thrust(unary与binary)中放复杂类到GPU加速

东方耀

thrust(unary与binary)中放复杂类到GPU加速

1. #include <thrust/device_vector.h>
   
2. #include <iostream>
   
3. #include <string.h>
   
4. 
   
5. using namespace std;
   
6. 
   
7. //thrust(unary与binary)中放复杂类到GPU加速
   
8. 
   
9. 
   
10. struct Teacher
    
11. {
    
12.   //string name;  // string类型不支持cuda吗？ string本身就是stl
    
13.   char name[64];  //基础数据类型
    
14.   int age;
    
15.   
    
16. 
    
17.   __host__ __device__
    
18.   Teacher() {
    
19.     // : name("xxx"),age(10)
    
20.     age = 10;
    
21.     strcpy(name, "XXX");
    
22.     //error: identifier "std::cout" is undefined in device code
    
23.     //cout << "无参构造" << endl;
    
24.   }
    
25.   
    
26.   __host__ __device__
    
27.   Teacher(const char *pname, int _age) {
    
28.     //warning: conversion from a string literal to "char *" is deprecated
    
29.     //在所有声明char *的地方都要加上const
    
30.     age = _age;
    
31.     strcpy(name, pname);   //这个支持gpu操作？
    
32.   }
    
33. 
    
34.   __host__
    
35.   void print_teacher_info(){
    
36.     //warning: address of a host variable "std::cout"
    
37.     std::cout << "name=" << name << ",age=" << age << std::endl;
    
38.   }
    
39. 
    
40. };
    
41. 
    
42. 
    
43. struct get_age_unary_op
    
44.   : public thrust::unary_function< Teacher, int >
    
45. {
    
46.   //  unary_op; //用于transform 一元操作
    
47.   
    
48.   __host__ __device__
    
49.   int operator()(const Teacher& t) const
    
50.   {
    
51.     //函数对象：()操作符的重载
    
52.     return t.age;
    
53.   }
    
54. };
    
55. 
    
56. struct jjj_plus : public thrust::binary_function<int,int,int>
    
57. {
    
58.     //自定义二元函数对象   这是基础数据类型的
    
59.   __host__ __device__
    
60.   int operator()(int a, int b)
    
61.   {
    
62.     return a+b;
    
63.   }
    
64. };
    
65. 
    
66. 
    
67. 
    
68. int main(){
    
69. 
    
70.     int a[6] = {1,2,3,4,5,6};
    
71.     thrust::device_vector<int> d1(a,a+6);
    
72.     thrust::copy(d1.begin(),d1.end(), std::ostream_iterator<int>(std::cout, "\n"));
    
73. 
    
74.     int sum1 = thrust::reduce(d1.begin(),d1.end());
    
75.     int sum2 = thrust::reduce(d1.begin(),d1.end(),0,thrust::plus<int>());
    
76.     int sum3 = thrust::reduce(d1.begin(),d1.end(),0,jjj_plus());
    
77. 
    
78.     std::cout << "求和1=" << sum1 << std::endl;
    
79.     std::cout << "求和2=" << sum2 << std::endl;
    
80.     std::cout << "求和3=" << sum3 << std::endl;
    
81. 
    
82.     //求老师的年龄之和呢？
    
83.     Teacher t1("zhang_san", 30), t2("li_si", 33), t3;
    
84.     t1.print_teacher_info();
    
85.     t2.print_teacher_info();
    
86.     t3.print_teacher_info();
    
87.     //Teacher t_array[3] = {t1,t2,t3};
    
88.     //thrust::device_vector<Teacher> d2(t_array,t_array+3);
    
89.     thrust::device_vector<Teacher> teacher_array;
    
90.     //thrust::system::system_error'
    
91.     //what():  for_each: failed to synchronize: cudaErrorLaunchFailure: unspecified launch failure
    
92.     teacher_array.push_back(t1);
    
93.     teacher_array.push_back(t2);
    
94.     teacher_array.push_back(t3);
    
95.     // 1元操作用来 转换 容器里的每个元素  2元操作用来操作相邻的value
    
96.     int result_sum = thrust::transform_reduce(teacher_array.begin(), teacher_array.end(), get_age_unary_op(), 0, jjj_plus());
    
97.     std::cout << "result_sum=" << result_sum << std::endl;
    
98. 
    
99.    
    
100.     return 0;
     
101. }
     
102. 
     
103. 
     
104. /**
     
105. for_each: failed to synchronize: cudaErrorLaunchFailure: unspecified launch failure
     
106. 已放弃 (核心已转储)
     
107. 
     
108. 
     
109. 具体代码：
     
110. thrust::device_vector<Teacher> teacher_array(3);
     
111. 
     
112. Teacher里有成员变量：
     
113. string name;  // string类型不支持cuda吗？
     
114. 
     
