Agner Fog's C++ Vector Class Library v2 (2019): Metaprogramming & Functionality Guide - CSDN文库

176 浏览量更新于2024-07-14 收藏 445KB PDF 举报

身份认证购VIP最低享 7 折!

30元优惠券

资源详情

资源推荐

for ( int i = 0; i < datasize /8; i ++) {

mydata [i] = mydata [ i ] * 0.1 f + 2.0 f ;

}

If you need an array of a size that is determined at runtime, then you will have a problem with

alignment. Each vector should preferably be stored at an address divisible by 16, 32 or 64 bytes,

according to its size. The compiler can do this when deﬁning a ﬁxed-size array, as in the above

example, but not necessarily with dynamic memory allocation. If you create an array of dynamic size

by using new, malloc or a container class template, or any other method, then you may not get the

proper alignment for vectors. A misaligned vector will likely cause the program to crash if compiled

for an instruction set less than AVX. See page 67 for details.

It is recommended to make an array of scalars instead of an array of vectors in order to avoid these

complications. If the datasize in the above example is variable, then the code could be implemented

in this way:

int datasize = 1024; // size of dataset , variable

float * mydata = new float [ datasize ]; // dynamic array

...

Vec8f x ;

for ( int i = 0; i < datasize ; i += 8) {

x. load ( mydata +i);

x = x * 0.1 f + 2.0 f ;

x. store ( mydata +i);

}

...

delete [] mydata ;

See page 67 for discussion of the case where the data size is not a multiple of the vector size.

2.6 Using a namespace

In general, there is no need to put the vector class library into a separate namespace. Therefore, the

use of a namespace is optional. You can give the vector class library a namespace with a name of

your choosing by deﬁning

VCL NAMESPACE

, for example:

# define VCL _NAM ESPA CE vcl

# include " vector class . h "

using na mes pac e vcl ;

// your vector code here ...

14

剩余87页未读，继续阅读

weixin_38645865

粉丝: 10
资源: 923

最新资源

资源上传下载、课程学习等过程中有任何疑问或建议，欢迎提出宝贵意见哦~我们会及时处理！点击此处反馈