|
From: <mma...@ny...> - 2005-08-29 19:21:19
|
Hello, I did a google search for this instruction it looks like it was never reported before. Can I work around this ? Using valgrind (gentoo portage as of today) on AMD64. Thanks, Mathieu |
|
From: John R.
|
0xF 0xD 0xF is "prefetchw (%rdi)". This is purely a performance optimization, so use a binary editor to change those three bytes to 0x90 0x90 0x90 which is "nop; nop; nop". There are likely to be more instances, so try to figure out how to tell the compiler to avoid that sequence (or remove or disable the optimization flag.) -- |
|
From: Mathieu M. <mma...@ny...> - 2005-08-30 01:39:15
|
John Reiser wrote: > 0xF 0xD 0xF is "prefetchw (%rdi)". This is purely a performance > optimization, so use a binary editor to change those three bytes > to 0x90 0x90 0x90 which is "nop; nop; nop". There are likely to > be more instances, so try to figure out how to tell the compiler > to avoid that sequence (or remove or disable the optimization flag.) Just for google cache: I am compiling my project with -g -O0 and I could not get rid of this prefetch instruction. Mathieu |
|
From: Nicholas N. <nj...@cs...> - 2005-08-29 20:15:28
|
On Mon, 29 Aug 2005, mma...@ny... wrote: > I did a google search for this instruction it looks like it was never > reported before. Can I work around this ? Using valgrind (gentoo portage > as of today) on AMD64. It is similar to this one: http://bugs.kde.org/show_bug.cgi?id=110478. You could try checking out the current SVN code, but it may not be fixed. If not, can you please file a Bugzilla report? Thanks. Nick |
|
From: Julian S. <js...@ac...> - 2005-08-30 00:25:31
|
On Monday 29 August 2005 21:15, Nicholas Nethercote wrote: > On Mon, 29 Aug 2005, mma...@ny... wrote: > > I did a google search for this instruction it looks like it was never > > reported before. Can I work around this ? Using valgrind (gentoo portage > > as of today) on AMD64. > > It is similar to this one: http://bugs.kde.org/show_bug.cgi?id=110478. > You could try checking out the current SVN code, but it may not be fixed. > If not, can you please file a Bugzilla report? Thanks. I'm pretty sure this is the same as 110478 and so the fix will appear in 3.0.1. J |
|
From: Mathieu M. <mma...@ny...> - 2005-08-30 01:40:24
|
Nicholas Nethercote wrote: > On Mon, 29 Aug 2005, mma...@ny... wrote: > >> I did a google search for this instruction it looks like it was never >> reported before. Can I work around this ? Using valgrind (gentoo portage >> as of today) on AMD64. > > > It is similar to this one: http://bugs.kde.org/show_bug.cgi?id=110478. > You could try checking out the current SVN code, but it may not be fixed. > If not, can you please file a Bugzilla report? Thanks. Really cool ! It works (at least on the small example I tested), I'll see tomorrow the results on the full nightly test suite. Thanks Mathieu |