Installation of pentium4 optimized packages

andreas_baumann · 2019-05-01 19:24:02

If your CPU is one of the never kind (supporting SSE2), you can benefit from the extra 2% you get in speed
by switching to optimized packages with an architecture called 'pentium4'.

The installation and transition steps are:

- Install the i686 version from the i686 ISO (there is no pentium4 ISO yet)
- Change the architecture to 'arch=pentium4' in '/etc/pacman.conf'
- Do a full update to transition to the pentium4 packages with

pacman -Qqn | pacman -S -

Slacka · 2019-05-02 19:39:23

I was able to do a clean pentium4 install with the i686 install iso. I made the following tweaks to a normal Arch install.
1. Edit /etc/pacman.conf ->

Architecture = pentium4

2. After, arch-chroot /mnt , again /etc/pacman.conf ->

Architecture = pentium4

3.

pacman -Syy

For those that want to transition an existing i686, there are lots of good tips here:

https://wiki.archlinux.org/index.php?ti … did=464524

Also, Why the snarky 2% comment? For multimedia apps, performance can be nearly doubled[1] Maybe that's true for server / network apps, but for those of us using Arch 32 on the desktop, SSE2 can greatly improve our multimedia performance. In addition, many engineering and scientific applications need 64-bit floating-point precision. SSE2 extends MMX integer SIMD instruction processing to 128 bits from 64 bits, so two 64-bit integer operations can be performed at once. SSE2 is essentially the successor to x87 floating point. Why have all that good silicon, sitting dark when nearly all modern software is built to take advantage of it? There is no downside, no wasted RAM from 64-bit pointers. Just taking advantage of what we (desktop/netbook users) nearly all have.

[1] https://www.geeks3d.com/20100711/test-s … plication/

Last edited by Slacka (2019-05-02 19:44:41)

andreas_baumann · 2019-05-02 19:43:24

Sorry about the snarky comment. :-)

I should have added a wink smiley there..

Just because taking advantage of SSE2 is so important, there are two architectures (i686 and pentium4) now.
Most distros now go for SSE2 as minimal requirement, so I think there is a niche for i686 for those not having SSE2.
i486 is for unfortunate people having old AMDs not supporting SSE, MMX and for people
playing with vintage machines.

levi · 2019-05-03 04:08:32

Yes, I also had to do a pacman -Sy before I could do the pacman -Qqn... step, otherwise it reports the architecture is invalid.

Edit: Also, I noticed it said it was downgrading some packages, as well as reinstalling the majority, and upgrading a few (I'm not sure offhand when I last updated the i686 install). From what I can tell, all of the downgrades I've checked thus far are packages whose build number has gone down; presumably the build number gets reset for the new platform, so a -2 i686 package and a -0 pentium4 package might be functionally equivalent.

Last edited by levi (2019-05-03 05:19:15)

andreas_baumann · 2019-05-03 07:20:10

The reason for downgrades is a little bit weird to me: all 686 where frozen in time and just refurbushed to be penrium4 packages (changing the architecture
inside .PKGINFO automatically). So, can you provide an example of a downgrade? I didn't see any when testing this..

As for the architecture: true, one has to sync the databases first with 'pacman -Sy' after setting the 'arch=pentium4' in '/etc/pacman.conf', thanks levi.

levi · 2019-05-03 17:24:05

The one I checked first of all which is my python3 installation shows this pattern I think. According to my cached pkgs, the i686 package I had installed was 3.7.3-1.1-i686 and now it's 3.7.3-1.0-pentium4.

Is it still worth updating your instructions in the first post to include the 'pacman -Sy' step? I guess I'm a little late to this party (I had mistakenly thought that i686 properly included SSE2, and that pentium4 must not have it- entirely backwards until I checked the architecture spec linked from the front page), but I reckon there might be more people still to do this after me.

Edit: FWIW, since updating, I can no longer run mplayer; it complains about missing libcdio.so.18 while fulfilling its runtime dependencies; I have libcdio.so.19, coming from package libcdio-2.1.0-1.0-pentium4. Before the sysupgrade, I came from libcdio-2.0.0-1.2-i686 which does seem to include libcdio.so.18, so I guess the mplayer package needs a rebuild.

