This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [ppc] Include hwcap as ifunc argument
On Mon, May 21, 2012 at 4:36 PM, David Miller <davem@davemloft.net> wrote:
> From: Richard Henderson <rth@twiddle.net>
> Date: Mon, 21 May 2012 13:29:59 -0700
>
>> On 05/21/12 12:42, David Miller wrote:
>>> And on this topic I frankly think, as I did when I made the original
>>> sparc change, that we should pass the hwcaps into the IFUNC resolvers
>>> on every target not just a select few.
>>
>> I'm ok not doing this for x86. ?The hwcap word is totally useless
>> for x86_64, and next to useless for i686. ?Users will have to go
>> to the other cpuid words for anything interesting anyway.
>
> This argument against doing it on x86 is circular, let me tell you
> why.
>
> The only reason it's useless on x86, is because x86 ran into the
> same problem I'm about to run into on sparc. ?Namely, running out
> of bits in the HWCAP aux vector entry.
>
> So the only reason it's really useless is because it only reports
> the first cpuid word.
>
> If the aux vector hwcaps were extended to be able to report more
> words, and then subsequently propagate them into the ifunc, then
> we'd be solving two problems.
>
> First, I'd have a solution for the sparc hwcap issue.
>
> Second, we really could start to move away from all of the cpuid
> probing stuff that has to be done on x86.
I couldn't agree more. It would be great if we could avoid the cpuid probing.
Cheers,
Carlos.