this is very weird. If I use the current julia binary, everything works
fine. there may be something wrong with the current master, which is what I
used before. problem is that this is pretty hard to track down...
On Friday, 27 June 2014 11:25:14 UTC+1, Florian Oswald wrote:
>
> Actually, let me correct the profiling times - I did that again, and times
> are much larger in session two. pretty much everhthing takes longer in
> session 2. here's an example for the function "ufun":
>
> 275 ...julia/src/solver.jl; ufun; line: 397 # session 1
> 1206 ...julia/src/solver.jl; ufun; line: 397 # session 2
>
> function ufun(x::Float64,own::Int,mc::Float64,def::Bool,p::Param)
> r = p.imgamma * x^p.mgamma + own*p.xi - def*p.lambda - mc
> end
>
>
>
>
> On Friday, 27 June 2014 11:20:58 UTC+1, Florian Oswald wrote:
>>
>> Hi Tomas
>> we might just have missed each other, but I added that I actually took
>> care of this by running things twice: no change.
>>
>> On Friday, 27 June 2014 11:15:26 UTC+1, Tomas Lycken wrote:
>>>
>>> Are you sure you're not including compilation in those timings? Since
>>> Julia is JIT-compiled ("just-in-time"), the funcitons aren't compiled until
>>> you use them the first time. Thus, you need to do a warmup call first, that
>>> calls the method with arguments of the same type as those you want to time
>>> the function for.
>>>
>>> // T
>>>
>>> On Friday, June 27, 2014 12:11:02 PM UTC+2, Florian Oswald wrote:
>>>>
>>>> Hi all,
>>>> I found some strange behaviour when running my code. I open 2 clean
>>>> julia sessions and do the following:
>>>>
>>>>
>>>>
>>>> # session 1
>>>> home = ENV["HOME"]
>>>> cd("$home/git/migration/julia")
>>>>
>>>> include("src/mig.jl")
>>>> p = mig.Param(2)
>>>> m = mig.Model(p)
>>>> @time mig.solve!(m,p)
>>>>
>>>> elapsed time: 9.358865389 seconds (236857932 bytes allocated)
>>>>
>>>> # session 2
>>>> home = ENV["HOME"]
>>>> cd("$home/git/migration/julia")
>>>> include("src/mig.jl")
>>>>
>>>> mig.mywrap()
>>>>
>>>> elapsed time: 18.051932747 seconds (226522796 bytes allocated)
>>>>
>>>> the definition of mywrap is
>>>>
>>>> function mywrap()
>>>> p = mig.Param(2)
>>>> m = mig.Model(p)
>>>> @time mig.solve!(m,p)
>>>> end
>>>>
>>>> I thought this is identical? I looked at the @profile of both sessions,
>>>> and the profile timings of mig.solve! (the main computation function) are
>>>> actually lower in session 2 (inside the wrap) than in session 1. I would
>>>> have expected the reverse?
>>>>
>>>>
