GPU CUDA编程中threadIdx, blockIdx, blockDim, gridDim之间的区别与联系

2024-04-02 12:58
文章标签 编程 区别 gpu 联系 之间 cuda threadidx blockidx blockdim griddim

本文主要是介绍GPU CUDA编程中threadIdx, blockIdx, blockDim, gridDim之间的区别与联系，希望对大家解决编程问题提供一定的参考价值，需要的开发者们随着小编来一起学习吧！

在启动kernel的时候，要通过指定gridsize和blocksize才行，举下面的例子说说：
dim3 gridsize(2,2);
dim3 blocksize(4,4);
gridsize相当于是一个2*2的block，gridDim.x，gridDim.y，gridDim.z相当于这个dim3的x，y，z方向的维度，这里是2*2*1。序号从0到3，且是从上到下的顺序，就是说是下面的情况:
grid中的blockidx序号标注情况为:

0	21	3

blocksize则是指里面的thread的情况，blockDim.x，blockDim.y，blockDim.z相当于这个dim3的x，y，z方向的维度，这里是4*4*1.序号是0-15，也是从上到下的标注：
block中的threadidx序号标注情况：

0	4	8	12
1	5	9	13
2	6	10	14
4	7	11	15

具体：
threadIdx是一个uint3类型，表示一个线程的索引。
blockIdx是一个uint3类型，表示一个线程块的索引，一个线程块中通常有多个线程。
blockDim是一个dim3类型，表示线程块的大小。
gridDim是一个dim3类型，表示网格的大小，一个网格中通常有多个线程块。
下面这张图比较清晰的表示的几个概念的关系：
在这里插入图片描述
cuda 通过<<< >>>符号来分配索引线程的方式，我知道的一共有15种索引方式。
在这里插入图片描述

