Closed msurkovsky closed 9 years ago
Octave API seems broken in thread backend. With OM we have investigated the problem this month, and it seems that the serialization of matrices are not thread-safe. I am not 100% sure, but if it is true, then there is no easy fix. It would probably need to migrate threads to processes, but it would be a good change anyway because it would solve some other problems and mimics MPI more precisely. So for now I am considering to temporary disable Octave API.
Merged via a85672f6cf2aae77d3c37776be37becd1509a18c
Please can you check tests? There is a problem with
matmul_cc
, but I don't know where. For example, when I change a doubles and call there all_processes (as init function of a new place) it works including matmul_cc. When it remains as it is now, then it ends with an error. My observation is that it does not work when such methods are called on blue are, includingca::range
. I think that it can be caused by some bad order of operation calling or something like that. Thematmul-cc
in samples seems to work; tested in release and simulation.