-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
On Tuesday 24 August 2004 18:01, Ronny V. Vindenes wrote:
tir, 24,.08.2004 kl. 13.40 -0700, skrev Ian Romanick:
Ronny V. Vindenes wrote: Of course, once we get to that point, I think we should get the existing TCL paths and run everything as a vertex program. I guess we can cross (and debate) that bridge when we come to it...
I agree, my concern was that particularly games test for certain extensions then pick a suitable code path, either because it's theoretically faster or because it allows more/better eye-candy, under the assumption that any supported extension is fully accelerated. I don't know if GL_ARB_vertex_program is actually used like that, it just happened to make me think (out loud).
I've now briefly tested Philipp's patches on x86-64; didn't see anything break and the GL_ARB_vertex_program tests in progs/{demos,tests} all seemed to work.
NV_vertex_program however...
The codegen version of NV_vertex_program is busted.
The code you're referring to is only supposed to work with fragment programs, and ARB_fragment_program in particular.
It's experimental in nature and will break in many many different ways, even for ARB_fragment_program.
Let me list some of them:
- It's not thread safe.
- It probably won't work with programs with more than one fragment program.
- TCC has all sorts of odd issues being used this way, and blows up spectacularly sometimes.
Also, from a performance point of view, what this code does highlight when it works is that the actual execution of fragment programs is only a fraction of the overhead we encounter when running ARB_fp programs. We lose as much or more time simply as a result of the non-optimal triangle scan-conversion, construction of span data, etc, which is all done in a very general way when running fragment programs. These issues need to be addressed before its worthwhile investing any great effort in the FP execution itself.
Ideally, you'd codegen the whole triangle routine, but there might be a good intermediate tradeoff point.
Keith
------------------------------------------------------- SF.Net email is sponsored by Shop4tech.com-Lowest price on Blank Media 100pk Sonic DVD-R 4x for only $29 -100pk Sonic DVD+R for only $33 Save 50% off Retail on Ink & Toner - Free Shipping and Free Gift. http://www.shop4tech.com/z/Inkjet_Cartridges/9_108_r285 -- _______________________________________________ Dri-devel mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/dri-devel
