Linaro has had a project under development to help port 1400 instances of code in GNU/Linux which have ARM assembly code in them.
After an analysis by Steve McIntyre, it was observed that some of these modules were written a long time ago and assembly may no longer be appropriate for them, some could really use an intrinsic in the compiler and some might be better with an algorithm change. It was determined to extend the idea of "porting" to "performance enhancement", with a strong sub-goal to getting all 1400 modules "ported" in a short period of time.
Several people have been working on this for some time, and the contest/porting effort is almost ready to go.
This session will:
- list the goals of this effort
- explain the rules and entry paths to the contest, and prizes
- solicit comments and ideas on those rules and paths
- solicit ideas for categories of performance improvement (cache utilization, memory size reduction, etc).
- solicit ideas for porting and measuring tools
- ask for help in creating a few "example ports and performance gains" to act as examples for entries
- explain judging rules, and seek volunteers to act as mentors, judges, etc.
Later in the week hacking sessions that may generate examples of porting and performance submissions to the contest will be encouraged.
Link to Hangout event: https://plus.google.com/events/c62lkf3vdpv4k3nsee2acfe2df0?partnerid=gplp0
Link to Etherpad: http://pad.linaro.org/p/LCU13-ARM-performance-optimizations