Monday 28th November 2011

This month's meetings

WorkingGroups/ToolChain/Meetings
<< <  2011 / 11 >  >>
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        

Attendees

Agenda

  • Review action items from last meeting
  • Release week next week
  • Michael is away
    • Check process over this week
  • BIGGEST_ALIGNMENT

  • OpenOCD for Cortex-A9 status and the card

  • Next steps in 64 bit sync primitives
  • QEMU work between Dave and Peter
  • 64 bit operations
  • Backport requests:
    • 714921 memory usage due to -fvar-tracking (from Fedora)

    • 675347 Qt and volatile bitfields

Action Items from this Meeting

  • ACTION: Ira to send an email to gcc@ querying BIGGEST_ALIGNMENT
  • ACTION: Dave to merge 64 bit operations for this month's release
  • ACTION: Andrew to ping Julian on 675347

Action Items from Previous Meeting

  • DONE: See how many versions of gas have support

Minutes

Next week is release week:

  • Peter: QEMU should be OK, basically the same as last month
  • Ulrich: hopes that remote core generation will be done over that time
    • Many things tied together
    • Has a plan and is discussing on the list
    • What parts are being touched?
    • One piece in bfd, and the common core, and the ARM specific
  • Michael is away, Andrew to run
    • Do the release merges this week
    • Michael needs to write up the checksheet, ubutest
  • Bugs are OK, blueprints are OK

BIGGEST_ALIGNMENT:

  • The vectoriser tries to force alignment of arrays
  • Checks if the required alignment is less than MAX OFILE ALIGNMENT
  • ...which falls back to BIGGEST_ALIGNMENT
  • ...which is 64 - can never force alignment for quad words
  • Richard S: feels wrong that there's no definition of MAX_OFILE_ALIGNMENT, should be for ELF targets
  • Used for increasing alignment
  • Currently telling that you can't go past 64 bits
  • ACTION: Ira to send an email to gcc@ querying

OpenOCD:

  • Ulrich hasn't had a chance to look into it yet unfortunately
  • Current 7.4 / core file work is more important

Next steps in 64 bit sync primitives:

  • Three things to do:
  • Get membase to pass it's test suite
  • Look around for anything else that uses the primitives
  • Repeat pre-ARMv7 build test
  • Last membase one seems to be a timer problem
  • Has a timer related testsuite to narrow it down
  • BTW, was a good testsuite as it showed the futex fault, and this may be kernel related as well
  • ACTION: Dave to merge for this month's release

QEMU work between Dave and Peter:

  • Dave has: string routines, sync primitives, STM
  • SD card QEMU work
  • QEMU related issues:
    • Code store buffer location, etc that go together with running QEMU on an ARM host
    • There are work-arounds for most of these
  • Priority?
    • Looking at the makecontext() one first - perhaps just add to GLIBC - longest lead time
    • Well-defined, reasonable amount of work
    • Shifting the code store is easy?
    • The x86 bug is unknown - can boot vexpress on ARM...
  • Juggling things?
    • Already juggling many things

64 bit operations:

  • Has patterns that do the 64 bit shifts
  • Sort out problems with sign extend, looked ugly due to extending 32 bit shift amount
  • Doing in core registers first, then to NEON
  • Handle left/right x constant/register
  • Ends up clobbering CC, even though it doesn't
  • Note register cases
  • Other operations?
    • Extend?
    • Bitwise seem to be present, correct
    • Increment/decrement

Backports:

  • 714921 memory usage due to -fvar-tracking (from Fedora)

    • Had response, has been backported
  • 675347 Qt and volatile bitfields

    • No response from Julian, will just do it

WorkingGroups/ToolChain/Meetings/Archive/2011-11-28 (last modified 2013-08-30 11:49:08)