Hi Samuel and Brice,
I made the changes to the code. I tried running it and it crashed around line
53 with this statement
(I debugged it by stepping through the program).
hwloc_set_thread_cpubind(topology, handle,
hwloc_get_obj_by_type(topology, HWLOC_OBJ_CORE, t)->cpuset,
0);
I also wanted
Ask nvidia to make their cuda driver free, and to add them to sysfs :)
There's a cuda hwloc branch that will solve this. In the meantime, there are no
nvidia osdevs.
Maybe look at hwloc/cuda.h and cudart.h, they give cuda device affinity without
osdevs.
Brice
Guillermo Miranda
Hi Brice,
I made the corrections in the program below. The thing I noticed now is that
the cores used by the
program switch around. There are two cores used at 100% but after a few seconds
another core
takes over at 100% and there were always two cores used at maximum, while the
other 6 (I