Re: [boost] Stack-based vector container

Subject: Re: [boost] Stack-based vector container
From: David Bergman (David.Bergman_at_[hidden])
Date: 2010-12-28 18:36:46

Greg,

What you are doing here is to mix two (presumably...) orthogonal aspects, in that you bake in the allocation policy in the vector interface.

In fact, a "static allocator" to be used in cases like std::vector was discussed at length on this list a year ago (or was it two years?)

/David

On Dec 28, 2010, at 6:03 PM, Gregory Crosswhite wrote:

> Hey everyone,
> 
> In the course of my work as a computational scientist I have written a container that emulates std::vector, but which is backed by an array whose size is fixed at compile time.  Concretely, the class looks something like the following:
> 
>    template<typename T,unsigned int buffer_size> class static_vector {
>        ...
>        protected:
>            T data[buffer_size];
>            T* end_ptr;
>            size_t current_size;
>    }
> 
> where in the [...] there are methods that emulate (at this time, partially) the interface of std::vector, save that the size of the static_vector is not allowed to grow beyond the buffer_size.
> 
> The advantage of using a static_vector over vector is performance.  In my code I have performed benchmarks that show up to a 30% performance boost of static_vector over vector;  the benefit occurs most stongly when working with lots of arrays of small objects, so that the use of dynamic memory rather than stack memory appears to add significant overhead to the computation.  (Note that I when running these benchmarks I would use std::vector::reserve() to pre-allocate the array so that std::vector would not perform multiple allocations as the container grew, so that should not have been a source of overhead.)
> 
> It seems to me that it would be useful to have this functionality generally available as a library, and in particular it might be a good fit for the boost family of libraries.  Is there any interest out there for having such a library in boost?  If so, I would be happy to do the work of taking what I have and fleshing it out so that it can function as a drop-in replacement for vector, which means modeling the Random Access Container and Back Insertion Sequence concepts as well as adding capacity() and reserve(size_t) (no-op) methods.
> 
> Thoughts?
> 
> Cheers,
> Gregory Crosswhite
> _______________________________________________
> Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost