Alberto Invernizzi
Alberto Invernizzi
Despite an important part was missing in the implementation of the tridiagonal solver (see #960), tests were not able to spot the problem. We are aware that current unit tests...
Current implementation of Permute distributed, i.e. `permuteOnCPU`, currently allocates memory at schedule-time for: - `inverse_perms` - `local2global_index` - `packing_index` - `unpacking_index` - `mat_send` - `mat_recv` If the ones for indices...
As quickly proposed here https://github.com/eth-cscs/DLA-Future/pull/880#discussion_r1221364476, even if their nature is almost the same, it might be good to have tune parameters separated from debugging parameters, making clear the distinction between...
PR #835 split the management of permutations available locally from the ones that needs the communication to be carried out. Currently, local permutations skips the communication part, but they are...
It might be worth tracking also improvements in terms of memory usage. Under the radar we are also working for reducing memory usage that sometimes represents a limit. Keeping track...
This issue just for trying keeping track of this detail that does not seem to represent a problem at the moment, but it might be in the future. https://github.com/eth-cscs/DLA-Future/blob/e50b4fa5041ac77f4a7c50f4be785eca704004f7/include/dlaf/permutations/general/impl.h#L284-L285 https://github.com/eth-cscs/DLA-Future/blob/e50b4fa5041ac77f4a7c50f4be785eca704004f7/include/dlaf/permutations/general/impl.h#L296-L297...
Check has been disabled in 6873c3268070c2872b952df39923a95bf49f34e9 for removing the dependency from `DLAF_test`.
While working on #704 I stumbled upon a deadlock problem with `test_gen_eigensolver`. I report here some results of the problem investigation. # Test-cases and configurations Apparently, of all test-cases currently...
An implementation detail of the distributed back transform tridiagonal algorithm has been chosen without performing any check. Operations are already scheduled scanning the matrix with a smart ordering by so...
8baa1ea18359b11e1741246750f0d964cc512501 here the trigger specialisation has been implemented and documented. It is a quite specific workaround and it might still create some problem https://github.com/eth-cscs/DLA-Future/blob/8baa1ea18359b11e1741246750f0d964cc512501/include/dlaf/eigensolver/reduction_to_band/impl.h#L1165 I will ask you reviewers @msimberg...