[Mlir-commits] [mlir] [mlir][ArmSME] Use liveness information in the tile allocator (PR #90448)

[Andrzej Warzyński]

================
@@ -126,26 +184,46 @@ func.func @constant_loop_init_with_multiple_users(%a: vector<[4]xf32>, %b: vecto
 
 // -----
 
-// Incorrect result! Everything assigned to tile 0 (which means values that are still live are overwritten).
-//
-// CHECK-BAD-LABEL: @run_out_of_tiles_but_avoid_spill
-// CHECK-BAD: arm_sme.zero {tile_id = 0 : i32}
-// CHECK-BAD-COUNT-4: arm_sme.move_vector_to_tile_slice {{.*}} {tile_id = 0 : i32} : vector<[4]xf32> into vector<[4]x[4]xf32>
+//  CHECK-LIVE-RANGE-LABEL: @run_out_of_tiles_but_avoid_spill
+//        CHECK-LIVE-RANGE: ========== Coalesced Live Ranges:
+//        CHECK-LIVE-RANGE: ^bb2:
+//   CHECK-LIVE-RANGE-NEXT: |S    arm_sme.copy_tile
+//   CHECK-LIVE-RANGE-NEXT: ||S   arm_sme.copy_tile
+//   CHECK-LIVE-RANGE-NEXT: |||S  arm_sme.copy_tile
+//   CHECK-LIVE-RANGE-NEXT: ||||S arm_sme.copy_tile
+//   CHECK-LIVE-RANGE-NEXT: EEEEE cf.br
+
+// Note in the live ranges (above) there is five tile values, but we only have four tiles.
----------------
banach-space wrote:

> // Note in the live ranges (above) there is five tile values, but we only have four tiles.

Indeed, so how do we manage to avoid spilling?

https://github.com/llvm/llvm-project/pull/90448