Re: [RFC][PATCH] generic hp_timing

[Mike Frysinger <vapier at gentoo dot org>]

On 31 Aug 2015 17:07, Maria Guseva wrote:
> The attached patch is a draft implementation of generic timing: via
> clock_gettime, vdso or system call, where available and via gettimeofday
> function call otherwise.
> 
> In order to eliminate performance impact in existing usages of HP_TIMING_NOW
> Âthe HP_TIMING_AVAIL is still set to 0. The generic timing is enabled only
> in DL_DEBUG_STATISTICS cases where performance is not critical. 
> However I suggest it may be wrong and we can use at least vdso clock_gettime
> as real HP_TIMING. And the rest probably shouldn't be even considered as
> high-precision timers and should be moved from hp-timing header to some new
> one. 
> 
> The need of generic timing was discussed in "LD_DEBUG option to measure init
> time" patch review:
> https://sourceware.org/ml/libc-alpha/2015-08/msg00084.html
> Proposed there new LD_DEBUG option needs HP_TIMING implementation to be
> usable on Arm platform in particular.

also worth noting that we now require __NR_clock_gettime everywhere,
so you can assume that rather than "#ifdef __NR_clock_gettime".
-mike

Attachment: signature.asc
Description: Digital signature