Use tail-call function-barrier in foldl for array

[tkf]

Performance of bench_missing_argmax.jl can be improved by using tail-call function-barrier.

However, it breaks inference of foldl(right, TeeZip(Filter(isodd) |> Map(inc)), xs): https://travis-ci.com/tkf/Transducers.jl/jobs/226136474#L321

This patch probably is not a good idea since TeeZip is pretty important.

Running bench_missing_argmax.jl

runv(cmd) = (println('$', " ", cmd); run(cmd); println())
runv(`git rev-parse HEAD`)
runv(`git rev-parse --abbrev-ref HEAD`)
runv(`git status --short`)

using BenchmarkTools
results = run(include(joinpath(pwd(), "benchmark/bench_missing_argmax.jl")))
show(IOContext(stdout, :compact => false), results)
println()
println("xf vs man")
display(judge(minimum.((results["xf"], results["man"]))...))
println("rf vs man")
display(judge(minimum.((results["rf"], results["man"]))...))

master

f87867e

$ `git rev-parse HEAD`
f87867ef9d15abdfb5a10e4ea6c5f34fd21f610b

$ `git rev-parse --abbrev-ref HEAD`
master

$ `git status --short`

┌ Warning: `Initializer(T -> ...)` is deprecated.  Please use `OnInit(() -> ...)`.
│   caller = argext_init at bench_missing_argmax.jl:28 [inlined]
└ @ Core ~/.julia/dev/Transducers/benchmark/bench_missing_argmax.jl:28
┌ Warning: `Initializer(T -> ...)` is deprecated.  Please use `OnInit(() -> ...)`.
│   caller = argext_init(::Function) at bench_missing_argmax.jl:28
└ @ Main.BenchMissingArgmax ~/.julia/dev/Transducers/benchmark/bench_missing_argmax.jl:28
3-element BenchmarkTools.BenchmarkGroup:
  tags: []
  "xf" => BenchmarkTools.Trial:
          memory estimate:  32 bytes
          allocs estimate:  1
          --------------
          minimum time:     1.156 μs (0.00% GC)
          median time:      1.289 μs (0.00% GC)
          mean time:        4.194 μs (0.00% GC)
          maximum time:     28.787 ms (0.00% GC)
          --------------
          samples:          10000
          evals/sample:     1
  "rf" => BenchmarkTools.Trial:
          memory estimate:  32 bytes
          allocs estimate:  1
          --------------
          minimum time:     1.160 μs (0.00% GC)
          median time:      1.303 μs (0.00% GC)
          mean time:        4.756 μs (10.70% GC)
          maximum time:     28.718 ms (0.00% GC)
          --------------
          samples:          10000
          evals/sample:     1
  "man" => BenchmarkTools.Trial:
          memory estimate:  32 bytes
          allocs estimate:  1
          --------------
          minimum time:     1.012 μs (0.00% GC)
          median time:      1.130 μs (0.00% GC)
          mean time:        2.021 μs (0.00% GC)
          maximum time:     8.435 ms (0.00% GC)
          --------------
          samples:          10000
          evals/sample:     1
xf vs man
BenchmarkTools.TrialJudgement:
  time:   +14.23% => regression (5.00% tolerance)
  memory: +0.00% => invariant (1.00% tolerance)
rf vs man
BenchmarkTools.TrialJudgement:
  time:   +14.62% => regression (5.00% tolerance)
  memory: +0.00% => invariant (1.00% tolerance)

noouttype (#31)

36f8211

+ git checkout noouttype
Switched to branch 'noouttype'
Your branch is up to date with 'origin/noouttype'.
+ julia-dev --startup-file=no --color=yes /home/takafumi/junk/2019/08/18-094411.run-bench_missing_argmax.jl
$ `git rev-parse HEAD`
36f82116729b958a56b3e6828051514a1ee8f4a8

$ `git rev-parse --abbrev-ref HEAD`
noouttype

$ `git status --short`

3-element BenchmarkTools.BenchmarkGroup:
  tags: []
  "xf" => BenchmarkTools.Trial:
          memory estimate:  32 bytes
          allocs estimate:  1
          --------------
          minimum time:     2.125 μs (0.00% GC)
          median time:      2.529 μs (0.00% GC)
          mean time:        8.633 μs (0.00% GC)
          maximum time:     60.267 ms (0.00% GC)
          --------------
          samples:          10000
          evals/sample:     1
  "rf" => BenchmarkTools.Trial:
          memory estimate:  32 bytes
          allocs estimate:  1
          --------------
          minimum time:     2.268 μs (0.00% GC)
          median time:      2.633 μs (0.00% GC)
          mean time:        8.413 μs (5.80% GC)
          maximum time:     51.188 ms (0.00% GC)
          --------------
          samples:          10000
          evals/sample:     1
  "man" => BenchmarkTools.Trial:
          memory estimate:  32 bytes
          allocs estimate:  1
          --------------
          minimum time:     1.042 μs (0.00% GC)
          median time:      1.355 μs (0.00% GC)
          mean time:        3.090 μs (19.96% GC)
          maximum time:     9.498 ms (0.00% GC)
          --------------
          samples:          10000
          evals/sample:     1
xf vs man
BenchmarkTools.TrialJudgement:
  time:   +103.93% => regression (5.00% tolerance)
  memory: +0.00% => invariant (1.00% tolerance)
rf vs man
BenchmarkTools.TrialJudgement:
  time:   +117.66% => regression (5.00% tolerance)
  memory: +0.00% => invariant (1.00% tolerance)

tail-call-function-barrier (this PR)

40a9f5c

$ `git rev-parse HEAD`
40a9f5c1c2cf60d36bbeaaa725f432ce80e84e04

$ `git rev-parse --abbrev-ref HEAD`
tail-call-function-barrier

$ `git status --short`

3-element BenchmarkTools.BenchmarkGroup:
  tags: []
  "xf" => BenchmarkTools.Trial:
          memory estimate:  32 bytes
          allocs estimate:  1
          --------------
          minimum time:     1.220 μs (0.00% GC)
          median time:      1.362 μs (0.00% GC)
          mean time:        6.334 μs (0.00% GC)
          maximum time:     49.570 ms (0.00% GC)
          --------------
          samples:          10000
          evals/sample:     1
  "rf" => BenchmarkTools.Trial:
          memory estimate:  32 bytes
          allocs estimate:  1
          --------------
          minimum time:     1.222 μs (0.00% GC)
          median time:      1.445 μs (0.00% GC)
          mean time:        7.110 μs (0.00% GC)
          maximum time:     56.443 ms (0.00% GC)
          --------------
          samples:          10000
          evals/sample:     1
  "man" => BenchmarkTools.Trial:
          memory estimate:  32 bytes
          allocs estimate:  1
          --------------
          minimum time:     1.004 μs (0.00% GC)
          median time:      1.105 μs (0.00% GC)
          mean time:        1.986 μs (0.00% GC)
          maximum time:     8.331 ms (0.00% GC)
          --------------
          samples:          10000
          evals/sample:     1
xf vs man
BenchmarkTools.TrialJudgement:
  time:   +21.51% => regression (5.00% tolerance)
  memory: +0.00% => invariant (1.00% tolerance)
rf vs man
BenchmarkTools.TrialJudgement:
  time:   +21.71% => regression (5.00% tolerance)
  memory: +0.00% => invariant (1.00% tolerance)