valgrind-3.0.0 crashes with SIGILL [illegal instruction] at startup on AMD Athlon ["plain"]. Apparently valgrind-3.0.0 assumes SSE x86 hardware ["ldmxcsr" instruction introduced with SSE]. So if "cat /proc/cpuinfo | grep flags" does not show "sse", then valgrind-3.0.0 will not work your CPU. http://bugs.kde.org/show_bug.cgi?id=110274 -- |
|
From: Julian S. <js...@ac...> - 2005-08-06 08:59:00
|
Correct. SSE1 is now a minimum requirement. Supporting non-SSE variants is too much hassle and everybody, more or less, has at least a Pentium-III or equivalent CPU anyway. J On Saturday 06 August 2005 03:54, John Reiser wrote: > valgrind-3.0.0 crashes with SIGILL [illegal instruction] > at startup on AMD Athlon ["plain"]. Apparently valgrind-3.0.0 > assumes SSE x86 hardware ["ldmxcsr" instruction introduced with SSE]. > So if "cat /proc/cpuinfo | grep flags" does not show "sse", > then valgrind-3.0.0 will not work your CPU. > > http://bugs.kde.org/show_bug.cgi?id=110274 |
|
From: Oswald B. <os...@kd...> - 2005-08-06 09:47:49
|
On Sat, Aug 06, 2005 at 10:01:51AM +0100, Julian Seward wrote: > Correct. SSE1 is now a minimum requirement. Supporting non-SSE > variants is too much hassle > uhm, what is the problem? > and everybody, more or less, has at least a Pentium-III or equivalent > CPU anyway. > that's quite a statement. it may be even true - in the western world and possibly even globally in the professional sw devel world. -- Hi! I'm a .signature virus! Copy me into your ~/.signature, please! -- Chaos, panic, and disorder - my work here is done. |
|
From: Julian S. <js...@ac...> - 2005-08-06 10:02:32
|
> > Correct. SSE1 is now a minimum requirement. Supporting non-SSE > > variants is too much hassle > > uhm, what is the problem? It means everywhere where the host machine's sse state is messed with, we now have to have a conditional branch around that code. To be fair, this is fewer places with 3.0 than with 2.4.X. It is certainly possible, just extra clutter and more code paths/ variants to verify. > > and everybody, more or less, has at least a Pentium-III or equivalent > > CPU anyway. > > that's quite a statement. it may be even true - in the western world and > possibly even globally in the professional sw devel world. Fair enough. Let's wait to see if this causes many complaints in practice. J |
|
From: Oswald B. <os...@kd...> - 2005-08-06 12:12:47
|
On Sat, Aug 06, 2005 at 11:05:34AM +0100, Julian Seward wrote: > > > and everybody, more or less, has at least a Pentium-III or > > > equivalent CPU anyway. > > > > that's quite a statement. it may be even true - in the western world > > and possibly even globally in the professional sw devel world. > > Fair enough. Let's wait to see if this causes many complaints in > practice. > ok, but note that this group is underrepresented here, because many of them have very limited internet access or get their software offline even. -- Hi! I'm a .signature virus! Copy me into your ~/.signature, please! -- Chaos, panic, and disorder - my work here is done. |
|
From: Julian S. <js...@ac...> - 2005-08-06 12:17:40
|
> ok, but note that this group is underrepresented here, Noted. J |
|
From: Avery P. <ape...@ni...> - 2005-08-07 17:43:28
|
On Sat, Aug 06, 2005 at 11:05:34AM +0100, Julian Seward wrote: > > > Correct. SSE1 is now a minimum requirement. Supporting non-SSE > > > variants is too much hassle > > > > uhm, what is the problem? > > It means everywhere where the host machine's sse state is messed > with, we now have to have a conditional branch around that code. > To be fair, this is fewer places with 3.0 than with 2.4.X. > It is certainly possible, just extra clutter and more code paths/ > variants to verify. You could do the trick that the openssl people do: modern versions of libc can load a different .so file