[mlir][Interfaces] ValueBoundsOpInterface: Fix typo

[matthias-springer]

This was likely a copy-and-paste typo.

[llvmbot]

@llvm/pr-subscribers-mlir

Author: Matthias Springer (matthias-springer)

Changes

This was likely a copy-and-paste typo.

---

Full diff: https://github.com/llvm/llvm-project/pull/87976.diff

1 Files Affected:

• (modified) mlir/lib/Interfaces/ValueBoundsOpInterface.cpp (+1-1)

diff --git a/mlir/lib/Interfaces/ValueBoundsOpInterface.cpp b/mlir/lib/Interfaces/ValueBoundsOpInterface.cpp
index 1f3f1bb863c554..05b65f7460edcc 100644
--- a/mlir/lib/Interfaces/ValueBoundsOpInterface.cpp
+++ b/mlir/lib/Interfaces/ValueBoundsOpInterface.cpp
@@ -205,7 +205,7 @@ int64_t ValueBoundsConstraintSet::insert(bool isSymbol) {
 int64_t ValueBoundsConstraintSet::insert(AffineMap map, ValueDimList operands,
                                          bool isSymbol) {
   assert(map.getNumResults() == 1 && "expected affine map with one result");
-  int64_t pos = insert(/*isSymbol=*/false);
+  int64_t pos = insert(isSymbol);
 
   // Add map and operands to the constraint set. Dimensions are converted to
   // symbols. All operands are added to the worklist (unless they were already

[joker-eph]

Can you provide a test that exercises this? Thanks!

[matthias-springer]

It is currently not possible to trigger this bug. New columns are always added to the constraint set as symbols. Only the first added SSA value (for which we compute a bound) is added as a dimension. There is currently no API to compute a bound for an "AffineMap + operands" combination. (But in anticipation of adding such an API in the future and for consistency reasons, I had already added an isSymbol argument to all ValueBoundsConstraintSet::insert overloads.) I have a PR in preparation that extends the API and will require this fix here.