[cfe-commits] [PATCH] AArch64 backend: Clang changes

[Tim Northover]

Hi John,

Thanks very much for your detailed comments. Most things are either
straightforward or I'm asking colleagues about (ABI!), which will have
to wait until next week. I'll give a more detailed reply then.

Meanwhile, there is one point that could be clarified today:

> -    case BuiltinType::SChar:     EltName = "poly8_t"; break;
> -    case BuiltinType::Short:     EltName = "poly16_t"; break;
> +    case BuiltinType::UChar:     EltName = "poly8_t"; break;
> +    case BuiltinType::UShort:     EltName = "poly16_t"; break;
>
> It's not appropriate for this patch to change the element type of these
> vectors for existing platforms.  If you really want to do this — you believe
> that these vectors should have unsigned char elements even on ARM32 —
> then please prepare a separate patch for that change and start a new thread,
> because that will require discussion.

It's more a case of convenience, as I recall. The AAPCS leaves the
signedness of the polynomial types unspecified (they're over GF(2)
donchaknow!) so the choice is arbitrary, but the AArch64 PCS requires
that they should be unsigned.

I'm happy to split this off for separate review as you think best, but
the justification for the AArch32 change only exists in AArch64.

Cheers.

Tim.