improve performance of parse

[KristofferC]

Improves performance by preventing everything getting lifted to the Char domain.

Benchmark

function parseintperf(t)
   local n, m
   for i=1:t
       n = rand(UInt32)
       s = string(n, base = 16)
       m = UInt32(parse(Int64,s, base = 16))
       @assert m == n
   end
   return n
end


strings = [string(rand(UInt32)) for i in 1:10^4];
f(strings) = for s in strings; parse(Int, s) end;

Before

julia> @btime parseintperf(1000);
  189.958 μs (2999 allocations: 109.36 KiB)

julia> @btime f(strings)
  848.226 μs (0 allocations: 0 bytes)

After

julia> @btime parseintperf(1000)
  170.692 μs (2999 allocations: 109.36 KiB)

julia> @btime f(strings)
  709.092 μs (0 allocations: 0 bytes)

Still not completely caught up with 0.6 for parseintperf but getting close.

[MasonProtter]

What were the numbers for 0.6, for comparison?

[StefanKarpinski]

Ideally this should really be necessary since the comparison operation between characters is just an integer comparison so it seems like something else is not quite right here if this is necessary for good performance.

[KristofferC]

It is not the comparison that is expensive, it is the arithmetic (- in this case), see #28671.

However, hoisting the conversion like this still has a ~4% performance improvement so might as well leave it?

[KristofferC]

@MasonProtter

0.6:

julia> @btime parseintperf(1000);
  143.515 μs (2000 allocations: 93.75 KiB)

julia> @btime f(strings)
  793.624 μs (0 allocations: 0 bytes)

[quinnj]

Parsers.jl hereby challenges you to a duel:

julia> strings = [string(rand(UInt32)) for i in 1:10^4];

julia> f(strings) = for s in strings; Parsers.parse(s, Int) end;

julia> @btime f(strings)
  549.167 μs (30000 allocations: 1.37 MiB)

[KristofferC]

# Base
julia> @btime f(strings)
  whatevs (0 allocations: 0 bytes)

# Parsers.jl
julia> @btime f(strings)
  whatevs (30000 allocations: 1.37 MiB)

Well?

[KristofferC]

Even though this comes with a small readability regression, I think it is worth it for such a basic function (we also show parseintperf on the benchmark plot on the homepage).

[KristofferC]

@quinnj Getting closer ;)

julia> @btime f(strings)
  595.071 μs (0 allocations: 0 bytes)