diff options
| author | Julien Duraj <julien.duraj@linaro.org> | 2017-02-01 16:19:10 +0000 |
|---|---|---|
| committer | Julien Duraj <julien@duraj.fr> | 2017-02-02 19:29:54 +0100 |
| commit | 0e1a3fc44ba4e2e7ea36de9271dd0af7dbf94f9d (patch) | |
| tree | 8960cd4b15013f0623a9f34936ea447add919a8a /framework | |
| parent | 6225f82f9deb28218ee09a1347f7e52001395a80 (diff) | |
| download | art-testing-0e1a3fc44ba4e2e7ea36de9271dd0af7dbf94f9d.tar.gz | |
Fix Benchmark json output for linux devices
For linux devices, there is some extra output from dex2oat when running the benchmarks.
The filtering we had in place was not taking this into account.
As a result, this output was interpreted as a valid benchmark result and
some erratic benchmark results were included in the json.
This patch changes the filtering to include only the lines which end with 'per
iteration'. These are the only lines which contain benchmark results and
which we are interested in.
Change-Id: I91f5898d08937b6eede077a8f6a037b5d67a8255
Diffstat (limited to 'framework')
| -rw-r--r-- | framework/org/linaro/bench/RunBench.java | 1 |
1 files changed, 1 insertions, 0 deletions
diff --git a/framework/org/linaro/bench/RunBench.java b/framework/org/linaro/bench/RunBench.java index 64a03e4..8d595bb 100644 --- a/framework/org/linaro/bench/RunBench.java +++ b/framework/org/linaro/bench/RunBench.java @@ -121,6 +121,7 @@ public class RunBench { log.info(benchmarkIdentifier(method) + ": " + duration + " ns for " + iterations + " iterations"); + // The runner expects each output line to end with "per iteration" System.out.printf("%-40s%.2f ns per iteration\n", benchmarkIdentifier(method) + ":", iterationTime); } |