[pytorch] Buffer in Pickler to improve performance. #28043
Closed
Conversation
This file contains hidden or 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
This change adds a small fixed-size buffer to Pickler to avoid calling writer_() and the associated downstream checks on a per-opcode/per-byte basis. We end up still doing a bounds check in the common case, but the memcpy() is a fixed size. And we reduce the number of backend calls. In practice, this change speeds up the Pickle1MInts benchmark for me locally from roughly 56msec to 22msec. Additionally, in this change we convert a few pushIValue() on typed lists, where we know the type to be double/int/boot to be pushInt() to bypass a bit of logic. We should additionally change the Unpickler, though keeping this separate, since the std::function<> prototype needs to be changed to do buffering (return value needs to change from bool to size_t) Differential Revision: [D17939311](https://our.internmc.facebook.com/intern/diff/D17939311/) [ghstack-poisoned]
facebook-github-bot
added
the
oncall: jit
jjlilley
pushed a commit
that referenced
this pull request
Oct 15, 2019
This change adds a small fixed-size buffer to Pickler to avoid calling writer_() and the associated downstream checks on a per-opcode/per-byte basis. We end up still doing a bounds check in the common case, but the memcpy() is a fixed size. And we reduce the number of backend calls. In practice, this change speeds up the Pickle1MInts benchmark for me locally from roughly 56msec to 22msec. Additionally, in this change we convert a few pushIValue() on typed lists, where we know the type to be double/int/boot to be pushInt() to bypass a bit of logic. We should additionally change the Unpickler, though keeping this separate, since the std::function<> prototype needs to be changed to do buffering (return value needs to change from bool to size_t) Differential Revision: [D17939311](https://our.internmc.facebook.com/intern/diff/D17939311/) ghstack-source-id: 91964964 Pull Request resolved: #28043
jjlilley
changed the title
Author
|
This change is identical to (approved) 27720, except that
There was a tooling failure this afternoon which made it tricky for me to update 27720 (particularly, I used the web-based export-from-phabricator rather than ghexport, which has an infra issue now, and precludes using ghexport on followup diffs). If it's possible to stamp this, that would be great! |
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.
Stack from ghstack:
This change adds a small fixed-size buffer to Pickler to
avoid calling writer_() and the associated downstream checks
on a per-opcode/per-byte basis.
We end up still doing a bounds check in the common case,
but the memcpy() is a fixed size. And we reduce the number
of backend calls.
In practice, this change speeds up the Pickle1MInts benchmark
for me locally from roughly 56msec to 22msec.
Additionally, in this change we convert a few pushIValue() on
typed lists, where we know the type to be double/int/boot to be
pushInt() to bypass a bit of logic.
We should additionally change the Unpickler, though keeping
this separate, since the std::function<> prototype needs to be
changed to do buffering (return value needs to change from bool
to size_t)
Differential Revision: D17939311