r205076 regresses mips 32 direct object emitter

[Rafael Espíndola]

Awesome!

On 31 March 2014 11:22, Daniel Sanders <Daniel.Sanders at imgtec.com> wrote:
> This parsing bug was fixed in r205200.
>
>> -----Original Message-----
>> From: Daniel Sanders
>> Sent: 31 March 2014 15:54
>> To: 'Rafael Espíndola'
>> Cc: Reed Kotler; llvm-commits at cs.uiuc.edu; Doug Gilmore
>> Subject: RE: r205076 regresses mips 32 direct object emitter
>>
>> The crash in xgot.s in r205042 is a bug in the assembly parser that doesn't
>> affect 'llc -filetype=obj'. There are a few kinds of relocations that aren't
>> recognised in MipsAsmParser::getVariantKind() and it ends up passing
>> VK_Mips_None into the MC layer instead of asserting or reporting an error.
>> I've got a fix for it in my working copy.
>>
>> > -----Original Message-----
>> > From: Rafael Espíndola [mailto:rafael.espindola at gmail.com]
>> > Sent: 31 March 2014 15:48
>> > To: Daniel Sanders
>> > Cc: Reed Kotler; llvm-commits at cs.uiuc.edu; Doug Gilmore
>> > Subject: Re: r205076 regresses mips 32 direct object emitter
>> >
>> > On 31 March 2014 09:48, Daniel Sanders <Daniel.Sanders at imgtec.com>
>> > wrote:
>> > > Hi,
>> > >
>> > >
>> > >
>> > > I'm not sure if this is related or not yet but I noticed that if I
>> > > resurrect xgot.ll (converted to an assembly test in
>> > > http://llvm-reviews.chandlerc.com/rL205042), it no longer passes.
>> >
>> > Unlikely. The .s test would already crash llvm-mc in r205042:
>> >
>> >
>> >
>> > Cheers,
>> > Rafael