-1

I've searched the the whole internet concerning this but I haven't found a reliable answer. I need to point out that I'm a newbie to GPU programming but I've just ordered my Jetson TK1 unit because I love Simulations and I love Python. I have tried out Numpy on PC but I want to bring these simulations to Android. I've read about PyCuda and Numba but I prefer the latter because it is entirely in python, I won't have to write kernels in C. What I don't seem to grasp is how (after I install numbapro on my Jetson TK1) I will be able to direct my code at the multi-core ARM CPU of the Tegra K1 SOC and how I will be able to use the Unified Memory available in CUDA 6. Mind you, Numba doesn't support ARM LLVM yet. Or do I need to start a compulsory C++ and CUDA C++ classes to be able to access the CPU cores/threads.

groxli
  • 113
  • 1
  • 3
NyproTheGeek
  • 53
  • 2
  • 7
  • @talonmies Then how do I do GPGPU on existing architectures on Android. And is there a possibility that in the near future Android will be supported by CUDA perhaps maybe through an Android-based dev kit like Jetson TK1. – NyproTheGeek Oct 06 '14 at 10:04

1 Answers1

1

@NyproTheGeek,

I had the same question as you; I spent many hours trying to get NumbaPro running on my Jetson TK1 and kept running into dead ends. Since there is almost no NumbaPro + Jetson TK1 mention on the Internet I finally emailed the people at Continuum Analytics and they replied the Jetson TK1 is currently not supported. However, they also indicated Jetson TK1 is on the roadmap although no clear date was given.

Your observation about Numba (as opposed to NumbaPro) not supporting LLVM support is right. I believe Continuum Analytics offers that capability as part of the NumbaPro commercial offering.

You might have already trying the jit decorator available with Numba. That does offer significant performance gains by compiling code, although nothing near the firepower available through a GPU :-)

groxli
  • 113
  • 1
  • 3