[Mlir-commits] [mlir] [mlir][ArmSME] Add comments in tile-spills-and-fills.mlir (PR #91450)

[Benjamin Maxwell]

================
@@ -92,8 +102,14 @@ func.func @use_too_many_tiles() {
 // AFTER-LLVM-LOWERING-NEXT:   "arm_sme.intr.ld1h.horiz"({{.*}}, %[[SLICE_PTR]], {{.*}}) <{tile_id = 0 : i32}>
 // AFTER-LLVM-LOWERING-NEXT:   vector.store %[[SLICE]], %[[TILE_ALLOCA]]
 // AFTER-LLVM-LOWERING-NEXT: }
-//                           Note: 85 is the mask for the 16-bit tile 0.
+
+///     4. Create/allocate %2
+///        Note: 85 is the mask for the 16-bit tile 0.
+
 //      AFTER-LLVM-LOWERING: "arm_sme.intr.zero"() <{tile_mask = 85 : i32}>
+
+///     5. Re-load %0
----------------
MacDue wrote:

Not a reload this is the same code as before `swap(<inMemoryTile>, tile zero)`. The two swaps cancel out. 

https://github.com/llvm/llvm-project/pull/91450