Shrunk HTMLTokenizer.o by 28.2%

Shrunk HTMLTokenizer.o by 28.2%

This CL moves the cold code in InputStreamPreprocessor::peek out-of-line, which
shrunks HTMLTokenizer.o on Linux by 28.2%. The CL also makes the
html-parser.html performance test faster by 1.3% (average of 15 runs).

Raw data:

== Before ==

$ ls -larth ../../out/Release/obj/third_party/WebKit/Source/core/html/parser/webcore_html.HTMLTokenizer.o
-rw-r----- 1 abarth eng 142K May 13 11:54 ../../out/Release/obj/third_party/WebKit/Source/core/html/parser/webcore_html.HTMLTokenizer.o

Ignoring warm-up run (1837.2089999029413 ms)
1722.1780000254512 ms
1716.4099999936298 ms
1779.4659999199212 ms
1722.438000026159 ms
1706.023000064306 ms

Ignoring warm-up run (1813.4049999061972 ms)
1694.472000002861 ms
1694.9190000304952 ms
1693.461999995634 ms
1700.8790000109002 ms
1693.1179999373853 ms

Ignoring warm-up run (1838.1410000147298 ms)
1722.632999997586 ms
1716.6979999747127 ms
1720.1839999761432 ms
1714.3919999944046 ms
1701.933000003919 ms

== After ==

$ ls -larth ../../out/Release/obj/third_party/WebKit/Source/core/html/parser/webcore_html.HTMLTokenizer.o
-rw-r----- 1 abarth eng 102K May 13 11:47 ../../out/Release/obj/third_party/WebKit/Source/core/html/parser/webcore_html.HTMLTokenizer.o

Ignoring warm-up run (1822.2709998954087 ms)
1684.012999990955 ms
1687.383999931626 ms
1680.9989999746904 ms
1682.835999992676 ms
1679.3179999804124 ms

Ignoring warm-up run (1792.570999939926 ms)
1678.2200000016019 ms
1743.3040000032634 ms
1682.9719999805093 ms
1681.1179999494925 ms
1676.7090000212193 ms

Ignoring warm-up run (1845.1740000164136 ms)
1704.390999977477 ms
1690.7409999985248 ms
1690.9110000124201 ms
1693.4739999705926 ms
1696.5929999714717 ms

Committed: https://src.chromium.org/viewvc/blink?view=rev&revision=150256

[pdr., 2013-05-13 19:04:23]

On 2013/05/13 19:01:25, abarth wrote:

LGTM!!

[commit-bot: I haz the power, 2013-05-13 19:10:57]

No reviewers yet.

[tonyg, 2013-05-13 19:14:27]

lgtm2

Just curious, how did you discover that this method was 28% of the size? Just
intuition and experimentation?

[commit-bot: I haz the power, 2013-05-13 19:18:54]

CQ is trying da patch. Follow status at
https://chromium-status.appspot.com/cq/abarth@chromium.org/14885016/2001

[abarth-chromium, 2013-05-13 19:19:27]

On 2013/05/13 19:14:27, tonyg wrote:
> lgtm2
> 
> Just curious, how did you discover that this method was 28% of the size? Just
> intuition and experimentation?

HTMLTokenizer.o shows up as one of the biggest translation units in Blink on
binary size profiles.  I just started deleting code until the size when down. 
Then I binary searched the deletion to find what exactly was making the size
number move.

[commit-bot: I haz the power, 2013-05-13 20:28:03]

Change committed as 150256