[Mlir-commits] [mlir] [MLIR][SideEffects][MemoryEffects] Modified LICM to be more aggressive when checking movability of ops with MemWrite effects (PR #155344)

[Mo Bagherbeik]

================
@@ -1437,3 +1437,283 @@ func.func @do_not_hoist_vector_transfer_ops_memref(
   }
   func.return %final : vector<4x4xf32>
 }
+
----------------
mbagherbeikTT wrote:

I looked into the affine-LICM and currently the passes aren't quite the same, as the affine version doesn't check for speculability and most of the affine ops don't have the speculation interface which LICM requires. The ops require some modification to use this pass; it may be as simple as adding "AlwaysSpeculatable" to relevant ops but I can't speak to that with confidence.

https://github.com/llvm/llvm-project/pull/155344