Vectors test fails on AMD Ryzen 9 4900HS

[Zach Etienne]

Vectors, after correctly finding that CCTK_REAL_VEC_SIZE is in fact 4 on this CPU, with flags:

flags           : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good nopl nonstop_tsc cpuid extd_apicid aperfmperf pni pclmulqdq monitor ssse3 fma cx16 sse4_1 sse4_2 movbe popcnt aes xsave avx f16c rdrand lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs skinit wdt tce topoext perfctr_core perfctr_nb bpext perfctr_llc mwaitx cpb cat_l3 cdp_l3 hw_pstate ssbd mba ibrs ibpb stibp vmmcall fsgsbase bmi1 avx2 smep bmi2 cqm rdt_a rdseed adx smap clflushopt clwb sha_ni xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local clzero irperf xsaveerptr rdpru wbnoinvd arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic v_vmsave_vmload vgif v_spec_ctrl umip rdpid overflow_recov succor smca sme sev sev_es

segfaults on the vec_loadu_maybe[1] test in Vectors.

INFO (Vectors): Testing vectorisation... [errors may result in segfaults]
INFO (VecINFO (Vectors): Testing vectorisation... [errors may result in segfaults]
INFO (Vectors): Test vec_set1...
INFO (Vectors): Test vec_set...
INFO (Vectors): Test vec_elt[0]...
INFO (Vectors): Test vec_elt[1]...
INFO (Vectors): Test vec_elt[2]...
INFO (Vectors): Test vec_elt[3]...
INFO (Vectors): Test vec_load...
INFO (Vectors): Test vec_loadu[0]...
INFO (Vectors): Test vec_loadu[1]...
INFO (Vectors): Test vec_loadu[2]...
INFO (Vectors): Test vec_loadu[3]...
INFO (Vectors): Test vec_loadu_maybe[0]...
INFO (Vectors): Test vec_loadu_maybe[1]...
Rank 0 with PID 218092 received signal 11

After commenting out this test, the test suite runs with no problems.

[Erik Schnetter]

This test checks whether unaligned loads are working. It is likely that there is a bug in thorn vectors where it uses an aligned load where it shouldn’t.

This part of thorn Vectors is unused by most code. If you do not see a segfault at run time, then your code is very likely working correctly.

[Roland Haas]

@Zach Etienne can you grant access to the system showing the issue to Erik or I so that one could test this? Or do some debugging yourself if provided suggestions what to test?

[Zach Etienne]

I just recompiled with the latest version of the ET on that machine. It compiled without any issues, and was able to run a few parfiles that use Vectors . Marking the ticket as Resolved.

[Zach Etienne]

I can no longer reproduce the issue seen, with the latest ET.

[Roland Haas]

@Zach Etienne you say you can compile without issues. However looking at the issue description the problem was never a compile time problem but a runtime failure.

Can you try and run the code? Just the testsuite for thorn Vectors should do. Something like:

make sim-testsuite PROMPT=no CCTK_TESTSUITE_RUN_TESTS=Vectors/vectors

should do. Or you can leave out the PROMPT and CCTK_TESTSUITE_RUN_TESTS options and use the interactive menu to select which tests to run.

[Zach Etienne]

Sorry forgot to report that I ran the Vectors tests from the testsuite as well. Again no failures.

[Roland Haas]

RIght. Actually you did report that you could run parfiles “and was able to run a few parfiles that use Vectors “ which would have been good enough (the testsuite just runs the same self-tests for all possible architectures) but I missed that part of the sentence, sorry.

‌