[AutoScheduler] Register workload when deserializing tasks #6927
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
comaniac
force-pushed
the
ansor_fix_deserial
branch
from
59b7525 to
2aabc3b
Compare
merrymercy
requested changes
Nov 18, 2020
| func_name = json.loads(self.workload_key)[0] | ||
| except Exception: # pylint: disable=broad-except | ||
| raise RuntimeError("Invalid workload key %s" % self.workload_key) | ||
| register_workload_tensors(func_name, self.dag.tensors) |
There was a problem hiding this comment.
We should do more checks here.
There are two kinds of workloads.
You can find the note here
https://github.com/apache/incubator-tvm/blob/51d81fb71781eb606f4b563ec28290dbd8d04faf/python/tvm/auto_scheduler/workload_registry.py#L42-L52
We have to handle two cases correctly without incorrect overwriting
There was a problem hiding this comment.
Changed the logic accordingly. Now we only register workloads extracted from Relay programs.
comaniac
force-pushed
the
ansor_fix_deserial
branch
from
efe9ef6 to
ca1945e
Compare
merrymercy
approved these changes
Nov 20, 2020
trevor-m
pushed a commit
to trevor-m/tvm
that referenced
this pull request
Dec 2, 2020
* [AutoScheduler] Register workload when deserializing tasks * fix name * format * merge * fix test * more checks
trevor-m
pushed a commit
to trevor-m/tvm
that referenced
this pull request
Dec 4, 2020
* [AutoScheduler] Register workload when deserializing tasks * fix name * format * merge * fix test * more checks
trevor-m
pushed a commit
to neo-ai/tvm
that referenced
this pull request
Dec 4, 2020
* [AutoScheduler] Register workload when deserializing tasks * fix name * format * merge * fix test * more checks
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
One issue of the current auto_scheduler task serialization is that the workload won't be registered when deserializing a task so users have to manually call
register_workload_tensorsafter deserialization. However,register_workload_tensorsalways uses compute DAG hash as the workload key. This could cause workload key mismatching if the task was from a TE function that uses function name as the workload key. This PR makes the following changes to make sureWORKLOAD_FUNC_REGISTRYis consist after deserializing a task.register_workload_tensorsby moving the ComputeDAG out.cc @merrymercy @jcf94