Random Chitchat 2012-2016

NiLSPACE · 02-02-2015, 03:10 AM

If it was allowed then I'd do that, but it has to be simular to your fancyaddress website.

FakeTruth · 02-02-2015, 08:37 AM

What! But that hardly uses PHP...

NiLSPACE · 02-02-2015, 08:34 PM

Looks like Rasberry Pi is going to have an upgrade Smile

http://www.raspberrypi.org/raspberry-pi-2-on-sale/

tonibm19 · 02-03-2015, 02:25 AM

That's awesome! Last christmas I wanted to buy one, but I finally didn't. Maybe now its the moment.
Hopefully it runs MCServer and spigot smoothly.

worktycho · (This post was last modified: 02-03-2015, 03:05 AM by worktycho.)

@xoft. I've just tested a quick benchmark with clang 3.4 on my i5-2450M. If I remove the repeated value detection from CalcFloorFrac it speeds up generation by 4 - 6 times. It looks like clang can vectorise the loop when its simple.

gcc 4.8 is giving similar improvements.

tigerw · 02-03-2015, 05:37 AM

yay~~

xoft · 02-03-2015, 06:50 PM

It depends on what your benchmark was doing. Can you post the code? How wide was the range, in relation to the step used?

worktycho · 02-03-2015, 08:38 PM

Its an adaptation of the NoiseTest benchmark that's been siting in the repository for ages. Its posted to the noiseTest branch.

worktycho · 02-03-2015, 09:38 PM

I've just run it under cache-grind and it reports a significant number of L1 misses in the code with repetitions and none in the code without. Looks like the a_Same array just pushes the working data over the size of the L1.

worktycho · 02-05-2015, 09:17 AM

I've just tried the benchmark in VC++ 2013. I'm seeing the same performance improvements dispite the fact that VC++ is not vectorising suggesting that the main issue is cache misses. Since that is an extremely machine sensitive issue could someone else test it on their machine. Its on the NoiseTest branch in the tests folder.