Jerome Glisse wrote: > On Fri, 24 Oct 2008 15:17:06 -0500 > Ioannis Papadopoulos <[EMAIL PROTECTED]> wrote: > > >> If I'm not mistaken, the critical path is basically operations on >> vectors/matrices - I'm not an expert on graphics, but I thought that's >> the reason why the GPUs look a lot like vector processors. >> > > Well today GPU are lot more than just vertex, fragment shader are becoming > more complex every year up to the point that they can be considered as > a program by themself. In the end you often schedule out a quad and keep > going on others quads while doing texture fetch or waiting for some others > resources. So we might want to schedule by ourself what we want to run > in order to hide latency like texture fetch (this what gpu scheduler does > basicly), it seems that openmp is good when you know the exact things > you are going to do at compile time so it can do the complexe thread > scheduling for you but in this case you only know at runtime what kind > of fragment program you will have to run. > > Cheers, > Jerome Glisse <[EMAIL PROTECTED]> > Nevertheless, I'm not dogmatic on OpenMP, it's just that I'd like to use it for rapid prototyping and see if I can get some speedup out of it in a simple dual core laptop, without having to spend time on creating a RTS that may be never used.
It is more important to define the granularity and what are the actual parts that can be parallelized (and is actually worth parallelizing them) than debating on to-OpenMP or not-to-OpenMP. Till now, my experience with OpenMP is that it allows you to parallelize something really fast - then you might make your own RTS if it is actually worth it. Giannis ------------------------------------------------------------------------- This SF.Net email is sponsored by the Moblin Your Move Developer's challenge Build the coolest Linux based applications with Moblin SDK & win great prizes Grand prize is a trip for two to an Open Source event anywhere in the world http://moblin-contest.org/redirect.php?banner_id=100&url=/ _______________________________________________ Mesa3d-dev mailing list Mesa3d-dev@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/mesa3d-dev
