Enabling RISC-V Big-Endian Support in GCC Toolchain

[Utatsu00]

I'd like to use the -mbig-endian option, which is part of the RISC-V GCC toolchain.

The GCC manual states that the -mbig-endian option targets riscv32be or riscv64be.

When I attempt to compile a RISC-V program in big-endian mode using the -mbig-endian option, do I need any special configurations? Or can I simply build the toolchain as usual and then just include the -mbig-endian option during compilation?

[TommyMurphyTM1234]

Do you actually have hardware that implements BE RISC-V?

It used to be the case today BE support didn't play well (or at all?) with vector extensions but I'm not sure if that's still the case. Just mentioning it in case that's of any relevance to you.

This may also be of interest/relevance to you...

• https://riscv.org/ecosystem-news/2025/03/to-boldly-big-endian-where-no-one-has-big-endianded-before/

[Utatsu00]

Thank you for your comments.

I'm currently in the process of evaluating whether to design a RISC-V processor that supports big-endian.

Therefore, I'd like to confirm the byte order for both instructions and data when a program is compiled with a RISC-V big-endian capable GCC.

I anticipate that instructions will still be compiled in little-endian even when using a big-endian GCC, so my primary interest is in the data section.

I couldn't quite understand how to use the GCC big-endian option from the link you provided. Are you familiar with this?

[TommyMurphyTM1234]

I anticipate that instructions will still be compiled in little-endian even when using a big-endian GCC, so my primary interest is in the data section.

That's my understanding too. And memory mapped device I/O may be too I think?

As I've mentioned I'm not sure if vector extensions work well or at all when compiling in BE mode.

I couldn't quite understand how to use the GCC big-endian option from the link you provided.

What issues have you experienced when you tried?

Maybe if you ping the Codethink people they might be able to offer some advice/guidance?

[Utatsu00]

When I build riscv-gcc-toolchain, I set the --target option in configure to riscv32be--, but when I actually run make, the setting is not reflected and riscv32-- is set as the target.
Is this a problem, or can the compiler executable riscv32-unknown-elf-gcc also be compiled with the option -mbig-endian to compile with the big-endian I'm assuming?

[TommyMurphyTM1234]

When I build riscv-gcc-toolchain

You'll need to state more clearly exactly how you're building your toolchain - e.g. your git clone, configure, make commands and maybe capture, zip, and upload your build log.

Is this a problem, or can the compiler executable riscv32-unknown-elf-gcc also be compiled with the option -mbig-endian to compile with the big-endian I'm assuming?

I don't really understand. If you want to compile code in BE mode then do so as per the GCC documentation?

[TommyMurphyTM1234]

For what it's worth, if I build the toolchain in its default configuration (i.e. LE and default arch/abi of rv64gc/lp64d and with --enable-multilib) and compile a simple test source file to object it has the following in the disassembly:

test.o:     file format elf64-littleriscv

If I compile it with -mbig-endian it has:

test-be.o:     file format elf64-bigriscv

I don't see any difference in the actual code generated for a simple example but I'm not sure if that's significant for the trivial test code that I used.

However If I try to compile and link to ELF with -mbig-endian it fails:

./riscv64-unknown-elf-gcc test.c -mbig-endian
/home/user/Downloads/issue-1742/installed-tools/lib/gcc/riscv64-unknown-elf/15.1.0/../../../../riscv64-unknown-elf/bin/ld: /home/user/Downloads/issue-1742/installed-tools/lib/gcc/riscv64-unknown-elf/15.1.0/../../../../riscv64-unknown-elf/lib/crt0.o: ABI is incompatible with that of the selected emulation:
  target emulation `elf64-littleriscv' does not match `elf64-bigriscv'
...

which doesn't really surprise me since the toolchain (even with --enable-multilib) doesn't provide support for BE libs etc.

So while the toolchain itself can seemingly compile in BE mode, by default full support for BE (e.g. multilibs etc.) doesn't seem to be available automatically.

I can't find anything so far that explains if/how it's possible to, for example, build a toolchain that works by default in BE mode and provides all relevant libraries and other components that also support BE mode.

I would be inclined to ask the likes of the Codethink if they have any comments on how to build a BE capable toolchain (although their focus seems to be the Linux rather than bare metal toolchain in case that matters)...

[TommyMurphyTM1234]

Any update @Utatsu00?
Does compiling with -mbig-endian do what you need?
If not then what specific problems do you have?

If changes/enhancements are required to support BE targets in the toolchin then this issue might need to be logged in the relevant upstream projects - e.g. GCC, Binutils, Newlib etc. - where ongoing development is done (not in this repo).

[pz9115]

FYI,

An initial specification for RISC-V Big Endian ABI is under discussing in psABI group
riscv-non-isa/riscv-elf-psabi-doc#470

[bjdooks-ct]

We have a big-endian kernel, qemu and patches for buildroot from our work from the blog-post above. The project home is based at https://gitlab.com/CodethinkLabs/riscv_bigendian which has our work. We also have been in touch with MIPS who have done similar work. We'll be trying to sort out getting projects updated.

The TL;DR, glibc and uclibc require some patches to work properly (musl probably too) the rest of user-space should just worlk.

[TommyMurphyTM1234]

In case it's of any relevance in this context...

• Linus Torvalds Lashes Out At RISC-V Big Endian Plans