Test for LOBPCG fails on single-precision

Issue #7 resolved
Carl Yang created an issue
Ubuntu 16.04
gcc 4.9
magma 2.3.0
ATLAS 3.10.2

Running the following test hangs forever:

./run_tests.py --lobpcg --start testing_ssolver

When I use gdb to get backtrace, it gives me:

#0  0x00007fc278687b13 in ATL_sscal_xp1yp0aXbX () from /usr/lib/libatlas.so.3
#1  0x00007fc281a6fa9e in slarfg_ () from /usr/lib/liblapack.so.3
#2  0x00007fc281a93bfa in slatrd_ () from /usr/lib/liblapack.so.3
#3  0x00007fc281ae1984 in ssytrd_ () from /usr/lib/liblapack.so.3
#4  0x00007fc281ad738a in ssyev_ () from /usr/lib/liblapack.so.3
#5  0x00000000004af34c in magma_slobpcg ()
#6  0x0000000000460be5 in magma_s_solver ()
#7  0x000000000041c3dc in main ()

I get a similar error on sscal() that comes from the ssyev() call using OpenBLAS instead of ATLAS. However, this test passes when using testing_dsolver so my guess is it could be due to the test matrix being ill-conditioned.

Comments (4)

  1. Log in to comment