Last edited by levi (2019-05-03 17:57:30)

alf9000 · 2019-05-03 18:00:35

To be sure, all my machines which have a

$ lscpu | egrep -i 'sse2'

output, can be moved to arch=pentium4?
Seems I'm in luck with my Asus Eee 1000H Netbook :-)

Thanks to Andreas !

^^^ edit after hint from Andreas:

Thanks to deep42thought too !

Last edited by alf9000 (2019-05-03 18:12:14)

andreas_baumann · 2019-05-03 18:05:13

Ahem: deep42thought and Andreas. :-)

levi · 2019-05-03 19:55:50

alf9000 wrote:

Seems I'm in luck with my Asus Eee 1000H Netbook :-)

Yes, all Atom-based EeePCs can run sse2 opcodes, as well as sse3 and even ssse3 ones. Celeron-based EeePCs seems to be able to run sse2 opcodes, but not any later ones, but they'd still be a good fit for these pentium 4 package releases.

Off-hand I wonder if the number of Atom-based EeePC users in these parts might validate a release for sse3 and even ssse3 opcodes, although as is the manner in these things, I guess the resultant speedup would be a lot less than the few percentage points benefit these sse2 instructions supposedly give.

I'm looking forward to testing mplayer once it's been fixed, and see if it runs appreciably better in downscaling 720p and even 1080p videos on this hardware. As is, I can generally run them without losing sync too badly if they don't cut too often or have high action; so talking heads videos with occasional changes of background run fine, par example, but compilation videos that change entire scenes frequently can't be run at full speed otherwise the audio gets appreciably ahead of the video (e.g. anything from failarmy or similar).

bdlepla · 2019-05-04 18:02:22

Kodi is also complaining about libcdio.so.18 not being found.

Furthermore, on the package page for libcdio pentium4 I get this:

Error 500: Internal Server Error
Inconsistency in Database found:
buildmaster[version] != repositories[Version]:
"2.1.0-1.0" != "2.0.0-1.2"

Bryan

K8ebBM214bp4UJV · 2019-05-04 20:14:53

Hi all.

the last update seems to have caused a ruckus. i read a few threads but looks like everything is being worked on so big thanks to deep42thought and andreas_baumann. not sure if you got the message I tried to send within the forum system, so please consider lifting the registration restriction on the irc channel.

at my end, I upgraded my netbook with the pentium4 optimized packages, and almost everything seems to be working aside of a couple of things. lightdm, lightdm-gtk-greeter, openbox all working, but honestly i don't see much of an improvement. i know this hardware is old and we are lucky it is still somewhat useful. my netbook has a n270 and 2GB ram, so 2% improvement would be welcome.

currently broken is chromium 73 as it depends on icu 63, so i guess most things depending on icu 63 are broken as the package upgraded to icu 64. i'm currently using a firefox nightly binary i got directly from mozilla, but in this hardware it sucks. hopefully chromium gets updated soon. i would love to be able to compile chromium on my own and even apply ungoogle patches but simply it's impossible with 32 bit hardware. current chromium snapshot is at version 76.

mpv is broken too, but the one in testing fixed the issue. qtox is broken.

commenting on @levi's post, video streaming in this hardware is hell. anything above 360p stutters, 480p would be pushing it but depends on bitrate. as said before, only chromium seems to work best for this. firefox even stutters at 160p on twitch. grabbing links and playing on mpv or external players isn't always possible, and audio/video desynchronization is expected when 720p and up.

anyway, as always, thanks for your hard work.

levi · 2019-05-05 05:27:19

Yeah, I just installed vlc to get around the broken mplayer for the time being. That's working fine for me currently, although I still have to learn not to hit 'p' to pause the videos; I have now changed the hotkeys in vlc but I guess it needs a restart to take effect. I checked out a few different players that don't depend on mplayer, and vlc turned out to have the least bad dependency hell.

But I find this machine was able to play fullscreen videos just fine provided they don't have stupid framerates, so I wrote a script I run on a faster machine I run in another room to transcode my downloaded videos to 600p30 or below using a single call to ffmpeg. I try my best not to use sites I can't pull an html5 video link out of with relative ease, so I don't use twitch (instead I wait for the stream to turn up on youtube), or instagram or things like that. I'm yet to put vlc really through it's paces though to see if it's any better.

