X86: Bad peephole interaction between adc, MOV32r0

The peephole tries to reorder MOV32r0 instructions such that they are
before the instruction that modifies EFLAGS.

The problem is that the peephole does not consider the case where the
instruction that modifies EFLAGS also depends on the previous state of
EFLAGS.

Instead, walk backwards until we find an instruction that has a def for
EFLAGS but does not have a use.
If we find such an instruction, insert the MOV32r0 before it.
If it cannot find such an instruction, skip the optimization.

llvm-svn: 182184

2 files changed, 45 insertions, 3 deletions

diff --git a/llvm/lib/Target/X86/X86InstrInfo.cpp b/llvm/lib/Target/X86/X86InstrInfo.cpp
index 582add7854a..afa43c134f9 100644
--- a/llvm/lib/Target/X86/X86InstrInfo.cpp
+++ b/llvm/lib/Target/X86/X86InstrInfo.cpp
@@ -3449,10 +3449,25 @@ optimizeCompareInstr(MachineInstr *CmpInstr, unsigned SrcReg, unsigned SrcReg2,
 
   // The instruction to be updated is either Sub or MI.
   Sub = IsCmpZero ? MI : Sub;
-  // Move Movr0Inst to the place right before Sub.
+  // Move Movr0Inst to the appropriate place before Sub.
   if (Movr0Inst) {
-    Sub->getParent()->remove(Movr0Inst);
-    Sub->getParent()->insert(MachineBasicBlock::iterator(Sub), Movr0Inst);
+    // Look backwards until we find a def that doesn't use the current EFLAGS.
+    Def = Sub;
+    MachineBasicBlock::reverse_iterator
+      InsertI = MachineBasicBlock::reverse_iterator(++Def),
+                InsertE = Sub->getParent()->rend();
+    for (; InsertI != InsertE; ++InsertI) {
+      MachineInstr *Instr = &*InsertI;
+      if (!Instr->readsRegister(X86::EFLAGS, TRI) &&
+          Instr->modifiesRegister(X86::EFLAGS, TRI)) {
+        Sub->getParent()->remove(Movr0Inst);
+        Instr->getParent()->insert(MachineBasicBlock::iterator(Instr),
+                                   Movr0Inst);
+        break;
+      }
+    }
+    if (InsertI == InsertE)
+      return false;
   }
 
   // Make sure Sub instruction defines EFLAGS and mark the def live.
diff --git a/llvm/test/CodeGen/X86/pr16031.ll b/llvm/test/CodeGen/X86/pr16031.ll
new file mode 100644
index 00000000000..4721173cb67
--- /dev/null
+++ b/llvm/test/CodeGen/X86/pr16031.ll
@@ -0,0 +1,27 @@
+; RUN: llc < %s -mtriple=i386-unknown-linux-gnu -mcpu=corei7-avx | FileCheck %s
+
+; CHECK: main:
+; CHECK: pushl %esi
+; CHECK-NEXT: movl  $-12, %eax
+; CHECK-NEXT: movl  $-1, %edx
+; CHECK-NEXT: testb $1, 8(%esp)
+; CHECK-NEXT: cmovel    %edx, %eax
+; CHECK-NEXT: xorl  %ecx, %ecx
+; CHECK-NEXT: movl  %eax, %esi
+; CHECK-NEXT: addl  $-1, %esi
+; CHECK-NEXT: movl  $-1, %esi
+; CHECK-NEXT: adcl  $-1, %esi
+; CHECK-NEXT: cmovsl    %ecx, %eax
+; CHECK-NEXT: cmovsl    %ecx, %edx
+; CHECK-NEXT: popl  %esi
+define i64 @main(i1 %tobool1) nounwind {
+entry:
+  %0 = zext i1 %tobool1 to i32
+  %. = xor i32 %0, 1
+  %.21 = select i1 %tobool1, i32 -12, i32 -1
+  %conv = sext i32 %.21 to i64
+  %1 = add i64 %conv, -1
+  %cmp10 = icmp slt i64 %1, 0
+  %sub17 = select i1 %cmp10, i64 0, i64 %conv
+  ret i64 %sub17
+}