[PATCH v2] arm: Fix bugs with MVE vmov from two GPRs to vector lanes

[Richard Earnshaw]

On 18/05/2021 10:44, Alex Coplan via Binutils wrote:
> On 17/05/2021 14:40, Richard Earnshaw wrote:
>>
>>
>> On 17/05/2021 13:03, Alex Coplan via Binutils wrote:
>>> Hi all,
>>>
>>> This is a v2 of my previous patch:
>>> https://sourceware.org/pipermail/binutils/2021-May/116594.html
>>>
>>> This version uses a sensible sampling of the encoding space for the
>>> instruction in the test instead of testing every possible instance of
>>> the instruction.
>>>
>>> There is no code change wrt. the v1.
>>>
>>> Regtested on arm-eabi, no regressions.
>>>
>>> OK for trunk? What about backports?
>>>
>>> Thanks,
>>> Alex
>>>
>>
>> OK for trunk.
> 
> Thanks, pushed as e683cb412046b40085505f42dd141f542661a6ae.
> 
>>
>> Which branches are affected?  What's the impact on users in practice? eg,
>> does not having this fix cause gcc to fail?
> 
> Binutils branches back to 2.33 (when MVE supported was added) are
> affected. This does indeed cause GCC to fail, see:
> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100572
> 
> I wonder if backporting prior to 2.36 would be sensible?
> 
> Alex
> 
>>
>> R.

Nick has given the OK for a backport to 2.36 on the first iteration of 
this patch.  So yes, you can do 2.36.

R.