REGRESSION(r116210 - r131528): Hindi line-breaking on TOT is 5% slower than Safari 6.0.1 Obviously the real regression range is much smaller. r116210 is when safari-536-branch (Safari 6) was made from trunk. The microbenchmark in question is attached to bug 100024. I wrote the microbenchmark to make sure that the 8-bit TextRun optimization (bug 99393) wouldn't be a slowdown for complex text for the harfbuzz ports. It's also possible that the microbenchmark a. doesn't test what I want it to. or b. isn't interesting and that this isn't a regression we care about.
On my rMBP: r131528: avg 95.18194456386033 runs/s median 95.1814733027575 runs/s stdev 0.22485624887911804 runs/s min 94.7867298578199 runs/s max 95.4653937947494 runs/s Safari 6.0.1 (8536.26.14): avg 104.41866244835714 runs/s median 104.43864229765013 runs/s stdev 0.22674389758002172 runs/s min 104.03120936280884 runs/s max 104.84927916120577 runs/s
<rdar://problem/12551884>