[Openib-windows] RE: [PATCH] cl_timer

Yossi Leybovich sleybo at mellanox.co.il
Sun Sep 18 01:07:09 PDT 2005


> -----Original Message-----
> From: Fab Tillier [mailto:ftillier at silverstorm.com] 
> Sent: Thursday, September 15, 2005 7:10 PM
> To: 'Yossi Leybovich'
> Cc: openib-windows at openib.org
> Subject: RE: [PATCH] cl_timer
> 
> 
> > From: Yossi Leybovich [mailto:sleybo at mellanox.co.il]
> > Sent: Thursday, September 15, 2005 3:14 AM
> > 
> > Fab
> > Attached patch for supporting nsec in timer stamp and 
> provide way to 
> > get the CPU freq We use this in our performance tests (I 
> hope that we 
> > upload them, to the OpenIB in the near future)
> > (I left the cl_get_time_stamp_sec as is so we will not 
> require to change old
> > code)
> 
> Why add the cl_get_time_stamp_usec function when 
> cl_get_time_stamp is identical? Why not just use the existing 
> function?  At a minimum, why not call the existing function 
> rather than duplicating it?
> 
I did not want to change the interface of existing code , 
but you right the cl_get_time_stamp and cl_get_time_stamp_usec do the same
I want new code to use cl_get_time_stamp_usec and later change existing code
to use it.

> Also, for performance tests it would probably be better 
> (lower impact of
> counters) to have two functions - one to return the raw tick 
> count, and the other to return the frequency.  That way, you 
> only perform the conversion to human readable time units 
> after the test completes, rather than at runtime.
> 
> If this is OK, I can code this up - it will add two 
> functions, cl_get_tick_count, and cl_get_tick_freq.
> 

OK its fine with me , the moment you check it in I will change the
perf_main.
(BTW I think I already add function to get the freq)

> Let me know.
> 
> Thanks,
> 
> - Fab 
> 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openfabrics.org/pipermail/ofw/attachments/20050918/fb46dab9/attachment.html>


More information about the ofw mailing list