Processes started concurrently could access shared memory before the
input matrix was loaded into it. Now, the matrix is loaded into a
different path than the one processes expect and hardlinked with the
proper one once the loading is finished. Processes may have to wait at
startup for this to finish, and an optional timeout parameter is added
to prevent them from waiting forever in case of error.
Processes started concurrently could access shared memory before the input matrix was loaded into it. Now, the matrix is loaded into a different path than the one processes expect and hardlinked with the proper one once the loading is finished. Processes may have to wait at startup for this to finish, and an optional timeout parameter is added to prevent them from waiting forever in case of error.