AdamW

I've built a kind of Vj-app using JItter and Lua that I'm looking to optimize. The patches and Lua-scripts are large and messy right now so I can't really post the whole thing here yet. What I've tried to do is a setup that is inspired by Jamie Lidells live performances (at least the one I saw at Sonar in 2002). Basically it's a videosampler with six buffers of 100 frames each. The buffers can be live with four seconds of offset or a recorded loop. The buffers are then accessed by either one of three layers and mixed in a global layer.

I've put together a pseudo patch with message boxes to illustrate the flow (in real life this is done in Lua where I got better performance).

I know it' hard to point at specifics when you can't see the real patch, but I'm wondering if there's any general design flaws that is keeping my framerate down. Right now fps is around 16 with no heavy fx going on (like 6x gaussian blur). And my hope is to get up to 20-25. The videobuffers have nothing to do with the framerate.

My hardware is a MacBook Pro, latest gen, GeForce 8600M GT, 512 MB VRAM.

Thanks in advance, and big thanks to Wes for jit.gl.lua and Vade for v001 and this list for all the help in the past!

I've built a kind of Vj-app using JItter and Lua that I'm looking to optimize. The patches and Lua-scripts are large and messy right now so I can't really post the whole thing here yet. What I've tried to do is a setup that is inspired by Jamie Lidells live performances (at least the one I saw at Sonar in 2002). Basically it's a videosampler with six buffers of 100 frames each. The buffers can be live with four seconds of offset or a recorded loop. The buffers are then accessed by either one of three layers and mixed in a  global layer.
I've put together a pseudo patch with message boxes to illustrate the flow (in real life this is done in Lua where I got better performance).
I know it' hard to point at specifics when you can't see the real patch, but I'm wondering if there's any general design flaws that is keeping my framerate down. Right now fps is around 16 with no heavy fx going on (like 6x gaussian blur). And my hope is to get up to 20-25. The videobuffers have nothing to do with the framerate.
My hardware is a MacBook Pro, latest gen, GeForce 8600M GT, 512 MB VRAM.

slow-vj-app

Oh, and everything is in 768 by 432 pixels.

How are the buffers stored? Are the matrices? Textures? UYVY

compressed? Are you only using one frame from the buffers or a range

On Sun, Oct 12, 2008 at 1:21 PM, Adam Wittsell wrote:

> Oh, and everything is in 768 by 432 pixels.

How are the buffers stored?  Are the matrices?  Textures? UYVY
compressed?  Are you only using one frame from the buffers or a range
of frames at a given draw call?

On Sun, Oct 12, 2008 at 1:21 PM, Adam Wittsell  wrote:
>
> Oh, and everything is in 768 by 432 pixels.
>
> //Adam
> --
> Adam Wittsell
> http://adamwittsell.com
>


The buffers are 4char matrixes. At any given draw call the videoin matrix is copied to six buffer matrices. Then each layer copies one matrix to it's texture in the same draw call.

If you use uyvy, you will halve your memory usage. Are you doing this?

On Sun, Oct 12, 2008 at 1:51 PM, Adam Wittsell wrote:

> The buffers are 4char matrixes. At any given draw call the videoin matrix is copied to six buffer matrices. Then each layer copies one matrix to it's texture in the same draw call.

If you use uyvy, you will halve your memory usage.  Are you doing this?
wes

On Sun, Oct 12, 2008 at 1:51 PM, Adam Wittsell  wrote:
>
> The buffers are 4char matrixes. At any given draw call the videoin matrix is copied to six buffer matrices. Then each layer copies one matrix to it's texture in the same draw call.
>
> //Adam
> --
> Adam Wittsell
> http://adamwittsell.com
>


I tried to change my jit.qt.grab to uyvy and put in a cc.uyvy2rgba.lite.jxs to do the conversion with a huge framerate drop as the result 6-8 fps. Since I'm more intereted in framerate then memory right now it doesn't quite fit, but then I'm sure I'm doing something wrong. 

I'm a little lost in yuv-land and don't understand why my already created 600 4 char matrices suddenly take up half the memory. How does it work?

I tried to change my jit.qt.grab to uyvy and put in a cc.uyvy2rgba.lite.jxs to do the conversion with a huge framerate drop as the result 6-8 fps. Since I'm more intereted in framerate then memory right now it doesn't quite fit, but then I'm sure I'm doing something wrong. 
I'm a little lost in yuv-land and don't understand why my already created 600 4 char matrices suddenly take up half the memory. How does it work?

The idea is to compress the chroma information since luma dominates

visual perception. the 'y' in uyvy is luma and uv are chroma. Look

at the related help objects to get more of an idea about this. For

your patch, the trial and error method would be to pare things down in

some regimented fashion in order to see what's hitting you.

Alternatively, you could profile it with things like shark or OpenGL

The idea is to compress the chroma information since luma dominates
visual perception.  the 'y' in uyvy is luma and uv are chroma.  Look
at the related help objects to get more of an idea about this.  For
your patch, the trial and error method would be to pare things down in
some regimented fashion in order to see what's hitting you.
Alternatively, you could profile it with things like shark or OpenGL
Debugger.

Slow VJ-app