Monday 23rd 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          



  • Review action items from last meeting
  • Review 11.11 blueprints
    • Work load vs uncertainty
    • Thumb-2 performance
    • NEON performance
    • Better intrinsics
    • Toolchain in Yocto/Android/Ubuntu
    • Yocto/Ubuntu as a testsuite
    • Good backtracing
  • Any that require specifications
    • In distributions?
    • Deeper validation?
    • Top level cross debug?
  • Public plan review
  • Should we be able to identify Linaro GCC?

Action Items from this Meeting

  • TBA

Action Items from Previous Meeting

  • DONE: Michael to shift the call an hour later starting next week
  • DONE: Michael to point Ramana at the logs for VFP vs NEON
  • DONE: Ramana to check into the bintuils vmov patch and recommend on the upgrade
  • DONE: Andrew to check inside CSL re: shrink-wrap clobbers to see what Bernd has done
  • DONE: Peter to try today's hwpack and rootfs on vexpress and OMAP3
  • DONE: gcc.dg/pr48335-2.c: Michael to check and log: is tracked as PR48493


Dave, Ken, Andrew, Chung-Lin, Richard, Ira, Matthias, Ramana, Revital

Stable branch:

  • Heard anything from Kate? Matthias: not yet

VFP vs NEON results:

  • Are a few differences. Overall in line with previous versions.
  • Not sure which is 'better'. Some may be test related.
  • Should do an ARM vs Thumb-2 build as well?

Binutils and vmov:

  • Ramana recommends binuitls patch for Natty as a SRU
  • No packages should be affected as all native builds with no 64 bit targets
  • Already in Oneiric

Cortex-A8 branch fix:

  • Is now now in Natty and Oneiric


  • Bernd is on other things at the moment
  • No upstream replies for a month. Bernd will have to ping soon.

QEMU 2011.04 and current hwpack:

  • Do work together, but you get more warnings than previousaly due to a kernel change
  • Periodic unfortunatly

Blueprint review (Thumb-2):

  • memset/memclr: tune the current versions, play with the macros that are currently there
  • Driven by the DENBench example
  • Also add unaligned support
  • Dave: going through SPEC2k6 at the moment, looking at these
    • Dynamic length ones are hard
  • How do we decide the way to implement these? What is the 'best' way?
  • How do we decide if we should align first?
  • When you don't know the alignment, NEON on A* handles unaligned much better
    • Ramana would rather not unless you know NEON is on

10 work items (10 weeks?) planned, want 40 weeks:

  • Ramana will have a look at the efforts
  • Revital has no more SMS work planned
  • Note also that -mtune=a9 often regresses compared to -mtune=a8 when running on an A9

Blueprints (NEON):

  • Split most of these into at least two - they're not a week, not a month
  • Add -fvect-cost-model causing a regression

Michael to talk to PM group about better estimates.

Better intrinsics:

  • Is there work to be done?
  • Richard doesn't know of any specific area where they need improvement
  • Will fix as found
  • Review upstream bugs first? Yes
  • Michael would like to do examples and proof in the future but it's wishlist


  • Ken to write up the specs

Good backtracing:

  • Needs a blueprint
  • Needs work items on the hard issues and investigations so that they're captured

Public plan review:

  • Middle of next week
  • Slides due in a few days

Should we be able to identify Linaro GCC?

  • From Matthias
  • Was a bug reported 780551 where on the FSF branch an ABI changed happened

  • Worked around by the reporter by checking the GCC version
  • Any other instances of a predefined value in other branches?
  • Some use the package string, but no macros
  • Andrew says 'bad bad bad wrong wrong wrong' :)

  • Correct way to fix is to do an autoconf test instead?
  • Continue with the discussion next week

WorkingGroups/ToolChain/Meetings/Archive/2011-05-23 (last modified 2013-08-30 11:47:26)