AMP批次中的FFT [英] FFT in AMP batches

问题描述

我有数据数组:

数据[1024] [1024]

data[1024][1024]

我想对每一行进行FFT.因此，我需要对每个FFT大小= 1024进行1024个FFT.如何通过在GPU中使用AMP FFT库来做到这一点? CUDA具有类似的带有批处理参数的FFT库.如何分配批次"?用于AMP FFT?

I would like to do FFT for each row. So I need to do 1024 FFTs with each FFT size = 1024. How to do this by using AMP FFT library in GPU? CUDA has similar FFT library with batch parameters. How to assign "batches" for AMP FFT?

for(int i=0; i<1024; i++)
{
    Do 1D FFT;
}

上面的代码不是我所需要的.我想在GPU中全部并行执行FFT.

The above code is not what I need. I like to do FFT all parallel in GPU.

推荐答案

当前的AMP FFT库确实具有1D，2D和3D配置，但是正确现在，它在所有维度上执行FFT.例如，如果使用2D FFT，它将执行两个独立的1D批量FFT.当前的AMP FFT库没有公开让您指定的API 您要对其进行FFT转换的维子集.但是，如果这对您来说很重要，我们可以考虑将该功能添加到AMP FFT库中.请让我们进一步了解您的情况以及批处理FFT功能将如何 为您有所作为.谢谢！

The current AMP FFT library does have 1D, 2D and 3D configuration, but right now, it does FFT on all dimensions. For example, If you use 2D FFT, it will do two independent 1D-batch FFTs. The current AMP FFT library doesn't expose APIs to let you specify the subset of dimensions you would like to do FFT transformation on. But if it's something very important to you, we could consider to add that feature to the AMP FFT library. Please let us know more about your scenarios and how the batch FFT feature would make a difference for you. Thanks!

这篇关于AMP批次中的FFT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！