x86_64: poll for optional CFLAGS

AndyRTR · 2006-11-21 11:00:32

I think about adding SSE3 support to the default and developers CFLAGS/CXXFLAGS. I have found -mtune=nocona to get optional SSE3 support. But I don't know exactly how to test if it would become activ and if it would really speed up things - help is wanted.

If all our arch64 users have SSE3 capable processors I don't see a big problem to add it by default. Most new arch64 user will get a new cpu that will support it. Maybe we can even set -mtune=core2 to get even SSSE3 (aka SSE4) support.

I think about changing from CFLAGS="-march=x86-64 -O2 -pipe" to CFLAGS="-march=x86-64 -O2 -msse3 -pipe" or CFLAGS="-march=x86-64 -O2 -mtune=core2 -pipe" .

So tell me if your cpu supports pni (prescott new instructions) by a "cat /proc/cpuinfo | grep pni".

for further reading: http://gcc.gnu.org/onlinedocs/gcc/i386- … tions.html and http://gentoo-wiki.com/Safe_Cflags

possible gain: http://www.anandtech.com/cpuchipsets/sh … spx?i=2350

Neverth · 2006-11-21 11:08:04

Nope

skottish · 2006-11-21 12:46:21

Mine does.

multitronic · 2006-11-21 15:02:38

I don't think it is a good idea. All pre-venice cpu owners will be locked out from Arch64. I also think SSE3 will not give much (if any) speedup. SSE2 is ok.

Pierre · 2006-11-21 15:23:57

We are still using sse2 by default, don`t we? Do you know any application which would benefit from sse3 that much? If it could be enabled as "optional" it would be OK; otherwise I do not see any reason to include this flag.

I played a bit with nexuiz and compiled it with different flags. I did not notice any difference between "-march=nocona" and just leaving it blank.

Another point: Wouldn`t it be better to replace the chost "86_64-unknown-linux-gnu" with "86_64-pc-linux-gnu"?

ganja_guru · 2006-11-21 18:13:35

pni not supported on mine. neither is sse3. I have an Athlon FX-55 and that is fairly recent. Correct me if im wrong, but I presume only newer intel dual cores do these things, so im definitely voting this down.

Pierre · 2006-11-21 18:35:37

Strange; my Athlon 64 4000+ supports SSE3/PNI. Should be equivalent to a FX 53.

AndyRTR · 2006-11-21 19:00:27

I can already see that we have more user running old athlon64 cpus than I expected. So we really need to keep compatibility.

But I still want to play with -mtune=xxx to get optional SSE3 support. Only old 130µm athlon64 and sempron miss that feature. I think a good test will be to transcode(mencoder, lxdvdrip,...) some dvd movies. Somebody wants to do that? A Core2Duo would be nice but also athlon64 or PIV would be good.

javajunky · 2006-11-21 19:10:09

Hmm, I'm confused. I just bought a Laptop with a Turion64x2 cpu, I *thought* this was quite a modern 64bit chip, but it seems I don't have sse3 either

edit: Ignore me just been on a whirlwind tour of extended instructions, I guess pni superceeds/equivalences sse3 (CPU-Z under windows at least informs me my chip *does* support SSE3) I'm adding this note in case it confuses anyone else who's brain is as tiny as mine <g> Sorry!

AndyRTR · 2006-11-21 19:58:06

pni = prescott new instrucions = SSE3 :!:

ckristi · 2006-11-21 22:23:07

I voted pro. I have a Pentium D 915 and my guess is that every 64bit user has a processor that supports SSE3/PNI.

kozaki · 2006-11-22 00:52:53

AndyRTR wrote:

pni = prescott new instrucions = SSE3 :!:

1st I thought my amd 3200 Venice didn't have sse3, but since it shows pni I guess it has.
grep sse /proc/cpuinfo

flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 syscall nx mmxext fxsr_opt lm 3dnowext 3dnow up pni lahf_lm

I do transcode from time to time with compiled mencoder (also use avidemux2, mmg (Matroska), ogg & (too)lame) so it definitively would be worth give optimized flags... as an option if that would break arch64 for 130mm athlon users.

Also I remember read than amd sse3 instructions is partial (more marketing than practical optimization), at least up to Venice core.
That beeing said, from Gentoo Linux wiki "Safe Flags":

If you have a newer Athlon64("Venice" or "San Diego")/Athlon64-X2("Manchester" or "Toledo") (check for pni in cat /proc/cpuinfo) you can also add -msse3 to your CFLAGS to enable SSE3 support. Any chip using the 90nm process except for the "Winchester" class Athlon64 supports SSE3.
Also bear in mind that all Athlon64 X2 and Opteron 165, 170, 175, 180, and 185 processors are dual-core CPUs so make sure that you use MAKEOPTS="-j3" in your make.conf. This doesn't affect the code but might reduce compile times.
All socket AM2 processors are SSE3 capable as well.

