Never proceeds past benchmarks on Raspberry ARM64

[Botspot]

Describe the bug
A clear and concise description of what the bug is.

To Reproduce
Compile xmrig as shown in this script: https://github.com/Botspot/pi-apps/blob/master/apps/XMRig/install

Expected behavior
Running xmrig for the first time completes all benchmarks and then mining begins.

I think this is the same problem as what was encountered in issue #125.
Please respond. Let me know if you need additional context or information. I can provide you with remote access to my Raspberry Pi 5 for testing and debugging.

[realsetvin]

do you have hwloc installed? seems some kind of memory issue. Undoubtedly due to arm incompatibility

xmrig#3183

[realsetvin]

I know someone who is good with arm, will try to ask them to take a look later

[Botspot]

This info should help:

I copied a xmrig.json from another PC to skip the benchmarks, then tested one algorithm at a time.
It seems that all cn algos fail, while the remaining ones mine fine.

[realsetvin]

This info should help:

I copied a xmrig.json from another PC to skip the benchmarks, then tested one algorithm at a time. It seems that all cn algos fail, while the remaining ones mine fine.

do u have hwloc installed, ext?

[Botspot]

do u have hwloc installed, ext?

hwloc package is not installed. Should I install it?

[realsetvin]

do u have hwloc installed, ext?

hwloc package is not installed. Should I install it?

please do

[Botspot]

please do

OK, I have it installed now. Should I try recompiling xmrig now, or is there something from the output of hwloc-info that you need to see?

[Botspot]

It's worth noting that libhwloc-dev was installed prior to compilation. Was that the package you meant?
Also, I can offer you access to my Pi via remote desktop software.

[mickbot-92]

Hello 👋
When building the latest XMRig only with cmake .., the benchmark seems to work. After benchmarking, you can use the optimised one freely.
So this issue seems to be linked with the CMake flags 🤔
Compiled on RPi 400 (RPiOS Bookworm) then tested on RPi 5 (NextCloudPi, unofficial armbian, Bookworm).

[Botspot]

I had never updated this issue with my full findings. With the very helpful tip from @mickbot-92, it's worth sharing them now.

• After bypassing benchmarking, I found that all cn algorithms failed, and the others succeeded. I was able to make partial use of xmrig by setting all benchmark scores of the cn algos to 0.
• I got huge pages working after recompiling the kernel, which boosted performance significantly.
• Today after the comment from @mickbot-92, I just now found that the cause of all this trouble was the -ffast-math cmake flag. It seems that without that flag, xmrig performs fine.

[Botspot]

I have made a change to the Pi-Apps install script for XMRig which should solve the issue. Botspot/pi-apps@d6c9402