[llvm] [InstCombine] Avoid infinite loop in `foldSelectValueEquivalence` (PR #142754)

[via llvm-commits]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-llvm-transforms

Author: Yingwei Zheng (dtcxzyw)

<details>
<summary>Changes</summary>

Before this patch, InstCombine hung because it replaced a value with a more complex one:
```
%sel = select i1 %cmp, i32 %smax, i32 0 ->
%sel = select i1 %cmp, i32 %masked, i32 0 ->
%sel = select i1 %cmp, i32 %smax, i32 0 ->
...
```
This patch makes this replacement more conservative. It only performs the replacement iff the new value is one of the operands of the original value.

Closes https://github.com/llvm/llvm-project/issues/142405.


---
Full diff: https://github.com/llvm/llvm-project/pull/142754.diff


2 Files Affected:

- (modified) llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp (+5-1) 
- (modified) llvm/test/Transforms/InstCombine/select-cmp-eq-op-fold.ll (+20) 


``````````diff
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp b/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp
index d7d0431a5b8d0..257dc943bd5c1 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp
@@ -1312,7 +1312,11 @@ Instruction *InstCombinerImpl::foldSelectValueEquivalence(SelectInst &Sel,
 
       // If NewOp is a constant and OldOp is not replace iff NewOp doesn't
       // contain and undef elements.
-      if (match(NewOp, m_ImmConstant()) || NewOp == V) {
+      // Make sure that V is always simpler than TrueVal, otherwise we might
+      // end up in an infinite loop.
+      if (match(NewOp, m_ImmConstant()) ||
+          (isa<Instruction>(TrueVal) &&
+           is_contained(cast<Instruction>(TrueVal)->operands(), V))) {
         if (isGuaranteedNotToBeUndef(NewOp, SQ.AC, &Sel, &DT))
           return replaceOperand(Sel, Swapped ? 2 : 1, V);
         return nullptr;
diff --git a/llvm/test/Transforms/InstCombine/select-cmp-eq-op-fold.ll b/llvm/test/Transforms/InstCombine/select-cmp-eq-op-fold.ll
index 760b1ef055ad3..2e6bbdf5da74d 100644
--- a/llvm/test/Transforms/InstCombine/select-cmp-eq-op-fold.ll
+++ b/llvm/test/Transforms/InstCombine/select-cmp-eq-op-fold.ll
@@ -182,3 +182,23 @@ define i8 @replace_with_y_for_simple_binop_fail(i8 %x, i8 noundef %y, i8 %z, i8
   %sel = select i1 %cmp, i8 %mul, i8 %z
   ret i8 %sel
 }
+
+; Make sure we don't run into an infinite loop.
+define i32 @pr142405(i32 noundef %x) {
+; CHECK-LABEL: @pr142405(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[SMAX:%.*]] = call i32 @llvm.smax.i32(i32 [[X:%.*]], i32 0)
+; CHECK-NEXT:    [[MASKED:%.*]] = and i32 [[SMAX]], 65535
+; CHECK-NEXT:    [[CMP:%.*]] = icmp eq i32 [[X]], [[MASKED]]
+; CHECK-NEXT:    [[TMP0:%.*]] = and i32 [[SMAX]], 1
+; CHECK-NEXT:    [[AND:%.*]] = select i1 [[CMP]], i32 [[TMP0]], i32 0
+; CHECK-NEXT:    ret i32 [[AND]]
+;
+entry:
+  %smax = call i32 @llvm.smax.i32(i32 %x, i32 0)
+  %masked = and i32 %smax, 65535
+  %cmp = icmp eq i32 %x, %masked
+  %sel = select i1 %cmp, i32 %smax, i32 0
+  %and = and i32 %sel, 1
+  ret i32 %and
+}

``````````

</details>


https://github.com/llvm/llvm-project/pull/142754