You did use the vgaonly machine type did you?

Tried mainly the "svga_s3" video card.

For "Vanility", authors said it required a "Tseng Labts Et4000" so i tried "svga_et4000" with and without univbe but it always freeze at water effect.

I'll give a try with "vgaonly".

- Demo "Crystal Dream" from Triton
Choose Sound player Stereo, 24 KHz
It will sound as good as Soundblaster

- Demo "The bad, the good and the ugly"
ems=false
gus=true
cycles=5000
machine=vgaonly

When using core=dynamic: run loadfix -20 before gbu

Issues:
The first screen with the blue cubes is broken as there should be text
The wireframe scene should only display the wireframe, not the surrounding "border"

> *The winged creature have some bad pixels on the wings
This happens on real hardware too on most VGA cards, or depends on something else

Here's what happens :

with "vgaonly" :
- Reve : Status windows display "Exit to error: PIC: 4 byte interval not handled" then DosBox crash

- The bad, the good and the ugly : first plasma is again badly rendered but if frameskip is set to 2 or 3 it's almost good, the wobbling effect with winged creature is ok without any change to frameskip, the creature has the same pixel bugs on the wings.

- Vanility : freeze at water effect

- Crystal dream : same soundblaster bug

Read my post

@ h-a-l-9000 : sorry i was typing when you've posted your reply.

Thanks you've saved my ears with Crystal dream 😉

For GBU, ok it works better. 😊
Do you remember the text at first screen ?

It uses the text that was on the screen just before start.

Vanility runs to completion for me with memsize=48 and all other settings at default. The crash occurs when the interrupt table is overwritten.

The DOSBox site mentions an unofficial demo compatibility list on its links page. However, it's very much out-of-date.
http://pain.scene.org/service_dosbox.php

I'd appreciate any comments/suggestions/feedback regarding the following.

Ninja 2 by Melon dezign (resolved): It's listed as working in Ripsaw's link, but for me on .74 it immediately crashes leaving the display in an odd garbled low-res text mode. This requires using GUS on port 220, or deleting NINJA2.CFG to enable realignment of the sound config with DOSBox settings.

Uni by MFX: Starts, but displays only a black screen while playing a continuous off-key noise.

Transgression by MFX (resolved): This runs well, but doesn't output correctly when using it's high-color mode (i.e. -hc switch). This requires VGAONLY machine mode, it seems to use an odd half-width resolution.

Booth 2 by MFX: Seems to run well, but spews "DMA segbound wrapping" messages while doing so and and hangs on a oddly patterned screen at it's end (or what I presume is it's end.) The "DMA segbound wrapping" issue also effects: sexadelic by Byterapers, and Te-2rb by Tpolm.

Zilog by Sunflower (resolved): Runs for a bit and then crashes DOSBox at the brown mushrooms scene, unless "normal" core or a low cycle rate that can be continuously maintained is used.

Alien Sex Clone by Fudge (resolved): Complains of missing 512x256 video mode, unless UniVBE is loaded beforehand or a modified DOSBox with additional graphics modes is used.

Crystal Dream: SoundBlaster support is hard-coded to IRQ5. Even then, it doesn't play properly for me. SoundPlayer mode works nicely, though.
I haven't looked at any of the others yet.

Reve: disabling the 4 byte interval check allows the demo to run to completion. Why is the check there? Shouldn't 4 byte interrupt size be the default behaviour, and thus acceptable?

Reve: 4 byte interval check? Where is that?

Transgression (2?): Uses a very special graphics card, works only in some inofficial SVN builds. Its oddly stretched because the DOSBox renderer doesn't support 4x horizontal stretching. You maybe could fix it with some video editing software.

Crystal Dream: Triton uses/abuses the soundblaster in a way that probably saved them some code to support both parallel port sound devices and soundblaster. DMA is set up to read from one memory address only which is changed at every sample. DOSBox's soundblaster reads sound data in millisecond intervals, and that obviously doesn't work with their scheme. I/we don't feel like increading the complexity and CPU usage of the emulation only to support one group of programs that is off, especially since the sound player is a simple workaround.

This should be moved to the DOSBox forums btw.

4 byte interval is in pic.cpp.
if (val&0x04) E_Exit("PIC: 4 byte interval not handled");

Ninja 2: requires GUS on port 220.

Documentation says: The ADI bit (ICW1 bit 2) has no meaning in 8086/88 mode (a vector is transferred to the CPU instead of a call opcode), so the check should be removed.

if(val&0xe0) E_Exit("PIC: 8080/8085 mode not handled"); can be removed too as the bits have no meaning in 8086/88 mode.

Side note: The datasheet also says that writing ICW1 has side effects like IMR and IRR being cleared which I don't see in our code.

Ninja 2: requires GUS on port 220.

Only "requires" that if you leave the NINJA2.CFG file included in the archive. Works fine with SB or GUS on most ports, IRQs, and DMA channels if you delete the file and choose options in the MIDAS setup screen that match DOSBox settings. Probably was a mistake to include that file.

Documentation says: The ADI bit (ICW1 bit 2) has no meaning in 8086/88 mode (a vector is transferred to the CPU instead of a call opcode), so the check should be removed.

portslist seems wrong on this, IBM docs mention that it must be set to zero
(0==4 byte vectors) but it may not have any adversary effect if set to one.
Just be sure that the stuff that sets this bit doesn't do so because of some
other bug.

IBM? We better ask those who actually made the chip 😉

http://bochs.sourceforge.net/techspec/intel-8259a-pic.pdf.gz

With Zilog there is a PMODE exception that appears to go away with LOADFIX (should always try loadfix with various allocations before reporting a problem). Also saw DOSBox run out of cacheblocks one time, but only the once, so might be random. Normal core can't have a problem with cacheblocks, but might be too slow. The startup message had me worried it was going to try and defrag part of the HD that isn't really there, but I think it's just something it does to show that it hasn't hung up while loading.

Alien Sex Clone searches through all VESA modes for a 512x256x16bit mode, but DOSBox doesn't have one. Apparently a straight VESA mode, not some tweaked mode.