AndyRTR · 2006-11-23 07:12:35

-mtune/march=core2 is planned for gcc-4.3.x release http://gcc.gnu.org/gcc-4.3/changes.html

so for now we can only use -mtune=nocona to get SSE3/pni support. it would be nice if somebody can make some benchmarks for us.

javajunky · 2006-11-23 08:10:41

I would be happy to, but I don't think theres any point since I still can't boot Arch64 without 'noapic' so I'm only using one of my cpu cores so I doubt my benchmarks would be worth anything Sorry.

If there is point, let me know what you want, and I'll fire it off.. My laptop sits next to me at work so I can leave it doing stuff

AndyRTR · 2006-11-23 09:01:25

javajunky wrote:

I would be happy to, but I don't think theres any point since I still can't boot Arch64 without 'noapic' so I'm only using one of my cpu cores so I doubt my benchmarks would be worth anything Sorry.
If there is point, let me know what you want, and I'll fire it off.. My laptop sits next to me at work so I can leave it doing stuff

You can try to benchmark even with only one core. let's meet in IRC freenode.net #archlinux64 ;-)

iphitus · 2006-11-25 04:47:05

pentium D 820 has pni.

James

rbl · 2006-12-08 16:52:45

Well.. pni is there, but no sse3, so I did not vote.
flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt rdtscp lm 3dnowext 3dnow pni cx16 lahf_lm cmp_legacy svm cr8_legacy
model name : AMD Athlon(tm) 64 X2 Dual Core Processor 4200+

AndyRTR · 2006-12-08 16:58:19

PNI = SSE3 :!:

iBertus · 2006-12-08 17:06:31

My Palermo E6 stepping Sempron has PNI but it's my understanding that the prior core did not.

AndyRTR · 2006-12-08 19:44:23

ok, we need some benchmarks to decide if it's worth to add any mtune option. who's going to do it?

chrismortimore · 2006-12-12 16:26:44

AndyRTR wrote:

ok, we need some benchmarks to decide if it's worth to add any mtune option. who's going to do it?

I'm up for it. I'm in the process of migrating my desktop from Arch to Arch64 (although I do plan to run both for a while). Spec in the sig if it meets with your satisfaction.

AndyRTR · 2007-03-27 21:18:49

OK, wanna see some really fast stuff?

# $Id: PKGBUILD,v 1.5 2007/02/24 19:07:58 jgc Exp $
# Maintainer: Dale Blount <dale@archlinux.org>
# Contributer: Nick Penwarden <toth64@yahoo.com>

pkgname=firefox
pkgver=2.0.0.3
pkgrel=1
pkgdesc="Standalone web browser from mozilla.org"
arch=(i686 x86_64)
depends=('gtk2>=2.10.11' 'pango>=1.16.1' 'gcc' 'libxt' 'libidl2' 'mozilla-common' 'nss>=3.11.5' 'desktop-file-utils')
makedepends=('zip' 'imagemagick' 'pkgconfig')
replaces=('mozilla-firebird' 'phoenix' 'mozilla-firefox')
conflicts=('mozilla-firefox')
provides=('mozilla-firefox')
install=firefox.install
url="http://www.mozilla.org/projects/firefox"
source=(ftp://ftp.mozilla.org/pub/mozilla.org/${pkgname}/releases/${pkgver}/source/${pkgname}-${pkgver}-source.tar.bz2
    http://getswiftfox.com/source/swiftfox-$pkgver.patch
        mozconfig
           launcher.patch
        firefox-1.1-uriloader.patch
    mozilla-firefox-1.0-lang.patch
        moz310924.patch
    moz325644.patch
    firefox-nopangoxft.patch
    firefox-visibility.patch
    firefox-1.5-new-gtkim.patch
    firefox-1.5-pango-cursor-position.patch
    firefox-2.0-add-ldflags.patch
    firefox-2.0-buildversion.patch
        firefox.desktop
    firefox-safe.desktop)
md5sums=('24398e3d98673a2a92a01a8f771ca12a'
          'f42869b09cb23ae5c487213c2b126923'
          'b6b9a0977d33f79b79d58f0b501578e7'
          '224962b5f2446cab7727fdf07fd526de'
          '2bd0dd9035dcb875b8340be358347e8d'
          'bd5db57c23c72a02a489592644f18995'
          '29194973e2a535b460c6b7f92c635eaf'
          '2082c2a2d1cedd08e83179271aacf337'
          'f975d9aa6ec232da0dfa8f27398d2af3'
          '362f9e0b0f25b964f7120b68fb629ee0'
          '60b4bbe73d2e919ee4a6476dca6705b6'
          '288fb7db871700ff5cf7286db6192b45'
          '25f355113cdee6800380c6e1a4cd38f0'
          '11b221ff41078d97c131e17361072e47'
          '74ea70c9e935f0e7f7b75436fe33efd5'
          '5e68cabfcf3c021806b326f664ac505e')

