Intel HLS Installation not Found. Make sure 'i++' is on the path

[Viggo-Avila]

Hello, I've been trying to use hls4ml with the Intel De1-SoC CycloneV FPGA and I'm using Quartus Prime

When I tried to run hls_model.build(), that Error in the title shows.

I've made sure that quartus path is recognized in the environment since I can open quartus using the terminal. Addtionally, in the code I changed backend and part into backend=Quartus and part=CycloneV and there were no problems with that part.

Am I missing something? When I did this with the PYNQ-ZU board there was no problem.

[vloncar]

Make sure Intel HLS (not just Quartus) is installed and available.

[jmitrevs]

This seems to imply there is some support, though I haven't tried it: https://www.intel.com/content/www/us/en/developer/articles/system-requirements/oneapi-dpcpp/2025.html

[Viggo-Avila]

I see. Our current version of quartus is 20.1, and from the intel website it saya thay cycloneV is supported at quartus prime standard ed. 22.1, so does this mean that we just have to install that version of quartus?
Then how about in the hls4ml documentation where it said that the support for intel hls is only until 20.1 to 21.4? Would intel hls match the version of the quartus we're using, or can we just choose what kind of version we want, which in this case would be what hls4ml is compatible with.

[Viggo-Avila]

I saw this,

It says that for cycloneV board to be supported by quartus I need quartus to be 22..1 and intel hls to be 23.3. But the problem is is that hls4ml only supports intel hls from 20.1 to 21.4. Does this mean I can't use my CycloneV board for hls4ml? Guess I'll just have to stick with oneAPI for now

[jmitrevs]

The supported Intel HLS and Quartus versions are just what has been tested. Newer stuff may work just fine. You can try it. Note, if you are using io_stream, the oneAPI backend should be better. For io_parallel, the results should be pretty similar. (The oneAPI backend hls4ml code started as a port of the Intel HLS version.)

[Viggo-Avila]

Ohh, I see. Since I'm going to be synthesizing a CNN then using io_stream is inevitable, so oneAPI should be better. We'll still try with the intel hls and quartus, and if that fails then we'll switch to oneAPI. Thank you for the help.

[jmitrevs]

You can alternately try the oneAPI backend, though for that, you have to make sure you have oneAPI setup. It's still experimental, and we haven't tried it with CycloneV.