Failures with compilation on raspberry-pi 5

[mycroft92]

Failures with compilation on raspberry-pi 5

[quark17]

You can define GHCOPTLEVEL to an empty string (or to -O0 in case the default is not zero). This would override the definition in bsc/src/comp/Makefile:

GHCOPTLEVEL ?= -O2

However, the error message is for opt which is an LLVM tool and not GHC, so I don't know if the GHC optimization level will affect the use of opt. A web search for that error message shows that it has been caused in the past by a segfault in opt, and I see a closed issue on GHC's repo, but the result was just to open a bug on LLVM's repo (which was later closed).

I'm surprised that you're running GHC on a raspberry pi. Are you able to successfully run GHC to compile other programs besides BSC? I would typically run BSC on a beefier system and then move the generated Verilog to where it's needed, but I'm also not familiar with raspberry pi 5.

[mycroft92]

Thanks for the reply.

Pi 5 is reasonably powerful to run GHC (with a 4GB RAM). I switched the OS to ubuntu server and the entire toolchain works (compilation included) without needing to change any customizations. You may please close this ticket.

I agree with your second comment (about moving the verilog files to pi) because the compile times are high on pi. My original intent was to have CI/CD flow on the raspberry pi with an FPGA connected to it. My current system (MacOS Sonoma) doesn't have support for Xilinx vivado.

[quark17]

I'm fortunately running on an old x86 Mac, so I am able to run Altera tools in a VirtualBox VM. But for ARM-based Mac, I'm not aware if an x86 emulator exists, and I'd guess it might be slow, if it does?

[mieszko]

I've used UTM (which wraps qemu) to run x86 Linux VMs on arm Macs (GUI-less). Don't expect stellar performance but I've run Modelsim and Quartus this way on some not-terribly-large designs.