Vits_HLS help

[Gabriel-MarcoSilva]

I used quick start and managed to run the build using the vitis backend, but how do I use it in Vivado now?

What I did: I created a project in Vivado and used the verilog files generated by build(), but when I go to synthesize, Vivado crashes.

How should I use the files generated in Vivado?

[bo3z]

Could you please post a more descriptive message of the crash? Without that it's hard to say what went wrong, but it could be any number of things related to how the Verilog files were connected to the rest of your design (non-hls4ml components).

If you're just interested in synthesis results (resource usage etc.), you don't need to import the hls4ml design into Vivado, you can just set vsynth=True in build(...) and this will run synthesis in Vivado as instructed by Vitis HLS (Vitis HLS has a lot of integration with the Vivado suite naturally, that is usually abstracted aways from the developer)

If you want to use the model in a larger design with some other, non-hls4ml components, you can export the HLS kernel as an IP, using the export=True option in the build(...) function. This will create an IP that can be imported into any Vivado project. If you need a different IP format (or any other options, like version number etc.), check the vitis_hls export_design documentation: https://docs.amd.com/r/en-US/ug1399-vitis-hls/export_design. You can modify the flags that hls4ml passes to this command in:

hls4ml/hls4ml/templates/vivado/build_prj.tcl

Line 232 in e690afe

export_design -format ip_catalog -version $version

[Gabriel-MarcoSilva]

Thanks for the answer, I initially used build(csim=True, synth=True, cosim=True), but how do I put it on the board?

Basically I went into Vivado, put the files from myproject_prj/solution1/syn/verilog and tried to do the synthesis and my laptop crashed, I went to do the bitstream and it crashed again...

Basically my question is: after build(), how do I send the code to the board?

[bo3z]

Check out part 7 of the tutorial, explaining how to deploy the model. To deploy a model, you need to integrate the model with the rest of the system, handle data movement etc. This is what the Accelerator backends do.