[gentoo-dev] Package version requiring sse

[gentoo-dev] Package version requiring sse

[Maurice van der Pot]

[-- Attachment #1: Type: text/plain, Size: 609 bytes --]

The new valgrind version (3.0.0) requires sse support. If you have a
processor without sse, you'll need to stay at 2.4.1.

To make people aware of this, I could use the sse use flag in 3.0.0 
and die if it is not present, telling people to mask versions 3.0.0 
and up if they have a processor without sse. I would also have to check
the arch before requiring sse.

Is there a better way to do this?

Thanks,
Maurice.

-- 
Maurice van der Pot

Gentoo Linux Developer   griffon26@gentoo.org     http://www.gentoo.org
Creator of BiteMe!       griffon26@kfk4ever.com   http://www.kfk4ever.com


[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: [gentoo-dev] Package version requiring sse

[Ivan Yosifov]

On Sat, 2005-08-06 at 16:15 +0200, Maurice van der Pot wrote:
> The new valgrind version (3.0.0) requires sse support. If you have a
> processor without sse, you'll need to stay at 2.4.1.
> 
> To make people aware of this, I could use the sse use flag in 3.0.0 
> and die if it is not present, telling people to mask versions 3.0.0 
> and up if they have a processor without sse. I would also have to check
> the arch before requiring sse.
> 
> Is there a better way to do this?
> 
> Thanks,
> Maurice.
> 

I don't think arch will help you much, at least not with x86 - x86
contains both CPUs like P4 that have SSE and i686 that does not.

I am not sure if it is better, but you can 
cat /proc/cpuinfo | grep flags | grep sse 
and die if not found.

Cheers,
Ivan Yosifov.

-- 
gentoo-dev@gentoo.org mailing list

Re: [gentoo-dev] Package version requiring sse

[Joshua Baergen]

Ivan Yosifov wrote:

>cat /proc/cpuinfo | grep flags | grep sse 
>  
>
This assumes that they have used SSE support during compilation though, 
which wouldn't be the case with -march=i686 or the like.  However, it'd 
be a good test.

Maybe a lookup table against -msse?  Is there a way to check -msse, 
whether explicit or implicit (IE, -march)?

--
Joshua Baergen
-- 
gentoo-dev@gentoo.org mailing list

Re: [gentoo-dev] Package version requiring sse

[Yuri Vasilevski]

Hi,

On Sat, 06 Aug 2005 20:04:20 +0300
Ivan Yosifov <ivan@yosifov.net> wrote:

> I am not sure if it is better, but you can 
> cat /proc/cpuinfo | grep flags | grep sse 
> and die if not found.

This will make packages dependant on the build system,
which will create inconsistencies in binary gentoo packages.

Yuri.
-- 
gentoo-dev@gentoo.org mailing list

Re: [gentoo-dev] Package version requiring sse

[Ciaran McCreesh]

[-- Attachment #1: Type: text/plain, Size: 967 bytes --]

On Sat, 6 Aug 2005 16:15:32 +0200 Maurice van der Pot
<griffon26@gentoo.org> wrote:
| The new valgrind version (3.0.0) requires sse support. If you have a
| processor without sse, you'll need to stay at 2.4.1.
| 
| To make people aware of this, I could use the sse use flag in 3.0.0 
| and die if it is not present, telling people to mask versions 3.0.0 
| and up if they have a processor without sse. I would also have to
| check the arch before requiring sse.
| 
| Is there a better way to do this?

We handle this kind of thing on SPARC (eg for vis and v9 code) by having
subprofiles for the different CPU capabilities, and using masks in the
appropriate subprofiles. Doing this could be a bit messy on x86 though,
at least until we get full profile inheritance capabilities...

-- 
Ciaran McCreesh : Gentoo Developer (Vim, Shell tools, Fluxbox, Cron)
Mail            : ciaranm at gentoo.org
Web             : http://dev.gentoo.org/~ciaranm


[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: [gentoo-dev] Package version requiring sse

[Jeff Walter]

Yuri Vasilevski wrote:
> Hi,
> 
> On Sat, 06 Aug 2005 20:04:20 +0300
> Ivan Yosifov <ivan@yosifov.net> wrote:
> 
> 
>>I am not sure if it is better, but you can 
>>cat /proc/cpuinfo | grep flags | grep sse 
>>and die if not found.
> 
> 
> This will make packages dependant on the build system,
> which will create inconsistencies in binary gentoo packages.
> 
> Yuri.

