[PATCH] D156467: [mlir][ArmSME] Add conversion from ArmSME to SCF to materialize loops

[Cullen Rhodes via Phabricator via llvm-commits]

c-rhodes marked an inline comment as done.
c-rhodes added inline comments.


================
Comment at: mlir/lib/Conversion/ArmSMEToSCF/ArmSMEToSCF.cpp:101-102
+    // loop to simplify the canonicalizations.
+    rewriter.replaceOpWithNewOp<arm_sme::CastTileToVector>(tileLoadOp, tileType,
+                                                           tileId);
+
----------------
awarzynski wrote:
> Is this 2nd cast really needed? Wouldn't this work:
> ```
> rewriter.replaceOp(tileLoadOp, tileInit);
> ```
> ? I'm probably missing something obvious 🤔 .
> Is this 2nd cast really needed? Wouldn't this work:
> ```
> rewriter.replaceOp(tileLoadOp, tileInit);
> ```
> ? I'm probably missing something obvious 🤔 .

You're right this isn't necessary, memory semantics prevent these from being reordered. I've removed it.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D156467/new/

https://reviews.llvm.org/D156467