Set the prof weight correctly for call instructions in DeadArgumentElimination.

[danielcdh]

Summary

In DeadArgumentElimination, the call instructions will be replaced. We also need to set the prof weights so that function inlining can find the correct profile.

[eraman]

lib/IR/Instruction.cpp

658

Nit: "...call and invoke instructions"

659

Why is this guard needed? Is there a difference in handling an instruction with no metadata and one with a call weight of 0?

test/Transforms/DeadArgElim/prof.ll

1 ↗

(On Diff #92362)

Nit: the file name is not very meaningful.

3 ↗

(On Diff #92362)

reserved?

[danielcdh]

update

[danielcdh]

lib/IR/Instruction.cpp

659

It should not be needed, removed.

[eraman]

LGTM after you fix the test case.

lib/IR/Instruction.cpp

658

Nit: "call and invoke instructions"

test/Transforms/DeadArgElim/call_profile.ll

6

Could you add the test case for the non va_arg case as well? There are two places in the code where a new instruction is created and you should check both the cases (or optionally refactor the code so that the new instruction creation and metadata copying is done in one helper function)

[danielcdh]

update

include/llvm/IR/Instruction.h

public:
   /// Retrieve total raw weight values of a branch.
   /// Returns true on success with profile total weights filled in.
   /// Returns false if no metadata was found.
   bool extractProfTotalWeight(uint64_t &TotalVal) const;
 
   /// Updates branch_weights metadata by scaling it by \p S / \p T.
   void updateProfWeight(uint64_t S, uint64_t T);
 
+  /// Sets the branch_weights metadata to \p W for CallInst.
+  void setProfWeight(uint64_t W);
+
   /// Set the debug location information for this instruction.
   void setDebugLoc(DebugLoc Loc) { DbgLoc = std::move(Loc); }
 
   /// Return the debug location for this node as a DebugLoc.
   const DebugLoc &getDebugLoc() const { return DbgLoc; }
 
   /// Set or clear the nsw flag on this instruction, which must be an operator
   /// which supports this flag. See LangRef.html for the meaning of this flag.

lib/IR/Instruction.cpp

APInt Val(128, mdconst::dyn_extract<ConstantInt>(ProfileData->getOperand(i))
                        ->getValue()
                        .getZExtValue());
     Val *= APInt(128, S);
     Weights.push_back(Val.udiv(APInt(128, T)).getLimitedValue());
   }
   MDBuilder MDB(getContext());
   setMetadata(LLVMContext::MD_prof, MDB.createBranchWeights(Weights));
 }
+
+void Instruction::setProfWeight(uint64_t W) {
+  assert((isa<CallInst>(this) || isa<InvokeInst>(this)) &&
+         "Can only set weights for call and invoke instruciton");
+  SmallVector<uint32_t, 1> Weights;
+  Weights.push_back(W);
+  MDBuilder MDB(getContext());
+  setMetadata(LLVMContext::MD_prof, MDB.createBranchWeights(Weights));
+}

lib/Transforms/IPO/DeadArgumentElimination.cpp

for (Value::user_iterator I = Fn.user_begin(), E = Fn.user_end(); I != E; ) {
     } else {
       New = CallInst::Create(NF, Args, OpBundles, "", Call);
       cast<CallInst>(New)->setCallingConv(CS.getCallingConv());
       cast<CallInst>(New)->setAttributes(PAL);
       cast<CallInst>(New)->setTailCallKind(
           cast<CallInst>(Call)->getTailCallKind());
     }
     New->setDebugLoc(Call->getDebugLoc());
+    uint64_t W;
+    if (Call->extractProfTotalWeight(W))
+      New->setProfWeight(W);
 
     Args.clear();
 
     if (!Call->use_empty())
       Call->replaceAllUsesWith(New);
 
     New->takeName(Call);
 
while (!F->use_empty()) {
     } else {
       New = CallInst::Create(NF, Args, OpBundles, "", Call);
       cast<CallInst>(New)->setCallingConv(CS.getCallingConv());
       cast<CallInst>(New)->setAttributes(NewCallPAL);
       cast<CallInst>(New)->setTailCallKind(
           cast<CallInst>(Call)->getTailCallKind());
     }
     New->setDebugLoc(Call->getDebugLoc());
+    uint64_t W;
+    if (Call->extractProfTotalWeight(W))
+      New->setProfWeight(W);
 
     Args.clear();
 
     if (!Call->use_empty()) {
       if (New->getType() == Call->getType()) {
         // Return type not changed? Just replace users then.
         Call->replaceAllUsesWith(New);
         New->takeName(Call);

test/Transforms/DeadArgElim/call_profile.ll

+; RUN: opt -deadargelim -S < %s | FileCheck %s
+
+; Checks if !prof metadata is corret in deadargelim.
+
+; Function Attrs: uwtable
+define void @_Z2f2v() #0 {
+; CHECK: call void @_ZL2f1iz(), !prof ![[PROF:[0-9]]]
+  call void (i32, ...) @_ZL2f1iz(i32 1), !prof !0
+  ret void
+}
+
+; Function Attrs: nounwind uwtable
+define internal void @_ZL2f1iz(i32, ...) #1 {
+  ret void
+}
+
+; CHECK:![[PROF]] = !{!"branch_weights", i32 30}
+!0 = !{!"branch_weights", i32 30}