fix scale only lop3 tensorize instructions.

microsoft · Aug 19, 2024 · 67ad761 · 67ad761
1 parent 95c93b7
commit 67ad761
Showing 1 changed file with 2 additions and 2 deletions.
diff --git a/bitblas/gpu/intrin/lop3.py b/bitblas/gpu/intrin/lop3.py
@@ -1127,7 +1127,7 @@ def fast_decode_desc(compressed: T.handle, decompressed: T.handle, scale: T.hand
                     1,
                 ],
                 dtype=target_dtype,
-                scope="global",
+                scope="local",
             )
             with T.block("root"):
                 T.reads(Compressed[0:n_storage_elems], Scale[0:1])
@@ -1173,7 +1173,7 @@ def fast_decode_impl(compressed: T.handle, decompressed: T.handle, scale: T.hand
                 dtype=target_dtype,
                 offset_factor=1,
                 strides=[s0],
-                scope="global",
+                scope="local",
             )
             with T.block("root"):
                 T.reads(Compressed[0:n_storage_elems], Scale[0:1])