C++ OO approach to use multi-dim array for HPC - 未名空间MITBBS历史存档

国际科技财经博客移民网络热点娱乐民生时事公众号

Redian新闻

>未名空间

>Programming - 葵花宝典

C++ OO approach to use multi-dim array for HPC

C++ OO approach to use multi-dim array for HPC# Programming - 葵花宝典

c*d2005-11-01 08:11

1 楼

I would like to hear your comments on my design of this array class. The
purpose is to provide a fast, easy to use multi-dimensional array container.
/*
* A C/C++ OO approach to use multi-dimenisional array in HPC environment.
*
* Goal: speed, ease of use, robustness
*
* Design advantage:
* By using 1D array as data container, CPU prefetching, caching, TLB
* features are used to make data access faster. By providing error checking
* in access methods, bounds check can be easily done. Access me

s*c2005-11-01 08:11

2 楼

Why not put ndim in template also?
template class array;
Too many constructor. Either use stl:vector for dimlen, or the constructor
parameter is size_t *. Same can be done for ele. You could use a stl:vector, or an array (correct size is ensured by your main program). You could also define another template class called pixel, or point, or vector:
template class pixel
{
T data[N]
...
...
}

【在 c*******d 的大作中提到】

: I would like to hear your comments on my design of this array class. The
: purpose is to provide a fast, easy to use multi-dimensional array container.
: /*
: * A C/C++ OO approach to use multi-dimenisional array in HPC environment.
: *
: * Goal: speed, ease of use, robustness
: *
: * Design advantage:
: * By using 1D array as data container, CPU prefetching, caching, TLB
: * features are used to make data access faster. By providing error checking

c*d2005-11-01 08:11

3 楼

the purpose of this class is for high performance computing, therefore
performance and ease of use is very important, especially for people
with a heavy fortran background.
for example, the declaration should be simple:
array a3d(3,3,3);
not array a3d(vector(3,3,3)); // also overhead of creating vector
class

, or an array (correct size is ensured by your main program). You could also
define another template class called pixel, or point, or vector:

【在 s*****c 的大作中提到】

: Why not put ndim in template also?
: template class array;
: Too many constructor. Either use stl:vector for dimlen, or the constructor
: parameter is size_t *. Same can be done for ele. You could use a stl:vector, or an array (correct size is ensured by your main program). You could also define another template class called pixel, or point, or vector:
: template class pixel
: {
: T data[N]
: ...
: ...
: }