Optimizing Closure :: pouët.net

Optimizing Closure

category: code [glöplog]

las: your ranting against me is unnecessary, stop that already :p you do it everywhere i appear, aint funny anymore! its just my 2cents/thinkings i put, and almost everytime i get a counter by you! wtf? i have the fucking right to have my own thoughts without you always having a different mindset! nah, go on, you have the right to express your mindset aswell, but why the fuck do you concentrate that hard on me?

added on the 2012-09-18 15:40:07 by ɧ4ɾɗվ.

maybe because he's right?

added on the 2012-09-18 16:26:30 by xTr1m

so you, too. whats wrong with my mindset then, plz?

added on the 2012-09-18 16:34:36 by ɧ4ɾɗվ.

simple: it's wrong. you're communicating false information.

added on the 2012-09-18 16:36:27 by Gargaj

you aswell? damn, i need to reconfigure then! 3 is more than 1 atleast!

added on the 2012-09-18 16:56:51 by ɧ4ɾɗվ.

concerning the info i spread in this thread, as said, just infos on glow, in my case hypnoGlow, my framerate stays the same (240fps vs 240fps in my testCase), if applied or not. so no falseInfo here atleast!

added on the 2012-09-18 16:59:51 by ɧ4ɾɗվ.

1. http://www.mvps.org/directx/articles/fps_versus_frame_time.htm
2. "it runs fast here" is NOT an argument.

added on the 2012-09-18 17:01:23 by Gargaj

Code:


"float4 zz(float2 v:texcoord):color"
"{"
	"float4 c=0.;"

	"for(int i=-128;i<=128;i++)"
	"{"
		"c+=tex2D(s,v+i*xy.xy)*(.13*(1./(abs(i)+2)));"
	"}"

"return c*xy.w;"
"}"

xy being fed with texture-sizes.

added on the 2012-09-18 17:03:02 by ɧ4ɾɗվ.

Quote:

this is almost for free

ALMOST! ;) i didnt say its for free completely! but here on my setup it doesnt count anything. so was before, the last 4 years, with 3 different computers. i am not talking about onboard-GPUs at all, as i was just reacting on mu6x Question! i havent answered to the whole Thread!

added on the 2012-09-18 17:06:33 by ɧ4ɾɗվ.

xy.w is just some multiplicator for brightness btw.

added on the 2012-09-18 17:09:55 by ɧ4ɾɗվ.

so how's your other testcase look like?

added on the 2012-09-18 17:10:53 by Gargaj

So time for some rough language - next escalation level is switching to german - be prepared! Dramatime.

You must be f**king kidding me. 256 samples per pixel and you have zero performance impact? Congratulations you've just found the holy grail of doing convolution on GPUs!

Sorry for being an asshole - but I'm certain you are doing it wrong.
Proof me wrong.

added on the 2012-09-18 17:14:24 by las

i dont double.test if its not for work! so maybe my testcase is wrong, but i tested with my 4k-solution like 200 times already, switching POST_FX (which is #defined and sorts my code) on and off all the time, to see if theres a diff! and the fps stay the same, all the time!

added on the 2012-09-18 17:15:06 by ɧ4ɾɗվ.

hardy: what res are you running - 1080p or 640x480? are you actually timing the gpu or are you just timing the cpu?

added on the 2012-09-18 17:15:58 by smash

its just a minor impact on rendering! as said: here the fps dont change at all! measuring with fraps btw.

added on the 2012-09-18 17:18:14 by ɧ4ɾɗվ.

Let me guess... ~60 fps? VSYNC.

added on the 2012-09-18 17:19:44 by las

timing is done by directX-API, yes i cheat, as everyone does! i set it to D3DPRESENT_INTERVAL_IMMEDIATE for DEBUG and D3DPRESENT_INTERVAL_DEFAULT for RELEASE-builds.

added on the 2012-09-18 17:20:17 by ɧ4ɾɗվ.

Quote: