When we look at instructions to convert to setting the 's' flag, we need to look

at more than those which define CPSR. You can have this situation:

(1)  subs  ...
(2)  sub   r6, r5, r4
(3)  movge ...
(4)  cmp   r6, 0
(5)  movge ...

We cannot convert (2) to "subs" because (3) is using the CPSR set by
(1). There's an analogous situation here:

(1)  sub   r1, r2, r3
(2)  sub   r4, r5, r6
(3)  cmp   r4, ...
(5)  movge ...
(6)  cmp   r1, ...
(7)  movge ...

We cannot convert (1) to "subs" because of the intervening use of CPSR.

llvm-svn: 117950

lib/CodeGen/PeepholeOptimizer.cpp

@@ -50,6 +50,10 @@ static cl::opt<bool>
 Aggressive("aggressive-ext-opt", cl::Hidden,
            cl::desc("Aggressive extension optimization"));
 
+static cl::opt<bool>
+DisablePeephole("disable-peephole", cl::Hidden, cl::init(false),
+                cl::desc("Disable the peephole optimizer"));
+
 STATISTIC(NumReuse,      "Number of extension results reused");
 STATISTIC(NumEliminated, "Number of compares eliminated");
 
@@ -276,11 +280,9 @@ bool PeepholeOptimizer::runOnMachineFunction(MachineFunction &MF) {
 
       if (MI->getDesc().isCompare() &&
           !MI->getDesc().hasUnmodeledSideEffects()) {
-#if 0
-        if (OptimizeCmpInstr(MI, MBB, MII))
+        if (!DisablePeephole && OptimizeCmpInstr(MI, MBB, MII))
           Changed = true;
         else
-#endif
           ++MII;
       } else {
         Changed |= OptimizeExtInstr(MI, MBB, LocalMIs);

lib/Target/ARM/ARMBaseInstrInfo.cpp

@@ -21,7 +21,6 @@
 #include "llvm/Constants.h"
 #include "llvm/Function.h"
 #include "llvm/GlobalValue.h"
-#include "llvm/ADT/STLExtras.h"
 #include "llvm/CodeGen/LiveVariables.h"
 #include "llvm/CodeGen/MachineConstantPool.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
@@ -34,6 +33,7 @@
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
+#include "llvm/ADT/STLExtras.h"
 using namespace llvm;
 
 static cl::opt<bool>
@@ -1557,10 +1557,10 @@ OptimizeCompareInstr(MachineInstr *CmpInstr, unsigned SrcReg, int CmpMask,
 
     for (unsigned IO = 0, EO = Instr.getNumOperands(); IO != EO; ++IO) {
       const MachineOperand &MO = Instr.getOperand(IO);
-      if (!MO.isReg() || !MO.isDef()) continue;
+      if (!MO.isReg()) continue;
 
-      // This instruction modifies CPSR before the one we want to change. We
-      // can't do this transformation.
+      // This instruction modifies or uses CPSR after the one we want to
+      // change. We can't do this transformation.
       if (MO.getReg() == ARM::CPSR)
         return false;
     }

test/CodeGen/ARM/arm-and-tst-peephole.ll

@@ -1,10 +1,15 @@
-; RUN: llc < %s -march=arm
-; FIXME: llc < %s -march=thumb | FileCheck -check-prefix=THUMB %s
-; FIXME: llc < %s -march=thumb -mattr=+thumb2 | FileCheck -check-prefix=T2 %s
+; RUN: llc < %s -march=arm | FileCheck -check-prefix=ARM %s
+; RUN: llc < %s -march=thumb | FileCheck -check-prefix=THUMB %s
+; RUN: llc < %s -march=thumb -mattr=+thumb2 | FileCheck -check-prefix=T2 %s
+
+; FIXME: The -march=thumb test doesn't change if -disable-peephole is specified.
 
 %struct.Foo = type { i8* }
 
-define %struct.Foo* @_ZN3Foo7collectEj(%struct.Foo* %this, i32 %acc) nounwind readonly align 2 {
+; ARM:   foo
+; THUMB: foo
+; T2:    foo
+define %struct.Foo* @foo(%struct.Foo* %this, i32 %acc) nounwind readonly align 2 {
 entry:
   %scevgep = getelementptr %struct.Foo* %this, i32 1
   br label %tailrecurse
@@ -18,8 +23,8 @@ tailrecurse:                                      ; preds = %sw.bb, %entry
   %tmp2 = load i8** %scevgep5
   %0 = ptrtoint i8* %tmp2 to i32
 
-; CHECK:      ands r12, r12, #3
-; CHECK-NEXT: beq
+; ARM:      ands r12, r12, #3
+; ARM-NEXT: beq
 
 ; THUMB:      movs r5, #3
 ; THUMB-NEXT: mov r6, r4
@@ -66,7 +71,7 @@ sw.epilog:                                        ; preds = %tailrecurse.switch
 
 %struct.S = type { i8* (i8*)*, [1 x i8] }
 
-; CHECK: bar
+; ARM: bar
 ; THUMB: bar
 ; T2: bar
 define internal zeroext i8 @bar(%struct.S* %x, %struct.S* nocapture %y) nounwind readonly {
@@ -74,7 +79,7 @@ entry:
   %0 = getelementptr inbounds %struct.S* %x, i32 0, i32 1, i32 0
   %1 = load i8* %0, align 1
   %2 = zext i8 %1 to i32
-; CHECK: ands
+; ARM: ands
 ; THUMB: ands
 ; T2: ands
   %3 = and i32 %2, 112
@@ -85,7 +90,7 @@ bb:                                               ; preds = %entry
   %5 = getelementptr inbounds %struct.S* %y, i32 0, i32 1, i32 0
   %6 = load i8* %5, align 1
   %7 = zext i8 %6 to i32
-; CHECK: andsne
+; ARM: andsne
 ; THUMB: ands
 ; T2: andsne
   %8 = and i32 %7, 112

test/CodeGen/ARM/long_shift.ll

@@ -1,6 +1,4 @@
 ; RUN: llc < %s -march=arm | FileCheck %s
-; XFAIL: *
-; FIXME: Fix after peephole optimizer is fixed.
 
 define i64 @f0(i64 %A, i64 %B) {
 ; CHECK: f0