WinUAE 5.3.0 beta series

[Cowcat]

@AC/DC b4 seems to fix the OS 4.1 issues I had. Just saying.

[Toni Wilen]

Originally Posted by AC/DC HACKER!

With OS 4, after installing fresh (Quickstart), I get a funky screen of the normal OS 4 'splash' screen, I see the drive light flicker for HDD, ...I don't see the Workbench. CSPPC reflects (same info as my real Amiga) CVPPC.

Is this really specific to 5.3 betas? Thats the most important bit in beta threads.
(Also don't use Quickstart if you want to use OS4. It isn't that simple)

[mark_k]

I booted NetBSD 5.2 and noticed a small issue regarding the Merlin.

When set to Zorro II mode, this text appears during boot:

grfet0 at zbus0 pa 0x200000 man/pro 2117/3
grf6 at grfet0: width 640 height 480 colors 16
...
grfet: 2MB Merlin with Tseng ETW32 and BrookTree Bt482 DAC being used
grfet1 at zbus0 pa 0xe90000 man/pro 2117/4

However when the Merlin is set to Zorro III mode (2MB or 4MB RAM):

grfet0 at zbus0 pa 0x40000000 man/pro 2117/3
grf6 at grfet0: width 640 height 480 colors 16
...
grfet: 32MB Merlin with Tseng ETW32 and BrookTree Bt482 DAC being used
grfet1 at zbus0 pa 0xe90000 man/pro 2117/4

Notice the bogus "32MB" memory size shown.

Check the NetBSD source, grf_et.c. Starting from line 513, code can set et_fbsize to some value. I'm assuming that whatever check it does, it never actually sets et_fbsize so that variable remains uninitialised.

case MERLIN:
		vgaw(ba, GREG_SEGMENTSELECT2, 0x00);
		if (((vgar(ba, GREG_FEATURE_CONTROL_R) & 12) |
		     (vgar(ba, GREG_STATUS0_R) & 0x60)) == 0x24) {
			WCrt(ba, CRT_ID_VIDEO_CONFIG2, 0x07);	/* 1Mx4 RAM */
			et_fbsize = 0x400000;			/* 4 MB */
		}
		else {
			/* check for 1MB or 2MB board (crest) */
			/* has there a 1MB Merlin ever been sold ??? */
			volatile unsigned long *et_fbtestaddr;
			et_fbtestaddr = (volatile unsigned long *)gp->g_fbkva;
			*et_fbtestaddr = 0x0;
			vgaw(ba, GREG_SEGMENTSELECT2, 0x11); /* 1MB offset */
			*et_fbtestaddr = 0x12345678;
			vgaw(ba, GREG_SEGMENTSELECT2, 0x00);
			if (*et_fbtestaddr == 0x0)
				et_fbsize = 0x200000;		/* 2 MB */
			else
				et_fbsize = 0x100000;		/* 1 MB */
		}
		/* ZorroII can map 2 MB max ... */
		if (!iszthreepa(kvtop(__UNVOLATILE(gp->g_fbkva))) &&
		    et_fbsize == 0x400000)
			et_fbsize = 0x200000;
		etctype = ETW32;
		etdtype = MERLINDAC;
		break;

Could it be that the first if condition never matches due to emulated register bits not being correct?

[Toni Wilen]

It looks like text outputs Z3 board physical size (32M) instead of logical size (VRAM size). It probably is designed to output board size. Board size is almost always larger than VRAM size if Z3 board. (Min Z3 physical board size is 16M)

If all tests in source snipped would failt, it would return 1M board.

[thomas]

CVisionPPC: the mirrored native display (if no RTG software is installed) is vertically squished. It takes up only the upper half of the screen. Not a big deal, hovewer the mouse pointer is at the right position. So it is almost impossible to drive the GUI because the mouse pointer's vertical position is twice the way towards the bottom as the GUI elements.

On a real Amiga the display is correctly stretched to full screen.

[Toni Wilen]

Originally Posted by thomas

CVisionPPC: the mirrored native display (if no RTG software is installed) is vertically squished. It takes up only the upper half of the screen. Not a big deal, hovewer the mouse pointer is at the right position. So it is almost impossible to drive the GUI because the mouse pointer's vertical position is twice the way towards the bottom as the GUI elements.

