[Mlir-commits] [mlir] [mlir] Extend affine.min/max ValueBoundsOpInterfaceImpls (PR #118840)

[Matthias Springer]

================
@@ -67,6 +67,27 @@ struct AffineMinOpInterface
           expr.replaceDimsAndSymbols(dimReplacements, symReplacements);
       cstr.bound(value) <= bound;
     }
+    // Get all constant lower bounds, choose minimum, and set lower bound to it.
+    MLIRContext *ctx = op->getContext();
+    AffineMap map = minOp.getAffineMap();
+    SmallVector<Value> mapOperands = minOp.getOperands();
+    std::optional<int64_t> minBound;
+    for (AffineExpr expr : map.getResults()) {
+      auto exprMap =
+          AffineMap::get(map.getNumDims(), map.getNumSymbols(), expr, ctx);
+      ValueBoundsConstraintSet::Variable exprVar(exprMap, mapOperands);
+      FailureOr<int64_t> exprBound =
+          cstr.computeConstantBound(presburger::BoundType::LB, exprVar,
----------------
matthias-springer wrote:

This is a bit misleading: You're actually calling a static function `ValueBoundsConstraintSet::computeConstantBound` here. That means building a brand new constraint set, which can be expensive because the same IR may be re-analyzed.

Can you try using `cstr.populateAndCompare` instead? And keep track of which AffineExpr result is the current minimum one. Keep comparing against that one to find the overall smallest one. That should also work with non-constant bounds then.


https://github.com/llvm/llvm-project/pull/118840