Re: [Starpu-devel] Assert failure with regenerate flag enabled

[Jeff Hand <jeffrey.hand@gmail.com>]

Experimenting with regenerate etc. I found the following...

https://gforge.inria.fr/scm/viewvc.php/tags/starpu-1.1.3/tests/Makefile.am?view=markup&revision=13471&root=starpu

- change 9918 omited

https://gforge.inria.fr/scm/viewvc.php/trunk/tests/Makefile.am?root=starpu&view=log

- has change 9918

As a result of change 9918,

=================================================
am__main_subgraph_repeat_regenerate_SOURCES_DIST =  \
    main/subgraph_repeat.c main/increment.cu
=================================================

This causes subgraph_repeat.c to be compiled instead of subgraph_repeat_regenerate.c. 

Was this intentional?

On Wed, Nov 19, 2014 at 10:22 AM, Jeff Hand <jeffrey.hand@gmail.com> wrote:

On Wed, Nov 19, 2014 at 4:54 AM, Samuel Thibault <samuel.thibault@ens-lyon.org> wrote:

Jeff Hand, le Tue 18 Nov 2014 09:45:59 -0600, a écrit :
> On a slightly different topic, I'm wondering: How can dependencies be changed?

> I'm not aware of a way change or clear the dependencies.

We haven't really thought about an interface for this.

With the chain A->B->C->D, it will likely be a long term requirement to change to A->C->B->D quickly. 

Why was I using dependencies in the first place?
1. make D dependant on A rather than the previous task (priming the pipeline)
2. declaring the dependency of a task output routed to multiple task inputs

Is the solution to enable data consistency and have the callback submit the next task in the pipeline?

Can I guarantee the dependencies of the output to many inputs occurs?  Would it require a StarPU pause, submit the many tasks, and resume StarPU?