View Single Post
Old 24 March 2016, 23:00   #11
Join Date: Jan 2010
Location: Kansas
Posts: 1,284
Originally Posted by eXeler0 View Post
However, when the FPU *does* appear in the apollo-core, I expect it to be much faster than the one in 040/060.
I worry more about the FPU compatibility than the speed. I would prefer an extended precision FPU rather than a double precision FPU which was last I heard (faster and saves logic but potentially less compatible). I doubt the loss of precision would affect the Quake source but it could very well affect fp support like the 68040sp/68060sp/mu68040.library/mu68060.library and compiler support which may need major rewrites for the Apollo core FPU (algorithms for extended precision can be simpler). When I left, it didn't sound like there would be enough room in the FPGA for all 68060 FPU instructions. Missing instructions could be trapped but would further reduce performance for existing code. Also, I have seen no attempt to create or document standards or gain support for changes outside of a few people.

Originally Posted by eXeler0 View Post
(Also, by then we might have some other interesting screenmodes to try out. I believe Apollo also has superior memory bandwidth and speed compared to something like Permedia 2 cards of the time (AGP1 maxes out at 250MB/s or something like that, and most Cards had 4 or 8MB RAM))
However, to get the most out of the Apollo-core I suspect parts would need to be re-written in assembler to make full use of the additional Apollo instructions?
SIMD/vector instructions could give a huge speed boost to the Apollo gfx. They would be like a fast Amiga blitter but with more versatility. The performance potential of enhancements is definitely there but Gunnar's "standards" seem to always change, never are documented and have no 3rd party support. He is a one man show and you get what he gives you.
matthey is offline  
Page generated in 0.03962 seconds with 10 queries