__device__ __constant__ const [英] __device__ __constant__ const

问题描述

有什么区别？在CUDA程序中定义设备常数的最佳方法是什么？在C ++，主机/设备程序中，如果我想将常量定义为设备常量内存中的任何一个，我都可以做

Is there any difference and what is the best way to define device constants in a CUDA program? In the C++, host/device program if I want to define constants to be in device constant memory I can do either

__device__ __constant__ float a = 5;
__constant__ float a = 5;

问题1.在设备2.x和CUDA 4上，它是否相同，

Question 1. On devices 2.x and CUDA 4, is it the same as,

__device__ const float a = 5;

问题2。为什么在PyCUDA SourceModule（ ...）中，即使执行以下工作，它也只能编译设备代码？

Question 2. Why is it that in PyCUDA SourceModule("""..."""), which compiles only do device code, even the following works?

const float a = 5;

推荐答案

在CUDA中 __ constant __ 是变量类型限定符，它指示要声明的变量将存储在设备常量存储器中。引用第节href = https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html rel = nofollow noreferrer> CUDA编程指南

In CUDA __constant__is a variable type qualifier that indicates the variable being declared is to be stored in device constant memory. Quoting section B 2.2 of the CUDA programming guide

__ constant __ 限定词，可以与 __ device __ 一起使用，
声明了一个变量：

The __constant__ qualifier, optionally used together with __device__, declares a variable that:

• 驻留在恒定的内存空间中，


• 具有生存期


• 可从网格中的所有线程
  以及主机通过运行时库
  （ cudaGetSymbolAddress（） cudaGetSymbolSize（） / cudaMemcpyToSymbol（） /
cudaMemcpyFromSymbol（） API和 cuModuleGetGlobal（） 用于
驱动程序API）。



• Resides in constant memory space,
• Has the lifetime of an application,
• Is accessible from all the threads within the grid and from the host through the runtime library (cudaGetSymbolAddress() / cudaGetSymbolSize() / cudaMemcpyToSymbol() / cudaMemcpyFromSymbol() for the runtime API and cuModuleGetGlobal() for the driver API).

CUDA，常量存储器是通过高速缓存（为此目的有一组专用的PTX加载指令）访问的专用，静态，全局存储区。 rm和只读，表示正在运行的内核中的所有线程。但是，恒定内存的内容可以在运行时通过使用上面引用的主机端API进行修改。这与使用 const 声明向编译器声明变量不同，后者在声明的范围内为变量添加了只读特征。两者根本不是同一回事。

In CUDA, constant memory is a dedicated, static, global memory area accessed via a cache (there are a dedicated set of PTX load instructions for its purpose) which are uniform and read-only for all threads in a running kernel. But the contents of constant memory can be modified at runtime through the use of the host side APIs quoted above. This is different from declaring a variable to the compiler using the const declaration, which is adding a read-only characteristic to a variable at the scope of the declaration. The two are not at all the same thing.

这篇关于__device__ __constant__ const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！