[FlattenCFG] Fix the miscompilation where phi nodes exist in the merge point

[dtcxzyw]

When there are phi nodes in the merge point of the if-region, we cannot do the merge.
Alive2: https://alive2.llvm.org/ce/z/DbgEan
Fixes #70900.

[llvmbot]

@llvm/pr-subscribers-llvm-transforms

Author: Yingwei Zheng (dtcxzyw)

Changes

When there are phi nodes in the merge point of the if-region, we cannot do the merge.
Alive2: https://alive2.llvm.org/ce/z/DbgEan
Fixes #70900.

---

Full diff: https://github.com/llvm/llvm-project/pull/81987.diff

2 Files Affected:

• (modified) llvm/lib/Transforms/Utils/FlattenCFG.cpp (+4)
• (modified) llvm/test/Transforms/Util/flatten-cfg.ll (+9-6)

diff --git a/llvm/lib/Transforms/Utils/FlattenCFG.cpp b/llvm/lib/Transforms/Utils/FlattenCFG.cpp
index db6ee39baece7b..5a444c503fc4e9 100644
--- a/llvm/lib/Transforms/Utils/FlattenCFG.cpp
+++ b/llvm/lib/Transforms/Utils/FlattenCFG.cpp
@@ -407,6 +407,10 @@ bool FlattenCFGOpt::CompareIfRegionBlock(BasicBlock *Block1, BasicBlock *Block2,
 /// form, by inverting the condition and the branch successors. The same
 /// approach goes for the opposite case.
 bool FlattenCFGOpt::MergeIfRegion(BasicBlock *BB, IRBuilder<> &Builder) {
+  // We cannot merge the if-region if the merge point has phi nodes.
+  if (isa<PHINode>(BB->front()))
+    return false;
+
   BasicBlock *IfTrue2, *IfFalse2;
   BranchInst *DomBI2 = GetIfCondition(BB, IfTrue2, IfFalse2);
   if (!DomBI2)
diff --git a/llvm/test/Transforms/Util/flatten-cfg.ll b/llvm/test/Transforms/Util/flatten-cfg.ll
index d6e34bda1cc60f..038dcaa47419ad 100644
--- a/llvm/test/Transforms/Util/flatten-cfg.ll
+++ b/llvm/test/Transforms/Util/flatten-cfg.ll
@@ -77,13 +77,16 @@ define void @test_not_crash3(i32 %a) #0 {
 ; CHECK-SAME: (i32 [[A:%.*]]) {
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[A_EQ_0:%.*]] = icmp eq i32 [[A]], 0
+; CHECK-NEXT:    br i1 [[A_EQ_0]], label [[BB0:%.*]], label [[BB1:%.*]]
+; CHECK:       bb0:
+; CHECK-NEXT:    br label [[BB1]]
+; CHECK:       bb1:
 ; CHECK-NEXT:    [[A_EQ_1:%.*]] = icmp eq i32 [[A]], 1
-; CHECK-NEXT:    [[TMP0:%.*]] = or i1 [[A_EQ_0]], [[A_EQ_1]]
-; CHECK-NEXT:    br i1 [[TMP0]], label [[BB2:%.*]], label [[BB3:%.*]]
+; CHECK-NEXT:    br i1 [[A_EQ_1]], label [[BB2:%.*]], label [[BB3:%.*]]
 ; CHECK:       bb2:
 ; CHECK-NEXT:    br label [[BB3]]
 ; CHECK:       bb3:
-; CHECK-NEXT:    [[CHECK_BADREF:%.*]] = phi i32 [ 17, [[ENTRY:%.*]] ], [ 11, [[BB2]] ]
+; CHECK-NEXT:    [[CHECK_BADREF:%.*]] = phi i32 [ 17, [[BB1]] ], [ 11, [[BB2]] ]
 ; CHECK-NEXT:    ret void
 ;
 entry:
@@ -278,9 +281,9 @@ define i1 @test_cond_multi_use(i32 %x, i32 %y, i32 %z) {
 ; CHECK-NEXT:  entry.x:
 ; CHECK-NEXT:    [[CMP_X:%.*]] = icmp ne i32 [[X]], 0
 ; CHECK-NEXT:    [[CMP_Y:%.*]] = icmp eq i32 [[Y]], 0
-; CHECK-NEXT:    [[TMP0:%.*]] = xor i1 [[CMP_Y]], true
-; CHECK-NEXT:    [[TMP1:%.*]] = or i1 [[CMP_X]], [[TMP0]]
-; CHECK-NEXT:    br i1 [[TMP1]], label [[IF_THEN_Y:%.*]], label [[EXIT:%.*]]
+; CHECK-NEXT:    [[CMP_Y_NOT:%.*]] = xor i1 [[CMP_Y]], true
+; CHECK-NEXT:    [[TMP0:%.*]] = or i1 [[CMP_X]], [[CMP_Y_NOT]]
+; CHECK-NEXT:    br i1 [[TMP0]], label [[IF_THEN_Y:%.*]], label [[EXIT:%.*]]
 ; CHECK:       if.then.y:
 ; CHECK-NEXT:    store i32 [[Z]], ptr @g, align 4
 ; CHECK-NEXT:    br label [[EXIT]]

[nikic]

Can the code that was added by d98cb81 be dropped now?

[dtcxzyw]

Can the code that was added by d98cb81 be dropped now?

Yeah, the original patch(https://reviews.llvm.org/D68032) just fixed the crash in an incorrect way.

[nikic]

LGTM

[dtcxzyw]

@arsenm Any comments?