计算线程index

最新推荐文章于 2025-07-05 17:27:35 发布

JUAN425

最新推荐文章于 2025-07-05 17:27:35 发布

阅读量928

点赞数

CC 4.0 BY-SA版权

分类专栏： Computer Vision OpenCV DeepLearning 文章标签：库接口

本文链接：https://blog.youkuaiyun.com/a130737/article/details/43603259

DeepLearning 同时被 3 个专栏收录

21 篇文章

订阅专栏

OpenCV

13 篇文章

订阅专栏

Computer Vision

9 篇文章

订阅专栏

本文详细介绍了CUDA编程中不同维度的网格与块所对应的全局线程索引计算方法，并提供了9种常见情况的具体实现代码。此外，还概述了CUDA中常用的两大库：BLAS和FFT的功能。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

CUDA相关知识整理

CUDA libraries主要有两个：
（1） BLAS：全称是Basic Linear Algebra Subprograms。是一个用于线性代数运算的应用程序接口。
（2） FFT：即使得能够在NVIDIA GPU上高效执行FFT变换的库。
当我们launch了a grid of blocks of threads 的时候，我们在kernel中需要指定CUDA的thread indexing。关于CUDA线程模型如下图：

我们的计算thread的global index的情况有如下种9种：
(1). 1D grid of 1D blocks

__device__ int getGlobalIdx_1D_1D() {
    return blockIdx.x * blockDim.x + threadIdx.x;
}

(2) 1D grid of 2D blocks

__device__ int getGlobalIdx_1D_2D() {
    return blockIdx.x * blockDim.x * blockDim.y + threadIdx.y * blockDim.x + threadIdx.x;
}

(3). 1D grid of 3D blocks

 __device__ int getGlobalIdx_1D_3D() { 
  return blockIdx.x * blockDim.x * blockDim.y * blockDim.z + threadIdx.z * blockDim.y * blockDim.x 
 + threadIdx.y * blockDim.x + threadIdx.x;  
}

(4) 2D grid of 1D blocks

__device__ int getGlobalIdx_2D_1D() { 
  int blockId = blockIdx.y * gridDim.x + blockIdx.x; 
 int threadId = blockId * blockDim.x + threadIdx.x; 
  return threadId;  
}

(5) 2D grid of 3D blocks

__device__ int getGlobalIdx_2D_3D() { 
  int blockId = blockIdx.x  + blockIdx.y * gridDim.x; 
 int threadId = blockId * (blockDim.x * blockDim.y * blockDim.z)  + (threadIdx.z * (blockDim.x * blockDim.y))  + (threadIdx.y * blockDim.x)  + threadIdx.x; 
  return threadId;  
}

(6) 3D grid of 1D blocks

__device__ int getGlobalIdx_3D_1D() { 
  int blockId = blockIdx.x + blockIdx.y * gridDim.x 
 + gridDim.x * gridDim.y * blockIdx.z; 
 int threadId = blockId * blockDim.x + threadIdx.x; 
  return threadId;  
}

(7) 3D of 2D blocks

__device__ int getGlobalIdx_3D_2D() { 
  int blockId = blockIdx.x  + blockIdx.y * gridDim.x 
 + gridDim.x * gridDim.y * blockIdx.z; 
 int threadId = blockId * (blockDim.x * blockDim.y) 
 + (threadIdx.y * blockDim.x) + threadIdx.x; 
  return threadId;  
}

(8) 3D grid of 3D blocks

__device__ int getGlobalIdx_3D_3D() { 
  int blockId = blockIdx.x + blockIdx.y * gridDim.x 
 + gridDim.x * gridDim.y * blockIdx.z; 
 int threadId = blockId * (blockDim.x * blockDim.y *       blockDim.z)  + (threadIdx.z * (blockDim.x * blockDim.y))  + (threadIdx.y * blockDim.x)  + threadIdx.x; 
 return threadId;  
}