Re: [Starpu-devel] hwloc CUDA plugin missing

[Alfredo Buttari <alfredo.buttari@enseeiht.fr>]

not sure I understand.

The documentation says:

"Note that the first workers correspond to the CUDA workers, then come
the OpenCL workers, and finally the
CPU workers. For example if we have STARPU_NCUDA=1 , STARPU_NOPENCL=1
, STARPU_NCPU=2
and STARPU_WORKERS_CPUID = "0 2 1 3" , the CUDA device will be
controlled by logical CPU #0,
the OpenCL device will be controlled by logical CPU #2, and the
logical CPUs #1 and #3 will be used by the
CPU workers."

so if I set

STARPU_NCUDA=4
STARPU_NCPU=20
STARPU_WORKERS_CPUID="0 1 12 13 2-11 14-23"

I would expect the four GPU workers to be bound to cpuids 0, 1, 12 and
13 and the CPU workers to be bound to cpuids 2-11,14-23. Instead here
is what the instructions in my previous email print

>>>  0   0  20
>>>  1   1  21
>>>  2  12  22
>>>  3  13  23
>>>  4   2   0
>>>  5   3   1
>>>  6   4  12
>>>  7   5  13
>>>  8   6   2
>>>  9   7   3
>>> 10   8   4
>>> 11   9   5
>>> 12  10   6
>>> 13  11   7
>>> 14  14   8
>>> 15  15   9
>>> 16  16  10
>>> 17  17  11
>>> 18  18  14
>>> 19  19  15
>>> 20  20  16
>>> 21  21  17
>>> 22  22  18
>>> 23  23  19

the second column, i.e., workers_bindid corresponds to what I set in
the STARPU_WORKERS_CPUID variable, the third, i.e., workers[i].bindid
does not and I don't understand what motivates the difference.

On Mon, Feb 11, 2019 at 3:59 PM Samuel Thibault
<samuel.thibault@inria.fr> wrote:
>
> Alfredo Buttari, le lun. 11 févr. 2019 14:59:52 +0100, a ecrit:
> > digging into starpu I found that starpu_config.topology.workers_bindid[i] 
> > and
> > _starpu_config.workers[i].bindid are not consistent; is this normal?
>
> workers_bindid is either set explicitly in the conf passed to
> starpu_init(), or filled from the STARPU_WORKERS_CPUID environment
> variable, i.e. it is only what was requested, while workers[i].bindid
> is what is actually used during execution, which might differ if a GPU
> worker got bound to one of them for instance.
>
> Samuel



-- 
-----------------------------------------
Alfredo Buttari, PhD
CNRS-IRIT
2 rue Camichel, 31071 Toulouse, France
http://buttari.perso.enseeiht.fr