#include "cuda_runtime.h"
#include "device_launch_parameters.h"#include <stdio.h>
#include <stdlib.h>
#include <iostream>using namespace std;//thread 1D
__global__ void testThread1(int *c, const int *a, const int *b)
{int i = threadIdx.x;c[i] = b[i] - a[i];
}
//thread 2D
__global__ void testThread2(int *c, const int *a, const int *b)
{int i = threadIdx.x + threadIdx.y*blockDim.x;c[i] = b[i] - a[i];
}//thread 3D
__global__ void testThread3(int *c, const int *a, const int *b)
{int i = threadIdx.x + threadIdx.y*blockDim.x + threadIdx.z*blockDim.x*blockDim.y;c[i] = b[i] - a[i];
}//block 1D
__global__ void testBlock1(int *c, const int *a, const int *b)
{int i = blockIdx.x;c[i] = b[i] - a[i];
}//block 2D
__global__ void testBlock2(int *c, const int *a, const int *b)
{int i = blockIdx.x + blockIdx.y*gridDim.x;c[i] = b[i] - a[i];
}//block 3D
__global__ void testBlock3(int *c, const int *a, const int *b)
{int i = blockIdx.x + blockIdx.y*gridDim.x + blockIdx.z*gridDim.x*gridDim.y;c[i] = b[i] - a[i];
}//block-thread 1D-1D
__global__ void testBlockThread1(int *c, const int *a, const int *b)
{int i = threadIdx.x + blockDim.x*blockIdx.x;c[i] = b[i] - a[i];
}//block-thread 1D-2D
__global__ void testBlockThread2(int *c, const int *a, const int *b)
{int threadId_2D = threadIdx.x + threadIdx.y*blockDim.x;int i = threadId_2D+ (blockDim.x*blockDim.y)*blockIdx.x;c[i] = b[i] - a[i];
}//block-thread 1D-3D
__global__ void testBlockThread3(int *c, const int *a, const int *b)
{int threadId_3D = threadIdx.x + threadIdx.y*blockDim.x + threadIdx.z*blockDim.x*blockDim.y;int i = threadId_3D + (blockDim.x*blockDim.y*blockDim.z)*blockIdx.x;c[i] = b[i] - a[i];
}//block-thread 2D-1D
__global__ void testBlockThread4(int *c, const int *a, const int *b)
{int blockId_2D = blockIdx.x + blockIdx.y*gridDim.x;int i = threadIdx.x + blockDim.x*blockId_2D;c[i] = b[i] - a[i];
}//block-thread 3D-1D
__global__ void testBlockThread5(int *c, const int *a, const int *b)
{int blockId_3D = blockIdx.x + blockIdx.y*gridDim.x + blockIdx.z*gridDim.x*gridDim.y;int i = threadIdx.x + blockDim.x*blockId_3D;c[i] = b[i] - a[i];
}//block-thread 2D-2D
__global__ void testBlockThread6(int *c, const int *a, const int *b)
{int threadId_2D = threadIdx.x + threadIdx.y*blockDim.x;int blockId_2D = blockIdx.x + blockIdx.y*gridDim.x;int i = threadId_2D + (blockDim.x*blockDim.y)*blockId_2D;c[i] = b[i] - a[i];
}//block-thread 2D-3D
__global__ void testBlockThread7(int *c, const int *a, const int *b)
{int threadId_3D = threadIdx.x + threadIdx.y*blockDim.x + threadIdx.z*blockDim.x*blockDim.y;int blockId_2D = blockIdx.x + blockIdx.y*gridDim.x;int i = threadId_3D + (blockDim.x*blockDim.y*blockDim.z)*blockId_2D;c[i] = b[i] - a[i];
}//block-thread 3D-2D
__global__ void testBlockThread8(int *c, const int *a, const int *b)
{int threadId_2D = threadIdx.x + threadIdx.y*blockDim.x;int blockId_3D = blockIdx.x + blockIdx.y*gridDim.x + blockIdx.z*gridDim.x*gridDim.y;int i = threadId_2D + (blockDim.x*blockDim.y)*blockId_3D;c[i] = b[i] - a[i];
}//block-thread 3D-3D
__global__ void testBlockThread9(int *c, const int *a, const int *b)
{int threadId_3D = threadIdx.x + threadIdx.y*blockDim.x + threadIdx.z*blockDim.x*blockDim.y;int blockId_3D = blockIdx.x + blockIdx.y*gridDim.x + blockIdx.z*gridDim.x*gridDim.y;int i = threadId_3D + (blockDim.x*blockDim.y*blockDim.z)*blockId_3D;c[i] = b[i] - a[i];
}void addWithCuda(int *c, const int *a, const int *b, unsigned int size)
{int *dev_a = 0;int *dev_b = 0;int *dev_c = 0;cudaSetDevice(0);cudaMalloc((void**)&dev_c, size * sizeof(int));cudaMalloc((void**)&dev_a, size * sizeof(int));cudaMalloc((void**)&dev_b, size * sizeof(int));cudaMemcpy(dev_a, a, size * sizeof(int), cudaMemcpyHostToDevice);cudaMemcpy(dev_b, b, size * sizeof(int), cudaMemcpyHostToDevice);testThread1<<<1, size>>>(dev_c, dev_a, dev_b);//testThread1<<<1, size>>>(dev_c, dev_a, dev_b);//uint3 s;s.x = size/5;s.y = 5;s.z = 1;//testThread2 <<<1,s>>>(dev_c, dev_a, dev_b);//uint3 s; s.x = size / 10; s.y = 5; s.z = 2;//testThread3<<<1, s >>>(dev_c, dev_a, dev_b);//testBlock1<<<size,1 >>>(dev_c, dev_a, dev_b);//uint3 s; s.x = size / 5; s.y = 5; s.z = 1;//testBlock2<<<s, 1 >>>(dev_c, dev_a, dev_b);//uint3 s; s.x = size / 10; s.y = 5; s.z = 2;//testBlock3<<<s, 1 >>>(dev_c, dev_a, dev_b);//testBlockThread1<<<size/10, 10>>>(dev_c, dev_a, dev_b);//uint3 s1; s1.x = size / 100; s1.y = 1; s1.z = 1;//uint3 s2; s2.x = 10; s2.y = 10; s2.z = 1;//testBlockThread2 << <s1, s2 >> >(dev_c, dev_a, dev_b);//uint3 s1; s1.x = size / 100; s1.y = 1; s1.z = 1;//uint3 s2; s2.x = 10; s2.y = 5; s2.z = 2;//testBlockThread3 << <s1, s2 >> >(dev_c, dev_a, dev_b);//uint3 s1; s1.x = 10; s1.y = 10; s1.z = 1;//uint3 s2; s2.x = size / 100; s2.y = 1; s2.z = 1;//testBlockThread4 << <s1, s2 >> >(dev_c, dev_a, dev_b);//uint3 s1; s1.x = 10; s1.y = 5; s1.z = 2;//uint3 s2; s2.x = size / 100; s2.y = 1; s2.z = 1;//testBlockThread5 << <s1, s2 >> >(dev_c, dev_a, dev_b);//uint3 s1; s1.x = size / 100; s1.y = 10; s1.z = 1;//uint3 s2; s2.x = 5; s2.y = 2; s2.z = 1;//testBlockThread6 << <s1, s2 >> >(dev_c, dev_a, dev_b);//uint3 s1; s1.x = size / 100; s1.y = 5; s1.z = 1;//uint3 s2; s2.x = 5; s2.y = 2; s2.z = 2;//testBlockThread7 << <s1, s2 >> >(dev_c, dev_a, dev_b);//uint3 s1; s1.x = 5; s1.y = 2; s1.z = 2;//uint3 s2; s2.x = size / 100; s2.y = 5; s2.z = 1;//testBlockThread8 <<<s1, s2 >>>(dev_c, dev_a, dev_b);//uint3 s1; s1.x = 5; s1.y = 2; s1.z = 2;//uint3 s2; s2.x = size / 200; s2.y = 5; s2.z = 2;//testBlockThread9<<<s1, s2 >>>(dev_c, dev_a, dev_b);cudaMemcpy(c, dev_c, size*sizeof(int), cudaMemcpyDeviceToHost);cudaFree(dev_a);cudaFree(dev_b);cudaFree(dev_c);cudaGetLastError();}
int main()
{const int n = 1000;int *a = new int[n];int *b = new int[n];int *c = new int[n];int *cc = new int[n];for (int i = 0; i < n; i++){a[i] = rand() % 100;b[i] = rand() % 100;c[i] = b[i] - a[i];}addWithCuda(cc, a, b, n);for (int i = 0; i < n; i++)printf("%d %d\n", c[i], cc[i]);delete[] a;delete[] b;delete[] c;delete[] cc; return 0;}

