Do not optimize (or (and X,Y), Z) into BFI and other sequences if the AND ISDNode has more than one user.

rdar://11876519

llvm-svn: 161775

2 files changed, 22 insertions, 1 deletions

diff --git a/llvm/lib/Target/ARM/ARMISelLowering.cpp b/llvm/lib/Target/ARM/ARMISelLowering.cpp
index 52bdd528870..c66618a8ef5 100644
--- a/llvm/lib/Target/ARM/ARMISelLowering.cpp
+++ b/llvm/lib/Target/ARM/ARMISelLowering.cpp
@@ -7422,8 +7422,12 @@ static SDValue PerformORCombine(SDNode *N,
       return COR;
   }
 
+
+  // The code below optimizes (or (and X, Y), Z).
+  // The AND operand needs to have a single user to make these optimizations
+  // profitable.
   SDValue N0 = N->getOperand(0);
-  if (N0.getOpcode() != ISD::AND)
+  if (N0.getOpcode() != ISD::AND || !N0.hasOneUse())
     return SDValue();
   SDValue N1 = N->getOperand(1);
 
diff --git a/llvm/test/CodeGen/ARM/2012-08-13-bfi.ll b/llvm/test/CodeGen/ARM/2012-08-13-bfi.ll
new file mode 100644
index 00000000000..8263833d987
--- /dev/null
+++ b/llvm/test/CodeGen/ARM/2012-08-13-bfi.ll
@@ -0,0 +1,17 @@
+; RUN: llc -march=thumb -mcpu=cortex-a8 < %s | FileCheck %s
+
+target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
+target triple = "x86_64-apple-macosx10.8.0"
+
+; CHECK: foo
+; CHECK-NOT: bfi
+; CHECK: bx
+define i32 @foo(i8 zeroext %i) nounwind uwtable readnone ssp {
+  %1 = and i8 %i, 15
+  %2 = zext i8 %1 to i32
+  %3 = icmp ult i8 %1, 10
+  %4 = or i32 %2, 48
+  %5 = add nsw i32 %2, 55
+  %6 = select i1 %3, i32 %4, i32 %5
+  ret i32 %6
+}