Monday 7th Feb 2011

This month's meetings

WorkingGroups/ToolChain/Meetings
<< <  2011 / 2 >  >>
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            

Attendees

Agenda

Action Items from this Meeting

Action Items from Previous Meeting

Minutes

Realview PBX:

  • Needs a block device, any one that works
  • Currently only have NFS root
  • A9 seems to work well enough for the kernel to boot, run normal code, pass kernel lock tests
  • More fundamental missing things:
    • QEMUs SMP support is basic - all cores in one thread, so slow. Hard to change.
    • Hard to use for validation as the host memory systems are different to target
  • Dave: prefer some other board rather than fix Realview
  • Michael: a bit useless from a end-user point of view
  • Dave: we might want easier support of making NFS roots in l-m-c

QEMU release:

  • Done a rc2
  • Would like to release rc2 as the actual release
  • Has a couple of known bugs, probably not regressions
  • Push back the mising serial console problem to next release (seems to be a fault in the image)
  • Steve will make packages hopefully on the same date
  • Not planning to do release candidates next time, this time to shake things out and to seed Steve

GCC:

  • x86_64 and i686 look fine
  • There are some failures - see the spreadsheet above
  • Might be related to address space randomisation
  • Might be due to the change of host
  • Will see after the builds on other hosts

Dave and testing EGLIBC:

  • Wants to make sure that he doesn't break old architectures
  • How can we test/ensure?
  • Richard: readelf -p should tell you if it's a ARMv5 binary. Keeps the build attributes all the way down
  • Might try running on QEMU locked down to an older CPU
    • Peter: it does try to reduce the instruction set to what's there
  • Ramana: building with a ARMv5 defaulting GCC should also warn of mixed code

GDB:

  • Did the release on Friday, testing looks good
  • Michael: please review the final builds, send me an email

Announcement:

  • Andrew asked if there should be a mega one instead. Michael will see how it looks.

Richard and NEON:

  • Where is he? Might get into the NEON side in parallel with IFUNC
  • Will read into the past emails, see where Julian was at, pick up

PGO:

  • It works as designed so we say that we support it
  • Dave: does it improve things? Have we measured it?
  • No from Ulrich and Michael

Standardising EEMBC:

  • Currently doesn't work out of the box
  • Should we standardise?
    • Dave would like a set of instructions to run
    • Andrew is concerned that you can't pick a random set of iterations due to data phases
    • Ramana has a list of iterations
      • On simulators can have a low number (<100) that gives a good result

      • On Linux, higher is better to amortize
    • Revital still sees cases where the helper functions are marked as 'hot'
      • Are some of the functions getting inlined into the helpers?
  • Andrew: for some in the suite, the number of iterations matter
    • For example, may have one with three images so a multiple of three is needed
    • Ramana: matrix01 benchmarks is an example

Benchmark runs:

  • Michael will take the history from the 4.4 release up to today on trunk, divide it, and keep dividing

WorkingGroups/ToolChain/Meetings/Archive/2011-02-07 (last modified 2013-08-30 11:49:15)