This is true, and there's no good way around the issue.  I had written a small 
script to actually search for the flag (grep'ing for sse will go true for sse2 
as well), we I noticed this.

Will valgrind 3.0.0 ever work on systems without sse?  If not, the USE flag 
might be your best bet.

--
Jeff Walter
-- 
gentoo-dev@gentoo.org mailing list

[gentoo-dev] Re: Package version requiring sse

[R Hill]

Maurice van der Pot wrote:
> The new valgrind version (3.0.0) requires sse support. If you have a
> processor without sse, you'll need to stay at 2.4.1.
> 
> To make people aware of this, I could use the sse use flag in 3.0.0 
> and die if it is not present, telling people to mask versions 3.0.0 
> and up if they have a processor without sse. I would also have to check
> the arch before requiring sse.

use flags are usually meant for optional support for a package, aren't they?

> Is there a better way to do this?

compile a small test program containing SSE specific intrinsics and die 
if it returns an error code?  does valgrind's configure check for sse?

--de.

-- 
gentoo-dev@gentoo.org mailing list

Re: [gentoo-dev] Re: Package version requiring sse

[Diego 'Flameeyes' Pettenò]

[-- Attachment #1: Type: text/plain, Size: 506 bytes --]

On Saturday 06 August 2005 21:24, R Hill wrote:
> compile a small test program containing SSE specific intrinsics and die
> if it returns an error code?  does valgrind's configure check for sse?
That will break while preparing packages for another machine (cross-compile 
the "simple" way).
Checks that looks at execution capabilities for the system are *bad*.

-- 
Diego "Flameeyes" Pettenò
Gentoo Developer - http://dev.gentoo.org/~flameeyes/
(Gentoo/FreeBSD, Video, Gentoo/AMD64, Sound, PAM)

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

[gentoo-dev] Re: Package version requiring sse

[R Hill]

Diego 'Flameeyes' Pettenò wrote:
> On Saturday 06 August 2005 21:24, R Hill wrote:
>> compile a small test program containing SSE specific intrinsics and die
>> if it returns an error code?  does valgrind's configure check for sse?
> That will break while preparing packages for another machine (cross-compile 
> the "simple" way).
> Checks that looks at execution capabilities for the system are *bad*.

Hrm, good point.

--de.

-- 
gentoo-dev@gentoo.org mailing list

Re: [gentoo-dev] Re: Package version requiring sse

[Ivan Yosifov]

On Sat, 2005-08-06 at 21:37 +0200, Diego 'Flameeyes' Pettenò wrote:
> On Saturday 06 August 2005 21:24, R Hill wrote:
> > compile a small test program containing SSE specific intrinsics and die
> > if it returns an error code?  does valgrind's configure check for sse?
> That will break while preparing packages for another machine (cross-compile 
> the "simple" way).
> Checks that looks at execution capabilities for the system are *bad*.
> 

Then the only fail-safe way to do it is to print a big red warning and
pause for 10 seconds during ebuild emerge. Given the valgrind-3.0.0
keywords ( -* ~amd64 ~ppc ~x86 ) I guess most people on those arches
have SSE anyway. AFAIK all amd64's have SSE, I don't know about ppc, and
I doubt anyone is running Gentoo on i686. You should expect a few bugs
about it though.


-- 
Cheers,
Ivan Yosifov.

-- 
gentoo-dev@gentoo.org mailing list

Re: [gentoo-dev] Re: Package version requiring sse

[Diego 'Flameeyes' Pettenò]

[-- Attachment #1: Type: text/plain, Size: 1191 bytes --]

On Sunday 07 August 2005 12:03, Ivan Yosifov wrote:
> Then the only fail-safe way to do it is to print a big red warning and
> pause for 10 seconds during ebuild emerge.
Probably is the better plan, looking for tc-arch = x86.

> Given the valgrind-3.0.0 
> keywords ( -* ~amd64 ~ppc ~x86 ) I guess most people on those arches
> have SSE anyway. AFAIK all amd64's have SSE, I don't know about ppc, and
> I doubt anyone is running Gentoo on i686.
Well ppc doesn't have SSE but I think the SSE stuff is limited to x86 arch 
anyway, All AMD64 cpu that I know of have SSE support, but there are many 
people running Gentoo non machines that does *not* support SSE, like 
Athlon-Thunderbirds. I was one of them before replacing the main box, and I'm 
still using the tbird for G/FBSD.
I know of people running it on 300MHz P2s (without SSE, too), so better avoid 
assuming that "nobody is running on a i686".. there still are i586 out there 
(and distcc or binary package creation from another machine can do the trick 
for them usually).

