SimplifyCFG has a heuristics for out-of-order processors that decides when it is worthwhile to merge branches. It tries to estimate if the operands of the instruction that we want to hoist are ready. This commit marks function arguments as 'ready' because they require no calculation. This boosts libquantum and a few other workloads from the testsuite.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194346 91177308-0d34-0410-b5e6-96231b3b80d8

lib/Transforms/Utils/SimplifyCFG.cpp

@@ -2096,7 +2096,7 @@ bool llvm::FoldBranchToCommonDest(BranchInst *BI) {
       for (Instruction::op_iterator OI = BonusInst->op_begin(),
            OE = BonusInst->op_end(); OI != OE; ++OI) {
         Value *V = *OI;
-        if (!isa<Constant>(V))
+        if (!isa<Constant>(V) && !isa<Argument>(V))
           UsedValues.insert(V);
       }
 

test/Transforms/SimplifyCFG/common-dest-folding.ll

@@ -0,0 +1,28 @@
+; RUN: opt < %s -simplifycfg -S | FileCheck %s
+
+;CHECK: @foo
+;CHECK: and i32 %c1, %k
+;CHECK: icmp eq i32
+;CHECK: and i32 %c2, %k
+;CHECK: icmp eq i32
+;CHECK: or i1
+;CHECK: ret
+define i32 @foo(i32 %k, i32 %c1, i32 %c2) {
+  %1 = and i32 %c1, %k
+  %2 = icmp eq i32 %1, 0
+  br i1 %2, label %8, label %3
+
+; <label>:3                                       ; preds = %0
+  %4 = and i32 %c2, %k
+  %5 = icmp eq i32 %4, 0
+  br i1 %5, label %8, label %6
+
+; <label>:6                                       ; preds = %3
+  %7 = tail call i32 (...)* @bar() nounwind
+  br label %8
+
+; <label>:8                                       ; preds = %3, %0, %6
+  ret i32 undef
+}
+
+declare i32 @bar(...)