Functions in `Clash.Sized.Vector` should also be available for `BitVector`

[martijnbastiaan]

A user on the mailing list wanted to use the functions defined for Vec on BitVector. Unfortunately, on large vectors Clash can struggle to translate these constructs. We should therefore reimplement a lot of the functions defined for Vec for BitVector.

[jhmcstanton]

I'll pick this up. Clash.Sized.Vector exports quite a few functions, is there anything specific to prioritize?

Unfortunately, on large vectors Clash can struggle to translate these constructs

Is there an example usage somewhere of a vector large enough to cause a problem?

[christiaanb]

@jhmcstanton:

top of my head:

replicateB :: SNat n -> Bit -> BitVector n
mapB :: (Bit -> Bit) -> BitVector n -> BitVector n
zipWithB :: (Bit -> Bit -> Bit) -> BitVector n -> BitVector n -> BitVector n
foldlB :: (Bit -> Bit -> Bit) -> Bit -> BitVector n -> Bit
foldrB :: (Bit -> Bit -> Bit) -> Bit -> BitVector n -> Bit
iterateB :: SNat n -> (Bit -> Bit) -> Bit -> BitVector n
concatMapB :: (Bit -> BitVector n) -> BitVector m -> BitVector (n*m)
reverseB :: BitVector n -> BitVector n
scanlB :: (Bit -> Bit -> Bit) -> Bit -> BitVector n -> BitVector (n+1)
scanrB :: (Bit -> Bit -> Bit) -> Bit -> BitVector n -> BitVector (n+1)

Note that some of these must be implemented as "primitives", see http://hackage.haskell.org/package/clash-prelude-1.0.0/docs/Clash-Tutorial.html#g:14, to avoid the compile time "explosion" @martijnbastiaan was referring too. You can draw inspiration from Vec (see https://github.com/clash-lang/clash-compiler/blob/master/clash-lib/prims/vhdl/Clash_Sized_Vector.json) as to which functions must be actual primitives. Evaluation rules for these primitives must then be added to the evaluator,

clash-compiler/clash-ghc/src-ghc/Clash/GHC/Evaluator.hs

Line 118 in c5a4327

reduceConstant :: PrimEvaluator

, as well.

I could setup a branch tonight that just implements mapB, so you know what needs to be done for the rest of the functions. Unless you want to do it all by yourself, let me know.

[jhmcstanton]

I won't have a chance to dig around until this evening myself. If/when I run into issues I can let you know. Thanks for the guidance!

[jhmcstanton]

To clarify, don't worry about making your own branch just yet.

[christiaanb]

Bumping to 1.4 because I want to release 1.2 soon which contains a lot of fixes, and the implementation of this would take some effort: adding blackboxes, evaluator rules, etc.