I would like to second Reza's question about what to do when it
starts looking like your project may need wider, faster computation?
Specifically does anyone have suggestions about how to get the
computation done and also leverage a team's experience with avr-gcc
style development? I suppose that you could gang several AVRs
together with a shared memory architecture for example.
A key feature of avr-gcc that I need to keep is that the whole
build process is scriptable. I am just getting into the Xilinx tool
flow, but when I read through example projects that are basically all
screen shots of mousing around Windows GUIs I get a little nervous
that development would have to slow down to accommodate that style of
development.