This is the mail archive of the libc-alpha@sourceware.org mailing list for the glibc project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCHv2, MIPS] Add support for O32 FPXX and program header based ABI information


"Joseph S. Myers" <joseph@codesourcery.com> writes:
> On Mon, 2 Jun 2014, Richard Sandiford wrote:
>
>> > - are you saying there is some existing Solaris-style check somewhere
>> > (where?) that disallows such a combination from executing on hardware
>> > lacking the features for all of the objects, even if the relevant
>> > instructions are only executed conditionally?  Or that some other
>> > combinations get disallowed at static link time?
>> 
>> No, I was arguing that taking the superset was the right behaviour.
>> You seemed to be saying that we should allow -mmsa to be used for
>> individual objects without marking the linked output as -mmsa
>> (because the functions in the -mmsa input object might all be
>> protected by a runtime check for MSA).
>
> I'm talking about the overall effect for the toolchain as a whole: that it 
> should allow building and running code using such runtime checks, without 
> requiring it to use IFUNCs.

OK, but my point was that that AFAIK has never been the case for MIPS
when using command-line options.  You have never been able to compile
a TU with -march=sb1, a TU with -mips64r2 and a TU with -march=octeon
and link them together.  And I think there are good reasons for that.
Similarly if you link -mmdmx and -mno-mdmx code together you get an
-mmdmx output.  

Personally I'm not convinced we should change the behaviour, but if we
did, I think we should do it consistently for all ISAs and ISA extensions
(whether from processor-specific extensions or ASEs).  I don't think MSA
should be a special case.

Thanks,
Richard


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]