allow type of x and w to differ in conv

[jbrea]

Trying to address #39

[MikeInnes]

Thanks! Given that the type parameters are now not used in most cases, these can be written as A::AbstractArray, B::AbstractArray; hope you don't mind making that quick change.

[jbrea]

I simplified some instances, as you suggested. Do you prefer to do this replacement everywhere?
maxpool!(y::A, x::A, k; kw...) where A<:AbstractArray looks nicely succinct.

[JobJob]

IMO makes sense to make the same change everywhere relevant in conv.jl. Restricting args to be the exact same type sounds like it was just a historical accident, and similar for many AbstractArray types returns a different type (often a straight up Array) - which is relevant e.g. here:

NNlib.jl/src/conv.jl

Lines 154 to 157 in 085adb7

	maxpool(x::AbstractArray, k; pad = map(_->0,k), stride = k) =
	maxpool!(similar(x, pdims(size(x), k, expand(Val{length(k)}, pad),
	expand(Val{length(k)}, stride))), x, k, pad = expand(Val{length(k)}, pad),
	stride = expand(Val{length(k)}, stride))

[MikeInnes]

Agreed. I'll merge this once tests pass, and I'm happy to take more patches to clean this up.

[codecov-io]

Codecov Report

Merging #63 into master will not change coverage.
The diff coverage is 100%.

@@          Coverage Diff           @@
##           master     #63   +/-   ##
======================================
  Coverage    71.4%   71.4%           
======================================
  Files           9       9           
  Lines         619     619           
======================================
  Hits          442     442           
  Misses        177     177

Impacted Files	Coverage Δ
src/conv.jl	66.66% <100%> (ø)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update a470e99...b696529. Read the comment docs.