Tuesday 24 January 2012
This month's meetings
<< < 2012 / 1 > >>
- Andrew Stubbs
- Åsa Sandahl
- Ken Werner
- Michael Hope
- Ramana Radhakrishnan
- Ulrich Weigand
- Review action items from last meeting
- Connect week.
- 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]
- 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)