Maniaxx · 2019-05-10 20:12:15

Thunderbird/libxul can't find icu.63 libs. Its icu.64 after upgrading to P4. Not tested yet, but i fear when downgrading it will break other stuff. I hope both versions can live side by side.

levi · 2019-05-10 22:30:57

Yeah, I think it's a coincidence that we all noticed the libicu bug after upgrading to pentium4; reportedly it's broken on i686 as well. I'm not sure we need libicu 63 and 64 to live side by side, other than a reported usage of both by libreoffice-fresh, but hopefully that's noise that will disappear when everything that uses libicu gets upgraded.

rossboulet · 2019-05-10 23:08:35

andreas_baumann wrote:

If your CPU is one of the never kind (supporting SSE2), you can benefit from the extra 2% you get in speed
by switching to optimized packages with an architecture called 'pentium4'.
The installation and transition steps are:
- Install the i686 version from the i686 ISO (there is no pentium4 ISO yet)
- Change the architecture to 'arch=pentium4' in '/etc/pacman.conf'
- Do a full update to transition to the pentium4 packages with
pacman -Qqn | pacman -S -

Shouldn't this say 'Architecture=pentium4' rather than 'arch=pentium4' for the change to /etc/pacman.conf? I believe pacman reads the 'Architecture' entry and will then create the '$arch' variable for the repositories from that line.

levi · 2019-05-11 01:07:29

In short, yes it should.

rogerthat · 2019-05-28 18:43:41

Any news on if the libicu 63/64 problem is fixed? I'm holding off on updates for now...

levi · 2019-05-28 19:43:19

I did an update at the weekend, and got at the very least pacman and kernel updates which are always worth keeping on top of, but libreoffice is still broken for me with a libicu error (and mplayer is still looking for the wrong libcdio).

To be honest, libreoffice is probably the more important of the two, but the lack of that is largely covered by https://bugs.archlinux32.org/index.php? … &pagenum=2 . I don't know why we haven't had an mplayer rebuild yet, and to be honest I'm getting slightly fed up of vlc's foibles.

deep42thought · 2019-05-28 20:12:58

on pentium4, mplayer waits for libbluray, which currently fails due to some java/ant issue :-/
IIRC, something similar is the culprit for libreoffice

levi · 2019-05-28 20:56:05

Ah right, could well be the same bug.

Maybe I should look into unpacking the old i686 libs from my archive.

Yep, just unpacked using bsdtar libicuuc.so.63*, libicudata.so.63* and libicui18n.so.63* from icu.63.1-2.1.i686.pkg.tar.xz (which was the last icu release from before I upgraded to pentium4). Now libreoffice works again, although I'll have to remember to tidy all of that up when this gets built.

Mplayer just required the old libcdio.so.18.0.0, although this time the softlinks weren't included in the archive, so I had to make that manually (to libcdio.so.18). Presumably there's some metadata in the archive that makes that when pacman installs it.

Edit: But mplayer doesn't seem to work properly now. It plays back mp3s just fine, but mp4 video files first played back with excessive desync, and the second and third time I tried, it stops after a fraction of a second and just hangs there. Oh well, I guess I was too optimistic expecting my simple hack to get that to work well.

Last edited by levi (2019-05-28 21:15:55)

K8ebBM214bp4UJV · 2019-05-30 00:14:21

hello. i've been thinking a lot about this. wouldn't be easier to assume the default i686 arch as the pentium4 (sse2 supporting cpu)? and then rename the non sse2 as i686-no-sse2 or something? pentium4 is not an architecture, and not all sse2 supporting cpus are pentium4, also isn't pentium an intel trademark?

you see posts in the forum complaining some packages from AUR can't be build or issues when pentium4 is not a supported arch. i think people with a non sse2 cpu would find it easier to switch to "i686-no-sse2" and most people simply stay with "i686" as defaul 32 bit architecture. i would guess there are a lot more users in need of sse2 to justify i686 as default option.

don't know, though. just my 2c

levi · 2019-05-30 05:10:19

If you rename platforms, then everyone will find AUR packages won't build for them by default. All it takes it the redefinition of a couple of variables in the PKGBUILD, and as I see it people need to understand PKGBUILDs if they're using AUR; you do review them before you install everything like you should, don't you?

