On the D Blog: A Looat at Chapel, D, and Julia Using Kernel Matrix Calculations

[data pulverizer]

On Wednesday, 3 June 2020 at 16:15:41 UTC, jmh530 wrote:
> Also, I'm curious if you know how the Julia functions (like 
> pow/log) are implemented, i.e. are they also calling C/Fortran 
> functions or are they natively implemented in Julia?

It's not 100% clear but Julia does appear to implement a fair few 
mathematical functions apart from things like abs, sqrt, pow, 
which all come from C/llvm:

* 
https://github.com/JuliaLang/julia/blob/c3f6542aa3f90485f4b5fbac0486c390df7284d5/src/runtime_intrinsics.c#L902
* 
https://github.com/JuliaLang/julia/blob/be9ab4873d42f52bc776aa29d6e301d55b314033/src/julia_internal.h#L1006

However I can't see definitions for basic functions like log, exp 
there (not just special math functions):

* https://github.com/JuliaLang/julia/tree/v1.4.2/base/special

But in the math.jl file:

* https://github.com/JuliaLang/julia/blob/v1.4.2/base/math.jl all 
the basic math functions are imported from .Base:

Which might mean that some basic definition/declaration is 
imported from somewhere to be overridden by functions declared in 
special so it is entirely possible that things like sin, cos and 
tan defined in  special/trig.jl file are being used as the 
de-facto Julia trig functions, but I'm not 100% on that one. The 
long and short is that at least *some* basic math functions come 
from C/LLVM.

In addition Julia has fast math options - LLVM fast math 
implementation, which I only just remembered when looking at 
their code:

* 
https://github.com/JuliaLang/julia/blob/479097cf8c5a7675689cb069568d6b1077df8ba7/base/fastmath.jl

Which are obviously Clang/LLVM based.

I think it's a good idea to get std.math implementations more 
competitive performance wise because people will naturally 
gravitate towards that as the standard library for basic math 
functions.

> Typo (other than Mike's headline):
> "In our exercsie"
> "Chapel’s arrays are more difficult to get started with than 
> Julia’s but are designed to be run on single-core, multicore, 
> and computer clusters using the same or very similar code, 
> which is a good unique selling point." (should have comma 
> between Julia's and but)
>
> This is unclear:
> The chart below shows matrix implementation times minus ndslice 
> times; negative means that ndslice is slower, indicating that 
> the implementation used here does not negatively represent D’s 
> performance.

Fair point on the typo grammar.

Thanks