faster mapfoldl for tuples

[stevengj]

Fixes #30465: should speed up mapreduce, sum, and many other functions on tuples.

To do:

• Tests.

[stevengj]

2x speedup on @btime sum(abs, $((rand(5)...,))); compared to Julia 1.0, for example.

[stevengj]

Also a huge speedup for heterogeneous tuples. If we define gsum(x...) = sum(y -> abs2(float(y)), x), for example, then I get the following on Julia 1.0

julia> @btime gsum(1.,2.,3.,4.,5.,6,7.,8.)  # notice that 6th argument is an Int
  179.733 ns (13 allocations: 656 bytes)
204.0

whereas I get 1.89ns and 0 allocations with this PR, a 100x speedup.

@cossio, after this PR we might want to change hypot in #30301 to get rid of the promote call and to simply call sum and maximum on the (possibly) heterogeneous tuples, since they should now hopefully be fast.

[cossio]

Awesome! We can wait for this to get merged, or else I can do that in a separate PR.

[bramtayl]

ref #28614

[bramtayl]

aka this is the tip of the iceberg

[stevengj]

Okay to merge?

[fredrikekre]

@nanosoldier runbenchmarks(ALL, vs = ":master")

[stevengj]

@fredrikekre, is nanosoldier not running? I don't see any output.

[fredrikekre]

Hmm, lets try again:
@nanosoldier runbenchmarks(ALL, vs = ":master")

[stevengj]

Still nothing.

[fredrikekre]

@ararslan ?

[ararslan]

The server dies periodically with a "too many open files" error, which is what happened here, so I've restarted it. @nanosoldier runbenchmarks(ALL, vs=":master")

[nanosoldier]

Your benchmark job has completed - possible performance regressions were detected. A full report can be found here. cc @ararslan

[stevengj]

The nanosoldier benchmark differences seem to be in the noise?

Although BaseBenchmarks includes some tuple benchmarks, it seems to use its own mapreduce implementation for some reason, so it shouldn't be affected by this PR anyway. (cc @KristofferC)

[KristofferC]

I'm not sure if the base version was slow at that point, or if I copied it from the current StaticArrays implementation at the time.

[stevengj]

Ok to merge?