Great comparison. What does || mean in your code? 2012/9/2 Francesco Abbate <francesco.bbt@xxxxxxxxx> > Hi all, > > I've made some tests using the Julia's benchmark to see the results of > GSL Shell. The performance is largely determined by LuaJIT so this may > be interesting for the people here. > > Here the document: > > > https://docs.google.com/presentation/d/1Gwx8rYEV1FqaqALUvnE0VHMhm87CW121cJ0YAoQwfys/edit > > and here the results: > > Test Julia GSL Shell Python > ------------------------------------------ > fib 0.429 0.120 3.641 > parse_int 0.367 0.856 3.878 > mandel 0.911 0.171 5.727 > quicksort 0.733 1.114 14.062 > pi_sum 38.59 38.60 764.24 > rand_mat_stat 42.17 57.09 254.92 > rand_mat_mul 175.45 178.70 200.25 > printfd 66.83 74.18 > > The file used for the test is in attachment. It is standard Lua but it > does use some GSL Shell modules for some of the tests. > > The results are quite good but in some cases Julia is performing > slightly better than GSL Shell. In particular I was surprised to see > that it is faster for the quicksort tests where I was thinking that > LuaJIT was going to rock. Something funny to note is that the integer > division is needed in this test and both Lua and GSL Shell needs to > use a trick to do it! :-) > > Otherwise LuaJIT is doing a great job in the Mandel test where we have > a tight loop with complex numbers computations. I guess that in this > case the work of Mike for objects sinking make the difference for this > test. > > For other tests like rand_mat_stat probably the GSL Shell stuff is > playing a role by using some slightly suboptimal code. What is > embarassing here is that the GSL Shell code is not only slower but > also more ugly. In this case Julia is showing its advantage being > specialized for matrix computations. > > The code for the Julia test can be easily find in the > http://julialang.org page, there is a link to the perf code in the git > repository. > > Francesco >