[X86] Don't invert NewCC variable while processing the jcc/setcc/cmovcc instructions...

The NewCC variable is calculated outside of the loop that processes jcc/setcc/cmovcc instructions. If we invert it during the loop it can cause an incorrect value to be used by a later iteration. Instead only read it during the loop and use a new variable to store the possibly inverted value.

Fixes PR35399.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@318934 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86InstrInfo.cpp b/lib/Target/X86/X86InstrInfo.cpp
index 68f964cc8669b8ca371e03155488957265da765d..bd8d447fb883e7b3eca26327c90d8fb00b4fb96d 100644 (file)
--- a/lib/Target/X86/X86InstrInfo.cpp
+++ b/lib/Target/X86/X86InstrInfo.cpp
@@ -7475,6 +7475,7 @@ bool X86InstrInfo::optimizeCompareInstr(MachineInstr &CmpInstr, unsigned SrcReg,
       }
       if (OldCC == X86::COND_INVALID) return false;
     }
+    X86::CondCode ReplacementCC = X86::COND_INVALID;
     if (IsCmpZero) {
       switch (OldCC) {
       default: break;
@@ -7494,31 +7495,32 @@ bool X86InstrInfo::optimizeCompareInstr(MachineInstr &CmpInstr, unsigned SrcReg,
         default:
           return false;
         case X86::COND_E:
+          ReplacementCC = NewCC;
           break;
         case X86::COND_NE:
-          NewCC = GetOppositeBranchCondition(NewCC);
+          ReplacementCC = GetOppositeBranchCondition(NewCC);
           break;
         }
     } else if (IsSwapped) {
       // If we have SUB(r1, r2) and CMP(r2, r1), the condition code needs
       // to be changed from r2 > r1 to r1 < r2, from r2 < r1 to r1 > r2, etc.
       // We swap the condition code and synthesize the new opcode.
-      NewCC = getSwappedCondition(OldCC);
-      if (NewCC == X86::COND_INVALID) return false;
+      ReplacementCC = getSwappedCondition(OldCC);
+      if (ReplacementCC == X86::COND_INVALID) return false;
     }
 
-    if ((ShouldUpdateCC || IsSwapped) && NewCC != OldCC) {
+    if ((ShouldUpdateCC || IsSwapped) && ReplacementCC != OldCC) {
       // Synthesize the new opcode.
       bool HasMemoryOperand = Instr.hasOneMemOperand();
       unsigned NewOpc;
       if (Instr.isBranch())
-        NewOpc = GetCondBranchFromCond(NewCC);
+        NewOpc = GetCondBranchFromCond(ReplacementCC);
       else if(OpcIsSET)
-        NewOpc = getSETFromCond(NewCC, HasMemoryOperand);
+        NewOpc = getSETFromCond(ReplacementCC, HasMemoryOperand);
       else {
         unsigned DstReg = Instr.getOperand(0).getReg();
         const TargetRegisterClass *DstRC = MRI->getRegClass(DstReg);
-        NewOpc = getCMovFromCond(NewCC, TRI->getRegSizeInBits(*DstRC)/8,
+        NewOpc = getCMovFromCond(ReplacementCC, TRI->getRegSizeInBits(*DstRC)/8,
                                  HasMemoryOperand);
       }
 

diff --git a/test/CodeGen/X86/pr35399.ll b/test/CodeGen/X86/pr35399.ll
new file mode 100644 (file)
index 0000000..394c257
--- /dev/null
+++ b/test/CodeGen/X86/pr35399.ll
@@ -0,0 +1,22 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=x86_64-unknown -mattr=lzcnt | FileCheck %s
+
+; Make sure we emit opoosite setcc instructions.
+define i64 @pr35399(i64, i8*, i8*) {
+; CHECK-LABEL: pr35399:
+; CHECK:       # BB#0:
+; CHECK-NEXT:    lzcntq %rdi, %rax
+; CHECK-NEXT:    setae (%rsi)
+; CHECK-NEXT:    setb (%rdx)
+; CHECK-NEXT:    retq
+  %4 = tail call i64 @llvm.ctlz.i64(i64 %0, i1 false)
+  %5 = icmp ne i64 %0, 0
+  %6 = zext i1 %5 to i8
+  store i8 %6, i8* %1, align 1
+  %7 = xor i1 %5, true
+  %8 = zext i1 %7 to i8
+  store i8 %8, i8* %2, align 1
+  ret i64 %4
+}
+
+declare i64 @llvm.ctlz.i64(i64, i1)