Tuesday 24 January 2012

This month's meetings

WorkingGroups/ToolChain/Meetings
<< <  2012 / 1 >  >>
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

  • Andrew Stubbs
  • Åsa Sandahl
  • Ken Werner
  • Michael Hope
  • Ramana Radhakrishnan
  • Ulrich Weigand

Agenda

  • Review action items from last meeting
  • Connect week.
  • Sunspider
  • AOB .

Action Items from this Meeting

  • scheduler description for the A9.[ramana to look into that and get back to Uli]

Action Items from Previous Meeting

  • Action on Michael to document the benchmarks and ubutest [TODO].
  • Uli to look into sched-pressure (Richard's patch) and specify flags to Michael and Åsa. [TODO]
  • Åsa to send out about Sunspider results. [DONE] -
  • A15 backport should be done in time for the next release. [DONE]

Minutes

  • Dependencies on Sunspider for Chrome - standalone driver for Sunspider (Can do this with Chromium v8 engine)
    • We can also run the v8 benchmark suite and use that as well. Set up paths accordingly to run with a new toolchain.
  • Performance meeting - 9 a.m. Pacific time on the Tuesday of Connect >

  • -O3 regressions , pre-built binary toolchains and maybe perf profiles for the benchmarks we are interested in ?
  • Preliminary results of performance patches
    • Auto-inc-dec patches . scheduler description for the A9.[ramana to look into that and get back to Uli]
    • vld1 vs vldr
    • Ramana to do a brain-dump of some of the backlog on his plate.
    • 4.7 - tracking branch .( Performance comparisons for 4.7 with Linaro 4.6 and FSF 4.6) - Åsa to run some benchmarks.
    • Longer term benchmarking on FSF trunk.
  • Sunspider - up and building by that time ? 10% regressions compared to 4.4 . Lets see if we can find something. Get set up to run Sunspider at connect. Åsa to help set us up for the same.
  • Uli - vmov fwprop patch - 4.7 benchmarked Richard's patch to fwprop + Uli's patch to combine . libav - on mainline it showed improvements of upto 13% with no single test regression. No change in perf. in the 4.6 branch - registers tieable patch was needed but that showed a couple of regressions , no performance improvements - turns out to be register pressure . Ramana suspects something with respect to subregs and vec_select - to see if he can find a RichardS patch.
  • Investigate gcc.dg/vect test failures on mainline during connect.

WorkingGroups/ToolChain/Meetings/Archive/2012-01-24 (last modified 2013-08-30 11:48:42)