115. 分析：string是c++的stl
     
116. 
     
117. 改用 char name[64];
     
118. 
     
119. 
     
120.     string name = "jiangjinju";
     
121.     //字符串怎么放到gpu上去？
     
122.     thrust::device_vector<string> d1(1);  //会报错
     
123. */

复制代码

东方耀

cpu与gpu适配代码,thrust::reduce有点混乱啊

1. //#include <thrust/transform_reduce.h>
   
2. #include <thrust/device_vector.h>
   
3. //#include <thrust/pair.h>
   
4. #include <thrust/random.h>
   
5. //#include <thrust/extrema.h>
   
6. 
   
7. // This example shows how to compute a bounding box
   
8. // for a set of points in two dimensions.
   
9. 
   
10. //cpu与gpu适配代码,thrust::reduce有点混乱啊
    
11. 
    
12. struct point2d
    
13. {
    
14.   float x, y;   //二维坐标点
    
15.   
    
16.   __host__ __device__
    
17.   point2d() : x(0), y(0) {}
    
18.   
    
19.   __host__ __device__
    
20.   point2d(float _x, float _y) : x(_x), y(_y) {}
    
21. };
    
22. 
    
23. // bounding box type
    
24. struct bbox
    
25. {
    
26.   // construct an empty box
    
27.   __host__ __device__
    
28.   bbox() {}
    
29. 
    
30.   // construct a box from a single point
    
31.   __host__ __device__
    
32.   bbox(const point2d &point)
    
33.     : lower_left(point), upper_right(point)
    
34.   {}
    
35. 
    
36.   // construct a box from a single point    box1=box2=box3;
    
37.   __host__ __device__
    
38.   bbox& operator=(const point2d &point)
    
39.   {
    
40.     lower_left = point;
    
41.     upper_right = point;
    
42.     return *this;  //为了三连等
    
43.   }
    
44. 
    
45.   // construct a box from a pair of points
    
46.   __host__ __device__
    
47.   bbox(const point2d &ll, const point2d &ur)
    
48.     : lower_left(ll), upper_right(ur)
    
49.   {}
    
50. 
    
51.   point2d lower_left, upper_right; //一个box 是由两个点 组成的
    
52. };
    
53. 
    
54. // reduce a pair of bounding boxes (a,b) to a bounding box containing a and b
    
55. struct bbox_reduction : public thrust::binary_function<bbox,bbox,bbox>
    
56. {
    
57.   //由两个box 变成一个box （能够覆盖掉的最大box）
    
58.   __host__ __device__
    
59.   bbox operator()(bbox a, bbox b)
    
60.   {
    
61.     // lower left corner
    
62.     point2d ll(thrust::min(a.lower_left.x, b.lower_left.x), thrust::min(a.lower_left.y, b.lower_left.y));
    
63.    
    
64.     // upper right corner
    
65.     point2d ur(thrust::max(a.upper_right.x, b.upper_right.x), thrust::max(a.upper_right.y, b.upper_right.y));
    
66.    
    
67.     return bbox(ll, ur);
    
68.   }
    
69. };
    
70. 
    
71. int main(void)
    
72. {
    
73.   const size_t N = 40;
    
74.   thrust::default_random_engine rng(6666);
    
75.   thrust::uniform_real_distribution<float> u01(0.0f, 1.0f);
    
76.   
    
77.   // allocate storage for points  很多点的 一个数组
    
78.   thrust::device_vector<point2d> points(N);
    
79.   
    
80.   // generate some random points in the unit square
    
81.   for(size_t i = 0; i < points.size(); i++)
    
82.   {
    
83.       float x = u01(rng);
    
84.       float y = u01(rng);
    
85.       points[i] = point2d(x,y); // 这里是用copy构造 还是 operator=
    
86.   }
    
87.   
    
88.   // initial bounding box contains first point
    
89.   bbox init = bbox(points[0], points[0]);
    
90.   
    
91.   // binary reduction operation
    
92.   bbox_reduction binary_op;
    
93.   
    
94.   // compute the bounding box for the point set
    
95.   // 为啥：bbox_reduction 二元操作 要求是两个bbox啊？ 这里是点啊？
    
96.   // 虽然初始的值是bbox 但是在点上遍历啊？ 第二个如何由点变box呢？  这里感觉混乱？
    
97.   bbox result = thrust::reduce(points.begin(), points.end(), init, binary_op);
    
98.   
    
99.   // print output
    
100.   std::cout << "bounding box " << std::fixed;
     
101.   std::cout << "(" << result.lower_left.x  << "," << result.lower_left.y  << ") ";
     
102.   std::cout << "(" << result.upper_right.x << "," << result.upper_right.y << ")" << std::endl;
     
103.   
     
104.   return 0;
     
105. }
     

复制代码

zouqiqi

让天下人人学会人工智能！人工智能的前景一片大好！