Minutes of the meeting of the Parallelisation subgroup on March 16th 2006

Present Ian Watson, Ian Rogers, Andrew Dinn, Jisheng Zhao, Chris Kirkham

IR is continuing work on the x86 decoder so that parallelisation will apply to non-Java code.

The hardware workload measurer discussed last week was described. AD expressed the view that it was not needed, because relevant measures could be easily done using a software history mechanism.

IR raised the issue of benchmarks. Do we need to create synthetic ones? Is it possible to exploit JavaGrande ....

AD mentioned that multiple simulator bugs had delayed his JByte benchmark runs.

JZ has looked at cycle counts of loop bodies - but is currently getting strange results.

AD suggested that a combination of cycle counts and workload measurement should inform decisions on parallel (re)compilation.

Next meeting JZ will report his progress on cycle counts!

Minutes written by CCK, who welcomes comments/corrections.

Mar1606

Minutes of the meeting of the Parallelisation subgroup on March 16th 2006