-
Notifications
You must be signed in to change notification settings - Fork 88
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
PowerPC build: 14 tests failed out of 231 #1260
Comments
|
|
|
Then, some MPI-related tests seem to fail quite badly. Output is too long to post, but here is the full log. @upsj Could you take a look? This we would definitely want to have fixed. |
Thanks for the report! Most of the failures are just our test tolerances being a bit too tight. The Coo test failure looks concerning though, and the MPI tests look like they're not being run with MPI? |
Any ideas why it may fail?
We may have a fix for this, I am rerunning the build now and will update you.
|
Probably atomics, since this is to my knowledge the only kernel where we use float atomics in OpenMP (handling overlaps between threads). This is also compounded by the fact that the errors only happen at discrete locations (Hybrid uses Coo internally, so those are essentially the same issue) |
GCC should support atomics via builtins, but on 32-bit platforms there is no float128. |
Are those atomics also supported by the OpenMP implementation? We use |
@upsj Honestly I cannot say for sure. From what I understand, OpenMP is supported by GCC internally via For 32-bit platforms, GCC needs explicit linking to What do you think about endianness here, for floating point? |
Endianness only matters for the We rely on CMake's FindOpenMP module adding the right flags, so if Do the tests also fail if you run them with |
@upsj So when running under Here what failed:
(then in next comment) |
|
Notice, And Does not look better, I guess. |
@upsj I will try with |
@upsj I tried passing Results up to that:
|
I think it is indeed the big-endian lead the difference in extended_float test. |
the execution failure is on the exception from over-allocation. |
@barracuda156 if you have time, could you check the core/test/base/extended_float and core/test/base/half in the half_type branch https://github.com/ginkgo-project/ginkgo/tree/half_type? |
I have rebuilt
ginkgo
from the latest commit in master, same results as from the last release:The text was updated successfully, but these errors were encountered: