Google Groups

Re: Julia < 2x C

Viral Shah Aug 13, 2012 11:26 AM
Posted in group: julia-dev
Time to create the perf 2.0 testsuite, I say..


On Monday, August 13, 2012 10:21:20 PM UTC+5:30, Stefan Karpinski wrote:
I wanted to point out that thanks to some compiler optimizations which Jeff recently implemented, and to Tim Holy noticing that the C version of rand_mat_stat was sneakily doing half as much work as it should have been (my fault), Julia is now across the board faster than 2x C. Of course, that's only on our micro-benchmark suite, but I think that it genuinely reflects overall Julia performance being well within the range of compiled languages — C and Fortran being among the fastest of those.

This is a landmark event, and it has been a long time coming (three years next month). I remember timing the first Julia version that could manage to multiply two matrices (not using LAPACK yet), and finding that it was, oh, 1000 times slower than C. But that was actually pretty decent given that it was working with arrays of pointers to boxed, heap-allocated floating-point numbers, and there zero optimizations in place anywhere.