On 16 July 2012 13:51, Michael Hope michael.hope@linaro.org wrote:
We've just started running a weekly benchmark of GCC trunk and Linaro GCC tip. I've written a short script that compares against a baseline and spits out a graph: http://ex.seabright.co.nz/benchmarks/gcc-4.8~svn.png http://ex.seabright.co.nz/benchmarks/gcc-linaro-4.7%2bbzr.png
I'll switch the baseline to GCC 4.7.0 once the build and benchmark run completes. The gcc-linaro results need more data before they'll make sense.
Part way there. An automatic email would be next. We should check the graphs before each performance call.
The 4.7.0 baseline is done and the graph now has enough data points to show something: http://ex.seabright.co.nz/benchmarks/gcc-4.8~svn.png
I've extended the testsuite diff to also make benchmark diffs like: http://ex.seabright.co.nz/benchmarks/gcc-4.8~svn188808/logs/armv7l-precise-c... and http://ex.seabright.co.nz/benchmarks/gcc-4.8~svn189410/logs/armv7l-precise-c...
This suggests that the dip in SPECINT around svn188808 was largely due to mcf and largely corrected by the next run. Some good stuff happened between svn189410 and svn189575.
-- Michael