Separates faster and slower bench alerts; sorts by delta.

tools/tests/benchalerts/Perf-Android-Nexus7-Tegra3-Arm7-Release/output-expected/stderr

 Traceback (most recent call last):
-  File "bench/check_bench_regressions.py", line 218, in <module>
+  File "bench/check_bench_regressions.py", line 228, in <module>
     main()
-  File "bench/check_bench_regressions.py", line 214, in main
+  File "bench/check_bench_regressions.py", line 224, in main
     platform_and_alg)
-  File "bench/check_bench_regressions.py", line 161, in check_expectations
-    '\n'.join(exceptions))
+  File "bench/check_bench_regressions.py", line 171, in check_expectations
+    raise Exception('\n'.join(outputs))
 Exception: Bench values out of range:
-Bench desk_baidu.skp_record_,Perf-Android-Nexus7-Tegra3-Arm7-Release-25th value 0.83 out of range [5.155, 11.033].
+
+Slower:
+Bench desk_blogger.skp_record_,Perf-Android-Nexus7-Tegra3-Arm7-Release-25th off range [0.4, 0.6] (1.794 vs 0.5, 258.8%).
+Bench desk_amazon.skp_record_,Perf-Android-Nexus7-Tegra3-Arm7-Release-25th off range [-1.0, 1.2] (1.213 vs 1.1, 10.2727272727%).

[epoger, 2014/02/19 18:51:08]

we allow negative results?  (-1.0 as lower bound?)

[benchen, 2014/02/19 21:08:08]

Yes, the batch script algorithm could make the lower bound smaller than zero. I
think it's good to leave them as is since they reflect the actual calculated
ranges (not truncated by 0), so we still know if they would be too wide.
On 2014/02/19 18:51:08, epoger wrote:

+
+Faster:
+Bench desk_baidu.skp_record_,Perf-Android-Nexus7-Tegra3-Arm7-Release-25th off range [0.9, 1.0] (0.83 vs 0.939, -11.6080937167%).