[Mlir-commits] [mlir] [mlir][ArmSME] Switch to an attribute-based tile allocation scheme (PR #73253)

[Cullen Rhodes]

================
@@ -506,6 +496,7 @@ struct TileVectorPrintOpConversion : public OpRewritePattern<vector::PrintOp> {
       rewriter.setInsertionPointToStart(forOp.getBody());
       // Extract the current row from the tile.
       Value rowIndex = forOp.getInductionVar();
+      // FIXME: Forward tile IDs.
----------------
c-rhodes wrote:

> I think the correct fix here is just to more the vector.print lowering into `-convert-vector-to-arm-sme` as it's a vector op lowering, not an ArmSME op lowering (no extra op required).

Ok, I think that's fine for now. As you mentioned, this doesn't have to be part of this patch. We may also have to revisit loop materialization in `-convert-vector-to-arm-sme`. When loads/stores were first implemented loop materialization was done during lowering to intrinsics, `-convert-arm-sme-to-scf` was introduced to break that up such that the intrinsic lowerings were 1-1. It's not quite the same when lowering Vector ops to ArmSME, but there may be value in getting the granularity right, that could necessitate more ops in ArmSME however, so some cost there at least.

https://github.com/llvm/llvm-project/pull/73253