DeadArgumentElim] Clone function level metadatas

[xur]

Summary

Some Function level metadatas, such as function entry count, are not cloned in DeadArgumentElim. This happens a lot in lto/thinlto because of DeadArgumentElim after internalization.

This patch clones the metadatas in the original function to the new function.

[davidxl]

Also add an dead vararg elimination test case for better coverage.

test/Transforms/DeadArgElim/func_metadata.ll

13 ↗

(On Diff #137106)

dead arg elim is missing the test case.

27 ↗

(On Diff #137106)

Simplify the attribute.

[xur]

Integrated David's review comments: (1) add vararg test (2) simplify function attributes.

[xur]

ping

[davidxl]

lgtm

llvm/trunk/lib/Transforms/IPO/DeadArgumentElimination.cpp

bool DeadArgumentEliminationPass::DeleteDeadVarargs(Function &Fn) {
   // it, remove the dead arguments from the DeadArguments list.
   for (Function::arg_iterator I = Fn.arg_begin(), E = Fn.arg_end(),
        I2 = NF->arg_begin(); I != E; ++I, ++I2) {
     // Move the name and users over to the new version.
     I->replaceAllUsesWith(&*I2);
     I2->takeName(&*I);
   }
 
-  // Patch the pointer to LLVM function in debug info descriptor.
-  NF->setSubprogram(Fn.getSubprogram());
+  // Clone metadatas from the old function, including debug info descriptor.
+  SmallVector<std::pair<unsigned, MDNode *>, 1> MDs;
+  Fn.getAllMetadata(MDs);
+  for (auto MD : MDs)
+    NF->addMetadata(MD.first, *MD.second);
 
   // Fix up any BlockAddresses that refer to the function.
   Fn.replaceAllUsesWith(ConstantExpr::getBitCast(NF, Fn.getType()));
   // Delete the bitcast that we just created, so that NF does not
   // appear to be address-taken.
   NF->removeDeadConstantUsers();
   // Finally, nuke the old function.
   Fn.eraseFromParent();
bool DeadArgumentEliminationPass::RemoveDeadStuffFromFunction(Function *F) {
   NF->copyAttributesFrom(F);
   NF->setComdat(F->getComdat());
   NF->setAttributes(NewPAL);
   // Insert the new function before the old function, so we won't be processing
   // it again.
   F->getParent()->getFunctionList().insert(F->getIterator(), NF);
   NF->takeName(F);
 
-  // Patch the pointer to LLVM function in debug info descriptor.
-  NF->setSubprogram(F->getSubprogram());
-
   // Loop over all of the callers of the function, transforming the call sites
   // to pass in a smaller number of arguments into the new function.
   std::vector<Value*> Args;
   while (!F->use_empty()) {
     CallSite CS(F->user_back());
     Instruction *Call = CS.getInstruction();
 
     ArgAttrVec.clear();
for (BasicBlock &BB : *NF)
             }
         }
         // Replace the return instruction with one returning the new return
         // value (possibly 0 if we became void).
         ReturnInst::Create(F->getContext(), RetVal, RI);
         BB.getInstList().erase(RI);
       }
 
+  // Clone metadatas from the old function, including debug info descriptor.
+  SmallVector<std::pair<unsigned, MDNode *>, 1> MDs;
+  F->getAllMetadata(MDs);
+  for (auto MD : MDs)
+    NF->addMetadata(MD.first, *MD.second);
+
   // Now that the old function is dead, delete it.
   F->eraseFromParent();
 
   return true;
 }
 
 PreservedAnalyses DeadArgumentEliminationPass::run(Module &M,
                                                    ModuleAnalysisManager &) {

llvm/trunk/test/Transforms/DeadArgElim/func_metadata.ll

+; RUN: opt -deadargelim -S < %s | FileCheck %s
+
+; Check if function level metadatas are properly cloned.
+
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+@s = common dso_local local_unnamed_addr global i32 0, align 4
+
+define internal i32 @va_func(i32 %num, ...) !prof !28 !PGOFuncName !29{
+; CHECK: define internal void @va_func(i32 %num) !prof ![[ENTRYCOUNT:[0-9]+]] !PGOFuncName ![[PGOFUNCNAME1:[0-9]+]] {
+entry:
+  %0 = load i32, i32* @s, align 4, !tbaa !31
+  %add = add nsw i32 %0, %num
+  store i32 %add, i32* @s, align 4, !tbaa !31
+  ret i32 0
+}
+
+define internal fastcc i32 @foo() unnamed_addr !prof !28 !PGOFuncName !30 {
+; CHECK: define internal fastcc void @foo() unnamed_addr !prof ![[ENTRYCOUNT:[0-9]+]] !PGOFuncName ![[PGOFUNCNAME2:[0-9]+]] {
+entry:
+  %0 = load i32, i32* @s, align 4, !tbaa !31
+  %add = add nsw i32 %0, 8
+  store i32 %add, i32* @s, align 4, !tbaa !31
+  ret i32 0
+}
+
+!llvm.module.flags = !{!0, !1}
+
+!0 = !{i32 1, !"wchar_size", i32 4}
+!1 = !{i32 1, !"ProfileSummary", !2}
+!2 = !{!3, !4, !5, !6, !7, !8, !9, !10}
+!3 = !{!"ProfileFormat", !"InstrProf"}
+!4 = !{!"TotalCount", i64 2}
+!5 = !{!"MaxCount", i64 1}
+!6 = !{!"MaxInternalCount", i64 0}
+!7 = !{!"MaxFunctionCount", i64 1}
+!8 = !{!"NumCounts", i64 2}
+!9 = !{!"NumFunctions", i64 2}
+!10 = !{!"DetailedSummary", !11}
+!11 = !{!12, !13, !14, !15, !16, !17, !17, !18, !18, !19, !20, !21, !22, !23, !24, !25, !26, !27}
+!12 = !{i32 10000, i64 0, i32 0}
+!13 = !{i32 100000, i64 0, i32 0}
+!14 = !{i32 200000, i64 0, i32 0}
+!15 = !{i32 300000, i64 0, i32 0}
+!16 = !{i32 400000, i64 0, i32 0}
+!17 = !{i32 500000, i64 1, i32 2}
+!18 = !{i32 600000, i64 1, i32 2}
+!19 = !{i32 700000, i64 1, i32 2}
+!20 = !{i32 800000, i64 1, i32 2}
+!21 = !{i32 900000, i64 1, i32 2}
+!22 = !{i32 950000, i64 1, i32 2}
+!23 = !{i32 990000, i64 1, i32 2}
+!24 = !{i32 999000, i64 1, i32 2}
+!25 = !{i32 999900, i64 1, i32 2}
+!26 = !{i32 999990, i64 1, i32 2}
+!27 = !{i32 999999, i64 1, i32 2}
+!28 = !{!"function_entry_count", i64 1}
+; CHECK: ![[ENTRYCOUNT]] = !{!"function_entry_count", i64 1}
+!29 = !{!"foo.c:va_func"}
+; CHECK: ![[PGOFUNCNAME1]] = !{!"foo.c:va_func"}
+!30 = !{!"foo.c:foo"}
+; CHECK: ![[PGOFUNCNAME2]] = !{!"foo.c:foo"}
+!31 = !{!32, !32, i64 0}
+!32 = !{!"int", !33, i64 0}
+!33 = !{!"omnipotent char", !34, i64 0}
+!34 = !{!"Simple C/C++ TBAA"}