On Sun, Jul 6, 2014 at 2:13 PM, Sid <tmf...@gmail.com> wrote: > I think you made a typo -- you're allocating an array of size 100,000 > rather than 1 million. >
I just copied what you wrote in your original message. When I run with 1 million, I get something close to what you reported, but the details of that number have already been explained, I think. Cheers! Kevin > > On Sunday, July 6, 2014 5:04:46 PM UTC-4, Kevin Squire wrote: >> >> I believe there is some work afoot on getting memory usage on a per-line >> basis. Until then, @profile is usually a reasonably good proxy, at least >> on Linux and OSX--Windows has had some issues around this. >> >> That said, the original answer was the best one: this is a known issue, >> and the best solution is probably to inline functions. This is a >> non-trivial change, so unfortunately probably won't happen in the very near >> future. >> >> Incidently, the deprecation of the "order" keyword was removed on master, >> so you won't be seeing that warning anymore. >> >> So? Temporary allocations take time and space, too, and as soon as the >>> system starts swapping, both don't just add, but multiply. >>> >> >> Interestingly, I only get 61Mb allocated for the same test that @Sid ran >> (69Mb the first time, which includes compilation), which is much less than >> 768Mb: >> >> julia> a = rand(100000); >> >> julia> @allocated sort(a, by=x->x) >> 68840632 >> >> julia> @allocated sort(a, by=x->x) >> 60856624 >> >> >>> The correct question is: why does Julia allocate incredible amounts of >>> memory even for the in-place sorting of an array, sometimes? When exactly >>> does that happen, and how can we avoid it? >>> >> >> 60Mb isn't that much, but it is still much larger than 8Mb. Can you tell >> me how much memory is allocated by Java or C++ for this? >> >> Cheers, >> Kevin >> >> >> >>> >>> Am Sonntag, 6. Juli 2014 22:28:02 UTC+2 schrieb Kevin Squire: >>> >>>> It's likely that much of the memory was for temporary allocations that >>>> were reused. >>>> >>>> Cheers, >>>> Kevin >>>> >>>> >>
