Closed RMeli closed 1 year ago
cscs-ci run
cscs-ci run
cscs-ci run
cscs-ci run
Current CI failures should be fixed by #983.
cscs-ci run
cscs-ci run
cscs-ci run
cscs-ci run
Issue reproduced in CI.
cscs-ci run
cscs-ci run
cscs-ci run
cscs-ci run
cscs-ci run
If the input matrix is banded and has a band size smaller than the target band size for the reduction,
nan
s are produced due to a division by zero when definingtau
: https://github.com/eth-cscs/DLA-Future/blob/1fd315b966e97326da34ce0f3f1af43fd8f8a996/include/dlaf/eigensolver/reduction_to_band/impl.h#L102-L105This PR introduces an early termination when
x0_and_squares[1] == 0
by returningtau = 0
, thus circumventing the problem. Fix #974. Thanks @albestro for the help in identifying the issue.With this PR, the situation in CP2K is the following (using the DLA-Future eigensolver for every matrix of size
2x2
or higher):All regression tests still returning
nan
s (see #974) pass. Of the two remaining tests one has been fixed in CP2K (missing DLAF/pika initialization when using CP2K as a library), while the other one also fails with ScaLAPACK.