On a real Amiga the display is correctly stretched to full screen.

Fixed. Permedia 2 "graphics processor" mode Line doubling bit was not emulated.

[Toni Wilen]

https://download.abime.net/winuae/fi...nuae_5300b5.7z
https://download.abime.net/winuae/fi...ae64_5300b5.7z

Beta 5:

- Fixed pre-KS 1.2 HDF/directory filesystem boot. Two bugs: crash due to config handling changes (3.6) and KS 1.0/1.1 seems to have a bug where level 2 CIA keyboard interrupt handler returns "I handled this interrupt" status even if it was caused by external device, skipping all following handers, including UAE filesystem interrupt handler, causing a hang. UAE FS interrupt level was changed to -1 if KS 1.3 or lower (4.9) because there was some buggy (KS1.x compatible only) program that poked interrupt list and broke if CIA handler wasn't first in list. Now UAE FS handler priority is lower than CIA handler only if KS 1.2 or 1.3.
- Multi-monitor active state was not fully reset when system was hard reset. (Secondary window was open -> hard reset -> window didn't open anymore again)
- If x86 bridgeboard SVGA and also other hardware emulated RTG board was active and END+F9 was used to switch monitors: both monitors' output was drawn in same window, causing corrupted graphics and possibly also crash if monitor sizes were too different.
- Adjusted 520b5 audio period hack, do not change period immediately after first loop but only after it has looped multiple times. Does not anymore mess up weird audio test programs.
- Permedia 2 Video Control register video enabled/disabled (disabled = blanked) and Line doubling bits emulated.
- Show error message without resetting config back to defaults when GUI Load button could not open the file.
- GD-CL54xx LFB VRAM reads fixed. (For example NetBSD console scrolling was broken)
- Fixed S3 Virge MMIO address calculation (fix from 86box update). NetBSD "No mapping" log message fix.
- A2410 NetBSD text console works again.
- Autoswitch to hardware emulated RTG board also when text mode gets enabled (*nix boot consoles etc)

[AC/DC HACKER!]

"- Permedia 2 Video Control register video enabled/disabled (disabled = blanked) and Line doubling bits emulated."

Ahh, That helped a lot with Boot Menus options selections with OS 4.x and OS 3.1. Sweet.

[Toni Wilen]

https://download.abime.net/winuae/fi...nuae_5300b6.7z
https://download.abime.net/winuae/fi...ae64_5300b6.7z

Beta 6:

- Los Justicieros (Zorton Brothers) laserdisc arcade game left holster hooked up correctly. (Was stuck active previously)
- If programmed mode was enabled first and if programmed horizontal blanking got enabled slightly later, it didn't force display position recalculation. This is needed because programmed mode horizontal position also depend on horizontal blanking configuration. Real monitors also use both syncs and blanks to detect display position. (I noticed this in CyberVision PPC config where CVPPC gets activated at boot, then when END+F9 is used to switch back to native mode, display was horizontally shifted and clipped)
- In NTSC mode if bitplane DMA last fetch "overflowed" to cycle 0 (which is harmless and normal in overscanned modes) when line was short, it was incorrectly kept allocated for bitplane DMA also during long lines. Long line has 1 more cycle = no overflow. (NewTek demo reel 3 dynamic hires image corruption in NTSC mode, CPU didn't have enough time to update all colors quickly enough. Note: corruption at the bottom in PAL mode is demo bug. NTSC mode also has 2 slightly corrupted lines at the bottom, this is also normal.)
- Magic mouse mode without enabled mouse driver was broken in b1. Moving mouse outside of emulation window and back caused it to stop moving.

[Rotareneg]

That dynamic hires image is fixed, but unfortunately the demo now glitches while on the loading screens:

[Toni Wilen]

https://download.abime.net/winuae/fi...nuae_5300b7.7z
https://download.abime.net/winuae/fi...ae64_5300b7.7z

Beta 7:

- b6 NTSC fix corruption fixed. Fix was incomplete.

