RE: [boost] Boost.MPL: "Superlinear unrolling" to avoid recursion depth limits

From: Aleksey Gurtovoy (agurtovoy_at_[hidden])
Date: 2003-01-15 10:21:14

Vesa Karvonen wrote:
> Hi,

Hi Vesa,

> 
> The following is an example of using "superlinear unrolling". 
> It compiles with
> 
>   gcc version 2.95.4 20011002 (Debian prerelease)
> 
> using the default settings (max template recursion depth should be 
> 17). The example code essentially applies the inc template 262 times. 
> The example code uses partial specialization, but it should not be 
> extremely difficult to remove, if necessary.

Very interesting! The price for unrolling doesn't seem to be big either, for
VC 7.1, at least - ten compilation runs of "naive" 'until' implementation
are only by 2.5% faster than ten compilation runs of the unrolled version.

> I think that superlinear unrolling could be useful for avoiding 
> template recursion depth limitations, but the technique needs further 
> studying.

Thanks for sharing it!

[...]

> struct inc {
>   template<class T>
>   struct apply {
>     typedef typename T::next type;
>   };
> };

FYI, you can turn any MPL metafunction into a metafunction class by simply
not providing its template arguments; e.g. to get equivalent of the above
you just write 'mpl::next<>'.

> 
> struct always {
>   template<class T>
>   struct apply {
>     typedef typename mpl::bool_c<false> type;
>   };
> };

"boost/mpl/always.hpp", FYI.

> template<int X>
> struct is {
>   template<class T>
>   struct apply : mpl::bool_c<(T::value == X)> {
>   };
> };

MPL has this one too; it's called 'eq', see
"boost/mpl/comparison/equal_to.hpp".

Aleksey