-- 
Diego "Flameeyes" Pettenò
Gentoo Developer - http://dev.gentoo.org/~flameeyes/
(Gentoo/FreeBSD, Video, Gentoo/AMD64, Sound, PAM)

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: [gentoo-dev] Package version requiring sse

[Mike Frysinger]

On Saturday 06 August 2005 10:15 am, Maurice van der Pot wrote:
> The new valgrind version (3.0.0) requires sse support. If you have a
> processor without sse, you'll need to stay at 2.4.1.

that's pretty retarded imho ... this applies only to x86 right ?

> To make people aware of this, I could use the sse use flag in 3.0.0
> and die if it is not present, telling people to mask versions 3.0.0
> and up if they have a processor without sse. I would also have to check
> the arch before requiring sse.

seems the only sane way to handle this :(
-mike
-- 
gentoo-dev@gentoo.org mailing list

Re: [gentoo-dev] Re: Package version requiring sse

[Petteri Räty]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Ivan Yosifov wrote:
> 
> Then the only fail-safe way to do it is to print a big red warning and
> pause for 10 seconds during ebuild emerge. Given the valgrind-3.0.0
> keywords ( -* ~amd64 ~ppc ~x86 ) I guess most people on those arches
> have SSE anyway. AFAIK all amd64's have SSE, I don't know about ppc, and
> I doubt anyone is running Gentoo on i686. You should expect a few bugs
> about it though.
> 

I have Gentoo installed on a couple of machines without sse. They both
have Pentium 2 processors. Although I won't be running valgrind on those
machines (desktops for my grand parents) there could be someone else who
does.

I started thinking that one possible solution would be to have some kind
of a subprofile system for different kind of processors. Of course it is
a bit of an overkill for just one package but there could be other uses
like setting the default CFLAGS and turning on the right use flags for
those processors. With these profiles >=valgrind-3.0.0 could be by
default in package.mask and unmasked only on the profiles that can use it.

Regards,
Petteri Räty (Betelgeuse)
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.2 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQFC9eBJcxLzpIGCsLQRAsGnAJ9YwpC3C3YkATUmZWMAdgqAkYOuwwCeIm0D
5D2e0E3NZkCCR23gCBnsUew=
=zNnV
-----END PGP SIGNATURE-----
-- 
gentoo-dev@gentoo.org mailing list

Re: [gentoo-dev] Re: Package version requiring sse

[Ivan Yosifov]

On Sun, 2005-08-07 at 12:11 +0200, Diego 'Flameeyes' Pettenò wrote:
> On Sunday 07 August 2005 12:03, Ivan Yosifov wrote:
> Well ppc doesn't have SSE

Then why is the ppc keyword there ?

> but there are many people running Gentoo non machines that does *not* support SSE

That's why I suggested the big red warning and told you to expect
bugs. ;)


-- 
Cheers,
Ivan Yosifov.

-- 
gentoo-dev@gentoo.org mailing list

[gentoo-dev] Re: Re: Package version requiring sse

[Duncan]

Ivan Yosifov posted <1123410491.29945.3.camel@home.yosifov.net>, excerpted
below,  on Sun, 07 Aug 2005 13:28:11 +0300:

> On Sun, 2005-08-07 at 12:11 +0200, Diego 'Flameeyes' Pettenò wrote:
>> On Sunday 07 August 2005 12:03, Ivan Yosifov wrote: Well ppc doesn't
>> have SSE
> 
> Then why is the ppc keyword there ?

Well, he said, but those accustomed to only thinking x86 might have missed
it, the way it was worded... that ppc uses different software code in that
section, code which presumably takes care of its own ppc specific issues,
so the sse-only issue only applies to x86 (and amd64, but since all of
those have sse, that means only x86 has the issue since only on it is sse
required but not all hardware provides it).

