exxos's Atari, Amiga & retro forum

Finally managed to get some CT60-only issues sorted out so it should be in a state identical to how it works on 030 now.
Pro tip: don't put the MMU table in memory that has copy-back cache enabled :lol:

I've been wanting to avoid publishing a too broken build so it's mostly been a few private ones so far, mainly to iron out issues that has been happening on Falcons. At least it seems to be at a point where it runs, though with some yet to be solved glitches in video emulation, whereas before it was not launching at all no matter what.

It's starting to get to a point where I should upload a new build to get some more testing on the different platforms.

There's a new internal file cache too which will speed up using disk images and plain TOS.
It's still nowhere near as quick as on Mint or MagiC, or when using direct disk access, but it does cut down the bootup time a fair bit.

New version.
This merges some generic changes with separate test branches for Falcon030 (thanks @Badwolf!) and CT60 (thanks Swe/YesCREW)
Hopefully now working correctly on Falcons, and hopefully not breaking any other machine/config out there.. :hide:

You can grab it from here:
http://www.happydaze.se/basiliskii/


Changes:
- 040/060 video emulation fixes
- General Falcon030 fixes.
- Better gfxcard detection and palette handling
- Optional disk-image read cache for a bit better performance
(when on plain TOS, it's not really needed with MiNT or MagiC)
- Shutdown related bugs
(still not perfect)
- GUI related issues

The default basilisk.inf was modified, but your old one should work fine too.
If you have a "diskcache false" in your old basilisk.inf then you probably want to set it to "true" instead :)

Definitely boots a bit quicker on the TT.

Here are the benchmarks, firstly Basilisk 8 vs. the current version:


And now TT-Medium vs. ST-High resolution:

stephen_usher wrote: 06 Jun 2022 21:54 Definitely boots a bit quicker on the TT.

Nice, thank you for the report! :thanksyellow:


Quite funny that emulating 640x480x8bpp is faster than native 8bpp on my machine.
Yeah I know, it's not really a fair comparison. There's less than a quarter of the pixels in ST-Low, plus it skips every other frame so I guess that makes it 1/8th of st-bus access..
But still! It's either decently fast, or my graphics card is really bad, or a combination thereof :)

I did a quick test in TT-Low this morning... It's significantly slower! I'll post pictures tonight.

As promised, here are the comparative performance statistics for TT-Low vs. TT-Med resolution.

stephen_usher wrote: 07 Jun 2022 17:57 As promised, here are the comparative performance statistics for TT-Low vs. TT-Med resolution.

Interesting! I guess it's the same amount of data in the end, but 8bpp+ scale c2p ends up taking significantly more cpu time.
Cool to see the comparison :)

A few bits:-

Cache definitely improves things, but disc performance (per the benchmarks) are still ~0.4x (TOS) to ~2x (MiNT).

Still have the issue with the colour map seeming off for menus. Black on black when activated. Not NVDI. Tried booting to an empty desktop, running FASTRAM and then BASILISK straight away -- still observe it.

Performance comparing 8 plane v 4 plane emulation as attached (640x400RGB)


Similarly 8 plane v TC emulation:-


Finally, a quick comparison between Mint disk image (2x disc) and Mint raw disc access (7.6x) in 768x480@8bpp mode:



All the above done with MMU and CMP emulation active. 50MHz Falcon w/TT-RAM.

Cheers,

BW.

Just quickly on the palette issue, here's a compare and contrast in case it's useful.

Anyone have a picture of what the palette in 256 mode is meant to look like?

Cheers,

BW

Yay! Thanks @Badwolf I really appreciate it :)
I'm relieved the build is finally at a stage where it's working on ST, TT, Falcon030 and Falcon060.

Badwolf wrote: 08 Jun 2022 20:24 Cache definitely improves things, but disc performance (per the benchmarks) are still ~0.4x (TOS) to ~2x (MiNT).

Yeah I think that's as good as it'll get on plain TOS, at least for now.
It'll go faster the more of TOS's code+ram you manage to cram into fastram of course.

I can be remembering wrong, but I think Speedometer measures by doing a cycle of "large write, then a tiny read" over and over again. In such scenario the disk-image read cache wouldn't have that much of an impact on the score given by Speedometer.
The best indication is if startup time is faster and hopefully ok-ish, and icons in the Finder come up quicker.

MiNT will always be faster though. It'll be running entirely from fastram, plus have all of its work memory there too. And a pretty nice internal i/o block cache on top of that.

Badwolf wrote: 08 Jun 2022 20:24 Still have the issue with the colour map seeming off for menus. Black on black when activated.
....
All the above done with MMU and CMP emulation active. 50MHz Falcon w/TT-RAM.

This is super strange and it would be great to get that resolved somehow.
I'm surely doing something wrong somewhere falcon palette related, or c2p, or both perhaps.
It seems to work correctly on the only other Falcon I know has tested Basilisk, though that one has a CT60 so the codepath is not exactly the same.

Is it just the top-level menubar, or is it the same for any highlighted item in the dropped down menu?
What happens when the Apple logo is selected? Does the entire section turn black including the logo or does that one stay correct?

I take it there's no difference for the colour issue if you turn off CMP emulation?

This is how the palette looks on my machine. To me it looks the same as on your machine.