[GitHub] [tvm] AndrewZhaoLuo commented on a diff in pull request #14167: [DRAFT][CUDA][Schedule] Better Layout Transform Schedules

["AndrewZhaoLuo (via GitHub)" <gi...@apache.org>]

AndrewZhaoLuo commented on code in PR #14167:
URL: https://github.com/apache/tvm/pull/14167#discussion_r1122796544


##########
python/tvm/relay/op/_transform.py:
##########
@@ -94,7 +94,7 @@ def compute_strided_set(attrs, inputs, output_type):
 _reg.register_injective_schedule("strided_set")
 
 # layout_transform
-_reg.register_injective_schedule("layout_transform")
+_reg.register_strategy("layout_transform", strategy.layout_transform_strategy)

Review Comment:
   Hey @elvin-n. The plan is to make it so all implementations will have an injective like schedule.
   
   For non-cuda targets it will still use an injective schedule via the generic fallback strategy. 
   
   For cuda it will use the new schedule in the PR after using the autoinline injective schedule (I still need to figure out this part for cuda).



-- 
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: commits-unsubscribe@tvm.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org