Re: Something fishy with the toolchain in unstable

James Cowgill <[email protected]> Mon, 17 Aug 2015 10:21:03 +0100

Newsgroups	gmane.linux.debian.ports.mips
Message-ID	<[email protected]>

Hi,

On Sat, 2015-08-15 at 11:14 +0900, Mike Hommey wrote:
> On Sat, Aug 15, 2015 at 11:01:00AM +0900, Mike Hommey wrote:
> > The build error manifests itself in the form of multiple lines of the
> > form:
> > jsimd_mips_dspr2.S:: Error: float register should be even, was 
> > number>
> > 
> > I wish I could test a downgrade of binutils in the schroot on the porter
> > box, but I can't.
> > 
> > Could someone please look into this?
> 
> Okay, I found the difference:
> gcc passes -mips2 -mfpxx to as that it wasn't passing before. Removing
> -mfpxx fixes the issue.
> 
> Now the question is whether this is expected, and the libjpeg-turbo code
> changed accordingly, or whether gcc shouldn't pass those new flags?

Yes this is an expected change. For gcc-5, the MIPS ABI was altered
slightly by enabling FPXX, so gcc-5 passing -mfpxx to as is an expected
change. See #789612.

This page has all the gory details of the FPXX ABI:
https://dmz-portal.mips.com/wiki/MIPS_O32_ABI_-_FR0_and_FR1_Interlinking

Any C code and any assembly which does not use the FPU is not affected,
but some assembly which does use the FPU might need altering and it
looks like libjpeg-turbo is one of those cases. I'll have a look at it
in a moment.

Thanks,
James

signature.asc (application/pgp-signature, 819 B)