参考：https://www.cnblogs.com/rainbow70626/p/6498738.html?utm_source=itdadao&utm_medium=referral

https://www.cnblogs.com/tiandsp/p/9458734.html

这篇关于GPU CUDA编程中threadIdx, blockIdx, blockDim, gridDim之间的区别与联系的文章就介绍到这儿，希望我们推荐的文章对编程师们有所帮助！


http://www.chinasem.cn/article/870048。

相关文章

Java中ArrayList和LinkedList有什么区别举例详解

Java中ArrayList和LinkedList有什么区别举例详解

《Java中ArrayList和LinkedList有什么区别举例详解》：本文主要介绍Java中ArrayList和LinkedList区别的相关资料,包括数据结构特性、核心操作性能、内存与GC影... 目录一、底层数据结构二、核心操作性能对比三、内存与 GC 影响四、扩容机制五、线程安全与并发方案六、工程
阅读更多...
java父子线程之间实现共享传递数据

java父子线程之间实现共享传递数据

《java父子线程之间实现共享传递数据》本文介绍了Java中父子线程间共享传递数据的几种方法,包括ThreadLocal变量、并发集合和内存队列或消息队列,并提醒注意并发安全问题... 目录通过 ThreadLocal 变量共享数据通过并发集合共享数据通过内存队列或消息队列共享数据注意并发安全问题总结在 J
阅读更多...
Java文件与Base64之间的转化方式

