[MemUsage][BinarySize] More perfect hashes for CSS strings.

[MemUsage][BinarySize] More perfect hashes for CSS strings.

gperf generates some "perfect" hashes for CSS properties and CSS
values. Those are not very compact but can be more compact if gperf
is asked to try harder. This will cost 1 s build time and reduce binary
size by about 12 KB.

The flags changed are:

'-m 50' - Try 50 times more initial values to the search algorithm. This
will make the call 50 times slower but for the current input size that is
no problem.

Removal of ‘-s 2’ - -s 2 told gperf to generate a twice as large data table
to make it easier (faster) to find a solution.

Also '-D' moved around but was there both before and after so no addition
or removal.

In the longer run we should move away from gperf to a better algorithm (they
exist at the research level but not as easily installed programs)

BUG=351774
Committed: https://src.chromium.org/viewvc/blink?view=rev&revision=169111

[Daniel Bratell, 2014-03-12 17:56:42]

scottmg, I actually have no idea who to ask to change these python scripts, but
you were the last to touch it so maybe you can take a look, or redirect me to
someone else?

[scottmg, 2014-03-12 18:10:01]

Do we know if the duplicate hashes will affect runtime performance at all?

(+eseidel, I don't own any blink code)

[eseidel, 2014-03-12 18:21:28]

Personally I'm interested in us moving off of these hashes eventually, and using
AtomicString for these values like HTMLNames does, etc.

That said, I'm all for saving binary size.  12k may not be worth it though. :)

The perf bots should tell us if this changes anything.  I would try running the
html5-full-render locally first though:

run-perf-tests Parser/html5-full-render.html

I believe it's also possible to run those with telementry these days as well:
http://www.chromium.org/developers/telemetry

see./tools/perf/run_benchmark list

[eseidel, 2014-03-12 22:00:34]

lgtm

I'm happy to see us experiment here.

Please update your description to detail the options you're adding/removing
explicitly so that others don't have to go read the man page. :)

[Daniel Bratell, 2014-03-12 22:53:08]

On 2014/03/12 22:00:34, eseidel wrote:
> lgtm
> 
> I'm happy to see us experiment here.
> 
> Please update your description to detail the options you're adding/removing
> explicitly so that others don't have to go read the man page. :)

Thanks, I will do so.

gperf is actually rather bad for this. It happens to be easily available,
otherwise there are research level programs that can generate much better static
hash tables faster. I've experimented with one named "mph" (minimal perfect
hash) where the hash function will have slightly more instructions than gperf
but with more compact tables I think the end result might be faster anyway due
to more compact tables. I had to fix some (trivial) bugs in it to make it work
for 64-bit computers but it might be an option in the long run.

[Daniel Bratell, 2014-03-12 22:59:53]

The CQ bit was checked by bratell@opera.com

[commit-bot: I haz the power, 2014-03-12 23:00:03]

CQ is trying da patch. Follow status at
https://chromium-status.appspot.com/cq/bratell@opera.com/196413006/1

[commit-bot: I haz the power, 2014-03-12 23:38:36]

The CQ bit was unchecked by commit-bot@chromium.org

[commit-bot: I haz the power, 2014-03-12 23:38:37]

Try jobs failed on following builders:
  tryserver.chromium on mac_blink_rel

[Daniel Bratell, 2014-03-13 07:14:00]

The CQ bit was checked by bratell@opera.com

[commit-bot: I haz the power, 2014-03-13 07:14:08]

CQ is trying da patch. Follow status at
https://chromium-status.appspot.com/cq/bratell@opera.com/196413006/1

[commit-bot: I haz the power, 2014-03-13 08:23:21]

The CQ bit was unchecked by commit-bot@chromium.org

[commit-bot: I haz the power, 2014-03-13 08:23:21]

Try jobs failed on following builders:
  tryserver.chromium on linux_blink

[Daniel Bratell, 2014-03-13 08:39:19]

The CQ bit was checked by bratell@opera.com

[commit-bot: I haz the power, 2014-03-13 08:39:26]

CQ is trying da patch. Follow status at
https://chromium-status.appspot.com/cq/bratell@opera.com/196413006/1

[commit-bot: I haz the power, 2014-03-13 10:06:49]

Change committed as 169111