All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=759606

(For debugging:) Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?sid=67c4a1ba3fa5b2212569d7df7804e52e89c31370f21565f529d35c1da93c474f


Bot(s) for this bug's original alert(s):

chromium-rel-mac12-mini-8gb

Started bisect job https://chromeperf.appspot.com/buildbucket_job_status/8970006242755030544

=== Auto-CCing suspected CL author kojii@chromium.org ===

Hi kojii@chromium.org, the bisect results pointed to your CL, please take a look at the
results.


=== BISECT JOB RESULTS ===
Perf regression found with culprit

Suspected Commit
  Author : Koji Ishii
  Commit : 85b84a4d900effddac4b32e7774d23f51fd562e0
  Date   : Fri Aug 25 16:15:31 2017
  Subject: Fix mid-word-break to handle grapheme clusters

Bisect Details
  Configuration: mac_10_12_mini_8gb_perf_bisect
  Benchmark    : loading.desktop
  Metric       : cpuTimeToFirstMeaningfulPaint_avg/pcv1-warm/Pantip
  Change       : 8.72% | 919.9945 -> 1000.22033333

Revision             Result                  N
chromium@497381      919.995 +- 17.254       6      good
chromium@497422      898.597 +- 79.3852      6      good
chromium@497427      902.622 +- 54.5811      6      good
chromium@497428      307.752 +- 1010.88      6      good
chromium@497429      996.737 +- 61.2086      6      bad       <--
chromium@497430      986.147 +- 63.8192      6      bad
chromium@497432      993.448 +- 55.4447      6      bad
chromium@497442      988.999 +- 63.9991      6      bad
chromium@497462      1000.22 +- 12.9865      6      bad

To Run This Test
  src/tools/perf/run_benchmark -v --browser=release --output-format=chartjson --upload-results --pageset-repeat=1 --also-run-disabled-tests --story-filter=Pantip loading.desktop

More information on addressing performance regressions:
  http://g.co/ChromePerformanceRegressions

Debug information about this bisect:
  https://chromeperf.appspot.com/buildbucket_job_status/8970006242755030544


For feedback, file a bug with component Speed>Bisection

Discussed with ksakatomo@.

The CL changes mid-word-break to use grapheme cluster boundary rather than surrogate pair increment. The site <pantip.com> is a Thai site, and Thai script needs 'word-wrap: break-word' more often than other scripts due to its long text without spaces. Given most of Thai's grapheme cluster is more than 1 code point, we're reading/shaping more characters than before the fix.

8.7% is a bit more than expected, but given this is for the i18n correctness and trying to use optimized code paths don't apply to Thai, I think the fix is correct.