为什么atomicAdd没有实现双打？ [英] Why has atomicAdd not been implemented for doubles?

问题描述

为什么不将 atomicAdd（）作为CUDA 4.0或更高版本的一部分显式实现？

Why hasnt atomicAdd() for doubles been implemented explicitly as a part of CUDA 4.0 or higher?

从 CUDA计划指南4.1 的附录F第97页
atomicAdd的以下版本已经实现。

From the appendix F Page 97 of the CUDA programming guide 4.1 the following versions of atomicAdd have been implemented.

int atomicAdd(int* address, int val);
unsigned int atomicAdd(unsigned int* address,
                       unsigned int val);
unsigned long long int atomicAdd(unsigned long long int* address,
                                 unsigned long long int val);
float atomicAdd(float* address, float val)

一个小实现的atomicAdd双重如下
，我刚刚开始在我的项目中使用。

The same page goes on to give a small implementation of atomicAdd for doubles as follows which I have just started using in my project.

__device__ double atomicAdd(double* address, double val)
{
    unsigned long long int* address_as_ull =
                             (unsigned long long int*)address;
    unsigned long long int old = *address_as_ull, assumed;
    do {
        assumed = old;
old = atomicCAS(address_as_ull, assumed,
                        __double_as_longlong(val +
                               __longlong_as_double(assumed)));
    } while (assumed != old);
    return __longlong_as_double(old);
}

为什么不将上述代码定义为CUDA的一部分？

Why not define the above code as a part of CUDA ?

推荐答案

编辑：自CUDA 8起，双精度 atomicAdd（）在CUDA中使用SM_6X（Pascal）GPU中的硬件支持实现。

As of CUDA 8, double-precision atomicAdd() is implemented in CUDA with hardware support in SM_6X (Pascal) GPUs.

目前，没有CUDA设备支持 atomicAdd c> 如你所说，它可以用 atomicCAS 对64位整数，但是有一个非平凡的性能成本。

Currently, no CUDA devices support atomicAdd for double in hardware. As you noted, it can be implemented in terms of atomicCAS on 64-bit integers, but there is a non-trivial performance cost for that.

因此，CUDA软件团队选择将正确的实现文档作为开发人员的选项记录，而不是将其作为CUDA标准库的一部分。这样，开发人员不会在不知不觉中选择他们不明白的性能成本。

Therefore, the CUDA software team chose to document a correct implementation as an option for developers, rather than make it part of the CUDA standard library. This way developers are not unknowingly opting in to a performance cost they don't understand.

Aside：我不认为这个问题应该被封闭为不建设性 。我认为这是一个完全有效的问题，+1。

Aside: I don't think this question should be closed as "not constructive". I think it's a perfectly valid question, +1.

这篇关于为什么atomicAdd没有实现双打？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！