On Mon, Mar 19, 2012 at 6:14 AM, Christopher Sean Morrison <[email protected]> wrote: > > On Mar 18, 2012, at 8:32 PM, Christopher Sean Morrison wrote: > >> A couple ideas that immediately come to mind would be to implement an >> OpenCL-based ray-bundle shotline intersection routine for our LIBRT >> raytracing library. > > I just realized that was only one idea.. ;) > > Another idea would be to convert our raytracer into a phase-based tracer > where all rays are fired, then all booleans are weaved, then all all optics > are computed, then all pixels are colored. With that done, the boolean > weaving would benefit substantially from an OpenCL refactoring. > I am finally thinking of going primarily with the idea of optimizing LIBRT with OpenCL. And I plan to complete the proposal in a day or two.
Now, If we look at first idea. Will I be basically dealing with /src/librt/bundle.c and shooting a grid rays in batches to the the device program of OpenCL? What all other modules will need to support this? (like as you mentioned each primitives will need to support this independently). With the second idea, boolean weaving is a bit unclear. What I could make of it is, to implement the raytracer in phases with some sort of a pipline where all are fired first, then weaved, then optics computed, then finally pixels are colored. This would involve substantial work in various libraries (librt, liboptical, etc), writing more than one device program for different phases. Am I making sense? If you could give some links where I could read about it more, it will be more than helpful. Regards, Aniket ------------------------------------------------------------------------------ This SF email is sponsosred by: Try Windows Azure free for 90 days Click Here http://p.sf.net/sfu/sfd2d-msazure _______________________________________________ BRL-CAD Developer mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/brlcad-devel
