I do get a 50 times slower performance if I do a CopyPixel with a pixelmapping (glPixelMap) enabled or a scale in glPixelTransfer enabled compared to a “clean” CopyPixel from-to the backbuffer. No copying of data to memory is done.
Is this correct ? Using NVidia driver 29.42
Can I do it faster. Is it the HW transformation that makes it slow. Is it faster/slower on other HW ?
Sounds correct. Any manipulation like this probably requires a readback to system memory then CPU work to apply the operation (like a resize) then a copy back to the graphics card instead of a a fast oncard blit without it. So this functionality should be carefully avoided if you need it to be fast hopefully the next generation of hardware support this kind of thing in hardware.