Closed badshah400 closed 5 years ago
This is probably because I haven't kept up with the LU decomposition advances in GSL, so I will look into this next week. Thank you for pointing this out.
I made some progress on dtrsv(), but the LU updates will take me awhile. The LU decomposition works, it's just that the answers are a bit different because the algorithm is different (the test matrices are small, so I think GSL uses the level-2 blas results). Thus I've just changed the code that the tests don't fail on small finite-precision errors. I think that should fix it, but let me know if you still have problems. I'll close this issue in a couple weeks if I don't hear back that the problems continued. Thank you for reporting.
It turns out that part of the problem was the testing matrix was ill-conditioned, so the tests were too noisy. I just fixed this in 8ec947675d4c6e27c06e2fbf8fa541c7e0df4860 .
Oddly LU decomposition tests fail on x86_64 architectures while passing on i586 (gcc 9.2.1, boost 1.69, armadillo 9.700.2):
Full build log here: https://paste.opensuse.org/78728641