Re: aarch64 gcc kernel compilation

[Martin Husemann <martin%duskware.de@localhost>]

On Tue, Jul 17, 2018 at 02:23:36PM +0900, Ryo Shimizu wrote:
>  union fpelem {
>  	uint64_t u64[2];
> -	__uint128_t u128[1] __aligned(16);
> +	/* __uint128_t u128[1] __aligned(16); */
>  };

I like the alignement parts of the change, but I seriously do not
understand why the __uint128_t is a bad thing.

I'd be fine if that line is #if'd out for compilers not supporting
it, if they exist.

No support in printf and friends for this type is a strange justification
for removing a very natural representation for kernel purposes here
if it does not actually hurt (and we never printf it this way
at all, and if needed just a hexdump of the fpelem would be good enough).

I am not objecting this patch in general, but so far no coherent
proposal has been made why this should happen.

Martin