Skip to content
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

Developer documentation for InferBound pass. #3040

Merged
merged 1 commit into from Apr 28, 2019

Conversation

Projects
None yet
4 participants
@jdavies-huawei
Copy link
Contributor

commented Apr 17, 2019

An earlier draft of this documentation was discussed here https://discuss.tvm.ai/t/discuss-contributing-new-docs-for-inferbound/2151

@tqchen @jroesch please review, or please suggest reviewers.

@tqchen

This comment has been minimized.

Copy link
Member

commented Apr 17, 2019

We cannot check-in binary images into the tvm repo, please put that in a separate repo https://github.com/tvmai/tvmai.github.io

@jdavies-huawei

This comment has been minimized.

Copy link
Contributor Author

commented Apr 18, 2019

No problem. I created a pull request for the images: tvmai/tvmai.github.io#39

I'll need to update the URLs in the docs, once the images are merged.

@jdavies-huawei jdavies-huawei force-pushed the jdavies-huawei:tvm_docs branch from b2d3d00 to c1e46d2 Apr 18, 2019

@jdavies-huawei

This comment has been minimized.

Copy link
Contributor Author

commented Apr 22, 2019

Images have been moved under images/docs/inferbound.

@jdavies-huawei jdavies-huawei force-pushed the jdavies-huawei:tvm_docs branch from c1e46d2 to 3e2e0e2 Apr 23, 2019

@jdavies-huawei

This comment has been minimized.

Copy link
Contributor Author

commented Apr 23, 2019

I updated the URLs to their final versions. So from my perspective this is ready to merge now.


In InferRootBound, the goal is to determine Ranges for the root_iter_vars of a particular stage, C. Phases 1-2 of InferRootBound assign IntSets to the leaf IterVars of C's consumers, and then propagate those IntSets up to the consumers' root_iter_vars.

If there are no attachments, the Ranges already computed for the consumer's variables define how much of C is needed by the consumer. However, if the stage is actually inside the scope of one of the consumer's variables j, then only a single point from within the Range of j is needed at a time.

This comment has been minimized.

Copy link
@merrymercy

merrymercy Apr 25, 2019

Member
Suggested change
If there are no attachments, the Ranges already computed for the consumer's variables define how much of C is needed by the consumer. However, if the stage is actually inside the scope of one of the consumer's variables j, then only a single point from within the Range of j is needed at a time.
If there are no attachments, the Ranges already computed for the consumer's variables define how much of C is needed by the consumer. However, if the stage is actually inside the scope of one of the consumer's variables j, then only a single point within the Range of j is needed at a time.

This comment has been minimized.

Copy link
@jdavies-huawei

jdavies-huawei Apr 26, 2019

Author Contributor

I removed all uses of "from within" from the document.

@@ -241,3 +241,759 @@ The ``PackedFunc``'s overloaded ``operator()`` will be called, which in turn cal
};

This concludes an overview of how TVM compiles and executes a function. Although we did not detail TOPI or Relay, at the end all neural network operators go through the same compilation process as above. You are encouraged to dive into the details of the rest of the codebase.

*******************************************

This comment has been minimized.

Copy link
@merrymercy

merrymercy Apr 25, 2019

Member

We'd better move this doc into a single file because it does not fit into the "general codebase walkthrough by examples"

This comment has been minimized.

Copy link
@jdavies-huawei

jdavies-huawei Apr 26, 2019

Author Contributor

Ok, I moved this InferBound documentation to its own file. I also added a link from the codebase walkthrough, where it talks about bound inference.

@jdavies-huawei jdavies-huawei force-pushed the jdavies-huawei:tvm_docs branch from 3e2e0e2 to 7c38960 Apr 26, 2019

@merrymercy merrymercy merged commit 6636c42 into dmlc:master Apr 28, 2019

1 check passed

continuous-integration/jenkins/pr-merge This commit looks good
Details
@vinx13

This comment has been minimized.

Copy link
Member

commented Apr 28, 2019

@jdavies-huawei image files are still not removed from this pr?

@tqchen

This comment has been minimized.

Copy link
Member

commented Apr 28, 2019

I removed the commit from the master. please send another PR excluding images

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
You can’t perform that action at this time.