C ++ AMP中的全局变量 [英] Global variables in C++ AMP

问题描述

我正在尝试创建一个可用于C#的C ++ AMP dll.   相同的，相当大的矩阵用于所有计算，我希望仅将其复制到GPU一次.在CUDA中，我只能使用将设备指针保存为全局变量，但是由于不允许并行使用全局变量 因为我有点迷路.

I’m trying to create a C++ AMP dll to be used from C#.  The same, rather large, matrix is used for all calculation, and I would prefer to copy it to the GPU only once. In CUDA I can just use save a device pointer as a global variable, but since I’m not allowed to use global variables in parallel for I’m a bit lost.

任何建议将不胜感激.

推荐答案

在调用p_f_e之前，请创建对该全局变量的本地引用，并捕获对p_f_e的引用以进行计算.

Before the call to the p_f_e, create a local reference to that global variable and capture that reference to the p_f_e for your computations.

下面的简单代码可以解决您的问题:

Following a simple code that solves your problem :

extern array_view<int,1> *arr_view;
extern array<int, 1> *arr;

int test()
{
	arr_view = new array_view<int,1>(16); // Creating a global instance of Array_view
	arr = new array<int,1>(16); // Creating a global instance of Array
	array_view<int,1> &arrview_ref = *arr_view; // Creating Local reference to Array_view object
	array<int,1> &arr_ref = *arr;  // Creating Local reference to Array object
	
	parallel_for_each(arr_view->extent,[=,&arr_ref](index<1> idx) restrict(amp)
	{
		arrview_ref[idx] = idx[0];
		arr_ref[idx]= 16 - idx[0];
	});
	
	std::vector<int> output = *arr; // copy out;
	for(int i=0; i < 16; i++)
	{
		std::cout << (*arr_view)[i] << " " << output[i] <<  std::endl;
	}	
}

这篇关于C ++ AMP中的全局变量的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！