2008/8/21 Wayne Meissner [wmei...@gmail.com]:
2008/8/21 Timothy Wall <twal...@dev.java.net>:
In addition, there's a Pointer.read() method that reads into an existing
int[] to avoid allocating a new one every time.
That would be the best way if you're going to be accessing the pixels
from java (Pointer and ByteBuffer ops will be slower than array
element accesses).
Its not the copy that is killing performance, it's the garbage collector.
Thanks.
So now I do the following to create the memory region.
ByteBuffer bytes = ByteBuffer.allocateDirect(vidWidth
*
vidHeight * vidDepth);
IntBuffer data = bytes.asIntBuffer();
Pointer ptr =
Native.getDirectBufferPointer(data);
BufferedImage renderImage = (BufferedImage) createImage(width, height);
int [] vidBuffer = new int[vidWidth * vidHeight];
The ptr gets passed to the native library and on return I now do this in the
paintComponent function:
if (ptr != null) {
ptr.read(0, vidBuffer, 0, vidWidth * vidHeight);
renderImage.setRGB(0, 0, vidWidth, vidHeight, vidBuffer, 0,
vidWidth);
g2d.drawImage(renderImage, 0, 0, width, height, null);
}
I think the ptr.read() change is giving me the biggest boost. I can now
render SD video onto a 1280x720 frame (cpu @ ~65% on a 3GHz Pentium 4). HD
video is still taking too much cpu in the decode to keep the frame rate up.
I am still trying to figure out how to use the directly allocated buffer as
the backing store for a raster, but I can't see the path yet. This would
remove the need for the at least one of the copies.
12 messages in net.java.dev.jna.usersRE: [jna-users] Getting pointers from...
