ziggurat performance comparison (julia vs. C)

[ViralBShah]

The julia implementation of the ziggurat implementation is 2x slower than the C implementation for generating 1 million normally distributed random numbers.

julia, randn_zig,30.411958694458008
c,randn_zig,16.033888

[ViralBShah]

The difference is considerably reduced now.

c,randn_zig,18.842936
julia,randn_zig,23.12302589416504

Note that I get 14ms if I use clang -O4, but that uses LTO and is probably not a fair comparison.

[StefanKarpinski]

Very nice. I'm sure we'll gradually close the gap with C on many of these sorts of things, which will be lovely for now, we're getting close enough and language features and API stability are a higher priority.