-- 
Duncan - List replies preferred.   No HTML msgs.
"Every nonfree program has a lord, a master --
and if you use the program, he is your master."  Richard Stallman in
http://www.linuxdevcenter.com/pub/a/linux/2004/12/22/rms_interview.html


-- 
gentoo-dev@gentoo.org mailing list

Re: [gentoo-dev] Package version requiring sse

[Maurice van der Pot]

[-- Attachment #1: Type: text/plain, Size: 832 bytes --]

On Sun, Aug 07, 2005 at 06:13:56AM -0400, Mike Frysinger wrote:
> that's pretty retarded imho ... this applies only to x86 right ?

Yes, it does. The valgrind developers are currently waiting to see if it
causes many complaints to see if they can prevent multiple code paths.
 
> > To make people aware of this, I could use the sse use flag in 3.0.0
> > and die if it is not present, telling people to mask versions 3.0.0
> > and up if they have a processor without sse. I would also have to check
> > the arch before requiring sse.
> 
> seems the only sane way to handle this :(

Alright, this is what I'll do then.

Thanks,
Maurice.

-- 
Maurice van der Pot

Gentoo Linux Developer   griffon26@gentoo.org     http://www.gentoo.org
Creator of BiteMe!       griffon26@kfk4ever.com   http://www.kfk4ever.com


[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: [gentoo-dev] Package version requiring sse

[Paul de Vrieze]

[-- Attachment #1: Type: text/plain, Size: 1004 bytes --]

On Saturday 06 August 2005 20:18, Jeff Walter wrote:
> Yuri Vasilevski wrote:
> > Hi,
> >
> > On Sat, 06 Aug 2005 20:04:20 +0300
> >
> > Ivan Yosifov <ivan@yosifov.net> wrote:
> >>I am not sure if it is better, but you can
> >>cat /proc/cpuinfo | grep flags | grep sse
> >>and die if not found.
> >
> > This will make packages dependant on the build system,
> > which will create inconsistencies in binary gentoo packages.
> >
> > Yuri.
>
> This is true, and there's no good way around the issue.  I had written
> a small script to actually search for the flag (grep'ing for sse will
> go true for sse2 as well), we I noticed this.
>
> Will valgrind 3.0.0 ever work on systems without sse?  If not, the USE
> flag might be your best bet.

Put a check on /proc/cpuinfo in pkg_preinst. This should get executed on 
the final machine, so not when building binary packages.

Paul

-- 
Paul de Vrieze
Gentoo Developer
Mail: pauldv@gentoo.org
Homepage: http://www.devrieze.net

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: [gentoo-dev] Package version requiring sse

[Martin Schlemmer]

[-- Attachment #1: Type: text/plain, Size: 1112 bytes --]

On Wed, 2005-08-24 at 14:53 +0200, Paul de Vrieze wrote:
> On Saturday 06 August 2005 20:18, Jeff Walter wrote:
> > Yuri Vasilevski wrote:
> > > Hi,
> > >
> > > On Sat, 06 Aug 2005 20:04:20 +0300
> > >
> > > Ivan Yosifov <ivan@yosifov.net> wrote:
> > >>I am not sure if it is better, but you can
> > >>cat /proc/cpuinfo | grep flags | grep sse
> > >>and die if not found.
> > >
> > > This will make packages dependant on the build system,
> > > which will create inconsistencies in binary gentoo packages.
> > >
> > > Yuri.
> >
> > This is true, and there's no good way around the issue.  I had written
> > a small script to actually search for the flag (grep'ing for sse will
> > go true for sse2 as well), we I noticed this.
> >
> > Will valgrind 3.0.0 ever work on systems without sse?  If not, the USE
> > flag might be your best bet.
> 
> Put a check on /proc/cpuinfo in pkg_preinst. This should get executed on 
> the final machine, so not when building binary packages.
> 

Same thing (and probably better option) if you put it in pkg_setup() ...


-- 
Martin Schlemmer


[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: [gentoo-dev] Package version requiring sse

[Paul de Vrieze]

[-- Attachment #1: Type: text/plain, Size: 409 bytes --]

On Wednesday 24 August 2005 15:23, Martin Schlemmer wrote:
>
> Same thing (and probably better option) if you put it in pkg_setup()
> ...

Isn't pkg_setup run too when just building a binary package (-B) (then the 
check shouldn't be performed), and just before installing a binary 
package?

