Change transfers to do pulls on multi transfer #19931
Open
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
benvanik
requested changes
Feb 6, 2025
compiler/src/iree/compiler/Dialect/Stream/Analysis/Partitioning/ReferencePartitioning.cpp
Outdated
Show resolved
Hide resolved
rsuderman
force-pushed
the
partition_changes
branch
from
3227400 to
d47e8a1
Compare
rsuderman
force-pushed
the
partition_changes
branch
from
35147c1 to
0f8da4a
Compare
Transfers are serialized per-device. This makes cases of 0->1,2,3 to be serially move to device. In these cases it is better to transfer on consumer to enable parallel transfer between devices.
Signed-off-by: Rob Suderman <[email protected]>
rsuderman
force-pushed
the
partition_changes
branch
from
c48543b to
f195fed
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Outdated suggestions cannot be applied.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.
Scatter-gather reduction operations are preferable due to non-parallel data transfers. To adjust for this we should only push tensors when they are mono-transfers. When distributing to multiple devices it is preferable to have the device fetch as the case of 1...8 -> 1 -> 1...8 we would prefer each device to handle its own data transfers.