Tonight or some time tomorrow, I'll check in a rewritten PCI target
that is designed to be synthesizable.  I have to run it through more
testing to make sure I didn't break anything in the rewrite, but in
simple tests, it still works.  In the process, I noticed and fixed
some weirdnesses and inconsistencies, so this one won't be
clock-for-clock identical for the simulation model.  It may be
appropriate to back-port the changes into the simulation model.  The
synthesizable version is truly an unreadable sight to be beheld.  No
amount of commenting will make the code "easy to understand," to
anyone, even myself, and I wrote it.

I've done my timing analysis using the Xilinx tools (because that's
what I have for Linux), and for a Spartan 3, the best it can do is
about 75MHz.  This all comes down to I/O buffer delays, and it'll be
better or worse in the Lattice part, depending on how fast its I/O
buffers are.  Howard may be able to work some magic to improve things
a little further, but the only effect will be to make it easier to hit
66MHz when more logic is included in the same chip.

Because the hopes of doing 133MHz are shattered, I'm sorely tempted to
try to remove the extra wait state on address decode.  It'll make PIO
accesses slightly faster.  On the other hand, since we generally
discourage PIOs anyway, in favor of DMA, I'm not sure I should even
worry about it.

--
Timothy Miller
http://www.cse.ohio-state.edu/~millerti
_______________________________________________
Open-graphics mailing list
[email protected]
http://lists.duskglow.com/mailman/listinfo/open-graphics
List service provided by Duskglow Consulting, LLC (www.duskglow.com)

Reply via email to