16bit CGImage problems

Question

Hello,
I'm working on 2D support for our framework and we're having framerate "issues". The framerate won't go any higher then 30fps on a simple program and it drops fast to 20fps.
At the moment I have a 32bit CGImage on which I draw everything.
To get some perfomance improvements I want to create a 16bit CGImage but I get some errors. I use the following code to create the 16bit CGImage:
CGColorSpaceRef colorspace = CGColorSpaceCreateDeviceRGB();
CGDataProviderRef provider = CGDataProviderCreateWithData(NULL, buffer, yres * lpitch, NULL);
surf = CGImageCreate(xres, yres, 5, 16, lpitch, colorspace, kCGImageAlphaNoneSkipLast, provider, NULL, FALSE, kCGRenderingIntentDefault);
CGColorSpaceRelease(colorspace);
CGDataProviderRelease(provider);
The problems I have are:
- red and green are swapped
- I can only draw the image once to the screen. It's is not possible to draw the image again. It does draw the stuff on the image and it does call the drawRect method where it is drawn to the screen but it isn't updated.
As soon as I change it back to 32bits it works again.
Because OpenGL uses 16bits I assume the colors are internally converted back to 16bits. To avoid this I want to create a 16bits CGImage, hoping for a little performance increasement.

Answer 1

It sounds like you have several issues confused.
EXR allows many depths of data to be embedded in it - but Photoshop uses EXR for HDR data, which needs to be 32 bit floating point.
Also, the most common data depth in EXR is 16 bits/channel floating point -- which is not the same as the 16 bit/channel integer format in Photoshop, and has to be converted up to 32 bit/channel floating point to prevent data loss.
Yes, 32 bit/channel data is always gamma 1.0 in Photoshop (and should be elsewhere) and that changes the values when converting colorspaces for 8 and 16 bit/channel.
A gamma 1.0 profile can be set as a working space profile - but it just isn't a good idea in 8 or 16 bit/channel.