Strange phenomenon when running FFT on GP

Question

anaconpa 2021-11-22

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1592554-strange-phenomenon-when-running-fft-on-gp

评论： anaconpa 2021-11-22

在 MATLAB Online 中打开

Parallel Computing Toolbox 7.4

MatlabR2021a

GPU Memory 5GB

I ran the following code.

    clear
    Nx = 1281;
    Ny = 340;
    Nz = 340;
    A = gpuArray(single(rand(Nx, 1)));
    B = single(rand(Nx, Ny, Nz));
    C = gpuArray(single(rand(Nx, 1)) + i*single(rand(Nx, 1)));
   % ↓ The 9th line ↓
   answer = A .* B -   real(ifftn(C .* gpuArray(B)));

Then, the first time on line 9, the following error occurs: "An invalid FFT plan was used on the GPU."

Next, I execute only the 9th line as, it was executed normally without any error.

Between the first and second time, GPU memory is not consumed by anything outside. I will get the same result no matter how many times you try

I have no idea what's going on.

Can you explain what is happening? thank you.

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Walter Roberson 2021-11-22

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1592554-strange-phenomenon-when-running-fft-on-gp#answer_837314

http://acarus.uson.mx/docs/cuda-5.5/CUFFT_Library.pdf

"CUFFT provides a simple configuration mechanism called a plan that pre-configures internal building blocks such that the execution time of the transform is as fast as possible for the given configuration and the particular GPU hardware selected. Then, when the execution function is called, the actual transform takes place following the plan of execution. The advantage of this approach is that once the user creates a plan, the library retains whatever state is needed to execute the plan multiple times without recalculation of the configuration."

So that is what an FFT plan is .

Why does it complain about invalid FFT plan? Why does it work afterwards?

Over in https://www.mathworks.com/matlabcentral/answers/786066-fftn-on-gpu-array#comment_1424841 we see a hint that maybe you ran out of GPU memory. The comment I link to shows some software to help keep track of GPU variables.