How to config the array_partition size in Vivado_HLS?

[CHFang731]

Hi, I am new to using hls4ml.
When I was executing part7a_bitstream, I encountered the following error.

And I found this discussion in hls4ml
https://github.com/fastmachinelearning/hls4ml/issues/1157

So I tried to modify the code in part7a_bitstream to configure the size of array_partition, as shown below

When I configured MaximumSize to 4096 and compiled the model, my hls4ml_proj.yml was successfully configured, but my project.tcl was not configured.
project.zip

Does anyone know how to correctly configure the array_partition size?
Or can I just manually rewrite the project.tcl file and then build it?

thanks.

[CHFang731]

my env:

hls4ml == v1.0.0

Vivado 2020.1

Vivado HLS 2020.1

[bo3z]

Make sure your hls4ml is fully up-to-date or try installing from main. I recall this being a bug in hls4ml sometime ago but looking at the code base I think this was fixed before the last hls4ml release, but I might be wrong and this fix isn't yet reflected on PyPi.

Anyway, the quickest solution is to hardcode the value in hls4ml/templates/vivado/build_prj.tcl on line 164 to 4096 (https://github.com/fastmachinelearning/hls4ml/blob/88f1c5f3667b485873871578c819b17ff52b8bf8/hls4ml/templates/vivado/build_prj.tcl#L164C46-L164C58)

[CHFang731]

Hi, Thank you for your reply.

I think I found my problem. The problem is that the error may not have been reflected on PyPi. When I use pip3 install hls4ml to install hls4ml, the error still exists, but when I use pip3 install git+https://github.com/fastmachinelearning/hls4ml.git@main to reinstall and execute, the value of maximum_size has been correctly changed to.

By the way, it is also effective to modify the build_prj.tcl file directly and then execute hls4ml.build().

And I would like to ask if I can modify the value of Maximum_size directly in config? For example, using config['MaximumSize'] = 8192 or config['Model']['MaximumSize'] = 8192 or other methods?

[bo3z]

You can yes, and I think the latter option is the correct one but I am not fully sure. Worth trying to see both, to see which one is the correct one.

[vloncar]

config['Model']['MaximumSize'] = 8192 should work. Be very careful with this. AMD set it to a low default of 1024 in Vivado HLS for a reason 😉.

[CHFang731]

Hi😀

config['Model']['MaximumSize'] = 8192 doesn't seem to work on my hls4ml , not sure if it's because it's not defined in https://github.com/fastmachinelearning/hls4ml/blob/main/hls4ml/utils/config.py ,but I don't think it has much impact on me at this stage.

What interests me is why AMD presets the value of Vivado HLS to 1024. Is it because of the resource allocation strategy? Or is it that the experiment shows that 1024 is a more stable resource allocation strategy? Or is there another reason? This is something worth thinking about for me.