[cfe-commits] [patch] Make ARM NEON polynomial types unsigned

[Bob Wilson]

On Jan 14, 2013, at 11:00 AM, Tim Northover <t.p.northover at gmail.com> wrote:

> Hi Bob,
> 
>> What is the motivation for switching to unsigned?
> 
> I personally believe it's a more sane for a polynomial, by quite a
> margin. For example it means that casting a poly8_t to a poly16_t will
> use the natural embedding rather than converting (say) x^7 into x^15 +
> x^14 + ... + x^7.

OK.  That makes sense.

> 
>> Is there something different about AArch64 that makes you
>> want to diverge from the de facto standard for those types?
> 
> I wouldn't call 2/3 a de facto standard, just a mess.
> 
>> If not, it really sounds to me like it might be best to just
>> switch everything (including your AArch64 ABI) to use
>> signed poly types.
> 
> That's not really a viable option at this stage. We're fixing bugs
> (not that we expect any, fingers crossed), but not changing the ABI
> for things like this.

So you're more willing to change the 32-bit ABI(s) than the 64-bit ABI?  That doesn't make sense to me.  I'm really not necessarily opposed to the change and your explanation above makes a lot of sense, but I don't think AArch64 compatibility is a good justification for an ABI-breaking change to the 32-bit targets.