improve concrete-foldability of core math functions

[aviatesk]

By making more use of the @assume_effects annotation. This commit adds
several :consistent annotations to certain functions that access to
constant global table (e.g. INV_2PI and J_TABLE), because currently
our effect analysis can't prove the consistency in the presence of
@inbounds and :boundscheck.

In the long term, we can revert this commit once we improve the effect
analysis so that it can properly reason about safe tuple indexing.

---

• these commits should be merged separately
• @nanosoldier runbenchmarks(ALL, vs=":master")

[oscardssmith]

This looks good to me.

[nanosoldier]

Something went wrong when running your job:

NanosoldierError: error when preparing/pushing to report repo: failed process: Process(setenv(`git push`; dir="/nanosoldier/workdir/NanosoldierReports"), ProcessExited(1)) [1]

Unfortunately, the logs could not be uploaded.

[aviatesk]

@Keno could I hear your thought on this? Especially I'm wondering about:

because currently our effect analysis can't prove the consistency in the presence of @inbounds and :boundscheck.

In the long term, we can revert this commit once we improve the effect analysis so that it can properly reason about safe tuple indexing.

[vtjnash]

https://github.com/JuliaCI/NanosoldierReports/blob/master/benchmark/by_hash/9be47a4_vs_54b92a7/report.md

[aviatesk]

Seems like some of scalar benchmarks get improved by these annotations.

[vtjnash]

win32 bots appear to be mad at this PR

      From worker 6:	┌ Error: bad effects found for sin(::Float32)
      From worker 6:	│   eff = (+c,+e,?n,?t,+s)
      From worker 6:	└ @ Main.Test85Main_math C:\buildbot\worker-tabularasa\tester_win32\build\share\julia\test\math.jl:1464
      From worker 6:	┌ Error: bad effects found for cos(::Float32)
      From worker 6:	│   eff = (+c,+e,?n,?t,+s)
      From worker 6:	└ @ Main.Test85Main_math C:\buildbot\worker-tabularasa\tester_win32\build\share\julia\test\math.jl:1464
      From worker 6:	┌ Error: bad effects found for tan(::Float32)
      From worker 6:	│   eff = (+c,+e,?n,?t,+s)
      From worker 6:	└ @ Main.Test85Main_math C:\buildbot\worker-tabularasa\tester_win32\build\share\julia\test\math.jl:1464
math                                     (6) |         failed at 2022-06-11T03:45:08.240
Test Failed at C:\buildbot\worker-tabularasa\tester_win32\build\share\julia\test\math.jl:1465
  Expression: false
Test Failed at C:\buildbot\worker-tabularasa\tester_win32\build\share\julia\test\math.jl:1465
  Expression: false
Test Failed at C:\buildbot\worker-tabularasa\tester_win32\build\share\julia\test\math.jl:1465
  Expression: false

[vchuravy]

As well as gnuassert https://buildkite.com/julialang/julia-master/builds/12828#01815e2f-ebc6-4410-9e4e-dd29dd1c8e90

[aviatesk]

These failures seem to happen non-deterministically... For now I'd like to just skip these tests on x86 machines.