StealthHawk
06-30-03, 09:34 PM
http://www.rage3d.com/#1056987428
Some of you might have noticed that when you're running the Dawn demo on your Radeon, Dawn doesn't have eyelashes. We recently received a new version of the wrapper which fixes that, and got a few additional comments from "The Guy" on how the wrapper works:
How it is done.
NV programs are parsed and tokenized then ARB programs are generated.
It is mostly straightforward instruction to instruction mapping.
Fragment programs in Dawn are short, mostly FX12 and no sweat for R300.
Reads from normalization cube map are detected and replaced with DP3/RSQ/MUL.
It improves quality because it removes graininess. Speedwise its about the same,
look at JC comments.
Vertex programs in Dawn are quite long, use lots of matrices, array
addressing and many vertex attributes. They map extremely well from NV to ARB,
even some ARB GL tokens have the same value as NV tokens.
Now that it renders everything there is no excuse for nvidia cards to be running Dawn slower than ATI cards :p
Some of you might have noticed that when you're running the Dawn demo on your Radeon, Dawn doesn't have eyelashes. We recently received a new version of the wrapper which fixes that, and got a few additional comments from "The Guy" on how the wrapper works:
How it is done.
NV programs are parsed and tokenized then ARB programs are generated.
It is mostly straightforward instruction to instruction mapping.
Fragment programs in Dawn are short, mostly FX12 and no sweat for R300.
Reads from normalization cube map are detected and replaced with DP3/RSQ/MUL.
It improves quality because it removes graininess. Speedwise its about the same,
look at JC comments.
Vertex programs in Dawn are quite long, use lots of matrices, array
addressing and many vertex attributes. They map extremely well from NV to ARB,
even some ARB GL tokens have the same value as NV tokens.
Now that it renders everything there is no excuse for nvidia cards to be running Dawn slower than ATI cards :p