|
From: Ulrich D. <dr...@re...> - 2006-11-07 07:33:50
Attachments:
d-valgrind-core2
|
Cachegrind needs some more support to automatically detect cache size=20 for more recent Core 2 processors. The latest revision of application=20 note 485 from Intel shows the values. The only complication is that Intel reused value 0x49. Patch is attached. --=20 =E2=9E=A7 Ulrich Drepper =E2=9E=A7 Red Hat, Inc. =E2=9E=A7 444 Castro St = =E2=9E=A7 Mountain View, CA =E2=9D=96 |
|
From: Nicholas N. <nj...@cs...> - 2006-11-22 11:52:56
|
On Mon, 6 Nov 2006, Ulrich Drepper wrote: > Cachegrind needs some more support to automatically detect cache size for > more recent Core 2 processors. The latest revision of application note 485 > from Intel shows the values. > > The only complication is that Intel reused value 0x49. > > Patch is attached. Turns out Josef added code to handle this, it's in the trunk but not the 3.2.X branch, it probably should be. Josef handled the 0x49 case slightly differently -- Josef, do you have any comments about this? Nick |
|
From: Andi K. <ak...@su...> - 2006-11-24 19:50:43
|
Nicholas Nethercote <nj...@cs...> writes: > On Mon, 6 Nov 2006, Ulrich Drepper wrote: > > > Cachegrind needs some more support to automatically detect cache size for > > more recent Core 2 processors. The latest revision of application note 485 > > from Intel shows the values. > > > > The only complication is that Intel reused value 0x49. > > > > Patch is attached. > > Turns out Josef added code to handle this, it's in the trunk but not the > 3.2.X branch, it probably should be. Josef handled the 0x49 case slightly > differently -- Josef, do you have any comments about this? Longer term it would be better to switch to CPUID 4. That is the new "intel way" to detect caches and should not require changes for each new CPU. Unfortunately it is also harder to parse. -Andi |
|
From: Josef W. <Jos...@gm...> - 2006-11-22 21:31:31
|
On Wednesday 22 November 2006 12:52, Nicholas Nethercote wrote: > On Mon, 6 Nov 2006, Ulrich Drepper wrote: > > > Cachegrind needs some more support to automatically detect cache size for > > more recent Core 2 processors. The latest revision of application note 485 > > from Intel shows the values. > > > > The only complication is that Intel reused value 0x49. > > > > Patch is attached. > > Turns out Josef added code to handle this, it's in the trunk but not the > 3.2.X branch, it probably should be. Josef handled the 0x49 case slightly > differently -- Josef, do you have any comments about this? Hmm... as commit message r6211 says: I just used Intels newest version (v.22) of the x86 processor manuals, available at http://www.intel.com/products/processor/manuals/index.htm In Volume 2a, there is the documentation for the CPUID instruction, with cache parameters given in table 3-17. There is no note about the reuse of value 0x49. On page 26 of this manual (version 22), there is a "See also" section, referencing the application note 485 http://www.intel.com/support/processors/sb/cs-009861.htm Unfortunately, I only can get an old version of this note (from 2003); and 0x49 is not used at all there. What is the URL for the current version of application note 485? Josef > > Nick > > ------------------------------------------------------------------------- > Take Surveys. Earn Cash. Influence the Future of IT > Join SourceForge.net's Techsay panel and you'll get the chance to share your > opinions on IT & business topics through brief surveys - and earn cash > http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV > _______________________________________________ > Valgrind-users mailing list > Val...@li... > https://lists.sourceforge.net/lists/listinfo/valgrind-users > |
|
From: Ulrich D. <dr...@re...> - 2006-11-22 21:48:12
|
Josef Weidendorfer wrote: > What is the URL for the current version of application note 485? Just use google instead of the lazyweb: http://www.google.com/search?hl=3Den&lr=3D&q=3Dapplication+note+485+pdf+s= ite%3Aintel.com&btnG=3DSearch --=20 =E2=9E=A7 Ulrich Drepper =E2=9E=A7 Red Hat, Inc. =E2=9E=A7 444 Castro St = =E2=9E=A7 Mountain View, CA =E2=9D=96 |
|
From: Josef W. <Jos...@gm...> - 2006-11-22 22:27:12
|
On Wednesday 22 November 2006 22:47, Ulrich Drepper wrote: > Josef Weidendorfer wrote: > > What is the URL for the current version of application note 485? > > Just use google instead of the lazyweb: OK, convinced. Intel seems to be reusing numbers if the cache parameters match exactly. Here, the reuse is for (4MB, 16-way associative, 64 byte line size) either for the 2nd or 3rd level (Xeon MP, Family F, Model 6). I think we should use your version. Nick? Josef > > http://www.google.com/search?hl=en&lr=&q=application+note+485+pdf+site%3Aintel.com&btnG=Search > |
|
From: Nicholas N. <nj...@cs...> - 2006-11-22 23:57:32
|
On Wed, 22 Nov 2006, Josef Weidendorfer wrote: > Intel seems to be reusing numbers if the cache parameters match exactly. > Here, the reuse is for (4MB, 16-way associative, 64 byte line size) either > for the 2nd or 3rd level (Xeon MP, Family F, Model 6). > > I think we should use your version. > Nick? You know better than I do :) Sounds good, Josef, would you mind committing the improvement to both Cachegrind and Callgrind? Thanks. Nick |