Merge CPUTime into JuliaLang

[zouhairm]

This pretty useful. It would be nice to have it as part of the master branch.

[schmrlng]

I'm hesitant to do this as modern linear algebra libraries (OpenBLAS) have really good support for multi-threading, and this can lead to incorrect conclusions being drawn about timing since CPU time is summed across all child threads of a process:

julia> n = 1000; A = rand(n,n); B = rand(n,n);
julia> @time @CPUtime C = A*B;     # matrix multiplication on a quad-core i7
elapsed CPU time: 0.178905 seconds
elapsed time: 0.045403309 seconds (8000488 bytes allocated)

See also JuliaLang/julia#10157. It might seem that a possible resolution would be counting only the CPU time used by the top-level process, but that can also be misleading - in this case it's probably 0. For multi-core machines, I still don't know a good way to reliably benchmark code when you've got other stuff going on in the background (the motivation of this package in the first place: http://stackoverflow.com/questions/24427185/measuring-elapsed-cpu-time-in-julia/). Oh well.

So unless you're running single-threaded code (scientific computing without any linear algebra??), the utility of this package is sort of nebulous, which is why I haven't bothered trying to register it with the Julia package manager. Putting it in Julia master would probably do more harm than help.

[zouhairm]

sounds reasonable. It might not be useful in the general sense, but definitely useful for the single threaded case.

But I understand what you mean about potentially doing more harm.

[mykelk]

Might be nice to have registered on METADATA.jl so people can just do Pkg.add("CPUTime").

[schmrlng]

Fine fine, but caveat emptor.

[chriselrod]

since CPU time is summed across all child threads of a process

FWIW, I searched for a CPU time Julia library because this is the behavior I wanted (and expected from a "CPUTime").