Add performance test skeleton.

Add performance test skeleton.

BUG=406405
LOG=n

TEST=tools/run_benchmarks.py --arch=ia32 test/perf-tests/perf-tests.json
TEST=tools/run-tests.py --mode=release --arch=ia32 --nopresubmit perf-tests --no-variants

[Michael Achenbach, 2014-08-24 20:37:57]

PTAL. I'd probably add one more level of hierarchy to the benchmark runner
before landing this.

https://codereview.chromium.org/503643002/diff/1/test/perf-tests/infra-test.cc
File test/perf-tests/infra-test.cc (right):

https://codereview.chromium.org/503643002/diff/1/test/perf-tests/infra-test.c...
test/perf-tests/infra-test.cc:11: printf("Trace(Constant), Result(10.0),
StdDev(1.0)\n");
I think it might be useful to have one level of hierarchy more here and return:
Graph(Infra), Trace(Constant), Result(10.0), StdDev(1.0)

And teach the benchmark driver to group these graphs and traces.

[Benedikt Meurer, 2014-08-25 04:16:26]

https://codereview.chromium.org/503643002/diff/1/test/perf-tests/infra-test.cc
File test/perf-tests/infra-test.cc (right):

https://codereview.chromium.org/503643002/diff/1/test/perf-tests/infra-test.c...
test/perf-tests/infra-test.cc:11: printf("Trace(Constant), Result(10.0),
StdDev(1.0)\n");
How about using the same output format as Chrome (as found in
testing/perf/perf_test.cc)?

<*>RESULT <graph_name>: <trace_name>= <value> <units>
<*>RESULT <graph_name>: <trace_name>= {<mean>, <std deviation>} <units>
<*>RESULT <graph_name>: <trace_name>= [<value>,value,value,...,] <units>

[Michael Achenbach, 2014-08-25 06:33:41]

PTAL

[Benedikt Meurer, 2014-08-25 06:55:43]

https://codereview.chromium.org/503643002/diff/20001/test/perf-tests/perf-tes...
File test/perf-tests/perf-tests.json (right):

https://codereview.chromium.org/503643002/diff/20001/test/perf-tests/perf-tes...
test/perf-tests/perf-tests.json:7: "unit": "ms"
The unit should be part of the RESULT output line.

[Michael Achenbach, 2014-08-25 08:15:56]

Done