[PATCH] Improve big-endian support for ARM and AArch64

[Eric Christopher]

On Fri, Sep 26, 2014 at 6:24 AM, Charlie Turner <charlie.turner at arm.com> wrote:
> Hi Renato,
>
> On 26 September 2014 08:51, Charlie Turner <Charlie.Turner at arm.com> wrote:
>>> I didn't see Eric's hint, but I assume it was something to do with adding REQUIRES: object-emission to stop the failing test on Hexagon. Is that right?
>> It was to do with %llc_dwarf. I'm not sure what that expands to on Windows.
>
> %llc_dwarf just replaces win32 with mingw32 in the OS part of the triple,
>
> lit.cfg
> 164:llc_dwarf = 'llc'
> 166:  llc_dwarf += ' -mtriple='+config.target_triple.replace('-win32', '-mingw32')
> 167:config.substitutions.append( ('%llc_dwarf', llc_dwarf) )
>
> COFF files still get generated with mingw32 as the target anyway, so the relocation will still fail to be computed and the test will hence fail.
>
> But %llc_dwarf isn't used in architecture specific directories like X86/, instead plain llc is used, not sure why.
>

Those usually have triples attached to them that are guaranteed to emit dwarf.

-eric