This adds two benchmarks, one for flights-app which doesn't use data binding
and one for city-list which uses lots of it.
1st runs in each benchmark contain the cost of collecting all the metadata
about bindings and templates in addition to creating instances. Later runs
only contain the cost of creating instances.
Overall it looks like SkyBinder is 24.4% - 35.5% faster depending on if
you're using data binding (the performance improvement is larger if you are).
It also appears to put less pressure on the GC.
== flights-app.sky benchmark ==
TemplateBinding:
flights-app.sky => [103.0, 27.0, 26.0, 26.0, 27.0, 26.0, 39.0, 29.0, 29.0, 72.0]
flights-app.sky => [101.0, 27.0, 27.0, 27.0, 27.0, 26.0, 37.0, 33.0, 26.0, 69.0]
flights-app.sky => [103.0, 27.0, 26.0, 26.0, 27.0, 26.0, 37.0, 27.0, 28.0, 71.0]
1st run avg = 102.33333333333333
avg = 33.148148148148145
SkyBinder:
flights-app.sky => [78.0, 25.0, 25.0, 25.0, 25.0, 30.0, 34.0, 26.0, 25.0, 32.0]
flights-app.sky => [77.0, 28.0, 25.0, 26.0, 26.0, 28.0, 31.0, 23.0, 26.0, 26.0]
flights-app.sky => [77.0, 25.0, 25.0, 24.0, 25.0, 28.0, 31.0, 23.0, 25.0, 24.0]
1st run avg = 77.33333333333333
avg = 26.51851851851852
24.4% reduction for 1st run, 20% reduction for normal runs. Normal runs
actually seem maybe only 5% faster, but it looks like pre SkyBinder there
were occasional runs that took 250% longer than normal from GC pauses. If I
increase the iteration count for SkyBinder I can see those long pauses
again, but they're much less frequent presumably because SkyBinder creates
less garbage.
== city-list.sky benchmark ==
TemplateBinding:
city-list.sky => [820.0, 609.0, 600.0, 529.0, 430.0, 437.0, 387.0, 462.0, 541.0, 415.0]
city-list.sky => [827.0, 599.0, 551.0, 481.0, 436.0, 425.0, 363.0, 366.0, 427.0, 364.0]
city-list.sky => [823.0, 574.0, 467.0, 483.0, 444.0, 429.0, 364.0, 363.0, 433.0, 367.0]
1st run avg = 823.3333333333334
avg = 457.25925925925924
SkyBinder:
city-list.sky => [521.0, 369.0, 320.0, 285.0, 319.0, 273.0, 267.0, 262.0, 272.0, 331.0]
city-list.sky => [553.0, 362.0, 314.0, 284.0, 327.0, 278.0, 263.0, 264.0, 249.0, 290.0]
city-list.sky => [556.0, 355.0, 317.0, 289.0, 333.0, 277.0, 269.0, 266.0, 264.0, 263.0]
1st run avg = 543.3333333333334
avg = 294.8888888888889
34% reduction for 1st run, 35.5% reduction for subsequent runs. This also
shows less variance in the later runs than pre SkyBinder which I think might
again be less GC pressure.
TBR=ojan@chromium.org
BUG=
Review URL: https://codereview.chromium.org/834983002
This script isn't quite working, but it's valuable to checkpoint the work at
this point. I'm coordinating with the chromeperf.appspot.com folks to get the
server to accept the data.
R=eseidel@chromium.org
Review URL: https://codereview.chromium.org/702603005
Address comments
Add a basic parser benchmark
This CL adds a basic benchmark for the parser. There isn't a direct way to
exercise the parser in Sky, so we use imports.
This CL also adds a load event to <import> to determine when the import is done
loading.
R=eseidel@chromium.org
parser benchmark wip
