KNOWN ISSUES

Some color mapping errors are happening with textured shapes in 16 bit
mode.  Use double-buffering and "blit_flip" to overcome this.  In fact,
there really is not a good reason to not use double-buffering for 16
bit mode anyway.

blit_transform - flip_horizontal - has known issues.  Try instead rotating
180 degrees and then flip_vertical.

Some framebuffer drivers are tearfully slow.  I have noticed that the Nouveau
framebuffer driver for NVidia is Molasses slow.  It's amazing how slow it is.
Other framebuffer drivers are amazingly fast.  It's clear the Nouveau
framebuffer driver is in dire need of optimization, or at the very least
hardware accelerated buffer copy, like CPU DMA.  I'll be looking into finding
ways to overcome this issue.  Maybe turning off the mirroring would help.

After more testing, it seems non-bulk read or write access is simply very
slow.  Odd really.  Operations in normal mode, like blit_write are very
fast, but any other mode like OR or Alpha, are molasses slow.  Normal mode
only writes as a large block of data, and takes no care about what it replaced.
However, other modes require bytes, words, or long words to be read in,
processed, then written back out.  That is what is not optimized in the
Nouveau driver.  Other drivers do this quickly, even the Intel driver.  I am
assuming the other drivers use a DMA routine to transfer the memory address
framebuffer to the real framebuffer, or map the actual framebuffer to real
memory, although I suspect most do it via DMA transfers.
