First post, by Folken
I believe there is a bug with the FPU change that was submitted back on June 1. I really don't know enough about the code to say what the bug is exactly, just that I've been noticing a number of issues with CVS builds for the past while and it seems like this is the change that is causing the issues.
I've been seeing the problems with a number of DOS demos, problems that I cannot reproduce in 0.65 release, or with CVS builds from before this change. The exact problem varies depending on the program - usually graphical glitches or sound issues. I haven't noticed problems with any games yet, but I haven't tried very many.
So I decided to compile my own build, and then I basically reverted changes back until the problems went away. And they did indeed go away when I reverted those FPU changes. At that point I went back to the latest CVS, and figured out I could just comment out the "#define X86_DYNFPU" line in cpu/core_dyn_x86/decoder.h. Making that one change is the easiest way to see the difference.
One specific example is the demo "The Fulcrum". It can be downloaded here: http://www.scene.org/file.php?file=/demos/gro … ix/mtx-fulc.zip
At the very start of this demo there is a robot walking around. Using latest CVS (with those FPU changes), after the robot walks for about six or seven steps, the shadows become corrupted. Using an earlier build (or if you comment out that #define) the shadows display fine.
I'm running an AMD64 dual-core under WinXP.