Re: [Starpu-devel] Failed assert in copy_driver.c

[Xavier Lacoste <xl64100@gmail.com>]

Hello again,

I went to 1.1 and got similar errors on some runs.
oops in copy_data_1_to_1_generic (datawizard/copy_driver.c:211)... 4: 
unspecified launch failure 

Moreover, 1.2 gives more FLOPS than 1.1 on my application (which is a good 
evolution :D). 
So I'll stay to 1.2 and rerun the failing runs which with some luck will not 
fail again.

For this experiment its not a problem rerunning the experiment until it does 
not fail, but when I'll have to use PaStiX with StarPU with GPUs in a "real 
world" application it will be more problematic...

Regards,

XL.

Le 20 oct. 2014 à 15:26, Xavier Lacoste <xl64100@gmail.com> a écrit :

> Hello,
> 
> With changing MAX_PENDING_REQUESTS_PER_NODE and 
> MAX_PENDING_PREFETCH_REQUESTS_PER_NODE. I got:
> oops in copy_data_1_to_1_generic (datawizard/copy_driver.c:200)... 4: 
> unspecified launch failure 
> 
> With revision 13658 I got:
> oops in _starpu_driver_test_request_completion 
> (datawizard/copy_driver.c:729)... 4: unspecified launch failure 
> 
> I'll go to 1.1 and see.
> 
> Thanks,
> 
> XL.
> 
> Le 20 oct. 2014 à 14:11, Samuel Thibault <samuel.thibault@ens-lyon.org> a 
> écrit :
> 
>> Xavier Lacoste, le Mon 20 Oct 2014 13:56:08 +0200, a écrit :
>>> I could use 1.1, the only thing I use from 1.2 are the commutable tasks 
>>> but in my code. The use of this feature does not change much the timing 
>>> of the execution, so I can go 1.1...
>>> it's just because I used 1.2 on an other machine (when I tested the 
>>> removal of sequential consistency flag on often used buffers) and wanted 
>>> the same configuration.
>> 
>> OK :)
>> 
>> Samuel
> 
> _______________________________________________
> Starpu-devel mailing list
> Starpu-devel@lists.gforge.inria.fr
> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/starpu-devel

Attachment: signature.asc
Description: Message signed with OpenPGP using GPGMail