[Toni Wilen]

https://download.abime.net/winuae/fi...nuae_5300b8.7z
https://download.abime.net/winuae/fi...ae64_5300b8.7z

Beta 8: (Final soon)

- Line 0 is also visible at the bottom of screen in OCS Agnus + OCS Denise combination, works like A1000 but without 1 line delayed vblank interrupt. ("Extra" lines are only visible if Overscan+ or higher mode)
- DMA debugger line 0 strobe type is now correct in OCS Agnus and A1000 configurations. (STRHOR if A1000, STRVBL if OCS Agnus, STREQU if ECS Agnus/AGA)
- Implemented Permedia 2 (CVPPC//BVPPC) LineCount register. Fixes OS4 boot screen hang. NOTE: Read Classic_FAQ on OS41FE CD root directory before attempting to use CVPPC/BVPPC with OS4.
- Flush log file before adding log to crash dump file.
- b5 audio fix (in some situations audio DMA slot was allocated in wrong position causing DMA debugger reported conflict)
- -theme command line parameter now also support "default" value: ignore GUI setting and follow Windows setting.

[freddy]

Toni
Entries do not exist for the translation of titles for the 2 dialog boxes.
Also, when you switch to dark mode, you switch from a button to a checkbox.
Thanks

[Toni Wilen]

Originally Posted by freddy

Entries do not exist for the translation of titles for the 2 dialog boxes.

Done.

Originally Posted by freddy

Also, when you switch to dark mode, you switch from a button to a checkbox.

It is normal. Those buttons are internally checkboxes that look like buttons (have both unpressed and pressed state). win32 dark mode hack does not support all possible element combinations. Functionally they are identical.

[paraj]

Maybe it's somehow just my setup, but it seems like a standard "blitz" looks wrong in 5.3.0 betas with A1200 quickstart? <= 5.2.0 final doesn't show anything like that. (64-bit version if it matters)

By blitz I mean ye olde:

start
        addq.w  #1,d0
        move.w  d0,$dff180
        bra     start

(Images removed, no longer matter)

EDIT: Bisecting, I can see it changed with https://github.com/tonioni/WinUAE/co...fc1cb67b27fe1a

Seems like you may want a v &= 0xFFF somewhere in the "aga_mode" block (that fixes it at least).

[elessar84]

Is it possible to fix this gfx bug in Zool? (AGA version).
The problem was probably fixed in 4.10, but since Winuae 5.0 it appears again.
Old thread for the described problem:
https://eab.abime.net/showthread.php?t=112888

My version of Zool AGA is an ipf image

[Toni Wilen]

Originally Posted by paraj

Maybe it's somehow just my setup, but it seems like a standard "blitz" looks wrong in 5.3.0 betas with A1200 quickstart? <= 5.2.0 final doesn't show anything like that. (64-bit version if it matters)

<snip>

Seems like you may want a v &= 0xFFF somewhere in the "aga_mode" block (that fixes it at least).

No, it wasn't that. Problem was genlock bit getting confused with special blanking bits (used same upper nybble of colors)

Fixed.

Originally Posted by elessar84

Is it possible to fix this gfx bug in Zool? (AGA version).

Yes but no one knows when. It depends on 68020 CPU speed and it can change between versions (there is not enough details to get it correct enough without breaking something else). It was never fixed. (also don't post non-beta introduced bug reports in beta thread, thanks)

[paraj]

Originally Posted by Toni Wilen

No, it wasn't that. Problem was genlock bit getting confused with special blanking bits (used same upper nybble of colors)
Fixed.

Yep, confirmed fixed with your latest commit. Thanks.

[Toni Wilen]

Originally Posted by elessar84

Is it possible to fix this gfx bug in Zool? (AGA version).
The problem was probably fixed in 4.10, but since Winuae 5.0 it appears again.
Old thread for the described problem:
https://eab.abime.net/showthread.php?t=112888

My version of Zool AGA is an ipf image

Still wrong thread but just wondering.. Is that top line really the only visible glitch now? This game has been known to have lots of blitter wait problems causing all kinds of glitches here and there randoomly (as shown in old thread).

[elessar84]

Originally Posted by Toni Wilen

Still wrong thread but just wondering.. Is that top line really the only visible glitch now? This game has been known to have lots of blitter wait problems causing all kinds of glitches here and there randoomly (as shown in old thread).

Sorry for the wrong thread, I thought it was something more related to the latest version of Winuae. Regarding Zool, this line is the only problem I noticed in the original version in the ipf file. There are definitely more graphics-related problems with the whdload version.