From Fedora Project Wiki

No edit summary
(→‎CMOV support: new section)
 
(3 intermediate revisions by 3 users not shown)
Line 1: Line 1:
'''We can support i586 processors by handling CMOV invalid instruction in kernel'''
* [[User:Dmalcolm|Dmalcolm]] 18:02, 15 June 2009 (UTC) If a Fedora user wants to verify that a CPU will work with F12, how should he/she query /proc/cpuinfo ?  (and does smolt have any of this data?)  Thanks
* [[User:Dmalcolm|Dmalcolm]] 18:02, 15 June 2009 (UTC) If a Fedora user wants to verify that a CPU will work with F12, how should he/she query /proc/cpuinfo ?  (and does smolt have any of this data?)  Thanks
** [[User:notting|Bill Nottingham]] Once a set has been decided on, this should be pretty trivial. With respect to the proposal, 'grep sse2 /proc/cpuinfo' should work.
** [[User:notting|Bill Nottingham]] Once a set has been decided on, this should be pretty trivial. With respect to the proposal, 'grep sse2 /proc/cpuinfo' should work.
* [[User:Bruno|Bruno]]: Is there a way to tell which binaries actually use SSE2 instructions so that a secondary arch targeting athlons would be able to inherit safe binaries from the i686+ arch?
* [[User:Bruno|Bruno]]: Is there a way to tell which binaries actually use SSE2 instructions so that a secondary arch targeting athlons would be able to inherit safe binaries from the i686+ arch?
** [[User:notting|Bill Nottingham]] This part of the change was reverted.


* [[User:Ngompa|Neal Gompa]]: Pentium II and lower do not support SSE or SSE2. Pentium III only supports SSE, not SSE2. Pentium 4 supports SSE2 and SSE3. However, all Pentium chips (Pentium Pro and higher) support i686 + MMX. If we wanted to move to i686, a secondary arch offering no SSE2 would be a good idea.
* [[User:Ngompa|Neal Gompa]]: Pentium II and lower do not support SSE or SSE2. Pentium III only supports SSE, not SSE2. Pentium 4 supports SSE2 and SSE3. However, all Pentium chips (Pentium Pro and higher) support i686 + MMX. If we wanted to move to i686, a secondary arch offering no SSE2 would be a good idea.
** [[User:akostadi|Aleksandar Kostadinov]]: I would second that SSE2 is a overkill. Why is not MMX enough? In fact I own an Athlon desktop that is pretty much enough for my work and I would have still worked on that if other factors wouldn't force me to entirely use my laptop (that's not much faster btw).
** [[User:akostadi|Aleksandar Kostadinov]]: I would second that SSE2 is a overkill. Why is not MMX enough? In fact I own an Athlon desktop that is pretty much enough for my work and I would have still worked on that if other factors wouldn't force me to entirely use my laptop (that's not much faster btw).
** [[User:notting|Bill Nottingham]] This part of the change was reverted.


* [[User:Michaelo|Michael Osborne]]: This change will break Fedora on Via C3 processors. These are used quite a bit for mini-itx boards for LTSP terminals as they don't require active cooling. A lot of breakage for little gain AFAICS.
* [[User:Michaelo|Michael Osborne]]: This change will break Fedora on Via C3 processors. These are used quite a bit for mini-itx boards for LTSP terminals as they don't require active cooling. A lot of breakage for little gain AFAICS.


* [[User:billmcgonigle|Bill McGonigle]]: There's an extensive thread on this on fedora-devel.  Many questioned the wisdom of abandoning so much hardware for a meager 1-2% performance gain.
* [[User:billmcgonigle|Bill McGonigle]]: There's an extensive thread on this on fedora-devel.  Many questioned the wisdom of abandoning so much hardware for a meager 1-2% performance gain.
* [[User:mikep|W. Michael Petullo]]: See also https://bugzilla.redhat.com/show_bug.cgi?id=523559.
** Why is my processor no longer supported in order to support a negligible performance gain (~1%)?
** Could we not support both new and old processors using two release architectures?
** Do we not want to support the embedded market?
** One historic advantage of Open Source was the ability to inject new life into old hardware. Do we want to lose this?
** Why is [https://www.redhat.com/archives/fedora-devel-list/2009-August/msg00780.html OLPC used as a benchmark] for worthy processors? If altruism is a motivation for this cut-off, then it seems allowing Fedora users to continue to use old hardware will do more then OLPC will (based on the OLPC statistics on Wikipedia -- enough have been donated to equip 0.0002% of the world population -- compared to VIA's processor sales).
== CMOV support ==
We could maintain support for i586/C6 by handling CMOV invalid instruction interrupt in kernel.

Latest revision as of 21:26, 15 October 2009

We can support i586 processors by handling CMOV invalid instruction in kernel

  • Dmalcolm 18:02, 15 June 2009 (UTC) If a Fedora user wants to verify that a CPU will work with F12, how should he/she query /proc/cpuinfo ? (and does smolt have any of this data?) Thanks
    • Bill Nottingham Once a set has been decided on, this should be pretty trivial. With respect to the proposal, 'grep sse2 /proc/cpuinfo' should work.
  • Bruno: Is there a way to tell which binaries actually use SSE2 instructions so that a secondary arch targeting athlons would be able to inherit safe binaries from the i686+ arch?
  • Neal Gompa: Pentium II and lower do not support SSE or SSE2. Pentium III only supports SSE, not SSE2. Pentium 4 supports SSE2 and SSE3. However, all Pentium chips (Pentium Pro and higher) support i686 + MMX. If we wanted to move to i686, a secondary arch offering no SSE2 would be a good idea.
    • Aleksandar Kostadinov: I would second that SSE2 is a overkill. Why is not MMX enough? In fact I own an Athlon desktop that is pretty much enough for my work and I would have still worked on that if other factors wouldn't force me to entirely use my laptop (that's not much faster btw).
    • Bill Nottingham This part of the change was reverted.
  • Michael Osborne: This change will break Fedora on Via C3 processors. These are used quite a bit for mini-itx boards for LTSP terminals as they don't require active cooling. A lot of breakage for little gain AFAICS.
  • Bill McGonigle: There's an extensive thread on this on fedora-devel. Many questioned the wisdom of abandoning so much hardware for a meager 1-2% performance gain.
  • W. Michael Petullo: See also https://bugzilla.redhat.com/show_bug.cgi?id=523559.
    • Why is my processor no longer supported in order to support a negligible performance gain (~1%)?
    • Could we not support both new and old processors using two release architectures?
    • Do we not want to support the embedded market?
    • One historic advantage of Open Source was the ability to inject new life into old hardware. Do we want to lose this?
    • Why is OLPC used as a benchmark for worthy processors? If altruism is a motivation for this cut-off, then it seems allowing Fedora users to continue to use old hardware will do more then OLPC will (based on the OLPC statistics on Wikipedia -- enough have been donated to equip 0.0002% of the world population -- compared to VIA's processor sales).

CMOV support

We could maintain support for i586/C6 by handling CMOV invalid instruction interrupt in kernel.