Thursday 26th May 2011

This month's meetings

<< <  2011 / 5 >  >>
Mon Tue Wed Thu Fri Sat Sun
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          



  • Standup call



  • Working on CSL things at the moment
  • Working on widening multiplies. Find a canonical form that matches with the patterns


  • Completed the latrace trace for SPEC. Bit of work
  • Some make a huge number of calls to libm or gfortran
  • Analysed, will write up
  • ltrace now passes its testsuite on ARM!
  • 64 bit primitives need a compare and swap and (another)


  • Having issues with the porter boxes being unresponsive
  • Looking into known SPEC hot functions
  • Doing another run, will post to benchmarks list when done
  • LP: #745743 looking at memsets
  • Continuing with handing debug information in SMS


  • Working on Panda, looking at running Android on it
  • Is OK with a non-LEB as the screen comes up
  • Found that DNS lookup is set through properties, not /etc/resolv.conf
  • Has built Firefox and Python 3 to measure unwind overhead
  • Michael: please record on disk size as well


  • Doing some GCC archeology to work out reason that a patch has issues
  • See PR42017 - fixes the use of the LR register in leaf functions
  • Tracking the 10 years of history to see why
  • Tried the CoreMark changes on 4.6. Improvements but a small regression.

  • Changing the aligned functions improved this
  • Small reduction in size exposed a alignment regression
  • Has traced down the bootstrap problem


  • Looking at auto increment/decrement. Going pretty well.
  • Expect performance numbers in the next few days


  • First half of the week on planning. Next is to transfer into blueprints
  • Ready to go for the cycle
  • Working on QEMU issue where kernel deliberatly access a register that doesn't exist
  • Asked upstream about performance improvements and many others are interested


  • Looking at blueprints, trying to break things down
  • Will create the engineering blueprints
  • We wern't generating vorn, vbic instructions. Will backport.
  • Is having trouble building GCC on a Oneiric or Natty chroot
  • Is missing /usr/include/asm
  • Revital sees similar
  • Ramana will raise a bug if apt-get build-dep gcc-4.6 isn't enough
  • Having trouble with SPEC as well especially in Perl with it's own build system
  • May have found a bug in length calculation involving the adr instruction (address of label)


  • Fixed a couple of bugs
  • Looking into widening multiplication
  • Works fine for signed types but not unsigned or constants
  • Doesn't match the correct pattern. May relate to Andrew's work


  • Blueprints, blueprints, blueprints
  • Public plan review preperation
  • Initial investigation into A8 vs A9, v5 vs v7, O2 vs O3

WorkingGroups/ToolChain/Meetings/Archive/2011-05-26 (last modified 2013-08-30 11:47:57)