[llvm-commits] [PATCH] SVR4 ABI support for the PowerPC backend

[Tilmann Scheller]

Hi Dale,

On Wed, Jul 1, 2009 at 11:26 PM, Dale Johannesen<dalej at apple.com> wrote:
> OK, I've looked these over.  They do not break any of the compatibility
> tests on Darwin, and I'm pretty sure none of the code changes will break
> Darwin.  I haven't looked over the SVR4-specific bits too carefully, but
> didn't see anything obviously wrong.  PPCCallingConv.td and
> PPCRegisterInfo.td say they have no newline at EOF, please fix those.
Ok

> One thing I'm concerned about is that you're replacing the existing "ELF"
> ABI with the "SVR4" ABI.  They seem to be different in a few areas, like
> callee-saved register usage.  Are you sure it is OK to destroy the older
> one?  It looks to me like somebody familiar with a spec had set up the
> register usage to be correct for their "ELF" target.  Nicolas Geoffray seems
> to have done most of what I'm talking about, so I think he should comment.
>  I think these might need to coexist as subdivisions of "elf", as on x86-32.
Actually the old ELF ABI already was the SVR4 ABI, as mentioned by
Nicolas in http://lists.cs.uiuc.edu/pipermail/llvmdev/2007-February/007911.html

All differences are due to missing or incorrect things in the original
implementation.

As ELF is not really an ABI but an object file format and there are
other ELF-based ABIs for PowerPC, like the EABI or the AIX ABI for
linux/ppc64, I did the rename from ELF to SVR4 to make this clear.

Just to be sure, did you take a look at the llvm-gcc patch as well?

Thanks for the fast review!

Greetings,

Tilmann