search

clash-lang / clash-prelude

CLaSH prelude library containing datatypes and functions for circuit design
http://www.clash-lang.org/
Other
31 stars 27 forks source link

Extend alteraPll to include all 9 pllout ports #130

Closed martijnbastiaan closed 6 years ago

martijnbastiaan commented 7 years ago

This commits extends alteraPll to include of all of the available 9 output ports.

Should be merged in tandem with: https://github.com/clash-lang/clash-compiler/pull/261

expipiplus1 commented 7 years ago

Out of interest, do you actually have a design using 9 different clocks?

martijnbastiaan commented 7 years ago

No, I have yet to design a circuit with more than 2 actually :-). We (me+@christiaanb) settled on just extending it to 9, to prevent a zoo of alteraPll{0,1,2,3,4,5...}. It shouldn't bother users too much, as Clash 0.99 is backwards incompatible anyway.

Come to think of it.. this PR should probably extend the Xilinx PLLs as well.

christiaanb commented 7 years ago

@martijnbastiaan If I understand te documentation it seems that for the Stratix series alteraPll can have up to 18 ports, although it "only" scales up to 9 for the Cyclone series. I don't want to merge this before we have a better solution.

martijnbastiaan commented 7 years ago

In that case I can see a few options, but none of them seem that great.

  1. Introduce alteraPll, alteraPll9, and alteraPll18. A balance between maximum usability, and satisfying ease of use for simple (sample) circuits.
  2. Introduce KnownNat n => alteraPll n, where the function will yield a 18-tuple, but only the first n are usable. This would require some blackbox template magic (which is not possible right now AFAIK). Either we have to introduce more flexible if statements, or introduce for loops. Alternatively, we could outsource our templating to, for example, heist or ginger. (Which would also force us to declare all variables used in the template upfront, which would probably be nice. Although this is more a comment on the ergonomics of the current template engine than anything else.)
  3. Introduce KnownNat n => alteraPll n, where the function will return a Vec n. The downside being that all clocks in the vector will need to have the same domain, forcing the user to use unsafe functions to cast clocks to other domains. This would require templating magic similar to (2).
  4. Introduce alteraPll{1,2,3,4....}. Lets not do this.