Public bug reported:

Whenever investigating test runs on hipblas it was proven that the tests
would occur in the form of:

```
[ RUN      ] _/getrs.solver/quick_getrs_a_type_f32_r_N_500_lda_601_ldb_700_F
./clients/gtest/../include/unit.h:79: Failure
Expected: (error) <= (tolerance), actual: -nan vs 0.0059604644775390625

./clients/gtest/../include/d_vector.hpp:148: Failure
Expected equality of these values:
  memcmp(host.data(), m_guard, m_guard_len)
    Which is: -180
  0

double free or corruption (out)

Aborting tests due to an alarm timeout.

This could be due to a deadlock caused by mutexes being left locked
after a previous test's signal was caught and partially recovered from.

```

After checking each individual error it was found that those would stem
from two main aspects, the first being the wrong use of `host_vector`
structures to hold status values (int) while their real type would match
the one of the matrix (T) used on the test, and the second being the
wrong assignment of LAPACK values from 32 to 64 bit addresses.

When comparing versions 7.1.0 to 7.2.0 the only difference between them
is the fact that on the latter the build forces linking of the
liblapack64-dev library instead of liblapack-dev, which means that this
bug is here for the fix on resolute release but most likely will not be
reflected on version 7.2.0 onwards.

** Affects: hipblas (Ubuntu)
     Importance: Undecided
     Assignee: Bruno Bernardo de Moura (bruno-bdmoura)
         Status: New

** Changed in: hipblas (Ubuntu)
     Assignee: (unassigned) => Bruno Bernardo de Moura (bruno-bdmoura)

-- 
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/2145046

Title:
  Fix tests for 7.1.0

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/hipblas/+bug/2145046/+subscriptions


-- 
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs

Reply via email to