anttir wrote on 2025-08-26, 16:51:

Hello! First time poster here. Just wanted to drop in to say that MartyPC is AWESOME!

Also, I really hope that someday MartyPC will support 286 and 386. You see, I've been doing a little assembly programming lately, and I've found that Dosbox is not accurate at all. For example, I made two versions of a the same smooth scrolling vga game "engine". On DosBox they were equally fast, but on real hardware the other one was waaay slower.

Thanks for your kind words. I have started a 286 branch, but work is very preliminary. I believe cycle-accurate emulation of the 286 is possible and it intrigues me to attempt it. I'll leave the debates about the usefulness of that to others.
If nothing else, being able to watch the progress of the prefetch and instruction queues on the 286 would be interesting.

I've been a bit distracted with the 386 lately and working on making emulator CPU tests for the 386. I hope to get back to MartyPC by the end of the year.

286 already gets complicated there. Things like various recursive descriptor loads (for example IDT->GDT/LDT recursive descriptors). One the 386 it gets even more crazier, as each access includes optional paging lookups as well (multi-level memory access trees and TLBs that cache them all).
And there's the different timings of the local caches (descriptor cache references) as well.
All built on top of the normal segmentation system (and yes, all segment descriptors are always active, even in real mode).

In my emulator's case, I simply make them all instant as it gets way too complicated to cycle-accurately implement those all (due to counting and stepping becoming very complicated due to various recursiveness).
So mine is effectively only 'cycle-accurate' in real mode (using officially documented timings however).