Recent

Author Topic: fpga programming with an arm controller using opencl  (Read 454 times)

Thaddy

  • Hero Member
  • *****
  • Posts: 16390
  • Censorship about opinions does not belong here.
fpga programming with an arm controller using opencl
« on: September 16, 2024, 10:14:39 am »
I came across this:
https://github.com/Er1cZ/Deploying_CNN_on_FPGA_using_OpenCL/tree/master
and this:
https://github.com/Er1cZ/Deploying_CNN_on_FPGA_using_OpenCL/blob/master/GettingStartedTutorial.md

Which suggests you can set up opencl to program fpga's.

Since fpc has some opencl support and already supports arm embedded, would it be feasable to write a freepascal front end to program fpga's in that manner?

I have used fpga's in the past using verilog but that was too difficult for me so our team at the time eventually corrected and implemented my design (high performance real-time ticker processing at the bank)

It would be fun if we could get that to work, but at first glance the opencl bindings are not really complete.

The getting started tutorial describes a convolutional neural network.

There is a free toolchain available from Intel:
« Last Edit: September 16, 2024, 10:30:17 am by Thaddy »
There is nothing wrong with being blunt. At a minimum it is also honest.

MarkMLl

  • Hero Member
  • *****
  • Posts: 8111
Re: fpga programming with an arm controller using opencl
« Reply #1 on: September 16, 2024, 10:48:07 am »
I've done a bit of "playing around" in this area, with Altera Cyclone and Tang Nano chips of various vintages which meant having to scrape around for support software.

In terms of loading, I'm not sure that OpenCL is the way to go: OpenFPGALoader might be a better bet since it appears to be where the love is going.

https://trabucayre.github.io/openFPGALoader/compatibility/fpga.html

I'd add that latterly I've been trying to put my FPGA and embedded development environments into Docker containers, and a major sticking point is the extent to which security has to be bypassed to allow the container to access the programming hardware. If nothing else, I feel that this argues that it's desirable to minimise the number of different loder tools being used, and to take the amount of community support and understanding into consideration.

MarkMLl
MT+86 & Turbo Pascal v1 on CCP/M-86, multitasking with LAN & graphics in 128Kb.
Logitech, TopSpeed & FTL Modula-2 on bare metal (Z80, '286 protected mode).
Pet hate: people who boast about the size and sophistication of their computer.
GitHub repositories: https://github.com/MarkMLl?tab=repositories

 

TinyPortal © 2005-2018