> SDL2 -video=accel should gain you more performance than -mt anyway. And it's not like
> -mt affects if e.g. Blitz/Gauntlet/Crusin' use multiple CPUs (they do even when it's
> off). Basically, -mt is known broken on baseline Windows, suspect broken on SDL
> Windows, known fatal to the OS on Mac at a may-crash-the-window-system level, and
> sometimes it works on Linux. I'm not sure why we even offer it as an option anymore.
I haven't had an issues with mt on Linux so far. The option still provides a slight performance advantage (7%) on my system:
./mame64 -mt -np 4 -str 60 -nothrottle radikalb -ad pulse -noautosave -video accel
Average speed: 174.01% (59 seconds)
./mame64 -nomt -np 4 -str 60 -nothrottle radikalb -ad pulse -noautosave -video accel
Average speed: 162.66% (59 seconds)
As the one who is originally responsible for "mt" in sdlmame I have to add that rendering in a background thread was not supported well on every OS in SDL1.2 and with SDL2 the situation seems to be worse. On the other hand I never really tested the SDL2 code on a windows build. In SDL1.2 there were differences between Linux and Windows around the initialization of certain systems either in the main thread or in the rendering thread. Maybe this has changed. But first I need to setup a working compile environment in my virtual machine again.
|
Index 
Flat 
Re: SDL Multithreading or build issue?
Reply
