(Experimental) Voodoo without thread

[luibgn]

I made some simple modifications to use the voodoo stuff without the dedicated thread but using only the bochs virtual timer. So removing all mutexes and semaphores.
Maybe it could be useful to check emulation correctness without bothering with threads sync and locking. Obviously there is a small performance loss but not so much.
Theoretically it could be possible to use both thread and non-thread versions in the same code with some #ifdef switches at compile time.
The thing could be refined, for example wirh start/stop of the timer and calculating the best timer period.
If there is interest I can post the modified files.

[Vort]

This is interesting topic.
I think same ideas can be applied to other subsystems as well.

When I started learning about Bochs, I thought for some reason that it allows to have full reproducibility, at least in some configurations.
Then after several attempts at tracking various bugs, I figured out practically that this is not the case.
However, pursuing such goal may make sense.

Obviously there is a small performance loss but not so much.

Can you test it with Ballance game?
I assume there will be 2x performance drop in single-threaded case.

[luibgn]

You can try anything as I know that you compile bochs by yourself.
I post the code, just some files in the iodev/display directory.
Save the originals if you have custom modifications.
The performance loss is variable. It is true that a machine has multiple processors, but both threads are cpu-bound and it is not sure that the application can have 2 cores for it.
Also I use bochs almost only on android where an arm chipset has few high performance cores and other power saving cores.
voodoo_single_thread.zip

[fysnet]

Can you test it with Ballance game?

Am I missing something? The .BIN file from that link isn't bootable.

On a side note, if you have the following line:
ata0-master: type=disk, mode="vmware3", path="ballance.bin"
and Bochs fails to open the file because it can't find the header, Bochs crashes.

[Vort]

Am I missing something?

This image is in .cue+.bin format.
There are tools which can convert such images into .iso format.

Also for Windows it is possible to mount .cue+.bin into Daemon Tools virtual drive on host and attach drive to guest afterwards.

[fysnet]

On a side note, if you have the following line: ata0-master: type=disk, mode="vmware3", path="ballance.bin" and Bochs fails to open the file because it can't find the header, Bochs crashes.

See #476

[vruppert]

After a quick test with the ValleyOfRa demo I can say that the performance loss in 3D mode is about 50% (IPS drop from about 64 to 29 Mips). Using this code to find out if an issue is caused by emulation or multi-threading is okay, but in general we need more usage of multi-treading, not less. In the early days of Bochs everything was done in a single thread and multi-threading was a feature request. Only the win32 and rfb guis were using a thread internally. In the meantime it has been added in some more guis and the lowlevel sound code. The Voodoo device is still the only device emulation using multi-threading.

[luibgn]

Yes. It is NOT supposed to replace the mt version, but can be useful to check something in the emulation.
For example, running the ballance game I see in the st version a super extra drop of fps at the start of the game when there is the small dynamic splash banner window with the ballance presentation.
I'm curious to know what it is that causes this super dropdown. Then in full 3D game it is just slower than the mt version.