[llvm] [bolt][aarch64] Add R_AARCH64_P32_ABS16/32 relocations (PR #143773)

[via llvm-commits]

whubeibei wrote:

> > > > I'd say it's more that this PR is still unfinished :)
> > > > 
> > > > * [[BOLT][AArch64] Add jump table support using .llvm_jump_table_info #132114](https://github.com/llvm/llvm-project/pull/132114)
> > > > 
> > > > Mystery solved then. Τhanks for being keen and trying it out! Would you mind if I go ahead and close #143709 ?
> > > > BTW, @aaupov is the diff rendering properly here?
> > > > 
> > > > * https://github.com/llvm/llvm-project/pull/132114/files
> > > > 
> > > > When I pull that PR locally I see changes on 17 files, including JumpTable.cpp, but online I only see 7 changed files.
> > > 
> > > 
> > > I agree with closing about #143709
> > 
> > 
> > thanks for your earlier comment!
> > based on the jump table support introduced in PR #132114.
> > As we noted, BOLT attempts to generate relocations for jump table entries, but currently there is no suitable relocation type for these entries on AArch64.
> > Since I’m still not very familiar with BOLT internals and ARM architecture specifics, I would like to seek more detailed guidance on how to properly improve or extend the implementation to handle this case.
> > Specifically:
> > 
> > * Should we define a new AArch64 relocation type within BOLT for jump table entries?
> 
> No, we just need to emit entry as symbol which should be encoded according to JT pattern
> 
> > * Or is there an alternative approach recommended for jump tables using `.byte` offset entries without relocations?
> > * Or are there any other best practices or existing mechanisms in BOLT that I might have overlooked?
> > 
> > I’m eager to learn and contribute to the fix once I better understand the recommended path forward.
> > Thank you very much for your advice and help!
> 
> I implemented JT for aarch64 but for GCC, I will help you here, we have a holidaynow but on the next week I will try to resolve this complicated moment

Thanks a lot for your quick reply and explanation!

I get that we don’t need new relocation types, and the jump table entries should just be emitted as symbols following the JT pattern.

Really appreciate that you’re willing to help with the AArch64 jump table stuff based on your GCC work. Enjoy your holiday, and I’m looking forward to working with you when you’re back.

If you have any docs or example code about the JT pattern or symbol encoding, could you please share? I’d like to start getting familiar with it while waiting.

Thanks again for your help!

https://github.com/llvm/llvm-project/pull/143773