Skip to content

Home

Jump to bottom
boonedoggle edited this page Sep 25, 2019 · 1 revision

Tutorial

A full tutorial of how to use this package to develop custom GNU Radio blocks that use CUDA may be found on the Deepwave Digital website:

https://deepwavedigital.com/tutorials/custom-gpu-signal-processing-with-gnu-radio-on-the-air-t/

 

https://deepwavedigital.com/wp-content/uploads/2019/09/gr-cuda-edited.gif

Demo GRC Block

The demo block provided in this toolkit inputs a signal, divides each sample by two on the GPU, and outputs the resulting signal with the amplitude cut in half. This is to be used as a baseline for you to create your own CUDA + GNU Radio processing blocks.

Settings

https://deepwavedigital.com/wp-content/uploads/2019/09/module.png  

  • GPU Device # - Note that the AIR-T only has one built-in GPU, which has index = 0

  • IO Type - Complex or Float

  • Vector Length - Number of samples for simultaneous computation on the GPU

  • Threads per Block - Tell the CUDA kernel how to split the workload into threads and blocks. The AIR-T's GPU can process up to 1024 threads per block.

Clone this wiki locally