It will be slower for a little while yet....., but we hope that it will be faster in the end.
A thought about decoding instructions and memory reads/writes (where we take a big hit right now). I know that the PII series of chips (and beyond) support the use of 16K (16384) memory segments of (2^32)-1 addresses (which may, or may not, overlap). If we can use this feature of the chip (Last I knew Windows and LINUX didn't by default--I don't know about >4GB RAM size machines though) then we can do 2 things in parallel to speed up plex86: 1) hard lock some RAM space to plex86 using the local kernel (or find some way to intelligently handle swapping the memory space) and keep that memory exclusively for plex86 2) Use the (2^32)-1 size linear address space of one segment just for plex86 and the memory space that we tell it to use (and find some way to make it reliably fault outside of that allowed range). This will allow us to have more than just ring3 register-->register code to execute without interruption. Thus far, as far as I have been able to tell, we've been using the old method of having all memory reads/writes emulated. Eliminating this emulation may allow us to speed things up (once we isolate and fix the problems with transitioning to protected mode and back). Roy Souther wrote: > > -----BEGIN PGP SIGNED MESSAGE----- > Hash: SHA1 > > How fast is plex86 compaired to bochs? > - -- > Roy Souther <[EMAIL PROTECTED]> > www.SiliconTao.com > > I am Linux MacLeod of the clan MacLeod, and I can not die! > -----BEGIN PGP SIGNATURE----- > Version: GnuPG v1.0.6 (GNU/Linux) > Comment: For info see http://www.gnupg.org > > iEYEARECAAYFAjx+iZIACgkQCbnxcmEBt41SNwCg0uB9V0J0bkj6NYSeCsa9ud93 > PFgAnArOyFGCTHNoQP+abz3n9maqgLM5 > =6ZMb > -----END PGP SIGNATURE----- -- |^^^ | | |^^| |^^^ Drew Northup, N1XIM |^^| | |^^^ \ / /^^\ /^^~ |__ | | | | | |__| | |___ \/ |__| |__ | | | | | | www.plex86.org | | | /\ | | | \ ___| |__| |__| |___ web.syr.edu/~suoc/ | |___ |___ / \ \__/ \__/
