CUDA学习-计算实际线程ID

转自：GPU CUDA编程中threadIdx, blockIdx, blockDim, gridDim之间的区别与联系_TH_NUM的博客-CSDN博客

最重要的就是这张图：

#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;}

CUDA学习-计算实际线程ID相关推荐

VS2017 CUDA编程学习8：线程同步-原子操作
文章目录前言 1. 原子操作的理解 2. C++ CUDA实现原子操作 3. 执行结果总结学习资料 VS2017 CUDA编程学习1:CUDA编程两变量加法运算 VS2017 CUDA编程学习2 ...
CUDA学习笔记 —— （三）GPU计算性能与线程关系
文章目录概述 CPU运行 GPU运行单线程 GPU 1block 多线程运行多block多线程结果比对概述我们通过一个例子,相同的计算,我们分别在CPU,GPU单线程,GPU单block多线 ...
CUDA学习笔记之程序优化
CUDA学习笔记之程序优化标签: cuda优化conflict存储算法数学计算 2010-01-05 17:18 5035人阅读评论(4) 收藏举报分类: CUDA(6) 版权声明:本文为博主 ...
深度学习（三十六）异构计算CUDA学习笔记（1）
异构计算CUDA学习笔记(1) 原文地址:http://blog.csdn.net/hjimce/article/details/51506207 作者:hjimce 近日因为感觉自己在深度学习工程化 ...
CUDA学习(十)：向量内积的多种方法实现
博主CUDA学习系列汇总传送门(持续更新):编程语言|CUDA入门文章目录一.CPU上实现向量内积二.GPU下单Block分散归约向量内积三.单Block低线程归约向量内积四.多block向 ...
CUDA学习（三）之使用GPU进行两个数组相加
传入两个数组,在GPU中将两个数组对应索引位置相加 #include "cuda_runtime.h" #include "device_launch_parameter ...
CUDA学习----sp, sm, thread, block, grid, warp概念
CUDA学习----sp, sm, thread, block, grid, warp概念 2017-01-11 17:14:28| 分类: HPC&CUDA优化 | 标签:cuda g ...
广告深度学习计算：异构硬件加速实践
丨目录: - 前沿 1. 算力需求与供给 1.1 算力需求:模型复杂度 1.2 算力供给:异构硬件计算能力 1.3 问题与优化方法 2. 算法-系统-硬件协同性能优化 2.1 算法优化 2.2 系统优 ...
CUDA学习笔记（持续更新——蜗速）
CUDA学习笔记(持续更新--蜗速) 1.CUDA 程序实现流程如下 2.内存管理 3.核函数 4.全局数据访问唯一索引 5.设备管理附录代码 1.CUDA 程序实现流程如下将数据从CPU内存拷贝 ...

CUDA学习-计算实际线程ID

CUDA学习-计算实际线程ID相关推荐

最新文章

热门文章