[Mlir-commits] [mlir] [mlir][ArmSME] Fold transpose into xfer read to enable in-flight transpose (PR #92562)

[Cullen Rhodes]

================
@@ -162,6 +162,21 @@ func.func @fold_transpose_into_load(%src : memref<?x?xf32>) {
 
 // -----
 
+// CHECK-LABEL: @fold_transpose_into_load_multi_use
+// CHECK: arm_sme.tile_load {{.*}} : memref<?x?xf32>, vector<[4]x[4]xf32>
+// CHECK: arm_sme.tile_store {{.*}} : memref<?x?xf32>, vector<[4]x[4]xf32>
+// CHECK: arm_sme.tile_load {{.*}} layout<vertical> : memref<?x?xf32>, vector<[4]x[4]xf32>
----------------
c-rhodes wrote:

> Would you mind adding more Ops? I'm actually curious to see the final "prevent.dce" Op.

👍 

> Also, somewhat tangential, we should probably switch to `"test.some_use"`.


I agree, still a few tests where we're using `prevent.dce` that need updating. I added the single `test.some_use` here as `prevent.dce` is misleading.

https://github.com/llvm/llvm-project/pull/92562