[Mlir-commits] [mlir] [mlir][vector] Teach `TransferOptimization` to forward masked stores (PR #87794)

[Andrzej Warzyński]

================
@@ -360,3 +360,51 @@ func.func @forward_dead_store_dynamic_non_overlap_trailing_dim(
   vector.transfer_write %x, %buffer[%i0, %i0] {in_bounds = [true]} : vector<4xf32>, memref<?x?xf32>
   return
 }
+
+// CHECK-LABEL: func @forward_dead_constant_splat_store_with_masking
+//   CHECK-NOT:   vector.transfer_write
+//   CHECK-NOT:   vector.transfer_read
+//       CHECK:   scf.for
+//       CHECK:   }
+//       CHECK:   vector.transfer_write
+//       CHECK:   return
----------------
banach-space wrote:

I think that it would be good to see how ` %cst = arith.constant dense<0.000000e+00> : vector<[8]x[8]xf32>` is used after the transformation.

https://github.com/llvm/llvm-project/pull/87794