Is cuFFTDx supported on the Jetson Xavier AGX platform?

Accelerated Computing GPU-Accelerated Libraries
1

The title of the topic pretty much says it all. The v0.3.0 documentation mentions architectures 6.0, 6.1, 6.2, 7.0, 7.5, and 8.0, but not 7.2 (which the Xavier AGX is). I wasn’t sure whether this implies that 7.2 is not supported. I tried running some of the examples and got mixed results:

  • block_fft_performance exits immediately with no output.
  • convolution/convolution_r2c_c2r exits immediately with no output.
  • All of the simple_fft_block* examples exit immediately with no output.
  • simple_fft_thread/simple_fft_thread_fp16 seem to run fine and report “Success” at the end.
  • nvrtc_fft_thread exits with an error about not being able to find the type_traits header.

I’m running with the latest software version from Jetpack. It would be a large benefit for my application if I could leverage cuFFTDx, so I’m hoping to figure out how to make it work. The above results maybe suggest that block execution method doesn’t work, but thread execution method does?

2

cuFFTDx yet support sm72. It will be in the next EA release.

3

Is there any estimate on the timeline/roadmap forward on when releases will occur, or when cuFFTDx will exit EA?

4

No ETA at this time. However, I will announce the releases to the EA members when they’re available.