On Sun, Dec 13, 2015 at 7:06 PM, Dominique Orban <[email protected]> wrote: > I'm writing Julia functions to perform a sparse matrix factorization and I'm > wondering if what I'm observing is a manifestation of type instability. I'm > using Julia 0.4 so I can use @code_warntype, but it's surprisingly quiet and > doesn't reveal much (there are no ANYs or UNIONs; the "Variables" section is > clean). But when I use ProfileView on my code, I observe a high peak. In it, > the functions called are > > - sys.dylib, typeinf_ext, typeinf, typeinf_uncached, abstract_eval, > abstract_call, abstract_call_gf, etc.
The high peak just mean that it is a really deep call and they are not taking much time. Those functions are in type inference so that usually mean the code isn't compiled when you called the method. (You should call the method once before benchmarking it.) > > Isn't that indicative of type instability? I've stared at the code for quite > a while and can't determine what could be type unstable. > > If that's useful, the flame graph is here: > https://gist.github.com/5bd3c6b6c6403dd141cb > > Thanks!
