svn commit: samba r21278 - in branches/SAMBA_3_0/source/smbd: .

Volker Lendecke Volker.Lendecke at SerNet.DE
Sat Feb 17 13:02:38 GMT 2007


On Sat, Feb 17, 2007 at 01:18:49PM +0100, Volker Lendecke wrote:
> On Sun, Feb 11, 2007 at 05:59:53PM -0800, Jeremy Allison wrote:
> > Can you run cachegrind on both methods to see why the
> > memcpy ends up taking less time ?
> 
> Ok, without the patch for 10000 runs cachegrind says:

Hmm. I take everything back :-)

I got fooled by -DDEVELOPER and clobber_region. Without that
both versions just don't show any significant difference:

Without

==9383== I   refs:      175,958,028
==9383== I1  misses:         70,890
==9383== L2i misses:          8,738
==9383== I1  miss rate:        0.04%
==9383== L2i miss rate:        0.00%
==9383== 
==9383== D   refs:      111,531,998  (66,671,301 rd + 44,860,697 wr)
==9383== D1  misses:        230,774  (   220,617 rd +     10,157 wr)
==9383== L2d misses:         25,086  (    21,384 rd +      3,702 wr)
==9383== D1  miss rate:         0.2% (       0.3%   +        0.0%  )
==9383== L2d miss rate:         0.0% (       0.0%   +        0.0%  )
==9383== 
==9383== L2 refs:           301,664  (   291,507 rd +     10,157 wr)
==9383== L2 misses:          33,824  (    30,122 rd +      3,702 wr)
==9383== L2 miss rate:          0.0% (       0.0%   +        0.0%  )


With patch

==9597== I   refs:      175,960,471
==9597== I1  misses:         49,092
==9597== L2i misses:          8,759
==9597== I1  miss rate:        0.02%
==9597== L2i miss rate:        0.00%
==9597== 
==9597== D   refs:      111,533,732  (66,672,336 rd + 44,861,396 wr)
==9597== D1  misses:        230,135  (   220,017 rd +     10,118 wr)
==9597== L2d misses:         25,003  (    21,299 rd +      3,704 wr)
==9597== D1  miss rate:         0.2% (       0.3%   +        0.0%  )
==9597== L2d miss rate:         0.0% (       0.0%   +        0.0%  )
==9597== 
==9597== L2 refs:           279,227  (   269,109 rd +     10,118 wr)
==9597== L2 misses:          33,762  (    30,058 rd +      3,704 wr)
==9597== L2 miss rate:          0.0% (       0.0%   +        0.0%  )
--9597-- 

Then the argument would be that with the patch we gain some
code simplicity, which was the original goal :-)

Volker
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
Url : http://lists.samba.org/archive/samba-technical/attachments/20070217/d498b53a/attachment.bin


More information about the samba-technical mailing list