Perf tests for Template Literals

Perf tests for Template Literals

[caitp (gmail), 2014-12-01 19:40:14]

PTAL

I'm having some trouble actually getting the NotifyResult callback to actually
be triggered, possibly because each step is happening too quickly. (I was hoping
I could get some help with that)

The test templates are fairly small, to simplify writing, but we could do really
long ones too --- it's just hard to do this nicely without the use of eval() or
similar.

[caitp (gmail), 2014-12-01 19:40:47]

caitpotter88@gmail.com changed reviewers:
+ arv@chromium.org, dslomov@chromium.org

[arv (Not doing code reviews), 2014-12-01 20:04:08]

https://codereview.chromium.org/769113002/diff/1/test/js-perf-test/Templates/...
File test/js-perf-test/Templates/Templates.json (right):

https://codereview.chromium.org/769113002/diff/1/test/js-perf-test/Templates/...
test/js-perf-test/Templates/Templates.json:6: "run_count": 5,
Try increasing this if the test is really fast

https://codereview.chromium.org/769113002/diff/1/test/js-perf-test/Templates/...
File test/js-perf-test/Templates/harmony-templates.js (right):

https://codereview.chromium.org/769113002/diff/1/test/js-perf-test/Templates/...
test/js-perf-test/Templates/harmony-templates.js:44: var expected = '' + SUBJECT
+ ' gives ' + TARGET + ' an ' + OBJECT + ' \\ud83c\\udf4f.';
long line

[caitp (gmail), 2014-12-01 20:11:22]

https://codereview.chromium.org/769113002/diff/1/test/js-perf-test/Templates/...
File test/js-perf-test/Templates/Templates.json (right):

https://codereview.chromium.org/769113002/diff/1/test/js-perf-test/Templates/...
test/js-perf-test/Templates/Templates.json:6: "run_count": 5,
On 2014/12/01 20:04:07, arv wrote:
> Try increasing this if the test is really fast

I don't think that solves the problem --- each step takes 1000ms to get through
(we elapse like 30,000+ runs in each step) but the reporting for those
iterations never happens. It looked to me like if it doesn't get through an
expected iteration count, it will report results, but if it does, it won't. I
feel like it's a bug in the benchmark itself, but I'm not positive. (I've tried
this with 100 steps instead of 5, same issue --- no results output).

[caitp (gmail), 2014-12-11 16:27:54]

On 2014/12/01 20:11:22, caitp wrote:
>
https://codereview.chromium.org/769113002/diff/1/test/js-perf-test/Templates/...
> File test/js-perf-test/Templates/Templates.json (right):
> 
>
https://codereview.chromium.org/769113002/diff/1/test/js-perf-test/Templates/...
> test/js-perf-test/Templates/Templates.json:6: "run_count": 5,
> On 2014/12/01 20:04:07, arv wrote:
> > Try increasing this if the test is really fast
> 
> I don't think that solves the problem --- each step takes 1000ms to get
through
> (we elapse like 30,000+ runs in each step) but the reporting for those
> iterations never happens. It looked to me like if it doesn't get through an
> expected iteration count, it will report results, but if it does, it won't. I
> feel like it's a bug in the benchmark itself, but I'm not positive. (I've
tried
> this with 100 steps instead of 5, same issue --- no results output).

So the materialized-literals-in-preparser thing fixed the segfault, and it's
working well now.

It would be good to measure the impact of hash collisions, and maybe add some
more complicated tests, but the simple cases seem to score pretty highly.

Any suggestions for better benchmarks?

[Dmitry Lomov (no reviews), 2014-12-11 19:24:39]

cc V8 per sheriffs.

[Dmitry Lomov (no reviews), 2014-12-11 19:24:57]

On 2014/12/11 19:24:39, Dmitry Lomov (chromium) wrote:
> cc V8 per sheriffs.
*perf* sheriffs

[Dmitry Lomov (no reviews), 2014-12-11 19:27:10]

Caitlin, could you rebase this patch?
We have recently moved all configurations to a common JSTests.json file from
individual files. Thanks!

[caitp (gmail), 2014-12-11 19:31:35]

On 2014/12/11 19:27:10, Dmitry Lomov (chromium) wrote:
> Caitlin, could you rebase this patch?
> We have recently moved all configurations to a common JSTests.json file from
> individual files. Thanks!

Yep just a sec

[Dmitry Lomov (no reviews), 2014-12-11 19:56:03]

lgtm

[arv (Not doing code reviews), 2014-12-11 21:05:47]

Can you add a large non tagged template too? The reason is that I'm curious to
see if skipping collecting the raw characters would make a noticeable
difference.

[caitp (gmail), 2014-12-11 23:15:29]

On 2014/12/11 21:05:47, arv wrote:
> Can you add a large non tagged template too? The reason is that I'm curious to
> see if skipping collecting the raw characters would make a noticeable
> difference.

Added one --- PS I don't think the pattern of verifying results in teardown
actually has any effect, base.js looks like it only has an impact if teardown
throws --- I copied that pattern from other perf tests which did it, but it
doesn't look meaningful

[Benedikt Meurer, 2014-12-12 04:46:57]

bmeurer@chromium.org changed reviewers:
+ bmeurer@chromium.org

[Benedikt Meurer, 2014-12-12 04:46:57]

The CQ bit was checked by bmeurer@chromium.org

[Benedikt Meurer, 2014-12-12 04:46:58]

LGTM.

[commit-bot: I haz the power, 2014-12-12 04:47:25]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/769113002/80001

[commit-bot: I haz the power, 2014-12-12 05:14:26]

Committed patchset #5 (id:80001)