Is there any way to get the clAMDFFT on MacOS? I run everything on linux but develop on MacOS.
BTW: The Apple FFT is radix 2 only and would require a different code base.
I've tried using the clAMDFFT on a Mac but I don't think it's possible.
I've been working on a non-power of two FFT based on the Apple FFT. Are you interested in 1D FFTs or do you need other dimensions, too?
My interest is for batched 2D, and possibly non-batched 3D.
VexCL library (https://github.com/ddemidov/vexcl) provides FFT that supports arbitrary vector expressions as input, multidimensional inputs (any number of dimensions), and arbitrary sizes. The performance is worse than CUFFT, but should be better than clAMDFFT. Some examples may be found here.
Although technically there may not be a problem with clAmdFFT using Apple products.
AFAIK, clAmdFFT is not supported on Apple platform. I would also recommend you to look for solution from Apple or OpenSource community.
Is there any way to get an unsupported compilation? We have 40 HD7970 at the moment and are looking to replace the 44 NVIDIA GTX cards we also have. We may have a couple hundred GPUs installed by the end of the year. We may even go with 7990 in a dense setup (8 cards in 4U).
If all AMD want is the gamer market, there is no doubt that Windows is the way to go. But if AMD is at all interested in the GPGPU market, Linux and MacOS should have tools on par with what is offered on Windows.
Linux is a fully supported platform, and we will fix any issue you report for it. But AFAIK apple has their own OpenCL implementation, and therefore AMD does not support it. You can contact apple, if you have any issues regarding running OpenCL on Mac.
Stepping through an OpenCL kernel with CodeXL on linux has been so unusably slow for me I gave up using it.
As you stated, OpenCL is already implemented on the Mac, hence AMD only would need to compile their libraries (i.e. FFT and BLAS), well short of the work NVidia put in:
Many Scientists use Macs, and scientist use GPGPUs.
Retrieving data ...