FFT library for OpenCL app Can cuFFT be used inside OpenCL app?

Accelerated Computing CUDA CUDA Programming and Performance
1

Is it possible to call CUFFT library from OpenCL-based program?
If not, does nVidia provide some FFT library for using with OpenCL ?

2

I’m not aware of any FFT library for OpenCL from NVIDIA, but maybe OpenCL_FFT from Apple will work for you.

3

I’m not aware of any FFT library for OpenCL from NVIDIA, but maybe OpenCL_FFT from Apple will work for you.

4

I’m not aware of any FFT library for OpenCL from NVIDIA, but maybe OpenCL_FFT from Apple will work for you.

5

Thanks, I’m already using this library with my OpenCL programs. But I would like to compare its performance with cuFFT lib. Looks like CUDA + CUFFT works faster in FFT part than OpenCL+Apple oclFFT.

6

Thanks, I’m already using this library with my OpenCL programs. But I would like to compare its performance with cuFFT lib. Looks like CUDA + CUFFT works faster in FFT part than OpenCL+Apple oclFFT.

7

Thanks, I’m already using this library with my OpenCL programs. But I would like to compare its performance with cuFFT lib. Looks like CUDA + CUFFT works faster in FFT part than OpenCL+Apple oclFFT.

8

I have made a few quick benchmarks (for my very specific case, i.e. 2D 1024x1024 and 2048x2048 complex FFT). Speed of opencl and cufft are quite similar (opencl seems to gain speed if it has more data to process).

Anyway, the tool I wrote uses both opencl fft and cufft so you should be able to do the same for your application, however if you meant that you wanted to call cufft from opencl code, I don’t think this is possible…

9

I have made a few quick benchmarks (for my very specific case, i.e. 2D 1024x1024 and 2048x2048 complex FFT). Speed of opencl and cufft are quite similar (opencl seems to gain speed if it has more data to process).

Anyway, the tool I wrote uses both opencl fft and cufft so you should be able to do the same for your application, however if you meant that you wanted to call cufft from opencl code, I don’t think this is possible…

10

I have made a few quick benchmarks (for my very specific case, i.e. 2D 1024x1024 and 2048x2048 complex FFT). Speed of opencl and cufft are quite similar (opencl seems to gain speed if it has more data to process).

Anyway, the tool I wrote uses both opencl fft and cufft so you should be able to do the same for your application, however if you meant that you wanted to call cufft from opencl code, I don’t think this is possible…

11

What I need is to perform FFT via cuFFT on data already stored on GPU in OpenCL cl_mem buffer object.

That is, not to copy it back to host memory and reload into CUDA context bound memory buffer.

12

What I need is to perform FFT via cuFFT on data already stored on GPU in OpenCL cl_mem buffer object.

That is, not to copy it back to host memory and reload into CUDA context bound memory buffer.

13

(double post)

14

(double post)

15

Can you post this tool, elhefe38 ?

Thanks

16

Can you post this tool, elhefe38 ?

Thanks