[Mlir-commits] [mlir] [mlir][Interfaces][NFC] `ValueBoundsConstraintSet`: Add columns for constant values/dims (PR #86097)

[Matthias Springer]

================
@@ -107,25 +107,47 @@ AffineExpr ValueBoundsConstraintSet::getExpr(Value value,
   assertValidValueDim(value, dim);
 #endif // NDEBUG
 
+  // Check if the value/dim is statically known. In that case, an affine
+  // constant expression should be returned. This allows us to support
+  // multiplications with constants. (Multiplications of two columns in the
+  // constraint set is not supported.)
+  std::optional<int64_t> constSize = std::nullopt;
   auto shapedType = dyn_cast<ShapedType>(value.getType());
   if (shapedType) {
-    // Static dimension: return constant directly.
     if (shapedType.hasRank() && !shapedType.isDynamicDim(*dim))
-      return builder.getAffineConstantExpr(shapedType.getDimSize(*dim));
-  } else {
-    // Constant index value: return directly.
-    if (auto constInt = ::getConstantIntValue(value))
-      return builder.getAffineConstantExpr(*constInt);
+      constSize = shapedType.getDimSize(*dim);
----------------
matthias-springer wrote:

> any such analysis is relative to its context

On second thought, that's actually how it is implemented at the moment. We provide a `ValueBoundsConstraintSet::computeBound(Value)` API. That function will build a constraint set and add the given value as a dimension, everything else as a symbol. When computing a bound for a different value, we build a brand new constraint set, this time with the new value being the only dimension. So this analysis is actually relative to a context and rooted at a certain op.

Computing multiple bounds back-to-back is slow at the moment because we have to re-analyze the entire IR. If we improve this at some point in the future, we will have to take a closer look at what's a symbol and what's a dimension, as you suggested.


https://github.com/llvm/llvm-project/pull/86097