Monday 30th January 2012

This month's meetings

WorkingGroups/ToolChain/Meetings
<< <  2012 / 1 >  >>
Mon Tue Wed Thu Fri Sat Sun
            1
2 3 4 5 6 7 8
9 10 11 12 13 14 15
16 17 18 19 20 21 22
23 24 25 26 27 28 29
30 31          

Attendees

  • Andrew Stubbs
  • Åsa Sandahl
  • Ken Werner
  • Michael Hope
  • Peter Maydell
  • Ramana Radhakrishnan
  • Richard Earnshaw
  • Rusty Russell
  • Ulrich Weigand
  • Matthias Klose

Agenda

  • Review action items from last meeting
  • Benchmarking
    • Can benchmark against trunk, release branches
  • PR50313 / Debian 657682 (Ramana, PIC)
  • 922474 (64 bit sync)

  • 923397 (alignment attribute ignored)

  • SunSpider

  • -O3 write up
    • Final graphs from Åsa
    • Package counts / sloc from Ken
  • Connect (MichaelHope/Sandbox/Q1.12Plans)

  • KVM
    • Michael will call Christoffer tomorrow, he might be at Connect

Action Items from this Meeting

  • TBA

Action Items from Previous Meeting

  • DONE: Peter and Rusty to talk re: QEMU release process and fallbacks
  • ACTION: Michael to add benchmarking to the jobs list
  • ACTION: Michael to reproduce the gdbserver libgcc step into
  • DONE: Michael to make a blueprint on enabling hot/cold partitioning in PGO
  • DONE: Peter to send Michael a couple of paragraphs on FMA testing
  • ACTION: Richard to think about soft float vs hard equivalence

Minutes

Actions:

  • Asa to add benchmarking description to the job page

Benchmarking:

  • Can submit merge requests against the Launchpad GCC project
  • Will run through the auto builders
  • Can be benchmarked afterwards
  • Asa:
    • Did spawn one against Ramana's branch on Friday
  • Asa to froward Michael's latest email to Ramana to see what went missing

PR50313 / Debian 657682 (Ramana, PIC):

  • Will apply to the FSF 4.6 branch
  • Has run, bootstrapped
  • 1 week shakedown on trunk
  • But how to get it into Linaro GCC?
  • Release merge hasn't been done yet
  • Ramana and Andrew to talk and hold the merge if it makes things simpler

922474 (64 bit sync):

  • Linaro GCC only? Yes
  • Was an intrinsic that we should have implemented in the backend
  • Falls back to memory barrier followed by 64 bit store
  • Could be some implementations with less than 64 bit interface, STRD may not be atomic
  • Why not in trunk?
    • Much work done on software transaction management
    • Legacy interfaces got implemented in a different fashion
    • But has never existed in a released FSF compiler
    • Would normally fail with a link error
    • Disabling would be a feature regression
    • May exist in FSF 4.6 as this has always fallen back to generation instead of libgcc
  • Fix, check in FSF, note
  • Thought experiment: could you write a binary scanner

923397 (alignment attribute ignored):

  • If semantically different and shouldn't link? Can't check
  • Else the change is in visibility
  • Need to check with the language lawyers
  • Joseph perhaps?

SunSpider:

  • Tried compiling on ARM, works fine, 18 minutes to build
  • Short enough, good
  • Can run the V8 benchmarks and SunSpider on ARM and x86 standalone

  • Has documented steps in her Sandbox
  • Next step is how to fiddle with scons
  • Michael and Asa will work on this at Connect

Connect:

KVM:

  • Michael will talk with Christoffer tomorrow
  • He's in the bay area at the end of Connect
  • Michael will re-jigger to the conference line and send Rusty the details

makecontext:

  • 2.15 vs 2.13 on Ubuntu? 2.15 does boot on Panda, AC100
  • Regressions due to the dynamic loader, seen on all archs
  • Regressions show through the testsuite, GDK loaders, Apache
  • A reduced testcase has been made
  • At least one is in IFUNC and only shows if the host has that feature
  • Grab the packages from Matthias's PPA
  • https://launchpad.net/~ubuntu-toolchain-r/+archive/glibc/+packages

  • ACTION: Michael to reproduce the fault
  • ACTION: Michael to decide or push what to on makecontext

WorkingGroups/ToolChain/Meetings/Archive/2012-01-30 (last modified 2013-08-30 11:50:04)