> > But I saw a discussion about using Intel's MKL for greater performance and > the Make.user options to use Intel compilers are meant to be supported by > Julia. Why if there is no advantage in using them?
Intel MKL only helps with faster linear algebra than the default OpenBLAS (in some cases). Not with runtime of pure-Julia code. On Tue, Apr 28, 2015 at 2:31 PM, Ángel de Vicente < [email protected]> wrote: > Hi, > > On Tuesday, April 28, 2015 at 3:36:48 PM UTC+1, Tim Holy wrote: >> >> Intel compilers won't help, because your julia code is being compiled by >> LLVM. >> > > But I saw a discussion about using Intel's MKL for greater performance and > the Make.user options to use Intel compilers are meant to be supported by > Julia. Why if there is no advantage in using them? > > >> It's still hard to tell what's up from what you've shown us. When you run >> @time, does it allocate any memory? (You still have global variables in >> there, >> but maybe you made them const.) >> > > I'm posting some numbers again in reply to Yuuki's mail. > > >> But you can save yourself two iterations through the arrays (i.e., more >> cache >> misses) by putting >> T[i-1,j-1,k-1] += RHS[i-1,j-1,k-1] >> inside the first loop and discarding the second loop (except for cleaning >> up >> the edges). Fortran may be doing this automatically for you? >> http://en.wikipedia.org/wiki/Polytope_model >> >> > I'm not sure if Fortran is doing that, but I certainly would not like to > implement those sort of low-level details in the code itself, since it > makes understanding the code quite more cumbersome... > > (But Yuuki's mail gave me the trick. I reply to his mail below) > > Thanks a lot (starting to get the feel for Julia...), > Ángel de Vicente >
