Prepare ompio frameworks for bigcount #12361
Conversation
Update io/ompio, common/ompio, fcoll, and sharedfp frameworks and components to use size_t instead of int for count types. Signed-off-by: Jake Tronge <[email protected]>
|
I can confirm that this PR is correct in terms of high level interface changes. We also still pass our testsuites. However, I am 99% sure that we will need to add more work to actually make bigcount work, including probably adding at least some tests to our testsuite. So if the plan is to get the interface changes in and work after that component by component to upgrade them to support bigcount (and I am happy to help with that), I would be ok with this pr. |
|
@edgargabriel do you approve this PR? |
|
@edgargabriel could you advise us on the module/component versioning? we think these should be change in fcoll and sharedfp? |
|
@hppritcha I think you are correct in that we should probably increase the module/component versions in the fcoll and sharedfp frameworks. Ping me if you make that change, and I will rerun our tests before merging again |
|
@edgargabriel could you double check @jtronge 's latest commit ? |
There was a problem hiding this comment.
You have removed all type-casts associated with mallocs, please add them back, e.g.
- sizes_old_group = (int*)malloc(num_merge_aggrs * sizeof(int));
+ sizes_old_group = malloc(num_merge_aggrs * sizeof(size_t));
please leave it as
+ sizes_old_group = (size_t *)malloc(num_merge_aggrs * sizeof(size_t));
throughout the code base. I will re-review once that is fixed everywhere, but I would have to add too many messages/comments right now because of this issue.
jtronge
force-pushed
the
io-framework-bigcount
branch
from
4a506ba
to
7a88f90
Compare
Signed-off-by: Jake Tronge <[email protected]>
jtronge
force-pushed
the
io-framework-bigcount
branch
from
7a88f90
to
03d8ba8
Compare
|
@edgargabriel I've added back all the malloc type-casts. When I was changing the versions in fcoll and sharedfp I noticed that I had missed some of the collective calls in |
|
@jtronge thank you, I think it looks good. I will run tomorrow morning our testsuites on this branch, and will let you know after that whether the tests expose something in addition. |
|
Tests look good, I think we are good to merge this pr. Thank you for this work @jtronge ! |
|
After merging the PR we should watch MTT for a while. If often catches corners cases that we don't normally test. |
|
(Clicked wrong button) |
|
@edgargabriel could you click the review again? |
|
canceled NVIDA CI as it seems to be unable to recover from power outage any time soon. |
This updates ompio-related frameworks and components to use size_t instead of int for counts.
I updated the version in the io framework from 2.0.0 to 3.0.0. fcoll and sharedfp both seem to be using different versions for the component and module structs, so I wasn't quite sure what version to use for those.