While it seems attractive in principle, in practice this will hurt performance. Because of the loop, there is no single call-site with consistent types for which the compiler can infer in advance which method you're going to call. Consequently, you're forcing julia to do dynamic lookup on each call. If you really need to do this, it would be much better if you unrolled that loop by hand.
See more detail here: http://docs.julialang.org/en/latest/manual/performance-tips/#types-with-values-as-parameters and in the following section: http://docs.julialang.org/en/latest/manual/performance-tips/#the-dangers-of-abusing-multiple-dispatch-aka-more-on-types-with-values-as-parameters Best, --Tim On Friday, June 10, 2016 8:39:35 PM CDT Po Choi wrote: > Thanks for pointing out the mistakes. > > I want to loop over the data types because I want to loop over functions. > I need to do something like this: > function mywork(x, y, z) > procedure1(x, y, z) > procedure2(x, y, z) > procedure3(x, y, z) > #... > procedure8(x, y, z) > procedure9(x, y, z) > procedure10(x, y, z) > end > > I feel it may be silly to write out all the procedurek(x, y, z) > Then I am thinking about the following: > function mywork(x, y, z) > for k in 1:10 > procedure(::Val{k}, x, y, z) > end > end > > I have not seen anyone doing this before. So I am not sure I am doing > something right. > That's why I am asking this question. > > I remember people like to use macro to define functions for different types. > In my case, should I use macro? Or am I fine to code as above? > > On Friday, June 10, 2016 at 7:34:09 PM UTC-7, Erik Schnetter wrote: > > Your code won't work. `Val{k}` is a type, which is an object in Julia. You > > are passing this object to `foo`. Thus in your declaration of `foo`, you > > need to list the type of `Val`, not just `Val` -- types have types as > > well: > > ``` > > foo(::Type{Val{1}}) = 1 > > ``` > > > > Of course you know that using `Val` in this way is nonsensical in a real > > program. I understand that you know this, as you're purposefully > > experimenting with Julia, but I'd still like to point it out for the > > casual > > reader of this example. > > > > Whether you encounter "performance issues" or not depends on what > > performance you need. If you compare this code to simple arithmetic > > operations (adding numbers), then it's slower. If you compare it to > > sending > > data across the network or accessing the disk, then it's faster. > > > > I assume that calling `foo` in the loop requires a hash table lookup at > > run time, and likely some memory allocation. > > > > -erik > > > > > > On Fri, Jun 10, 2016 at 9:40 PM, Po Choi <[email protected] > > > > <javascript:>> wrote: > >> Ops! I accidentally hit the post button! So the post is not completed! > >> > >> It is an example: > >> foo(::Val{1}) = 1 > >> foo(::Val{2}) = 2 > >> foo(::Val{2}) = 3 > >> > >> function bar() > >> > >> s = 0 > >> for t in Datatype[Val{k} for k in 1:3] > >> > >> s += foo(t) > >> > >> end > >> > >> end > >> > >> Will there be any performance issue if I loop over types? > >> I am still trying to understand how the multiple-dispatch works. > >> Sometimes I am confused! > >> > >> On Friday, June 10, 2016 at 6:37:31 PM UTC-7, Po Choi wrote: > >>> foo(::Val{1}) = 1 > >>> foo(::Val{2}) = 2 > >>> foo(::Val{2}) = 3 > >>> > >>> function bar() > >>> > >>> for t in Datatype[Val{k} for k in 1:3] > >>> > >>> end > >>> > >>> end
