Use Dict constructor for 1 and 2 arguments new_ordered_dict

[blegat]

Based on the following benchmark:

function slow_new_ordered_dict(::Type{K}, ::Type{V}, kv::Pair...) where {K,V}
    dict = OrderedDict{K,V}()
    sizehint!(dict, length(kv))
    for pair in kv
        JuMP.add_or_set!(dict, convert(K, pair.first), convert(V, pair.second))
    end
    return dict
end
function fast_new_ordered_dict(::Type{K}, ::Type{V}, kv::Pair) where {K,V}
    return OrderedDict{K,V}(kv)
end
function fast_new_ordered_dict(::Type{K}, ::Type{V}, kv1::Pair, kv2) where {K,V}
    if isequal(kv1.first, kv2.first)
        return OrderedDict{K,V}(kv1.first => kv1.second + kv2.second)
    else
        return OrderedDict{K,V}(kv1, kv2)
    end
end
using BenchmarkTools
using JuMP
function bench()
    model = Model()
    a = 2.0
    @variable(model, x)
    b = -1.0
    @variable(model, y)
    @btime slow_new_ordered_dict(JuMP.VariableRef, Float64, $x => $a)
    @btime fast_new_ordered_dict(JuMP.VariableRef, Float64, $x => $a)
    @btime slow_new_ordered_dict(JuMP.VariableRef, Float64, $x => $a, $y => $b)
    @btime fast_new_ordered_dict(JuMP.VariableRef, Float64, $x => $a, $y => $b)
    @btime slow_new_ordered_dict(JuMP.VariableRef, Float64, $x => $a, $x => $b)
    @btime fast_new_ordered_dict(JuMP.VariableRef, Float64, $x => $a, $x => $b)
end
bench()

which gives

  266.433 ns (8 allocations: 624 bytes)
  233.452 ns (7 allocations: 592 bytes)
  324.908 ns (9 allocations: 656 bytes)
  291.885 ns (9 allocations: 656 bytes)
  313.509 ns (9 allocations: 656 bytes)
  249.245 ns (9 allocations: 656 bytes)

The speedup is smaller than I would expect but it is still noticeable.

The timing with speed.jl is:

P-Median(100 facilities, 100 customers, 5000 locations) benchmark:
BenchmarkTools.Trial: 
  memory estimate:  714.37 MiB
  allocs estimate:  12030567
  --------------
  minimum time:     1.779 s (51.12% GC)
  median time:      1.941 s (52.86% GC)
  mean time:        2.082 s (56.42% GC)
  maximum time:     2.527 s (62.89% GC)
  --------------
  samples:          3
  evals/sample:     1
Cont5(n=500) benchmark:
BenchmarkTools.Trial: 
  memory estimate:  402.77 MiB
  allocs estimate:  6785839
  --------------
  minimum time:     849.008 ms (39.81% GC)
  median time:      955.320 ms (42.55% GC)
  mean time:        951.107 ms (43.90% GC)
  maximum time:     1.056 s (48.26% GC)
  --------------
  samples:          6
  evals/sample:     1

which is a slight improvement compared to the timing reported in #1639

[codecov]

Codecov Report

Merging #1640 into master will increase coverage by <.01%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master    #1640      +/-   ##
==========================================
+ Coverage   90.85%   90.86%   +<.01%     
==========================================
  Files          28       28              
  Lines        3751     3755       +4     
==========================================
+ Hits         3408     3412       +4     
  Misses        343      343

Impacted Files	Coverage Δ
src/aff_expr.jl	94.95% <100%> (+0.17%)	⬆️

---

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 616ad67...7c491c6. Read the comment docs.