comaniac commented on a change in pull request #8618:
URL: https://github.com/apache/tvm/pull/8618#discussion_r681126946
##########
File path: python/tvm/topi/cuda/reduction.py
##########
@@ -114,6 +114,7 @@ def schedule_reduce(outs):
outs = [outs] if isinstance(outs, te.tensor.Tensor) else outs
sch = te.create_schedule([x.op for x in outs])
scheduled_ops = []
+ enable_auto_inline = tvm.te.schedule.EnableAutoInline(sch)
Review comment:
Will this too aggressive for some other workloads? Since
`EnableAutoInline` returns true as long as there's an unscheduled internal
stage with reduce axis, I'm worrying whether it may be too aggressive if a
workload has many stages or has a large stage. In this case, the CUDA kernel
launching may fail in run time due to out of resource.
cc @vinx13 @icemelon @Hzfengsy @junrushao1994
--
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]
