[Mlir-commits] [mlir] [mlir][Vector] Reject scalable masks in maskedload/maskedstore emulation (PR #223316)

[Cheng-Chieh Liu]

================
@@ -145,3 +145,22 @@ func.func @vector_maskedstore_rank0(%arg0: memref<f32>, %arg3: vector<1xi1>, %ar
   vector.maskedstore %arg0[], %arg3, %arg4 : memref<f32>, vector<1xi1>, vector<1xf32>
   return
 }
+
+// A scalable mask cannot be unrolled at compile time, so these must not be
+// converted.
----------------
jay87406 wrote:

Good point, done — removed the comment and moved the info into the test names (`negative_maskedload_scalable_mask` / `negative_maskedstore_scalable_mask`), and switched to "scalarize" to match the wording already used for `vector.gather`'s similar pattern.

https://github.com/llvm/llvm-project/pull/223316