tested on sifive unmatched The following tests FAILED: 13 - test_nvector_mpi_4_1000_0 (Failed)[0;0m 14 - test_fnvector_parallel_mod_4 (Failed)[0;0m 15 - test_nvector_mpimanyvector_parallel1_4_1000_200_0 (Failed)[0;0m 16 - test_nvector_mpimanyvector_parallel2_4_200_1000_0 (Failed)[0;0m 18 - test_fnvector_mpimanyvector_mod_4 (Failed)[0;0m 19 - test_nvector_mpiplusx_4_1000_9 (Failed)[0;0m 21 - test_fnvector_mpiplusx_mod_4 (Failed)[0;0m 26 - test_nvector_parhyp_4_1000_0 (Failed)[0;0m 83 - test_sunlinsol_spgmr_parallel_100_1_1_50_1e-3_0 (Failed)[0;0m 84 - test_sunlinsol_spgmr_parallel_100_1_2_50_1e-3_0 (Failed)[0;0m 85 - test_sunlinsol_spgmr_parallel_100_2_1_50_1e-3_0 (Failed)[0;0m 86 - test_sunlinsol_spgmr_parallel_100_2_2_50_1e-3_0 (Failed)[0;0m 87 - test_sunlinsol_spfgmr_parallel_100_1_50_1e-3_0 (Failed)[0;0m 88 - test_sunlinsol_spfgmr_parallel_100_2_50_1e-3_0 (Failed)[0;0m 89 - test_sunlinsol_spbcgs_parallel_100_1_50_1e-3_0 (Failed)[0;0m 90 - test_sunlinsol_spbcgs_parallel_100_2_50_1e-3_0 (Failed)[0;0m 91 - test_sunlinsol_sptfqmr_parallel_100_1_50_1e-3_0 (Failed)[0;0m 92 - test_sunlinsol_sptfqmr_parallel_100_2_50_1e-3_0 (Failed)[0;0m 106 - test_sunlinsol_superlumt_300_0_1_0 (Failed)[0;0m 107 - test_sunlinsol_superlumt_300_1_1_0 (Failed)[0;0m Reproducible: Always
Created attachment 744411 [details] emerge --info
Created attachment 744414 [details] test output
Created attachment 744417 [details] build log compressed due to size limit, please use zcat to view it
This issue in upstream may be relevant. https://github.com/LLNL/sundials/issues/26
It seems the cause of most failure is just my testing machine is overloaded and there are not enough core/slots for openmpi, therefore openmpi gave explicit warning: "There are not enough slots available in the system to satisfy the 4 slots that were requested by the application:" Rerun the test manually with --oversubscribe solved most failures.
The only test failure left now is test_sunlinsol_superlumt. The error message is raised from file SRC/get_perm_c.c in sci-libs/superlu.
Looks like the problem with test_sunlinsol_superlumt is integer sizes. Sundials always use 8 bytes integer as index type, > if(HAS_${INT64_TYPE_NOSPACE} EQUAL "8") but superlu_mt is compiled with int_t default to int, which is 4 bytes on risv64 lp64. set -D_LONGINT will solve the problem > #ifdef _LONGINT > typedef long long int int_t; > #define IFMT "%lld" > #else > typedef int int_t; /* default */ > #define IFMT "%8d" > #endif
(In reply to Alex Fan from comment #7) > Sundials always use 8 bytes integer as index type, Not quite - the size of indices is controlled by the CMake option SUNDIALS_INDEX_SIZE, which defaults to 64 in our case but could be set to 32. > but superlu_mt is compiled with int_t default to int, which is 4 bytes on > risv64 lp64. set -D_LONGINT will solve the problem ...or emerge sci-libs/superlu_mt with USE=int64. I've got an idea on how to solve this but will probably not test it until tomorrow, my systems are fully loaded at present.
Reproduced the error on amd64.
Posting this manually because something seems to be wrong with the post-commit hook on g.g.o: commit dd47d1b3ec10aa6779430d1c1142e64d462ee707 Author: Marek Szuba Date: Tue Oct 12 21:55:29 2021 +0000 sci-libs/sundials: implement USE=int64 What it already does: * allows changing the index size from 64 (default) to 32 bits * if USE=superlumt is set makes sure the state of USE=int64 in sci-libs/superlu_mt is the same as here, so that the assumption about index-type compatibility made in sunlinsol_superlumt.h is correct * allows test_sunlinsol_superlumt_300_0_1_0 and test_sunlinsol_superlumt_300_1_1_0 to pass *if USE=-int64* What still needs work: * getting the two superlumt tests to pass for USE=int64 - they still fail with "COLAMD failed at line 58 in file get_perm_c.c" Bug: https://bugs.gentoo.org/817680 Signed-off-by: Marek Szuba <marecki@gentoo.org>
The bug has been closed via the following commit(s): https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=b25061d0cf0d41213c90ebab0845ed3f50403eef commit b25061d0cf0d41213c90ebab0845ed3f50403eef Author: Marek Szuba <marecki@gentoo.org> AuthorDate: 2021-11-28 20:48:56 +0000 Commit: Marek Szuba <marecki@gentoo.org> CommitDate: 2021-11-28 20:48:56 +0000 sci-libs/superlu_mt: apply Alex's PREDEFS patch New revision because it changes runtime behaviour for USE=int64 users. Closes: https://bugs.gentoo.org/817680 Closes: https://github.com/gentoo/gentoo/pull/23063 Signed-off-by: Marek Szuba <marecki@gentoo.org> sci-libs/superlu_mt/superlu_mt-3.1-r1.ebuild | 105 +++++++++++++++++++++++++++ 1 file changed, 105 insertions(+)