build() {
  cd ${startdir}/src/mozilla
  patch -Np0 -i ${startdir}/src/firefox-1.1-uriloader.patch || return 1
  patch -Np0 -i ${startdir}/src/moz310924.patch || return 1
  patch -Np0 -i ${startdir}/src/moz325644.patch || return 1
  patch -Np0 -i ${startdir}/src/launcher.patch || return 1
  patch -Np1 -i ${startdir}/src/mozilla-firefox-1.0-lang.patch || return 1
  patch -Np1 -i ${startdir}/src/firefox-nopangoxft.patch || return 1
  patch -Np1 -i ${startdir}/src/firefox-1.5-new-gtkim.patch || return 1
  patch -Np1 -i ${startdir}/src/firefox-1.5-pango-cursor-position.patch || return 1
  patch -Np0 -i ${startdir}/src/firefox-2.0-add-ldflags.patch || return 1
  patch -Np0 -i ${startdir}/src/firefox-2.0-buildversion.patch || return 1
  patch -Np0 -i ${startdir}/src/swiftfox-2.0.0.3.patch || return 1

  if [ "$CARCH" = "x86_64" ]; then
    patch -Np0 -i ../firefox-visibility.patch || return 1
  fi
  # this time we need -j1
#  export MAKEFLAGS="-j1"

#  sed "s/#CFLAGS#/${CFLAGS} -msse3/g" ${startdir}/src/mozconfig >.mozconfig
  sed "s/#CFLAGS#/-march=x86-64 -O3 -pipe -msse3/g" ${startdir}/src/mozconfig >.mozconfig

  export MOZ_PROJECT=browser

  make -f client.mk build || return 1
  make DESTDIR=${startdir}/pkg install || return 1

  cd ${startdir}/pkg/opt/mozilla/lib/firefox-${pkgver}
  export MOZ_DISABLE_GNOME=1
  export MOZTMP=`mktemp -d -p ${startdir}/src`
  LD_LIBRARY_PATH=`pwd` HOME=${MOZTMP} ./firefox-bin -register
  rm -rf ${MOZTMP}
  cd chrome
  find . -maxdepth 1 -type d -exec rm -rf {} \;

  #Remove mozilla devel stuff, this is in XULRunner now
  rm -rf ${startdir}/pkg/opt/mozilla/share
  rm -rf ${startdir}/pkg/opt/mozilla/include
  rm -rf ${startdir}/pkg/opt/mozilla/lib/pkgconfig


  cd ${startdir}/pkg/opt/mozilla/lib && ln -sf firefox-${pkgver} firefox

  rm -rf ${startdir}/pkg/opt/mozilla/bin/defaults

  mkdir -p ${startdir}/pkg/usr/share/applications
  mkdir -p ${startdir}/pkg/usr/share/pixmaps
  convert ${startdir}/src/mozilla/browser/app/default.xpm ${startdir}/pkg/usr/share/pixmaps/firefox.png
  install -m644 ${startdir}/src/firefox.desktop ${startdir}/pkg/usr/share/applications/
  install -m644 ${startdir}/src/firefox-safe.desktop ${startdir}/pkg/usr/share/applications/

  mkdir -p ${startdir}/pkg/opt/mozilla/lib/firefox/chrome/icons/default
  install -m644 ${startdir}/src/mozilla/browser/app/default.xpm ${startdir}/pkg/opt/mozilla/lib/firefox/chrome/icons/default/
  install -m644 ${startdir}/src/mozilla/browser/app/default.xpm ${startdir}/pkg/opt/mozilla/lib/firefox/icons/
}

It's based on Swiftfox patches and adds CLFAGS that i call really "nice" - there's really a difference in rendering speed.

But I'm still waiting for benchmarks for -msse3....