Paul

-- 
Paul de Vrieze
Gentoo Developer
Mail: pauldv@gentoo.org
Homepage: http://www.devrieze.net

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: [gentoo-dev] Package version requiring sse

[Brian Harring]

[-- Attachment #1: Type: text/plain, Size: 607 bytes --]

On Thu, Aug 25, 2005 at 01:41:00PM +0200, Paul de Vrieze wrote:
> On Wednesday 24 August 2005 15:23, Martin Schlemmer wrote:
> >
> > Same thing (and probably better option) if you put it in pkg_setup()
> > ...
> 
> Isn't pkg_setup run too when just building a binary package (-B) (then the 
> check shouldn't be performed), and just before installing a binary 
> package?
Yep, something that's rather unclean.
Reinitializing the env for the local box I have no issue with, I just 
dislike re-running pkg_setup which also set's up vars for building.

Alternatives welcome mind you...
~harring

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: [gentoo-dev] Package version requiring sse

[Paul de Vrieze]

[-- Attachment #1: Type: text/plain, Size: 948 bytes --]

On Thursday 25 August 2005 17:29, Brian Harring wrote:
> On Thu, Aug 25, 2005 at 01:41:00PM +0200, Paul de Vrieze wrote:
> > On Wednesday 24 August 2005 15:23, Martin Schlemmer wrote:
> > > Same thing (and probably better option) if you put it in pkg_setup()
> > > ...
> >
> > Isn't pkg_setup run too when just building a binary package (-B) (then
> > the check shouldn't be performed), and just before installing a binary
> > package?
>
> Yep, something that's rather unclean.
> Reinitializing the env for the local box I have no issue with, I just
> dislike re-running pkg_setup which also set's up vars for building.
>
> Alternatives welcome mind you...
> ~harring

Shouldn't pkg_preinst work? It should be run just before installation. The 
only thing is that the package is allready build before it's being tested.

Paul

-- 
Paul de Vrieze
Gentoo Developer
Mail: pauldv@gentoo.org
Homepage: http://www.devrieze.net

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: [gentoo-dev] Package version requiring sse

[Martin Schlemmer]

[-- Attachment #1: Type: text/plain, Size: 750 bytes --]

On Thu, 2005-08-25 at 13:41 +0200, Paul de Vrieze wrote:
> On Wednesday 24 August 2005 15:23, Martin Schlemmer wrote:
> >
> > Same thing (and probably better option) if you put it in pkg_setup()
> > ...
> 
> Isn't pkg_setup run too when just building a binary package (-B) (then the 
> check shouldn't be performed), and just before installing a binary 
> package?
> 

True, but usually you build whatever on a machine that have capabilities
to run it (not talking about cross-compiling).  And besides, I think its
bad style to build something, and then bail after its done about
something that could have been tested at setup time (think glibc testing
tls/nptl capabilities only during pkg_preinst ...).


-- 
Martin Schlemmer


[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: [gentoo-dev] Package version requiring sse

[Paul de Vrieze]

[-- Attachment #1: Type: text/plain, Size: 1552 bytes --]

On Thursday 25 August 2005 21:17, Martin Schlemmer wrote:
> On Thu, 2005-08-25 at 13:41 +0200, Paul de Vrieze wrote:
> > On Wednesday 24 August 2005 15:23, Martin Schlemmer wrote:
> > > Same thing (and probably better option) if you put it in
> > > pkg_setup() ...
> >
> > Isn't pkg_setup run too when just building a binary package (-B)
> > (then the check shouldn't be performed), and just before installing a
> > binary package?
>
> True, but usually you build whatever on a machine that have
> capabilities to run it (not talking about cross-compiling).  And
> besides, I think its bad style to build something, and then bail after
> its done about something that could have been tested at setup time
> (think glibc testing tls/nptl capabilities only during pkg_preinst
> ...).

That's a different issue. The best way would be for some variable to be 
set that indicates that a package is going to be build for installation 
on the current host and not in a ROOT environment (the latter is easy). I 
agree though that testing afterwards is also not a good solution from a 
user interface perspective, locking the building of packages for a 
differnent host is also unwanted though. Alternatively we could also 
agree on an environment variable name that would be set if these kinds of 
checks should be overridden. People knowing what they're doing could then 
just use this I_KNOW_WHAT_IM_DOING flag and be off.

Paul

-- 
Paul de Vrieze
Gentoo Developer
Mail: pauldv@gentoo.org
Homepage: http://www.devrieze.net

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]