nonblocking handles with RMA requests #53
Draft
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
Signed-off-by: Jeff Hammond <jeff.science@gmail.com>
Signed-off-by: Jeff Hammond <jeff.science@gmail.com>
…EFINED check Signed-off-by: Jeff Hammond <jeff.science@gmail.com>
Signed-off-by: Jeff Hammond <jeff.science@gmail.com>
Signed-off-by: Jeff Hammond <jeff.science@gmail.com>
Signed-off-by: Jeff Hammond <jeff.science@gmail.com>
Signed-off-by: Jeff Hammond <jeff.science@gmail.com>
Signed-off-by: Jeff Hammond <jeff.science@gmail.com>
Signed-off-by: Jeff Hammond <jeff.science@gmail.com>
with the separate memory model, there are scenarios where we want/need to synchronize the public and private window after all processes have completed ARMCI_AllFence. ideally, GA calls ARMCI_Barrier in GA_Sync but since that isn't the case, we will add an option to call ARMCII_Sync (which calls MPI_Win_sync on all windows) inside of armci_msg_barrier (and similar), since we know that GA always calls this after it fences. Signed-off-by: Jeff Hammond <jeff.science@gmail.com>
Signed-off-by: Jeff Hammond <jeff.science@gmail.com>
Fetch_and_op or Compare_and_swap plus Flush(_local) might be more expensive so we add an option to use Rget_accumulate (yes, way more arguments) and wait on the resulting request, which might be better in some cases.
Signed-off-by: Jeff Hammond <jeff.science@gmail.com>
no implementation of request-based RMA yet... Signed-off-by: Jeff Hammond <jeff.science@gmail.com>
Signed-off-by: Jeff Hammond <jeff.science@gmail.com>
this is not working for nonblocking vector ops, which fails in armci-test. all other tests pass, at least in shared-memory. Signed-off-by: Jeff Hammond <jeff.science@gmail.com>
Signed-off-by: Jeff Hammond <jeff.science@gmail.com>
Signed-off-by: Jeff Hammond <jeff.science@gmail.com>
Signed-off-by: Jeff Hammond <jeff.science@gmail.com>
Signed-off-by: Jeff Hammond <jeff.science@gmail.com>
Signed-off-by: Jeff Hammond <jeff.science@gmail.com>
Signed-off-by: Jeff Hammond <jeff.science@gmail.com>
Signed-off-by: Jeff Hammond <jeff.science@gmail.com>
Signed-off-by: Jeff Hammond <jeff.science@gmail.com>
Signed-off-by: Jeff Hammond <jeff.science@gmail.com>
Signed-off-by: Jeff Hammond <jeff.science@gmail.com>
Signed-off-by: Jeff Hammond <jeff.science@gmail.com>
This was referenced Oct 3, 2024
Signed-off-by: Jeff Hammond <jeff.science@gmail.com>
running NWChem generates a huge number of assertions/warnings about bogus handles. it would seem that GA does a bad job of initializing these. Signed-off-by: Jeff Hammond <jeff.science@gmail.com>
ARMCII_Warning was called before ARMCI_GROUP_WORLD was initialized, so warnings in init were printed by every rank. Signed-off-by: Jeff Hammond <jeff.science@gmail.com>
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 was a long-standing omission in the implementation. ARMCI nonblocking handles are similar to MPI RMA requests but are not 1:1 because aggregate request handles are 1:N.
This implements request handles using RMA requests, which replaces the prior implementation that just did
flush(_all)instead of individual handle completion. The old implementation is preserved via the preprocessor.This also adds a feature to switch to
Rget_accumulatefor atomics (all of which are blocking), which avoids a flush in this code path that might be slowed down by the need to complete more expensive, potentially non-hardware, operations.This has not been tested thoroughly. It will be merged after sufficient testing.
Tested with: