configuration OneAPI

[Gabriel-MarcoSilva]

Hello guys, how to configure oneAPI? Please give me the step by step, as I have tried many ways and nothing has worked.

[jmitrevs]

You need to install oneAPI 2025.0 and the fpga compiler. Note, the fpga compiler is transitioning from Intel to Altera, so you cannot use oneAPI 2025.1. https://www.intel.com/content/www/us/en/developer/tools/oneapi/fpga.html

[Gabriel-MarcoSilva]

Well, I had to ask chatGPT for help on how to configure OpenCLs, and he told me to install the FPGA SDK for OpenCL, I'm waiting for it to finish downloading, as it's a large file.

[jmitrevs]

And you installed the fpga compiler? It's either a separate download or with yum/dnf/apt.

[Gabriel-MarcoSilva]

these ones here?

[jmitrevs]

I think so.

[Gabriel-MarcoSilva]

Do you remember all the steps you used to use OneAPI?

I did it like this:

I installed HLS4ML and the other libraries
I used the code from the quickstart, changing the backend to OneAPI
I installed the OneAPI basetoolkit
I installed the DPC++ compiler

and that's it

[jmitrevs]

I am not sure I know the code from the quickstart. There is also a oneAPI branch of the hls4ml tutorial (https://github.com/fastmachinelearning/hls4ml-tutorial/tree/oneAPI) that you can try using (only parts 1-4 were ported, maybe parts of 6).

[Gabriel-MarcoSilva]

No, it wasn't, the terminal keeps saying that OpenCL is empty

[Gabriel-MarcoSilva]

@jmitrevs
The error occurs because of hls_model.compile(), when removing it it stops appearing, but I don't know if this solves the problem, as it now complains about a certain file: #include <sycl/ext/intel/prototype/interfaces.hpp>. Do you know how to solve it?

[jmitrevs]

It seems like you have a general oneAPI installation issue, not really an hls4ml issue. I am really not an expert on that. Some ideas are:

1. Try force setup of oneAPI (add --force), not just regular setup
2. Completely remove and reinstall oneAPI, potentially even removing the /opt/intel/oneapi directory before reinstalling.

[vloncar]

Hi Gabriel,

Indeed this is an oneAPI installation issue and not hls4ml issue. If you're using repository packages to install as opossed to the binaries from the website, here's what worked for me:

1. Uninstall everything (apt purge *oneapi*).
2. Install only 2025.0. I installed intel-oneapi-base-toolkit-2025.0 and intel-oneapi-compiler-fpga (which will install intel-oneapi-compiler-fpga-2025.0 as the only version available).
3. I still needed to do . /opt/intel/oneapi/setvars.sh --force to make the compiler properly available. I added . /opt/intel/oneapi/setvars.sh --force > /dev/null to ~/.profile to make it available every time I log in.

If none of this works, you'll need to ask in Intel's forums for help.

[Gabriel-MarcoSilva]

Thanks, but it didn't work, it keeps giving the same error regarding the OpenCL platform ID and name, the problem always occurs when running: make fpga_emu

[Gabriel-MarcoSilva]

@jmitrevs and @vloncar Could you show how OpenCL platform ID and name are configured? Because I don't want to use quartus, since it will be deprecated

[jmitrevs]

I am not familiar with configuring the OpenCL platform ID. The only other thing I can think of is if you are using other OpenCL software on the machine that is interfering? Is, for example, XRT also installed? I have seen messages related to that that before, but those messages could be ignored. It seems like multiple OpenCL libraries can mess things up.

[jmitrevs]

Another potential option is to work in a docker container.

[Gabriel-MarcoSilva]

How would I do this?

[jmitrevs]

For our testing environment we are setting up the container with:

FROM intel/oneapi-basekit:2025.0.1-0-devel-rockylinux9 AS hls4ml-test-base
LABEL [email protected]

# Install dependencies
RUN yum groupinstall -y "Development Tools" \
    && yum install -y sudo nano mc intel-oneapi-compiler-fpga \
    && yum clean all \
    && rm -rf /var/cache/yum
...

[Gabriel-MarcoSilva]

I don't know how to do this, can you help me with more details? I'm a beginner, I don't know how to use docker

[Gabriel-MarcoSilva]

@jmitrevs and @vloncar, Could you please do a step by step on another device, make a video and send it? Because I depend on this backend to use on the FPGA