computeTFactor parallelise computation with bulk (MC Local and Distributed) #798
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
albestro
force-pushed
the
alby/tfactor-bulk
branch
2 times, most recently
from
1af3047
to
c2d5c50
Compare
albestro
changed the title
albestro
commented
Apr 18, 2023
albestro
force-pushed
the
alby/tfactor-bulk
branch
from
0d944b5
to
26d72aa
Compare
rasolca
approved these changes
Apr 18, 2023
|
cscs-ci run |
|
The eigensolver test is deadlocking 😱 |
rasolca
added a commit
that referenced
this pull request
Apr 19, 2023
msimberg
approved these changes
Apr 19, 2023
|
cscs-ci run checking if the workaround works also in the CI. In case, I will document more the case and use it to investigate deeper the problem. |
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Outdated suggestions cannot be applied.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 PR improves computeTFactor computation by parallelising the computation that previously was serialised because of the reduction over a single tile.
The algorithm implementation has been split in two variants:
Backend::MCnow has thebulk-based implementationBackend::GPUstill use the previous version of the codeThis required a couple of changes to the interface structure of the algorithm, and I took the chance to fix also some problem with the documentation.
Note: Look at first three commits to see the main implementation change in a more "isolated" way from the rest.
TODO
nthreadsused for bulk tunable