{2023.06} rebuild Siesta/5.2.2-foss-2023a#1535
Conversation
Siesta/5.2.2-foss-2023a
|
bot: build repo:eessi.io-2023.06-software instance:eessi-bot-mc-aws for:arch=x86_64/generic |
|
New job on instance
|
|
New job on instance
|
|
New job on instance
|
|
New job on instance
|
|
New job on instance
|
|
New job on instance
|
|
New job on instance
|
|
New job on instance
|
|
New job on instance
|
|
New job on instance
|
|
New job on instance
|
|
New job on instance
|
|
New job on instance
|
|
New job on instance
|
|
failed on A64FX, let's try that again... bot: build repo:eessi.io-2023.06-software instance:eessi-bot-deucalion for:arch=aarch64/a64fx |
|
New job on instance
|
|
Does 5.2.2 need a rebuild? Because their the commit of elsi-interconnect is correct. |
A bunch of tests failed, first one ( So this is due to a lack of sufficient memory... All failing tests are @@AnthoniAlcaraz Anything we can tweak here to reduce memory requirements for these tests? |
Strictly speaking maybe not, but the easyconfig file was changed quite a bit in: |
|
Ok. Also why is that licence CI failing. Is that because it is a rebuild. This is also the case in the other Siesta pr. Is this because it is a rebuild? |
|
CI is now fixed so when you merge with main it should pass |
That siesta-00.BasisSets-default_basis_mpi4_omp1 should use very little resources as it is. Note that there have been quite a number of fixes and extensions between 5.2.2 and 5.4.2. I am not sure whether it might be worth using 5.4.2 for any further deployments. |
|
Just wanted to share a related observation, in case it's useful context for this rebuild. While preparing a local EasyBuild install of Root cause: The basis-generation routines use large Fortran automatic arrays (stack-allocated, sized at runtime based on basis-set parameters). With a more demanding basis like I confirmed this by re-running the exact same compiled binary with Executable : siesta
Version : 5.2.2
Architecture : x86_64
Compiler version: GNU-12.3.0
Compiler flags : -O3 -fopenmp -march=native -fno-math-errno -fallow-argument-mismatch
Parallelisations: MPI, OpenMP
* OpenMP version: 201511
NetCDF support
NetCDF-4 support
NetCDF-4 MPI-IO support
Lua support
Native ELPA support
ELSI support. Solvers:
ELPA (internal)
NTPoly
OMM
PEXSI
DFT-D3 support
Wannier90 wrapper support
Given that the comment above mentions a similar SIGSEGV on If it turns out to be the same root cause, adding |
|
I checked the ulimits on Deucalion (where our A64FX build node is): Soft limit: Hard limit: @boegel I guess we can increase the The open file limit is actually also quite tight, also the hard limit. I'm a bit surprised we've never hit that before, but it's irrelevant in the context of the issue in this PR. |
|
A64FX bot has been updated to set bot: build repo:eessi.io-2023.06-software instance:eessi-bot-deucalion for:arch=aarch64/a64fx |
|
New job on instance
|
fixes #1519