-
Notifications
You must be signed in to change notification settings - Fork 3.5k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[AutoScheduler] Add task scheduler #6663
Conversation
570e525
to
4cccb5d
Compare
050e351
to
cfd4b27
Compare
cfd4b27
to
316b31e
Compare
if cost < self.best_costs[task_idx]: | ||
self.best_costs[task_idx] = cost | ||
|
||
if len(measure_inputs) == 0: |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
if len(measure_inputs) == 0: | |
if not measure_inputs: |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
measure_inputs
is not a python list. It is a tvm::runtime::Array
, so we cannot use this idiom.
random_states = &temp_random_states; | ||
} else { | ||
random_states->clear(); | ||
std::pair<Array<MeasureInput>, Array<MeasureResult>> SketchPolicyNode::ContinueSearchOneRound( |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
From the functionality it doesn't look like "continue" search one round, but more like SearchOneRound
. On the other hand, the current search one round it is calling is more like TuneStates
.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I still prefer the current names.
a0cb698
to
364a7f1
Compare
a341e4b
to
c5b7842
Compare
c5b7842
to
d43e22c
Compare
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM. Thanks.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks!
* Add task scheduler * fix lint * fix tests * fix tests * fix tests * fix test cases * fix test cases * fix tests * address comments
* Add task scheduler * fix lint * fix tests * fix tests * fix tests * fix test cases * fix test cases * fix tests * address comments
* Add task scheduler * fix lint * fix tests * fix tests * fix tests * fix test cases * fix test cases * fix tests * address comments
* Add task scheduler * fix lint * fix tests * fix tests * fix tests * fix test cases * fix test cases * fix tests * address comments
python/tvm/auto_scheduler/task_scheduler.py
) to allocate time resources for optimizing all subgraphs in a DNN. It prioritizes important subgraphs (e.g. performance bottleneck, graphs with a large room for improvement) according to a given objective function.ContinueSearchOneRound
to the search policy to support the task scheduler.