AvsManDan
30th May 2004 06:17 UTC
Want a faster AVS???
Everybody knows AVS has a serious speed problem, but many don't know how much faster they can make it without damging the quality of a preset. I'm not talking about lowering resolution, pixel doubling, or disabling bilinear; because that just sucks.
Back in Winamp 2 I discovered the "Clear every frame" checkbox in the main and all effect lists. many presets had this on but didn't need it. Disabling the ones using DM that redraw the whole screen anyway gave a 5 fps boost.
Just today I found how much "Wait for retrace in the settings-display took up. Disabling this in one preset gave me a 20 fps boost!
So, anyone have any AVS/SSC/DM tips to make rendering faster?
Deamon
30th May 2004 10:11 UTC
GREAT! Finally a topic worth reading with useful tips :D. Let me add some speedup tips to your small list:
- Lowering the grid size in DM speeds things up.
- Lowering the n in the init of a superscope speeds things up.
- Don't use any DM/SSC at all, they usually slow things down anyways.
- Disabling effect lists speeds things up as well.
- Buying a faster processor speeds things up.
- drinking a lot of alcohol makes avs look a lot smoother.
- Sit closer to the screen, the bigger the avs window looks, the better.
damn... :hang:
Jaak
30th May 2004 11:53 UTC
dont use divisions at all, only time i use division is when i need to do:
z=1/rz;
x=rx*z;y=ry*z*asp;
there is no way getting around with this.
And dont use pow(a,b) it's slow as hell
if you have:
loop(n,
assign(a,b)+
assign(c,d)+
...
);
use this instead:
loop(n,
exec2(
exec3(
assign(a,b),
assign(c,d),
assign(e,f)
),
exec3(
...,
)
)
)
use ONLY powers of 2 in your convo filters.
use ifs a lot when possible, if you have:
if(a,b,c) and if a is true, only b is executed, and if a is false only c is executed and b statement is ignored totaly
TomyLobo
30th May 2004 12:07 UTC
Originally posted by Jaak
loop(n,
exec2(
exec3(
assign(a,b),
assign(c,d),
assign(e,f)
),
exec3(
...,
)
)
)
[advertisement]
to make such fine code check out my
AVS Code Translator
It is also useful for complex if-statements that Jaak mentioned.
[/advertisement]
AvsManDan
30th May 2004 16:44 UTC
Thanks Jaak,I never thought of those.
A question. Convultion is faster using certain numbers. That must mean it doesn't use lookup tables for its values. Would a new verion using this make it faster?
UnConeD
30th May 2004 18:25 UTC
Convolution compiles specific (MMX) assembly code for each kernel. There's more info about in on the forums somewhere (search for posts by cfp).
Please do enlighten us how a lookup table would help speed convolution up?
AvsManDan
31st May 2004 20:52 UTC
I looked up what you said and figured it wouldn't make it faster. Earlier I had a nice half page of what I ment but it was erased by the back button and stuff. New to the forum's great ways.:D
TomyLobo
31st May 2004 20:58 UTC
get Firefox
it doesnt delete the text in the input text fields if you click the "back" button :)
if you click "forward" after that, it is still there
(and it is HTML 4.0 compatible, safe against most dialers, trojans and stuff and it has an integrated popup blocker that even defeats the most pesky popup methods :))
to the topic:
Trans/Bump is also very slow (especially if used multiple times)
if you really need a bumpy texture try the following convolution filter: (i wrote only the center 3x3 numbers here)
-3 0 0
0 1 0
0 0 3
jheriko
6th June 2004 04:57 UTC
Originally posted by Deamon
- Don't use any DM/SSC at all, they usually slow things down anyways.
- Disabling effect lists speeds things up as well.
Surely you want to delete an effect list.. has the same effect as disabling but is probably a tiny tiny bit faster.
DM/SSC only slow things down if they are complex... simple DMs are lightning fast and even complex SSCs can be run on their own at blinding speeds.
I find that most of the 'trans' effects suck huge amounts of fps up compared to the simpler 'render' effects like superscopes, moving particle, simple etc.
Increasing line width slows stuff down too.
[Ishan]
6th June 2004 09:22 UTC
Also using Blend filter and Bilinear filtering at the same time slows things down drastically.....