Erlang RISC-V Architecture JIT Support

[yskelg]

We can develop and test RISC-V support with official Ubuntu spported Vision Five 2 RISC-V Board.

If we add architecture support for RISC-V, First of all, I think asmjit and
beam module support needed. Is that good starting point for RISC-V support?

If we need new architecture support, Can I ask is there a guide we can refer to?

[jhogberg]

RISC-V should already work out of the box, just not with the JIT. Have you had any issues building and running Erlang?

[yskelg]

@jhogberg thank you for your response. I'll build and test it on my ubuntu environment on my Vision Five 2 board to see if it works without JIT well. :)
Are there any plans to support JIT for RISC-V like arm, arm64 or x86, x86_64 architecture?

[yskelg]

I find some discussion threads about this topic.

• https://groups.google.com/g/erlang-programming/c/p6eTB8pe6Ls
• https://erlangforums.com/t/cosmopolitan-erlang-build-once-run-anywhere-possible/2004
• Add code path caching #6729 (comment)

[yskelg]

@jhogberg If it works well in the ubuntu environment on my Vision Five 2 board, would it be fine to submit a documentation or build script update PR for architectural support?

• https://github.com/erlang/otp/blob/OTP-26.0.2/erts/emulator/asmjit/core/environment.h#L328
• https://github.com/erlang/otp/blob/OTP-26.0.2/lib/erl_interface/configure#L4850
• https://github.com/erlang/otp/blob/OTP-26.0.2/lib/os_mon/src/memsup.erl#L571
• https://github.com/erlang/otp/blob/OTP-26.0.2/erts/autoconf/config.guess#L1494
• https://github.com/erlang/otp/blob/OTP-26.0.2/make/autoconf/otp.m4#L2959

[yskelg]

I tested erlang on real RISC-V board. It works fine!

starfive:~$ uname -ra
Linux starfive 5.15.0-starfive #1 SMP Mon Dec 19 07:56:37 EST 2022 riscv64 GNU/Linux
starfive:~$ erlang/bin/erl -noshell -eval 'io:fwrite("Hello, World!\n"), init:stop().'
Hello, World!

[yskelg]

@jhogberg Are there any plans to support JIT for RISC-V?

[garazdawi]

There are currently no plans to add RISC-V support for the JIT.

[adrianbrink]

What are the tradeoff of not having JIT support for RISC-V?

[garazdawi]

What are the tradeoff of not having JIT support for RISC-V?

Adding a new JIT backend is a lot of work and each backend adds to our maintainance burden as there is very little code that can be shared between the backends. So for a new backend to be added there needs to be a very strong usecase. Initially we had not planned to have any more than x86_64, but then apple went and changed the game by making aarch64 great again.