Thursday 8th March 2012

This month's meetings

WorkingGroups/ToolChain/Meetings
<< <  2012 / 3 >  >>
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

  • Åsa Sandahl
  • Ken Werner
  • Michael Hope
  • Peter Maydell
  • Ramana Radhakrishnan
  • Ulrich Weigand

Agenda

  • Stand up call

Minutes

Asa:

  • Triaging tickets
  • One left - failure on PowerPC with memory exhaustion
  • Michael will try to reproduce
  • Created a wiki page on running benchmarks with cbuild
  • Covers using the auto builders and running things yourself
  • V8/SunSpider as a benchmark
  • Has switched from scons to gyp, easier to pass flags via CXXFLAGS/CFLAGS
  • Writing a Makefile to drive it

Ramana:

  • Setting up SPEC 2000 runs for hot/cold partitioning
  • Testsuite showed no regressions, turns some unsupported into passes
  • VFP addressing modes
  • Bug with C++ templates, dynamic libraries, and unique objects

Peter:

  • Re-working the QEMU CP15 handling
  • Need to disentangle the current shortcuts from many CPUs into something clean
  • Making progress, three categories left
  • Getting ready for next release

Ken:

  • Removed linux-linaro, switched to linux-yocto
  • Updated wiki on how to build OE Core with meta-linaro
  • Created initial version of card

Ulrich:

  • Finishing usat/ssat patch. In mainline, Linaro 4.6, Linaro 4.7
  • Ran into spurious failures where branch name affects test case
  • ...due to the embedded ident string
  • Now that 4.8 is open, sending patches
  • First of fwprop subreg sent
  • Doing builds, test, benchmarks on Ira's vectoriser patches
  • Richard is/will pushing sched pressure
  • End of loop:
    • Works for this case
    • Regresses by inhibiting other optimisations
    • As unisgned can overflow, signed can't
    • So optimisers can't fire due to overflow

Michael:

  • KVM organisation
  • Changed SPEC 2000 to ref
  • Issues with DENBench and PSNR
  • Posted tiny patch to turn on longlong routines in libgcc
  • Backported makecontext() to Precise, will probably be accepted
  • Changed the version string in snapshots

WorkingGroups/ToolChain/Meetings/Archive/2012-03-08 (last modified 2013-08-30 11:50:03)