depending on your processor attributes. So you could have one valgrind library that manipulates SSE state, and another that doesn't, and load the right one depending on the processor type. No runtime conditionals necessary. > > > and everybody, more or less, has at least a Pentium-III or equivalent > > > CPU anyway. > > > > that's quite a statement. it may be even true - in the western world and > > possibly even globally in the professional sw devel world. > > Fair enough. Let's wait to see if this causes many complaints in practice. In my experience, if it comes up during the beta, it will *definitely* come up in practice. Have fun, Avery |
|
From: Julian S. <js...@ac...> - 2005-08-07 17:54:12
|
> > > > and everybody, more or less, has at least a Pentium-III or equivalent > > > > CPU anyway. > > > > > > that's quite a statement. it may be even true - in the western world > > > and possibly even globally in the professional sw devel world. > > > > Fair enough. Let's wait to see if this causes many complaints in > > practice. > > In my experience, if it comes up during the beta, it will *definitely* come > up in practice. I think Ossi's case is convincing -- requiring SSE means we exclude a potentially large set of under-represented users, which is obviously bad. I'll fix this later this evening. Not sure what you mean by beta though. This isn't a beta; 3.0.0 went live last Weds. J |
|
From: Maurice v. d. P. <gri...@ge...> - 2005-08-06 14:20:29
|
On Sat, Aug 06, 2005 at 11:05:34AM +0100, Julian Seward wrote: > > > and everybody, more or less, has at least a Pentium-III or equivalent > > > CPU anyway. > > > > that's quite a statement. it may be even true - in the western world and > > possibly even globally in the professional sw devel world. >=20 > Fair enough. Let's wait to see if this causes many complaints in practic= e. // received bug report from user with Athlon Thunderbird complaints++; Maurice. --=20 Maurice van der Pot Gentoo Linux Developer gri...@ge... http://www.gentoo.org Creator of BiteMe! gri...@kf... http://www.kfk4ever.com |
|
From: Stephen T. <st...@to...> - 2005-08-06 19:47:36
|
On Sat, 2005-08-06 at 16:20 +0200, Maurice van der Pot wrote: > On Sat, Aug 06, 2005 at 11:05:34AM +0100, Julian Seward wrote: > > > > and everybody, more or less, has at least a Pentium-III or equivale= nt > > > > CPU anyway. > > > > > > that's quite a statement. it may be even true - in the western world = and > > > possibly even globally in the professional sw devel world. > >=20 > > Fair enough. Let's wait to see if this causes many complaints in pract= ice. >=20 > // received bug report from user with Athlon Thunderbird > complaints++; Who will be keeping track of the actual number of complaints and why? Given maurice's code, humor noted, there is a race condition for multiple threads. Stephen |
|
From: Julian S. <js...@ac...> - 2005-08-08 00:45:56
|
Fixed (vex 1321, valgrind 4339). Please verify. J On Saturday 06 August 2005 03:54, John Reiser wrote: > valgrind-3.0.0 crashes with SIGILL [illegal instruction] > at startup on AMD Athlon ["plain"]. Apparently valgrind-3.0.0 > assumes SSE x86 hardware ["ldmxcsr" instruction introduced with SSE]. > So if "cat /proc/cpuinfo | grep flags" does not show "sse", > then valgrind-3.0.0 will not work your CPU. > > http://bugs.kde.org/show_bug.cgi?id=110274 |
|
From: John R.
|
>>valgrind-3.0.0 crashes with SIGILL [illegal instruction] >>at startup on AMD Athlon ["plain"]. Apparently valgrind-3.0.0 >>assumes SSE x86 hardware ["ldmxcsr" instruction introduced with SSE]. >>So if "cat /proc/cpuinfo | grep flags" does not show "sse", >>then valgrind-3.0.0 will not work your CPU. >> >>http://bugs.kde.org/show_bug.cgi?id=110274 > Fixed (vex 1321, valgrind 4339). Please verify. Verified (vex 1321, valgrind 4340). Thank you. -- |