performance suggestions

[johnnychen94]

The quantizers could possibly be improved, but doesn't have the bandwidth to check it 😢

---

Use MappedArrays to avoid allocating buffers for res at each iteration. This could be GC-heavy if the image itself is large.

GIFImages.jl/src/decode.jl

Lines 62 to 65 in c3b84d0

	res = map(x -> RGB{N0f8}(x.Red / 255, x.Green / 255, x.Blue / 255), colortypes)
	res = reshape(res, Int(loaded_gif.SWidth), Int(loaded_gif.SHeight))
	res = res'
	final[:, :, i] = res

[] is equivalent to Any[] and is slow. Use Concrete types, e.g., UInt8[]?

GIFImages.jl/src/encode.jl

Line 35 in c3b84d0

colors = []

colors should also be a concret type Dict{RGB{N0f8}, UInt8}()

GIFImages.jl/src/encode.jl

Lines 48 to 51 in c3b84d0

	mapping = Dict()
	for (i, j) in enumerate(colors)
	mapping[j] = UInt8(i-1)
	end

Computating order matters. Here we want to make sure pix is a dense memory with valid pointer. But we can compute it in one line, e.g., map(x->colors[x], @view(img[:, :, i])'). -- Vector and Matrix should be the same when you get the pointer of it so vec is needless.

GIFImages.jl/src/encode.jl

Lines 67 to 68 in c3b84d0

	img1 = vec(collect(@view(img[:, :, i])'))
	pix = map(x -> mapping[x], img1)