.:B:. · 2007-03-28 00:50:37

javajunky wrote:

Hmm, I'm confused. I just bought a Laptop with a Turion64x2 cpu, I *thought* this was quite a modern 64bit chip, but it seems I don't have sse3 either
edit: Ignore me just been on a whirlwind tour of extended instructions, I guess pni superceeds/equivalences sse3 (CPU-Z under windows at least informs me my chip *does* support SSE3) I'm adding this note in case it confuses anyone else who's brain is as tiny as mine <g> Sorry!

When you check /proc/cpuinfo you'll see pni, and if you google around for how to find out if your CPU has sse3 support using the proc interface, you'd find out that's what pni refers to .

I got a Turion ML-32 that supports SSE3 too, btw.

jb · 2007-03-28 05:23:48

I too, have a Sempron3100 with SSE3. However, as long as SSE3 or SSE4 aren't made a requirement for the x86_64 platform, since one doesn't imply the other, I'm all for it.

[vEX] · 2007-03-28 10:09:50

AndyRTR wrote:

OK, wanna see some really fast stuff?
It's based on Swiftfox patches and adds CLFAGS that i call really "nice" :D - there's really a difference in rendering speed.
But I'm still waiting for benchmarks for -msse3....

I was trying to compile this but dies pretty much right away:

==> Starting build()...
patching file uriloader/exthandler/Makefile.in
Hunk #1 succeeded at 103 (offset 1 line).
patching file uriloader/exthandler/unix/nsGNOMERegistry.cpp
patching file uriloader/exthandler/unix/nsGNOMERegistry.h
The next patch would create the file uriloader/exthandler/unix/nsMIMEInfoUnix.cpp,
which already exists! Skipping patch.
1 out of 1 hunk ignored -- saving rejects to file uriloader/exthandler/unix/nsMIMEInfoUnix.cpp.rej
The next patch would create the file uriloader/exthandler/unix/nsMIMEInfoUnix.h,
which already exists! Skipping patch.
1 out of 1 hunk ignored -- saving rejects to file uriloader/exthandler/unix/nsMIMEInfoUnix.h.rej
patching file uriloader/exthandler/unix/nsOSHelperAppService.cpp
==> ERROR: Build Failed. Aborting...

Arch Linux

#1 2006-11-21 11:00:32

x86_64: poll for optional CFLAGS

#2 2006-11-21 11:08:04

Re: x86_64: poll for optional CFLAGS

#3 2006-11-21 12:46:21

Re: x86_64: poll for optional CFLAGS

#4 2006-11-21 15:02:38

Re: x86_64: poll for optional CFLAGS

#5 2006-11-21 15:23:57

Re: x86_64: poll for optional CFLAGS

#6 2006-11-21 18:13:35

Re: x86_64: poll for optional CFLAGS

#7 2006-11-21 18:35:37

Re: x86_64: poll for optional CFLAGS

#8 2006-11-21 19:00:27

Re: x86_64: poll for optional CFLAGS

#9 2006-11-21 19:10:09

Re: x86_64: poll for optional CFLAGS

#10 2006-11-21 19:58:06

Re: x86_64: poll for optional CFLAGS

#11 2006-11-21 22:23:07

Re: x86_64: poll for optional CFLAGS

#12 2006-11-22 00:52:53

Re: x86_64: poll for optional CFLAGS

#13 2006-11-23 07:12:35

Re: x86_64: poll for optional CFLAGS

#14 2006-11-23 08:10:41

Re: x86_64: poll for optional CFLAGS

#15 2006-11-23 09:01:25

Re: x86_64: poll for optional CFLAGS

#16 2006-11-25 04:47:05

Re: x86_64: poll for optional CFLAGS

#17 2006-12-08 16:52:45

Re: x86_64: poll for optional CFLAGS

#18 2006-12-08 16:58:19

Re: x86_64: poll for optional CFLAGS

#19 2006-12-08 17:06:31

Re: x86_64: poll for optional CFLAGS

#20 2006-12-08 19:44:23

Re: x86_64: poll for optional CFLAGS

#21 2006-12-12 16:26:44

Re: x86_64: poll for optional CFLAGS

#22 2007-03-27 21:18:49

Re: x86_64: poll for optional CFLAGS

#23 2007-03-28 00:50:37

Re: x86_64: poll for optional CFLAGS

#24 2007-03-28 05:23:48

Re: x86_64: poll for optional CFLAGS

#25 2007-03-28 10:09:50

Re: x86_64: poll for optional CFLAGS

Board footer