Various ncclize improvements #6
Merged
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
Now sets of addresses that result in larger counts when contiguosly allocated are greedily allocated first.
This fixes a problem where nop ops with no dependencies were created.
Any addresses that are both input and output are now copied with the new instruction before the main algorithm starts. The greedy scratch allocation algorithm will also insert copies to make input/output addresses contiguous in scratch.
Also improve greedy scratch sorting to work better with this.
Also now postcopies are preferred, unless doing a precopy allows maintaining contiguity.
This was breaking extra memory constraints.
Now contiguous sends are split up as necessary to match concurrency. Also remove MaxConcurrency channel policy.
The old format has become entrenched.
New option is --no-scratch
Not added with a sledgehammer anymore, hooks into the same dependency tracking all other operations use.
olsaarik
force-pushed
the
feature/ncclize-improvements
branch
from
90da877 to
8406fe9
Compare
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Outdated suggestions cannot be applied.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.
This includes a collection of work on improving ncclize, including:
--old-formatand--use-scratch<copy/>tags that were never implemented in SCCL-RT in favor of newcpyoperations.