Yichao, Thanks for the response. I'm going to keep using rand() and not worry about it. The main downside I see that is that, when analyzing code with @code_warntype and similar, each use of rand() "pollutes" the result and makes it harder to understand.
-- mb On Sun, Mar 13, 2016 at 9:55 PM, Yichao Yu <yyc1...@gmail.com> wrote: > On Sun, Mar 13, 2016 at 9:47 PM, Miguel Bazdresch <eorli...@gmail.com> > wrote: > > While trying to optimize some code for performance, I noticed that rand() > > and randn() generate much more code than rand(1) and randn(1). In > addition, > > It generate more code since there are more inlining. > > > rand() and randn() have type instabilities, easily seen with > @code_warntype. > > These type instability actually doesn't matter since the result is > never used and IIRC it is also only in the slow path. > > > > > It seems to me than rand() should be faster and generate less code than > > rand(1), because rand(1) basically means "run rand() and put the result > in > > an array". > > > > Is this a bug, or at least a place where Julia's performance could be > > improved? > > The type stability could be fixed (although last time I tried it > there's no performance impact at all). > > > > > -- mb >
