SetVector and GPU render mode

Question

Hello,
If i use the setVector method on a BitmapData in GPU render mode on a mobile device, it works very slow.
This changes if the render mode is CPU. The method is fast enough.
The problem lies in the fact that if I use CPU render mode the app i;m developing is getting very slow, while on GPU it works great.
setPixel32 is even slower,it even won't work with lock/unlock.
Where the problem lies? How can I bipass this since setPixel32 or setVector is a must have when working with bitmaps.
UPDATE:
I've tested alot, and it seems the problem lies not with setVector method but with copying transparent pixels
You can find the code here since i've asked the question o stackexchange too:  http://stackoverflow.com/questions/21827049/gpu-vs-cpu-render-mode-adobe-air

Answer 1

Hi Lucian,
Unfortunately, I am unaware of any tutorials on the net I could recommend, though I am sure there are a few -- not on the subject of setting up render targets, specifically -- most likely-- but on setting up Stage3D + shaders.
One source I would recommend, though, if this is your first time setting up a Stage3D render pipeline, besides the code examples in the Stage3D Adobe reference, is this great book:
http://www.amazon.com/Adobe-Stage3D-Molehill-Programming-Beginners/dp/1849691681/ref=sr_1_ sc_1?ie=UTF8&qid=1392715725
It's a quick read.  Nicely presented and explained.  I think in 3-5 days you could be up and running.
I would also browse this site.  You might get some ideas:
http://www.flashandmath.com
Regarding using render targets to identify areas that have changed and need further processing:
This is a fairly common practice when interactivity is key, and works well in some use-cases.
For example, in Photoshop, there is a filter called "Liquify", with which you can stretch and compress an image in real-time.  There are 2 ways to do this effect, live: one is to map the source -- undistorted-- image onto a grid of polygons, and start pushing and pulling the vertices.  The other method -- same principle, but finer -- is to paint the compression field into a render target ( a texture ), temporarily, while you work in real time by 'painting' the effect ( note: the displacement vectors are based off of the gradient of the compression field ).  Conceivably, you could also paint the vectors directly, as if painting a normal map.  Once the user is happy with the look, then the user clicks 'apply', and the render target is used one last time to bake the displacements into the source image.
Other example:  I once worked on a game where the player could use a laser beam to scar all the buildings of a 3D city.  The way this was done was that each building had its own render target ( usually a 256x256 ), into which the user could 'draw' ( even though the player never saw this render target, which was just a grayscale mask ).  The render target was then fed into the building's exterior shader, which would use it to identify which areas to show the destruction effect into.
Overall, the idea is to use a render target ( usually smaller that the source texture it will end up modifying ) to paint either a mask, or vectors ( like a normal map ), which is then used along with the source texture in the final shader to create the look you want.
Although I haven't had to use render targets for this particular situation, lately, I use render targets frequently, otherwise, to draw my assets into and apply shaders, and then draw the render target into the backbuffer, at the very end. 
The reasons for this is explained here ( along with other tips ):
http://forums.adobe.com/thread/1399727?tstart=30