[Mlir-commits] [mlir] [mlir][IntRange] Poison support in int-range analysis (PR #152932)

Ivan Butygin llvmlistbot at llvm.org
Sat Aug 16 04:25:12 PDT 2025


================
@@ -306,7 +329,8 @@ static ConstantIntRanges inferDivURange(const ConstantIntRanges &lhs,
 
   // X u/ Y u<= X.
   APInt umax = lhsMax;
-  return ConstantIntRanges::fromUnsigned(umin, umax);
+  return propagatePoison(ConstantIntRanges::fromUnsigned(umin, umax),
----------------
Hardcode84 wrote:

Yeah, it's annoying and suboptimal (we should check for the poison args before trying to compute the ranges), but doing it this way was the least amount of code churn.

https://github.com/llvm/llvm-project/pull/152932


More information about the Mlir-commits mailing list