From: Mike G. <xo...@gm...> - 2010-07-23 08:58:52
|
Okay, well I'd definitely prefer feature support over performance (esp. directional light support). However in any case, if you've already identified areas which are open to parallelism it would be good for you to document these on the wiki to make life easier for anybody else who may want to work on this in future :) If you decide to go down the speed improvement route, I'd prefer OpenCL to just CPU multithreading too, in theory you should be able to get a much bigger maximum performance gain. OpenCL runs on the CPU as well (with no particular requirements other than a fairly modern chip AFAIK) so people without a GPGPU would still get some benefits. On 23 July 2010 06:35, Mohit Taneja <moh...@gm...> wrote: > @Mike : I ain't sure but I guess it shouldn't be very difficult. > > @res: Sure, simply running multiple threads to explore would be much easier > to implement, rather than using OpenCL, but I guess during our application > discussion period, most of the people were in favour of using OpenCL or CUDA > (more preferrence on OpenCL as they could explore ATI cards too), and in > principal they would be better even, as one could explore more computing > power. > > Regarding user friendliness, I strongly agree with you because for a first > time user it is generally very confusing to end up filling a random number > for most of the PM settings and then figure out that which one is best for > him/her (this used to happen with me too in the beginning), but I guess it > would be pretty tough to have hueristics to figure out those values. > > >> And: does PM work together with directional lightmaps? And 'specular' >> lightmaps? >> >> > I dont have an idea about this, probably Scott would be having a better > idea about this. > > -- - Mike |