Limit the icmp+phi merging optimization to the cases where it is profitable:
don't make i1 phis when it won't be possible to eliminate them.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52097 91177308-0d34-0410-b5e6-96231b3b80d8
diff --git a/lib/Transforms/Scalar/InstructionCombining.cpp b/lib/Transforms/Scalar/InstructionCombining.cpp
index 67c40fe..3b181c8 100644
--- a/lib/Transforms/Scalar/InstructionCombining.cpp
+++ b/lib/Transforms/Scalar/InstructionCombining.cpp
@@ -5043,8 +5043,12 @@
if (Instruction *LHSI = dyn_cast<Instruction>(Op0))
switch (LHSI->getOpcode()) {
case Instruction::PHI:
- if (Instruction *NV = FoldOpIntoPhi(I))
- return NV;
+ // Only fold fcmp into the PHI if the phi and fcmp are in the same
+ // block. If in the same block, we're encouraging jump threading. If
+ // not, we are just pessimizing the code by making an i1 phi.
+ if (LHSI->getParent() == I.getParent())
+ if (Instruction *NV = FoldOpIntoPhi(I))
+ return NV;
break;
case Instruction::SIToFP:
case Instruction::UIToFP:
@@ -5348,8 +5352,12 @@
break;
case Instruction::PHI:
- if (Instruction *NV = FoldOpIntoPhi(I))
- return NV;
+ // Only fold icmp into the PHI if the phi and fcmp are in the same
+ // block. If in the same block, we're encouraging jump threading. If
+ // not, we are just pessimizing the code by making an i1 phi.
+ if (LHSI->getParent() == I.getParent())
+ if (Instruction *NV = FoldOpIntoPhi(I))
+ return NV;
break;
case Instruction::Select: {
// If either operand of the select is a constant, we can fold the
diff --git a/test/Transforms/InstCombine/2008-06-08-ICmpPHI.ll b/test/Transforms/InstCombine/2008-06-08-ICmpPHI.ll
new file mode 100644
index 0000000..555f810
--- /dev/null
+++ b/test/Transforms/InstCombine/2008-06-08-ICmpPHI.ll
@@ -0,0 +1,49 @@
+; RUN: llvm-as < %s | opt -instcombine | llvm-dis | grep {phi i32} | count 2
+target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128"
+target triple = "i386-apple-darwin8"
+
+define void @test() nounwind {
+entry:
+ br label %bb
+
+bb: ; preds = %bb16, %entry
+ %i.0 = phi i32 [ 0, %entry ], [ %indvar.next, %somebb ] ; <i32> [#uses=1]
+ %x.0 = phi i32 [ 37, %entry ], [ %tmp17, %somebb ] ; <i32> [#uses=1]
+ %tmp = tail call i32 (...)* @bork( ) nounwind ; <i32> [#uses=0]
+ %tmp1 = tail call i32 (...)* @bork( ) nounwind ; <i32> [#uses=0]
+ %tmp2 = tail call i32 (...)* @bork( ) nounwind ; <i32> [#uses=1]
+ %tmp3 = icmp eq i32 %tmp2, 0 ; <i1> [#uses=1]
+ br i1 %tmp3, label %bb7, label %bb5
+
+bb5: ; preds = %bb
+ %tmp6 = tail call i32 (...)* @bork( ) nounwind ; <i32> [#uses=0]
+ br label %bb7
+
+bb7: ; preds = %bb5, %bb
+ %tmp8 = tail call i32 (...)* @bork( ) nounwind ; <i32> [#uses=0]
+ %tmp9 = tail call i32 (...)* @bork( ) nounwind ; <i32> [#uses=0]
+ %tmp11 = icmp eq i32 %x.0, 37 ; <i1> [#uses=1]
+ br i1 %tmp11, label %bb14, label %bb16
+
+bb14: ; preds = %bb7
+ %tmp15 = tail call i32 (...)* @bar( ) nounwind ; <i32> [#uses=0]
+ br label %bb16
+
+bb16: ; preds = %bb14, %bb7
+ %tmp17 = tail call i32 (...)* @zap( ) nounwind ; <i32> [#uses=1]
+ %indvar.next = add i32 %i.0, 1 ; <i32> [#uses=2]
+ %exitcond = icmp eq i32 %indvar.next, 42 ; <i1> [#uses=1]
+ br i1 %exitcond, label %return, label %somebb
+
+somebb:
+ br label %bb
+
+return: ; preds = %bb16
+ ret void
+}
+
+declare i32 @bork(...)
+
+declare i32 @bar(...)
+
+declare i32 @zap(...)