From e61e516a51f1211c1385a3043523552f4e56003d Mon Sep 17 00:00:00 2001
From: Evan Cheng <evan.cheng@apple.com>
Date: Tue, 9 Oct 2012 23:48:33 +0000
Subject: [PATCH] When expanding atomic load arith instructions, do not lose
 target flags. rdar://12453106

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@165568 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/Target/X86/X86ISelLowering.cpp      |   7 +-
 test/CodeGen/X86/atomic-minmax-i6432.ll | 100 ++++++++++++++----------
 2 files changed, 63 insertions(+), 44 deletions(-)

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 5e0b9c0034a..7e43e5432d8 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -12389,9 +12389,12 @@ X86TargetLowering::EmitAtomicLoadArith6432(MachineInstr *MI,
   // Hi
   MIB = BuildMI(thisMBB, DL, TII->get(LOADOpc), X86::EDX);
   for (unsigned i = 0; i < X86::AddrNumOperands; ++i) {
-    if (i == X86::AddrDisp)
+    if (i == X86::AddrDisp) {
       MIB.addDisp(MI->getOperand(MemOpndSlot + i), 4); // 4 == sizeof(i32)
-    else
+      // Don't forget to transfer the target flag.
+      MachineOperand &MO = MIB->getOperand(MIB->getNumOperands()-1);
+      MO.setTargetFlags(MI->getOperand(MemOpndSlot + i).getTargetFlags());
+    } else
       MIB.addOperand(MI->getOperand(MemOpndSlot + i));
   }
   MIB.setMemRefs(MMOBegin, MMOEnd);
diff --git a/test/CodeGen/X86/atomic-minmax-i6432.ll b/test/CodeGen/X86/atomic-minmax-i6432.ll
index 01a926489be..e3ef605f7f1 100644
--- a/test/CodeGen/X86/atomic-minmax-i6432.ll
+++ b/test/CodeGen/X86/atomic-minmax-i6432.ll
@@ -1,51 +1,67 @@
-; RUN: llc -march=x86 -mattr=+cmov -mtriple=i386-pc-linux < %s | FileCheck %s
+; RUN: llc -march=x86 -mattr=+cmov -mtriple=i386-pc-linux < %s | FileCheck %s -check-prefix=LINUX
+; RUN: llc -march=x86 -mtriple=i386-macosx -relocation-model=pic < %s | FileCheck %s -check-prefix=PIC
+
 @sc64 = external global i64
 
 define void @atomic_maxmin_i6432() {
-; CHECK: atomic_maxmin_i6432
+; LINUX: atomic_maxmin_i6432
   %1 = atomicrmw max  i64* @sc64, i64 5 acquire
-; CHECK: [[LABEL:.LBB[0-9]+_[0-9]+]]
-; CHECK: cmpl
-; CHECK: setl
-; CHECK: cmpl
-; CHECK: setl
-; CHECK: cmovne
-; CHECK: cmovne
-; CHECK: lock
-; CHECK-NEXT: cmpxchg8b
-; CHECK: jne [[LABEL]]
+; LINUX: [[LABEL:.LBB[0-9]+_[0-9]+]]
+; LINUX: cmpl
+; LINUX: setl
+; LINUX: cmpl
+; LINUX: setl
+; LINUX: cmovne
+; LINUX: cmovne
+; LINUX: lock
+; LINUX-NEXT: cmpxchg8b
+; LINUX: jne [[LABEL]]
   %2 = atomicrmw min  i64* @sc64, i64 6 acquire
-; CHECK: [[LABEL:.LBB[0-9]+_[0-9]+]]
-; CHECK: cmpl
-; CHECK: setg
-; CHECK: cmpl
-; CHECK: setg
-; CHECK: cmovne
-; CHECK: cmovne
-; CHECK: lock
-; CHECK-NEXT: cmpxchg8b
-; CHECK: jne [[LABEL]]
+; LINUX: [[LABEL:.LBB[0-9]+_[0-9]+]]
+; LINUX: cmpl
+; LINUX: setg
+; LINUX: cmpl
+; LINUX: setg
+; LINUX: cmovne
+; LINUX: cmovne
+; LINUX: lock
+; LINUX-NEXT: cmpxchg8b
+; LINUX: jne [[LABEL]]
   %3 = atomicrmw umax i64* @sc64, i64 7 acquire
-; CHECK: [[LABEL:.LBB[0-9]+_[0-9]+]]
-; CHECK: cmpl
-; CHECK: setb
-; CHECK: cmpl
-; CHECK: setb
-; CHECK: cmovne
-; CHECK: cmovne
-; CHECK: lock
-; CHECK-NEXT: cmpxchg8b
-; CHECK: jne [[LABEL]]
+; LINUX: [[LABEL:.LBB[0-9]+_[0-9]+]]
+; LINUX: cmpl
+; LINUX: setb
+; LINUX: cmpl
+; LINUX: setb
+; LINUX: cmovne
+; LINUX: cmovne
+; LINUX: lock
+; LINUX-NEXT: cmpxchg8b
+; LINUX: jne [[LABEL]]
   %4 = atomicrmw umin i64* @sc64, i64 8 acquire
-; CHECK: [[LABEL:.LBB[0-9]+_[0-9]+]]
-; CHECK: cmpl
-; CHECK: seta
-; CHECK: cmpl
-; CHECK: seta
-; CHECK: cmovne
-; CHECK: cmovne
-; CHECK: lock
-; CHECK-NEXT: cmpxchg8b
-; CHECK: jne [[LABEL]]
+; LINUX: [[LABEL:.LBB[0-9]+_[0-9]+]]
+; LINUX: cmpl
+; LINUX: seta
+; LINUX: cmpl
+; LINUX: seta
+; LINUX: cmovne
+; LINUX: cmovne
+; LINUX: lock
+; LINUX-NEXT: cmpxchg8b
+; LINUX: jne [[LABEL]]
+  ret void
+}
+
+; rdar://12453106
+@id = internal global i64 0, align 8
+
+define void @tf_bug(i8* %ptr) nounwind {
+; PIC: tf_bug:
+; PIC: movl _id-L1$pb(
+; PIC: movl (_id-L1$pb)+4(
+  %tmp1 = atomicrmw add i64* @id, i64 1 seq_cst
+  %tmp2 = add i64 %tmp1, 1
+  %tmp3 = bitcast i8* %ptr to i64*
+  store i64 %tmp2, i64* %tmp3, align 4
   ret void
 }
-- 
2.34.1