-
Notifications
You must be signed in to change notification settings - Fork 45
CI: add rpm build workflow #1244
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
base: master
Are you sure you want to change the base?
Conversation
|
@junghans Does not seem like it works unless I'm missing something. |
db41882 to
8d70ca1
Compare
|
|
@junghans Is it possible to run the CI based on the current branch? So that if I push here, it runs the change and not 2.0. Also, is there a way to speed it up, it seems to take 2+ hours? |
@aprokop it makes a tarball out of the current checkout, it is just the tarball is always named But maybe the easiest would to trying to just build it with same flags and CMake options: |
|
So it seems that the following change makes it pass: --- a/src/spatial/detail/ArborX_ExpandHalfToFull.hpp
+++ b/src/spatial/detail/ArborX_ExpandHalfToFull.hpp
@@ -50,19 +50,13 @@ void expandHalfToFull(ExecutionSpace const &space, Offsets &offsets,
"ArborX::Experimental::HalfToFull::counts");
Kokkos::parallel_for(
"ArborX::Experimental::HalfToFull::rewrite",
- Kokkos::TeamPolicy(space, n, Kokkos::AUTO, 1),
- KOKKOS_LAMBDA(
- typename Kokkos::TeamPolicy<ExecutionSpace>::member_type const
- &member) {
- auto const i = member.league_rank();
- auto const first = offsets_orig(i);
- auto const last = offsets_orig(i + 1);
- Kokkos::parallel_for(
- Kokkos::TeamVectorRange(member, last - first), [&](int j) {
- int const k = indices_orig(first + j);
- indices(Kokkos::atomic_fetch_inc(&counts(i))) = k;
- indices(Kokkos::atomic_fetch_inc(&counts(k))) = i;
- });
+ Kokkos::RangePolicy(space, 0, n), KOKKOS_LAMBDA(int i) {
+ for (int j = offsets_orig(i); j < offsets_orig(i + 1); ++j)
+ {
+ int const k = indices_orig(j);
+ indices(Kokkos::atomic_fetch_inc(&counts(i))) = k;
+ indices(Kokkos::atomic_fetch_inc(&counts(k))) = i;
+ }
});
Kokkos::Profiling::popRegion();
}I don't understand why. Both codes seem valid to me. It seems to only affect |
|
I am not sure either, but maybe @dalg24 knows.... |
|
Either way, I patched that in and rebuild: https://koji.fedoraproject.org/koji/taskinfo?taskID=131982908 |
|
Hmm, the latest patch failed in a different place: So, it seems, the failures are intermittent. I really need to be able to run things in a loop to properly debug this. |
|
The |
| auto const last = offsets_orig(i + 1); | ||
| Kokkos::parallel_for( | ||
| Kokkos::TeamVectorRange(member, last - first), [&](int j) { | ||
| Kokkos::TeamThreadRange(member, last - first), [&](int j) { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
TeamVectorRange looks more correct here (so that it would also work with vector_length > 1).
|
@aprokop let me know if you have a patch set I should test on Fedora again. |
|
@aprokop any update on this, anything I can help with. |
|
@junghans I think I'm essentially stuck here. None of it makes any sense to me. I will try to add more printouts and see if I can track it some more. I wonder if it is some optimizations again, and the issue would disappear with "-O0". |
|
Either the failure is intermittent (which it could be), or it is similar to #1186. |
|
@dalg24 any ideas? |
009c4a1 to
30529a5
Compare
|
@aprokop I gave this a fresh start, similar to what we did in |
|
I looked at the code again, and still no wiser. Need to be able to access arm64 without cross-compiling. |
|
On apple-silicon mac with docker you could use the following dockerfile: FROM registry.fedoraproject.org/fedora:latest
RUN dnf install -y fedpkg wget
RUN wget https://github.com/junghans/ArborX/archive/refs/heads/rpmbuild.zip
RUN unzip rpmbuild.zip
RUN mv ArborX-rpmbuild ArborX
RUN tar -cvzf ArborX-9999.tar.gz ArborX/
RUN mkdir ArborX.rpm
RUN mv ArborX-9999.tar.gz ArborX.rpm
WORKDIR ArborX.rpm
RUN wget https://github.com/junghans/ArborX/raw/refs/heads/rpmbuild/.github/workflows/ArborX.spec
RUN dnf -y builddep ArborX.spec
RUN fedpkg --verbose --debug local |
Thank you. I need to figure out how to bypass the proxies on my work mac, as it's running into |
|
You could add a line like |
I think I wrongly attributed the problem to proxy. It is the certificate replacement issue, where our deep inspection thingie replaces certificates. I slightly changed the recipe by this tweak: |
|
Running into issue on the last step: |
|
Ok, I not sure how that happened, but I was able to reproduce it on one of my machines, try with these 3 lines at the end: |
|
May still be a certificate problem |
|
How about ...
RUN dnf -y builddep ArborX.spec
RUN rpmbuild --define "_sourcedir $(pwd)" -ba ArborX.spec |
That works. I'm trying to build it. I'm getting warnings like this Is it possible to go into the build directory and try to compile manually? If I try |
|
Something like: |
|
Getting some errors when building inside the container But the two tests that fail in this PR were built ( |
For #1241