512, i16 1)
   call void @nvptx()
   call void @amdgcn()
   ret void
 }
 
+; REMARKS: remark: single_threaded_execution.c:1:0: Could not internalize function. Some optimizations may not be possible.
+
 ; CHECK-NOT: [openmp-opt] Basic block @nvptx entry is executed by a single thread.
 ; CHECK: [openmp-opt] Basic block @nvptx if.then is executed by a single thread.
 ; CHECK-NOT: [openmp-opt] Basic block @nvptx if.end is executed by a single thread.
 ; Function Attrs: noinline
 define internal void @nvptx() {
 entry:
   %call = call i32 @llvm.nvvm.read.ptx.sreg.tid.x()
   %cmp = icmp eq i32 %call, 0
   br i1 %cmp, label %if.then, label %if.end
 
 if.then:
   call void @foo()
   call void @bar()
+  call void @baz()
   br label %if.end
 
 if.end:
   ret void
 }
 
 ; CHECK-NOT: [openmp-opt] Basic block @amdgcn entry is executed by a single thread.
 ; CHECK: [openmp-opt] Basic block @amdgcn if.then is executed by a single thread.
 ; CHECK-NOT: [openmp-opt] Basic block @amdgcn if.end is executed by a single thread.
 ; Function Attrs: noinline
 define internal void @amdgcn() {
 entry:
   %call = call i32 @llvm.amdgcn.workitem.id.x()
   %cmp = icmp eq i32 %call, 0
   br i1 %cmp, label %if.then, label %if.end
 
 if.then:
   call void @foo()
   call void @bar()
+  call void @baz()
   br label %if.end
 
 if.end:
   ret void
 }
 
 ; CHECK: [openmp-opt] Basic block @foo entry is executed by a single thread.
 ; Function Attrs: noinline
 define internal void @foo() {
 entry:
   ret void
 }
 
 ; CHECK: [openmp-opt] Basic block @bar.internalized entry is executed by a single thread.
 ; Function Attrs: noinline
 define void @bar() {
 entry:
   ret void
 }
 
+; CHECK-NOT: [openmp-opt] Basic block @baz entry is executed by a single thread.
+; Function Attrs: noinline
+define weak void @baz() !dbg !8 {
+entry:
+  ret void
+}
+
 declare i32 @llvm.nvvm.read.ptx.sreg.tid.x()
 
 declare i32 @llvm.amdgcn.workitem.id.x()
 
 declare void @__kmpc_kernel_init(i32, i16)
 
 !llvm.dbg.cu = !{!0}
 !llvm.module.flags = !{!3, !4, !5, !6}
 !nvvm.annotations = !{!7}
 
 
 !0 = distinct !DICompileUnit(language: DW_LANG_C99, file: !1, producer: "clang version 12.0.0", isOptimized: false, runtimeVersion: 0, emissionKind: FullDebug, enums: !2, splitDebugInlining: false, nameTableKind: None)
 !1 = !DIFile(filename: "single_threaded_execution.c", directory: "/tmp/single_threaded_execution.c")
 !2 = !{}
 !3 = !{i32 2, !"Debug Info Version", i32 3}
 !4 = !{i32 1, !"wchar_size", i32 4}
 !5 = !{i32 7, !"openmp", i32 50}
 !6 = !{i32 7, !"openmp-device", i32 50}
 !7 = !{void ()* @kernel, !"kernel", i32 1}
+!8 = distinct !DISubprogram(name: "bar", scope: !1, file: !1, line: 8, type: !9, scopeLine: 1, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0, retainedNodes: !2)
+!9 = !DISubroutineType(types: !2)
+!10 = !DILocation(line: 5, column: 7, scope: !8)