| View previous topic :: View next topic |
| Author |
Message |
Thaidog
Veteran
Joined: 19 May 2004
Posts: 1053
|
 Posted: Mon Sep 06, 2010 10:52 pm Post subject: LinuxDNA high performance IPO enabled kernel patch |
 |
|
This time around Jeroen has managed to get the -ipo1 flag working with a lot of different parts of the kernel (but not all parts) Still it should provide a significant performance boost:
wget http://www.linuxdna.com/dna-ipo-2.6.34.1-intel64-1.patch
Also Luyi has created a 2.6.35 patch:
wget http://www.linuxdna.com/dna-2.6.35-intel64-2.patch
This one does not have any ipo support enabled yet.
These beta patches have been tested with icc v. 11.1.072 - Please post to the linuxdna google group for best support resolution! 
_________________
Registered Linux User: 437619
"I'm a big believer in technology over politics" - Linus Torvalds |
|
| Back to top |
|
 |
Letharion
Veteran
Joined: 13 Jun 2005
Posts: 1344
Location: Sweden
|
| Posted: Wed Sep 08, 2010 7:51 am Post subject: |
|
|
I'm glad to see you've gotten "up to speed" in the kernels. 
IIRC you only had a somewhat, in Gentoo terms, old release working last time I checked.
I saw https://forums.gentoo.org/viewtopic-t-814539-start-0.html, but those number don't mean anything to me.
It would be interesting to see some more benchmarks. Phoronix, anyone? |
|
| Back to top |
|
|
Thaidog
Veteran
Joined: 19 May 2004
Posts: 1053
|
| Posted: Thu Sep 09, 2010 3:38 am Post subject: |
|
|
| Letharion wrote: | I'm glad to see you've gotten "up to speed" in the kernels.
IIRC you only had a somewhat, in Gentoo terms, old release working last time I checked.
I saw https://forums.gentoo.org/viewtopic-t-814539-start-0.html, but those number don't mean anything to me.
It would be interesting to see some more benchmarks. Phoronix, anyone? |
LMbench would probably be the best thing to bench the kernels with because it is a more direct benchmark (and Torvalds uses it for his benchies  )
_________________
Registered Linux User: 437619
"I'm a big believer in technology over politics" - Linus Torvalds |
|
| Back to top |
|
|
Shining Arcanine
Veteran
Joined: 24 Sep 2009
Posts: 1110
|
| Posted: Thu Sep 09, 2010 4:02 am Post subject: |
|
|
| How is performance defined and what constitutes a boost? If these patches are an improvement, why have they not been accepted into the mainline kernel? |
|
| Back to top |
|
|
Ormaaj
Guru
Joined: 28 Jan 2008
Posts: 319
|
| Posted: Thu Sep 09, 2010 5:13 am Post subject: |
|
|
| Shining Arcanine wrote: | | How is performance defined and what constitutes a boost? If these patches are an improvement, why have they not been accepted into the mainline kernel? |
Because that would require a whole boatload of other modification in order to remove all GNUisms that prevent compilation with ICC. I would assume these patches are ICC specific and don't have anything to do with GCC's -fwhopr for instance. |
|
| Back to top |
|
|
Thaidog
Veteran
Joined: 19 May 2004
Posts: 1053
|
| Posted: Thu Sep 09, 2010 5:16 am Post subject: |
|
|
| Shining Arcanine wrote: | | How is performance defined and what constitutes a boost? If these patches are an improvement, why have they not been accepted into the mainline kernel? |
The patches are used to compile the kernel with icc instead of gcc.... therefore it's never going to be "mainline". We do submit bugs we find however in an effort to enhance the kernel source itself. Performance is simple... it makes the kernel do what it does faster. In some cases marginally faster in other cases noticeably faster.
_________________
Registered Linux User: 437619
"I'm a big believer in technology over politics" - Linus Torvalds |
|
| Back to top |
|
|
Shining Arcanine
Veteran
Joined: 24 Sep 2009
Posts: 1110
|
| Posted: Thu Sep 09, 2010 2:53 pm Post subject: |
|
|
| Thaidog wrote: | | Shining Arcanine wrote: | | How is performance defined and what constitutes a boost? If these patches are an improvement, why have they not been accepted into the mainline kernel? |
The patches are used to compile the kernel with icc instead of gcc.... therefore it's never going to be "mainline". We do submit bugs we find however in an effort to enhance the kernel source itself. Performance is simple... it makes the kernel do what it does faster. In some cases marginally faster in other cases noticeably faster. |
Is there some reason why the mainline kernel will not support icc in addition to gcc? I had heard of work that Intel had done a few years ago to fix issues in the kernel that kept it from being compiled in ICC and I thought that the mainline kernel would have merged that work by now.
By the way, are the LinuxDNA kernel patches exclusively targeted toward compilation with ICC? If that is the case, I might try them. It would be nice to compile the kernel with ICC. |
|
| Back to top |
|
|
Thaidog
Veteran
Joined: 19 May 2004
Posts: 1053
|
| Posted: Thu Sep 09, 2010 9:51 pm Post subject: |
|
|
| Shining Arcanine wrote: | | Thaidog wrote: | | Shining Arcanine wrote: | | How is performance defined and what constitutes a boost? If these patches are an improvement, why have they not been accepted into the mainline kernel? |
The patches are used to compile the kernel with icc instead of gcc.... therefore it's never going to be "mainline". We do submit bugs we find however in an effort to enhance the kernel source itself. Performance is simple... it makes the kernel do what it does faster. In some cases marginally faster in other cases noticeably faster. |
Is there some reason why the mainline kernel will not support icc in addition to gcc? I had heard of work that Intel had done a few years ago to fix issues in the kernel that kept it from being compiled in ICC and I thought that the mainline kernel would have merged that work by now.
By the way, are the LinuxDNA kernel patches exclusively targeted toward compilation with ICC? If that is the case, I might try them. It would be nice to compile the kernel with ICC. |
It's really more of a matter of icc supporting the kernel than the the other way around. Every release of icc needs to work with the code that was specifically written with gcc intrinsics in mind. This causes issues sometimes and we submit those bugs directly to Intel to make sure the next release of icc works correctly. So really we have two jobs in our project - make patch that compiles with icc and make sure icc will compile it. Because mainline kernel developers are most interested in making sure their code works on a variety of cpus and platforms gcc is the natural choice... our development focus is much more specific.
_________________
Registered Linux User: 437619
"I'm a big believer in technology over politics" - Linus Torvalds |
|
| Back to top |
|
|
Shining Arcanine
Veteran
Joined: 24 Sep 2009
Posts: 1110
|
| Posted: Fri Sep 10, 2010 4:06 am Post subject: |
|
|
| Is there an overlay in which ebuilds are available for these sources? |
|
| Back to top |
|
|
Thaidog
Veteran
Joined: 19 May 2004
Posts: 1053
|
|
| Back to top |
|
|
kernelOfTruth
Watchman
Joined: 20 Dec 2005
Posts: 6111
Location: Vienna, Austria; Germany; hello world :)
|
|
| Back to top |
|
|
|
Unsupported Software
