junrushao commented on code in PR #12564:
URL: https://github.com/apache/tvm/pull/12564#discussion_r953186348


##########
python/tvm/tir/schedule/schedule.py:
##########
@@ -2092,9 +2092,9 @@ def before_tensorize(
 
             @T.prim_func
             def mma_desc(a: T.handle, b: T.handle, c: T.handle) -> None:
-                A = T.match_buffer(a, (16, 16), align=128, offset_factor=1)
-                B = T.match_buffer(b, (16, 16), align=128, offset_factor=1)
-                C = T.match_buffer(c, (16, 16), align=128, offset_factor=1)
+                A = T.match_buffer(a, (16, 16), align=64, offset_factor=1)
+                B = T.match_buffer(b, (16, 16), align=64, offset_factor=1)
+                C = T.match_buffer(c, (16, 16), align=64, offset_factor=1)

Review Comment:
   mma does use `align=128`, so let's revert this change here :-)



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to