Re: [Starpu-devel] StarPU: questions about performance model

[Leo Li <luli757@student.liu.se>]

Hi, Sameul

I want to ask a silly question.

When I use @code{starpu_task_expected_length} from external
application, I have link error  undefined reference to
`starpu_task_expected_length(starpu_task*, starpu_perf_archtype)'.
Could you tell me how to use this function from an external
application?

Thanks in advance.


On Fri, Jun 17, 2011 at 11:51 AM, Samuel Thibault
<samuel.thibault@ens-lyon.org> wrote:
> Hello,
>
> Leo Li, le Fri 17 Jun 2011 11:43:20 +0200, a écrit :
>> 1) There is one API(starpu_load_history_debug) which can be used to
>> load history-based performance model. If the performance model have
>> multiple entries for cpu, and given the size and layout of parameters
>> I need to find the right entry. Does anybody know how to find the
>> right entry according to size and layout of parameters? Which APIs
>> should I use?
>>
>> 2) Could someone send me more information or an example about
>> regression model? How does it work?
>
> I have just added these yesterday to the documentation:
>
> “For instance, @code{tests/perfmodels/regression_based.c} uses a
> regression-based performance model for the @code{memset} operation”
>
> “The application can request time estimations from the StarPU performance
> models by filling a task structure as usual without actually submitting
> it. The data handles can be created by calling @code{starpu_data_register}
> functions with a @code{NULL} pointer (and need to be unregistered as usual)
> and the desired data sizes. The @code{starpu_task_expected_length} and
> @code{starpu_task_expected_power} functions can then be called to get an
> estimation of the task duration on a given arch. @code{starpu_task_destroy}
> needs to be called to destroy the dummy task afterwards.”
>
> Samuel
>



-- 
Faithful
Lu Li


Contact: +46704759692