search

cafana / OscLib

1 stars 7 forks source link

eigen fun with art 314 series #26

Open gavinsdavies opened 2 weeks ago

gavinsdavies commented 2 weeks ago

The saga continues. Redmine Issue 28782 requested stan_math and sundials versions to work with eigen v23_08_01_66e8f in art 314 series. Ran into issues...we dropped back to eigen v3_4_0... Meanwhile, stan_math was patched for the eigenv23_08_01_66e8f. I've tested this combination -- it works , for the e26 compiler.

c14 compiler is fussier:

/cvmfs/larsoft.opensciencegrid.org/products/stan_math/v4_9_0a/include/stan/math/prim/functor/operands_and_partials.hpp:57:1: error: 'ops_partials_edge' defined as a struct template here but previously declared as a class template; this is valid, but may result in linker errors under the Microsoft C++ ABI [-Werror,-Wmismatched-tags]

I only tested the e26 locally before pushing to jenkins. For now, on Jenkins I have turned off all compilers except e26.

OK, so we can bypass the mismatched-tags warning as an error easy enough but in interests of SBN folks having something to test, I'll make osclib v00.26 available for the e26 compiler, and revisit for c14 compiler, assuming SBN have success with e26. It does come with a warning label from SciSoft given the stan_math patch; buyer beware.

gavinsdavies commented 2 weeks ago

and note for myself to fix in Jenkins. it makes the build directories for the different qualifiers as defined by underlying machine now (ALMA9) so we get directories like: 'OS=ALMA9.e26.n316.prof.stan' instead of: slf7.x86_64.e26.n316.prof.stan which is not helpful for UPS -- manually fixed today's. Just need to fix the Jenkins config

PetrilloAtWork commented 2 weeks ago

Can we report that to stan-dev/math? If I don't misread it, it's just about replacing a struct with a class, or vice versa.