[llvm] r267206 - [AArch64] Fix optimizeCondBranch logic.

[Quentin Colombet via llvm-commits]

Hi Renato,

That helped!

I missed that the encoding of the X variant could not represent immediate smaller than 32, thus we have to stick to the W variant but take the right subreg.
Thanks to Tim for helping me on the encoding!

Will post a fixed patch shortly.

Cheers,
Q.
> On Apr 25, 2016, at 11:27 AM, Renato Golin <renato.golin at linaro.org> wrote:
> 
> Hi Quentin,
> 
> When I just removed llvm-tblgen and lib/TableGen and re-built, the
> binary was identical, but but cleaning and re-building, I got the
> issue (attached).
> 
> Hope that helps.
> 
> --renato
> 
> 
> 
> On 25 April 2016 at 18:15, Quentin Colombet <qcolombet at apple.com> wrote:
>> Could you send me the binary of llvm-tblgen generated with and without the patch?
>> (Or just a diff)
>> 
>> Thanks,
>> -Quentin
>>> On Apr 25, 2016, at 10:04 AM, Renato Golin <renato.golin at linaro.org> wrote:
>>> 
>>> On 25 April 2016 at 17:42, Quentin Colombet <qcolombet at apple.com> wrote:
>>>> Hi Renato,
>>>> 
>>>> Sorry about the breakage.
>>> 
>>> No worries, let me know if you need anything.
>>> 
>>> --renato
>> 
> <llvm-tblgen.aarch64.bug.zip>