Re: [Starpu-devel] Using STARPU_COMMUTE with starpu_mpi_task_build()

[Xavier Lacoste <xl64100@gmail.com>]

Ok the performances on one node are back (DMDA 88s, EAGER 90s), I accidentaly 
removed a 
"starpu_data_set_sequential_consistency_flag(blocktab_handles[iter], 0);" on 
a read only data used in all GEMM.

Two MPI nodes gives also encouraging results with 52s runtime.

Thanks,

XL.

Le 11 mars 2014 à 16:19, Xavier Lacoste <xl64100@gmail.com> a écrit :

> It was without GPUs but I must have broken something else in my application 
> (I have the same performances with old r12012 now...)
> 
> Le 11 mars 2014 à 14:17, Samuel Thibault <samuel.thibault@ens-lyon.org> a 
> écrit :
> 
>> Xavier Lacoste, le Tue 11 Mar 2014 12:30:54 +0100, a écrit :
>>> Indeed, the trunk seems to be slower with r13383 than with r12012 (820s 
>>> vs 100s
>>> on 12 cores). 
>> 
>> Is this without GPUs?
>> 
>> At any rate, please try trunk r12389, it should fix this kind of
>> performance regression.
>> 
>> Samuel
> 
> 
> _______________________________________________
> Starpu-devel mailing list
> Starpu-devel@lists.gforge.inria.fr
> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/starpu-devel