[B2-devel] Re: OS X

[Brian J. <bjj...@us...>]

--- Max Horn <ma...@qu...> wrote:
> At 10:46 Uhr +1000 12.10.2001, ni...@in... wrote:
> >...
> >>  BTW, Nigel, how is speed for you? On my G4/400, running Sys
> >>  7.5.3 in the emu, I can literally see it do redraws etc.
> >
> >	Same speed deficiencies. I haven't worked out if it is the
> >graphics environment (e.g. the barber pole, and having to have the
> >MacOS blit from an NSBitmap), or the emulation on a RISC
> >processor. I do remember that it was also quite slow on my Sun
> >Sparcstation 5 at work.
>
> Don't forget that you also have double buffering on from the MacOS,
> which in our case doesn't really help - turning on retained mode for
> the window might help.

A few general BasiliskII things to try:
- Get direct addressing working, if it isn't alreay.  It takes a _lot_
  of gunk out of the emulated memory reference path, and will improve
  performance dramatically.
- Get VOSF (video on seg fault) working, if it isn't already.  This
  lets you be _much_ smarter about blitting:  instead of redrawing the
  entire emulated frame buffer, you only need to redraw the parts
  which have been written since the last update.  If MacOS X's mmap()
  and signal handlers aren't up to the task, I'm sure you could do it
  with a Mach external pager.  That's exactly the sort of trick
  external pagers are good at (if I remember all those Mach papers I
  had to read in school.)
- Make sure that CPU_CAN_ACCESS_UNALIGNED ends up defined when you
  compile (PPC can do unaligned memory refs., can't it?  See
  Unix/sysdeps.h.)  That greatly simplifies the emulated memory
  reference path.
- Check out the optimized video blitter loops from video_blit.{cpp,h}
  in the CVS version.  If your blit is more than a simple bcopy, they
  should help somewhat.

On my SGI Octane with dual 195 MHz R10000 CPUs, BII with X11 (XShm
transport), direct addressing, VOSF, and MIPS-optimized unaligned
reference code is about as fast as a real IIci.  Not stellar, but very
usable, even for old games.  SGI's compilers are probably better than
GCC, but a 400MHz G4 has a lot more raw horsepower than my box... you
should be able to get BII running well.

> >
> >	It is fast enough to be usable in a pinch, though.
>
> Well if you compare it to the same version of Basilisk (latest
> release snapshot) but compiled with an X11 front end, it is quite a
> bit slower - this tells me we have a lot to improve :)

If there's some way you can map the frame buffer directly, rather than
going through the OS, that would give you the best performance.  Since
you're on a real Mac, you probably wouldn't have trouble with the
frame buffer layout (waving my hands wildly at this point....)

You might also want to try and write an OpenGL front end.  On SGI
boxes at least, OpenGL is the most optimized path to the frame buffer.
(And then I could steal it and use it on IRIX....)

						Brian J. Johnson

--------------------------------------------------------------------

   "There is the greatest practical benefit in making a few failures
    early in life."
                                           -- Thomas Henry Huxley