[OpenJDK 2D-Dev] RFR 8144938: Handle properly coordinate overflow in Marlin Renderer

[Laurent Bourgès]

Jim,

Yes these results were obtained with MapBench and a specific profile:

Loading profile file:
/home/marlin/mapbench/bin/../profiles/shared_nu-tx.properties
Profile properties (merged with defaults):
  doRotate=true
  doScale=true
  rotateAngle=3.123456789 (deg)
  scaleX=1.156
  scaleY=0.945

Moreover, I checked in the code (system.out.println) that this benchmark
really tests the modified pipeline (non-uniform case).

Cheers,
Laurent

2016-03-15 0:35 GMT+01:00 Jim Graham <james.graham at oracle.com>:

> Was this on a uniform scale?  Did you do a test with a non-uniform scale?
>
>                 ...jim
>
> On 3/11/2016 3:10 PM, Laurent Bourgès wrote:
>
>> Jim,
>>
>> Here are MapBench results comparing Marlin 0.7.3.2 (before) vs patched
>> 0.7.3.3 (latest patch) on JDK 1.8_72:
>>
>> Synthetic results:
>>
>> Tests    36    12    12    12
>> Threads    4    1    2    4
>>
>> Before = Marlin 0.7.3.2 <http://0.7.3.2>:
>> *Med    168.535    165.403    166.940    173.263 *
>> *Pct95    170.256    166.011    167.495    177.260
>> FPS    43.200    43.404    43.203    42.993 *
>>
>> Patched = Marlin 0.7.3.3 <http://0.7.3.3>:
>> *Med    169.481    166.344    168.096    174.004 *
>> *Pct95    170.295    166.800    168.751    175.335
>> FPS    42.194    42.340    42.211    42.031
>>
>> *
>> The difference is very minor ~ 1/2ms in global average (36 or 12 tests)
>> ie within error margin !
>> Only FPS (based on median timings) seem indicating that Marlin 0.7.3.2
>> is 'very slightly' faster.*
>> *
>> *
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.openjdk.java.net/pipermail/2d-dev/attachments/20160315/9f7ec660/attachment.html>