fix: fix gf1250 naming style (#787)

Author: jli-melchior

python/flydsl/expr/rocdl/cluster.py

@@ -73,7 +73,7 @@ def compute_mcast_masks(local_x, local_y, cluster_m: int, cluster_n: int):
     """Compute MCAST workgroup_mask values for A and B matrices.
 
     Hardware flat WG index within a cluster uses X-inner ordering
-    (MI400 Shader Programming, TTMP6 layout, section 3.5.5.1):
+    (gfx1250 Shader Programming, TTMP6 layout, section 3.5.5.1):
 
         flat_wg_id = wg_x + wg_y * nwg_x = local_x + local_y * cluster_m
 

tests/perf/bench_tdm_bandwidth_gfx1250.py

@@ -486,7 +486,7 @@ def main():
         print(f"ERROR: TDM benchmark requires gfx1250, got {arch}", file=sys.stderr)
         sys.exit(1)
 
-    # MI450: 8 XCDs × 256 CUs = 2048 CUs total
+    # gfx1250: 8 XCDs × 256 CUs = 2048 CUs total
     grid_sizes = [(4, 4), (8, 8), (16, 16), (32, 32), (32, 64), (64, 64), (64, 128), (128, 128), (256, 128), (256, 256)]
     # Multicast mode: square grids only (fewer combos, cleaner comparison)
     mcast_grid_sizes = [(8, 8), (16, 16), (32, 32), (64, 64), (128, 128)]