[HarfBuzz] Test please!

[Behdad Esfahbod]

On 15-11-06 03:35 AM, Simon Cozens wrote:
> On 05/11/2015 14:57, Behdad Esfahbod wrote:
>> Simon, this should speed up your use case more than anyone else's.  Can you
>> please share numbers?
> 
> With 1.0.6, test.sil (Latin text) took 6.1s; a stripped-down version of
> Khaled's Arefruqaa test file (30,000 lines of Arabic) took 59.1s
> 
> With HEAD, test.sil took 5.9s, Khaled's case took 60s. So no discernible
> advantage.
> 
> Even removing the "obeylines" from Khaled's case and shipping the whole
> lot of Arabic to the shaper at once did not show much of a speed
> advantage either...

Thanks for testing.  I was optimizing for the simplest fonts, ie those with no
or little OpenType lookups, as well as ASCII text.  As soon as the font has
multiple lookups, the lookup application time dominates the shaping.

Anyway, good to know it didn't make any dent in your tests :)).  I'm guiding
optimization work using Roboto, lots of ASCII text, and "perf stat" command
(which I recently discovered!).  I'll optimize the hb-ot functions a bit and
leave it there for now.


-- 
behdad
http://behdad.org/