Add support for digits keyword argument of round()

[kimikage]

This only supports the non-negative digits option. All other options throw an error.
This provides specialized implementations only for N0f8 and Fixed{Int8}. The other types use fallback with floating-point operations and therefore it is slower and less accurate.

Fixes #234

[codecov]

Codecov Report

Merging #235 into master will increase coverage by 0.23%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #235      +/-   ##
==========================================
+ Coverage   96.24%   96.47%   +0.23%     
==========================================
  Files           6        6              
  Lines         692      738      +46     
==========================================
+ Hits          666      712      +46     
  Misses         26       26              

Impacted Files	Coverage Δ
src/FixedPointNumbers.jl	96.69% <100.00%> (+0.20%)	⬆️
src/fixed.jl	98.11% <100.00%> (+0.19%)	⬆️
src/normed.jl	95.94% <100.00%> (+0.25%)	⬆️

---

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 f57e96b...23aa8e5. Read the comment docs.

[kimikage]

Benchmark

using ColorTypes, BenchmarkTools
image = rand(RGB{N0f8}, 1000, 1000);

function quantize(x::N0f8)
    floor(x, digits=1)
end
function quantize(color::AbstractRGB)
    mapc(quantize, color)
end


function quantize_f(x::N0f8)
    floor(float(x), digits=1) % N0f8
end
function quantize_f(color::AbstractRGB)
    mapc(quantize_f, color)
end

julia> @benchmark quantize.($image)
BenchmarkTools.Trial:
  memory estimate:  2.86 MiB
  allocs estimate:  2
  --------------
  minimum time:     1.504 ms (0.00% GC)
  median time:      1.532 ms (0.00% GC)
  mean time:        1.731 ms (10.89% GC)
  maximum time:     5.461 ms (71.47% GC)
  --------------
  samples:          2888
  evals/sample:     1

julia> @benchmark quantize_f.($image)
BenchmarkTools.Trial:
  memory estimate:  2.86 MiB
  allocs estimate:  2
  --------------
  minimum time:     18.977 ms (0.00% GC)
  median time:      19.034 ms (0.00% GC)
  mean time:        19.298 ms (1.17% GC)
  maximum time:     22.876 ms (16.72% GC)
  --------------
  samples:          260
  evals/sample:     1

n4f4  = rand(N4f4,  1000);
n0f8  = rand(N0f8,  1000);
n0f16 = rand(N0f16, 1000);

q3f4  = rand(Q3f4,  1000);
q0f7  = rand(Q0f7,  1000);
q0f15 = rand(Q0f15, 1000);

julia> @btime trunc.($n4f4,  digits=1); # not optimized
  13.199 μs (1 allocation: 1.06 KiB)

julia> @btime trunc.($n0f8,  digits=1);
  161.658 ns (1 allocation: 1.06 KiB)

julia> @btime trunc.($n0f8,  digits=2);
  158.163 ns (1 allocation: 1.06 KiB)

julia> @btime trunc.($n0f16, digits=2); # not optimized
  15.100 μs (1 allocation: 2.13 KiB)

julia> @btime trunc.($q3f4,  digits=1);
  221.577 ns (1 allocation: 1.06 KiB)

julia> @btime trunc.($q0f7,  digits=1);
  126.801 ns (1 allocation: 1.06 KiB)

julia> @btime trunc.($q0f7,  digits=2);
  222.530 ns (1 allocation: 1.06 KiB)

julia> @btime trunc.($q0f15, digits=2); # not optimized
  5.767 μs (1 allocation: 2.13 KiB)

[kimikage]

This changes the behavior as a result, but I think this is rather a bug fix. So, I will merge this.