Transparency slows game big time!

Question

When I make any object and set it to have transparency:
AlphaComposite myAlpha = AlphaComposite.getInstance(AlphaComposite.SRC_OVER, 0.5f);
g.setComposite(myAlpha);
or
Color newColor = new Color(255,0,0,100);
g.setColor(newColor)
then...
g.drawXXX( ... );
the game slows down hugely! I've noticed it doesnt happen for drawn objects with small dimensions. However, my game needs to have layered menus on top of the game animation which can be seen through it.
Help please....

Answer 1

It is activatable via flags though, as is the Java2D opengl pipeline.
if it doesnt get better, you might want to perform any transparency/translucency settings in an outside program (paint/photo applications),
so that your java application needs not worry about setting these values during runtimeThis will not speed upi your app. at all, the reason full alpha is slow, is because it requires read-back of pixels, to be able to perform the composition.
If you are rendering onto an unaccelerated surface (i.e. a BufferedImage backBuffer), the operation will be performed entirely in software.
1) the source pixel is read from main memory, (using up memory bandwidth)
2) the destination pixel is read from main memory (using up more memory bandwidth)
3) the composition operation performed by the CPU, (using CPU time)
4) finally the new value is written back to the destination image in main memory. (using yet more memory bandwidth)
This process is fairly slow, because it uses up CPU time, and chokes your memory bandwidth.
However, if you are rendering full alpha images onto an accelerated surface (such as a VolatileImage, or a BufferStrategy backBuffer), the overall speed will be ALOT SLOWER than performing everything in software!!
This is because the operation is still performed by the CPU, not the GPU.
So,
1) the source pixel must be read from the source image - this will either be a read of main memory (in the case of a BufferedImage), or it will be a VERY slow read-back from vram (in the case of a VolatileImage, or a BufferStrategy backbuffer)
2) the destination pixel must be read-back from the accelerated surface onto which you want to render - Either a BufferStrategy backBuffer, or a VolatileImage. As mentioned earlier, read-back from VRAM is a VERY slow operation.
3) The composition operation is performed by the CPU.
4) the new value is written back to the VRAM destination image. (this uses bus bandwidth)
So, to get faster fullAlpha rendering, you have 2 options.
1) Set the "sun.java2d.translaccel" flag to true, to enable fullAlpha acceleration.
This will hardware accelerate the operation, reducing the cost of the operation to very little.
2) do not use a vram surface for the source or destination images.
(basically do not use VolatileImage or BufferStrategy if you intend on using fullAlpha).
This will allow everything to be performed in software, as efficiently as it can be. The speed won't be impressive, and it will hog your memory bandwidth - but its still ALOT better than alpha compositing onto accelerated surfaces in software.