This is not the classical globals problem.
This difference is that our (map)reduce functions does a recursive
reduction (splits in half until the list is shorter than some threshold),
and BigInt is much more efficient with small numbers than with big numbers.
function mr_pairwise(f::Callable, op::Callable, A::AbstractArray, i1,n)
if n < 128
@inbounds v = f(A[i1])
for i = i1+1:i1+n-1
@inbounds v = op(v,f(A[i]))
end
return v
else
n2 = div(n,2)
return op(mr_pairwise(f,op,A, i1,n2), mr_pairwise(f,op,A, i1+n2,n-n2))
end
end
My statement was definitely wrong in some cases where a recursive algorithm
has advantages.
Ivar
kl. 11:49:53 UTC+1 lørdag 15. februar 2014 skrev harven følgende:
>
>
>
> Le samedi 15 février 2014 11:30:12 UTC+1, Tim Holy a écrit :
>>
>> You're running into the classic globals problem, see the performance tips
>> section of the manual.
>>
>>
> I am sorry, I don't understand your remark. There are no global variables
> in the code I posted previously.
> Also, wrapping the code in a function does not get better performance.
> Maybe I don't understand what is the classic globals problem?
>
> julia>
> function test () ;
> let biglist = [string(i) for i in 1:20000] , result = BigInt(1);
> @time for x in biglist ; result = result * BigInt(x) ; end ; result
> ; end
> end
>
> julia> test (generic function with 1 method)
>
> julia> test()
> elapsed time: 0.396508521 seconds (304614928 bytes allocated)
>
>
> julia> test()
> elapsed time: 0.393320767 seconds (304614928 bytes allocated)
>
>
>