Bob Hill · 2019-05-30 14:03:02

andreas_baumann wrote:

The reason for downgrades is a little bit weird to me: all 686 where frozen in time and just refurbushed to be penrium4 packages (changing the architecture
inside .PKGINFO automatically). So, can you provide an example of a downgrade? I didn't see any when testing this...

Today, attempting to transition from "i686" to "pentium4" would have downgraded the following packages:

bluez-libs:    5.50-6.0   ==> 5.50-5.0
glibc:         2.29-1.26  ==> 2.29-1.3
libbluray:     1.1.1-1.0  ==> 1.1.0-1.1
libjpeg-turbo: 2.0.2-1.1  ==> 2.0.2-1.0
libutil-linux: 2.33.2-1.2 ==> 2.33.2-1.1

Might it be possible to upgrade these five packages in the "pentium4" repositories to at least the same levels as in the "i686" repositories, so that a "i686"-to-"pentium4" transition does not result in downgrades?

Or have I misunderstood something?

Many thanks in advance!

levi · 2019-05-30 16:47:33

I don't think build numbers (the bit after the hyphen) can be compared across architectures like that - those downgrade messages are just noise. And I suspect the libbluray thing is a known bug where it's very tricky to build under sse2 architectures, which is stopping us getting a working mplayer currently.

andreas_baumann · 2019-05-30 19:39:14

bluez-libs: 5.50-6.0 ==> 5.50-5.0
glibc: 2.29-1.26 ==> 2.29-1.3
libbluray: 1.1.1-1.0 ==> 1.1.0-1.1
libjpeg-turbo: 2.0.2-1.1 ==> 2.0.2-1.0
libutil-linux: 2.33.2-1.2 ==> 2.33.2-1.1

for example 2.29.1.26 is:
- pkgver=2.29
- pkgrel=1
- the last number is the build number, in this case 26 tells you that glibc had quite some trouble to get released. :-)

Arch Linux

#1 2019-05-01 19:24:02

Installation of pentium4 optimized packages

#2 2019-05-02 19:39:23

Re: Installation of pentium4 optimized packages

#3 2019-05-02 19:43:24

Re: Installation of pentium4 optimized packages

#4 2019-05-03 04:08:32

Re: Installation of pentium4 optimized packages

#5 2019-05-03 07:20:10

Re: Installation of pentium4 optimized packages

#6 2019-05-03 17:24:05

Re: Installation of pentium4 optimized packages

#7 2019-05-03 18:00:35

Re: Installation of pentium4 optimized packages

#8 2019-05-03 18:05:13

Re: Installation of pentium4 optimized packages

#9 2019-05-03 19:55:50

Re: Installation of pentium4 optimized packages

#10 2019-05-04 18:02:22

Re: Installation of pentium4 optimized packages

#11 2019-05-04 20:14:53

Re: Installation of pentium4 optimized packages

#12 2019-05-05 05:27:19

Re: Installation of pentium4 optimized packages

#13 2019-05-10 20:12:15

Re: Installation of pentium4 optimized packages

#14 2019-05-10 22:30:57

Re: Installation of pentium4 optimized packages

#15 2019-05-10 23:08:35

Re: Installation of pentium4 optimized packages

#16 2019-05-11 01:07:29

Re: Installation of pentium4 optimized packages

#17 2019-05-28 18:43:41

Re: Installation of pentium4 optimized packages

#18 2019-05-28 19:43:19

Re: Installation of pentium4 optimized packages

#19 2019-05-28 20:12:58

Re: Installation of pentium4 optimized packages

#20 2019-05-28 20:56:05

Re: Installation of pentium4 optimized packages

#21 2019-05-30 00:14:21

Re: Installation of pentium4 optimized packages

#22 2019-05-30 05:10:19

Re: Installation of pentium4 optimized packages

#23 2019-05-30 14:03:02

Re: Installation of pentium4 optimized packages

#24 2019-05-30 16:47:33

Re: Installation of pentium4 optimized packages

#25 2019-05-30 19:39:14

Re: Installation of pentium4 optimized packages

Board footer