[PowerPC] fix a bug in redundant compare elimination

This patch fixes a bug in the redundant compare elimination reported in https://reviews.llvm.org/rL320786 and re-enables the optimization.

The redundant compare elimination assumes that we can replace signed comparison with unsigned comparison for the equality check. But due to the difference in the sign extension behavior we cannot change the opcode if the comparison is against an immediate and the most significant bit of the immediate is one.

Differential Revision: https://reviews.llvm.org/D41385

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@321147 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/PowerPC/PPCMIPeephole.cpp b/lib/Target/PowerPC/PPCMIPeephole.cpp
index a2640727f813818c9976450b017de60dbed30bdd..474661aaaee86eeae29b64c395dc6cd6c0aec35b 100644 (file)
--- a/lib/Target/PowerPC/PPCMIPeephole.cpp
+++ b/lib/Target/PowerPC/PPCMIPeephole.cpp
@@ -1025,9 +1025,6 @@ bool PPCMIPeephole::eliminateRedundantTOCSaves(
 //   bge    0, .LBB0_4
 
 bool PPCMIPeephole::eliminateRedundantCompare(void) {
-  // FIXME: this transformation is causing miscompiles. Disabling it for now
-  // until we can resolve the issue.
-  return false;
   bool Simplified = false;
 
   for (MachineBasicBlock &MBB2 : *MF) {
@@ -1087,10 +1084,21 @@ bool PPCMIPeephole::eliminateRedundantCompare(void) {
       // we replace it with a signed comparison if the comparison
       // to be merged is a signed comparison.
       // In other cases of opcode mismatch, we cannot optimize this.
-      if (isEqOrNe(BI2) &&
+
+      // We cannot change opcode when comparing against an immediate
+      // if the most significant bit of the immediate is one
+      // due to the difference in sign extension.
+      auto CmpAgainstImmWithSignBit = [](MachineInstr *I) {
+        if (!I->getOperand(2).isImm())
+          return false;
+        int16_t Imm = (int16_t)I->getOperand(2).getImm();
+        return Imm < 0;
+      };
+
+      if (isEqOrNe(BI2) && !CmpAgainstImmWithSignBit(CMPI2) &&
           CMPI1->getOpcode() == getSignedCmpOpCode(CMPI2->getOpcode()))
         NewOpCode = CMPI1->getOpcode();
-      else if (isEqOrNe(BI1) &&
+      else if (isEqOrNe(BI1) && !CmpAgainstImmWithSignBit(CMPI1) &&
                getSignedCmpOpCode(CMPI1->getOpcode()) == CMPI2->getOpcode())
         NewOpCode = CMPI2->getOpcode();
       else continue;

diff --git a/test/CodeGen/PowerPC/cmp_elimination.ll b/test/CodeGen/PowerPC/cmp_elimination.ll
index 3251ae2881b93681ba0728b3fb74d0f9f42726ea..6bc8b8a041c2da8026264e71b4231dfd30ee9e81 100644 (file)
--- a/test/CodeGen/PowerPC/cmp_elimination.ll
+++ b/test/CodeGen/PowerPC/cmp_elimination.ll
@@ -1,4 +1,3 @@
-; XFAIL: *
 ; RUN: llc -verify-machineinstrs < %s -mtriple=powerpc64-unknown-linux-gnu | FileCheck %s
 ; RUN: llc -verify-machineinstrs < %s -mtriple=powerpc64le-unknown-linux-gnu | FileCheck %s
 
@@ -748,6 +747,37 @@ do.end:
   ret void
 }
 
+define void @func29(i32 signext %a) {
+; We cannot merge two compares due to difference in sign extension behaviors.
+; equivalent C code example:
+;   int a = .. ;
+;   if (a == -1) dummy1();
+;   if (a == (uint16_t)-1) dummy2();
+
+; CHECK-LABEL: @func29
+; CHECK: cmp
+; CHECK: cmp
+; CHECK: blr
+entry:
+  %cmp = icmp eq i32 %a, -1
+  br i1 %cmp, label %if.then, label %if.else
+
+if.then:
+  tail call void @dummy1()
+  br label %if.end3
+
+if.else:
+  %cmp1 = icmp eq i32 %a, 65535
+  br i1 %cmp1, label %if.then2, label %if.end3
+
+if.then2:
+  tail call void @dummy2()
+  br label %if.end3
+
+if.end3:
+  ret void
+}
+
 declare void @dummy1()
 declare void @dummy2()
 declare void @dummy3()