Java文件与Base64之间的转化方式

《Java文件与Base64之间的转化方式》这篇文章介绍了如何使用Java将文件（如图片、视频）转换为Base64编码,以及如何将Base64编码转换回文件,通过提供具体的工具类实现,作者希望帮助读者... 目录Java文件与Base64之间的转化1、文件转Base64工具类2、Base64转文件工具类3、
阅读更多...
java中不同版本JSONObject区别小结

java中不同版本JSONObject区别小结

《java中不同版本JSONObject区别小结》本文主要介绍了java中不同版本JSONObject区别小结,文中通过示例代码介绍的非常详细,对大家的学习或者工作具有一定的参考学习价值,需要的朋友们... 目录1. FastjsON2. Jackson3. Gson4. org.json6. 总结在Jav
阅读更多...
C#多线程编程中导致死锁的常见陷阱和避免方法

C#多线程编程中导致死锁的常见陷阱和避免方法

《C#多线程编程中导致死锁的常见陷阱和避免方法》在C#多线程编程中,死锁（Deadlock）是一种常见的、令人头疼的错误,死锁通常发生在多个线程试图获取多个资源的锁时,导致相互等待对方释放资源,最终形... 目录引言1. 什么是死锁？死锁的典型条件：2. 导致死锁的常见原因2.1 锁的顺序问题错误示例：不同
阅读更多...
数据库使用之union、union all、各种join的用法区别解析

数据库使用之union、union all、各种join的用法区别解析

《数据库使用之union、unionall、各种join的用法区别解析》：本文主要介绍SQL中的Union和UnionAll的区别,包括去重与否以及使用时的注意事项,还详细解释了Join关键字,... 目录一、Union 和Union All1、区别：2、注意点：3、具体举例二、Join关键字的区别&php
阅读更多...
java中的HashSet与 == 和 equals的区别示例解析

java中的HashSet与 == 和 equals的区别示例解析

《java中的HashSet与==和equals的区别示例解析》HashSet是Java中基于哈希表实现的集合类,特点包括：元素唯一、无序和可包含null,本文给大家介绍java中的HashSe... 目录什么是HashSetHashSet 的主要特点是HashSet 的常用方法hasSet存储为啥是无序的
阅读更多...
PyCharm接入DeepSeek实现AI编程的操作流程

PyCharm接入DeepSeek实现AI编程的操作流程

《PyCharm接入DeepSeek实现AI编程的操作流程》DeepSeek是一家专注于人工智能技术研发的公司,致力于开发高性能、低成本的AI模型,接下来,我们把DeepSeek接入到PyCharm中... 目录引言效果演示创建API key在PyCharm中下载Continue插件配置Continue引言
阅读更多...
2.1/5.1和7.1声道系统有什么区别? 音频声道的专业知识科普

2.1/5.1和7.1声道系统有什么区别? 音频声道的专业知识科普

《2.1/5.1和7.1声道系统有什么区别?音频声道的专业知识科普》当设置环绕声系统时，会遇到2.1、5.1、7.1、7.1.2、9.1等数字，当一遍又一遍地看到它们时，可能想知道它们是什... 想要把智能电视自带的音响升级成专业级的家庭影院系统吗？那么你将面临一个重要的选择——使用 2.1、5.1 还是
阅读更多...
Python中@classmethod和@staticmethod的区别

Python中@classmethod和@staticmethod的区别

《Python中@classmethod和@staticmethod的区别》本文主要介绍了Python中@classmethod和@staticmethod的区别,文中通过示例代码介绍的非常详细,对大... 目录1.@classmethod2.@staticmethod3.例子1.@classmethod
阅读更多...

Copyright China编程 23002807@qq.com

沪ICP备2023033072号-2