This is the mail archive of the
mailing list for the glibc project.
Re: [PATCH 1/5] PowerPC: Consolidate NPTL/non versions of vfork
- From: Adhemerval Zanella <azanella at linux dot vnet dot ibm dot com>
- To: libc-alpha at sourceware dot org
- Date: Fri, 23 May 2014 15:42:08 -0300
- Subject: Re: [PATCH 1/5] PowerPC: Consolidate NPTL/non versions of vfork
- Authentication-results: sourceware.org; auth=none
- References: <537F4C7E dot 4020005 at linux dot vnet dot ibm dot com> <537F939A dot 8080002 at twiddle dot net>
On 23-05-2014 15:29, Richard Henderson wrote:
> On 05/23/2014 06:26 AM, Adhemerval Zanella wrote:
>> + /* Load the TCB-cached PID value and negates it. If It it is zero
>> + sets it to 0x800000. And then sets its value again on TCB field.
>> + See raise.c for the logic that relies on this value. */
>> + lwz r0,PID(r13)
>> + cmpwi cr0,r0,0
>> + neg r0,r0
>> + bne- cr0,1f
>> + lis r0,0x8000
>> +1: stw r0,PID(r13)
> I was thinking about this when I did the alpha port:
> Why don't we use ~pid instead of -pid? That way we still get high-bit set
> semantics, but don't have to special-case 0, and so don't need a branch here
> (or a conditional move on other hosts)?
> There is a tiny matter of updating all of the architectures all at once though...
I don't see why not, however I think such change would be better suited for a following
patch. Also, do you see this code sequence as a hotspot for such optimization?