Adds a couple more measures to the benchmarking app

apps/benchmark/benchmark_app.cc

Index: apps/benchmark/benchmark_app.cc
diff --git a/apps/benchmark/benchmark_app.cc b/apps/benchmark/benchmark_app.cc
index bd7c8a99b1269a2b751dc0f282efde5321e2a9c6..8c88d89415b89fd5b364e07f0d96ec959b0c69e2 100644
--- a/apps/benchmark/benchmark_app.cc
+++ b/apps/benchmark/benchmark_app.cc
@@ -116,13 +116,18 @@ class BenchmarkApp : public mojo::ApplicationDelegate,
     bool succeeded = true;
     Measurements measurements(events, time_origin_);
     for (const Measurement& measurement : args_.measurements) {
-      double result = measurements.Measure(measurement);
-      if (result >= 0.0) {
-        printf("measurement: %s %lf\n", measurement.spec.c_str(), result);
-      } else {
-        succeeded = false;
-        printf("measurement: %s FAILED\n", measurement.spec.c_str());
+      std::vector<double> results;
+      measurements.Measure(measurement, &results);
+      printf("measurement: %s", measurement.spec.c_str());
+      for (double result : results) {

[ppi, 2015/10/08 22:29:47]

As discussed offline, I think it would be good to stick with one result per
measurement. We could have the measurement take a percentile to compute as
parameter: percentile_duration/category/event_name/0.9.

[zra, 2015/10/09 14:51:15]

Done.

+        if (result >= 0.0) {
+          printf(" %lf", result);
+        } else {
+          succeeded = false;
+          printf(" FAILED");
+        }
       }
+      printf("\n");
     }
 
     // Scripts that run benchmarks can pick this up as a signal that the run