Supermodel Forum

deleted

Thanks to the 820 update, I can finally compile my own version of Supermodel for my post! Bart, what is that NEW_FRAME_TIMING option feature?

It no longer does anything and ought to be removed.

YAY! i can compile again. As said before, adding "-j plus the number of cpu cores" to "mingw32-make -f Makefiles\Makefile.Win32" helps speed up the compile time.

Don't forget the SDL2_net.dll file when you build if you have the NET_BOARD = 1 variable on Options.inc. I tried using the -j command at the end of "mingw32-make -f Makefiles\Makefile.Win32" (no quotes) command will speed up the compilation time. This method is way faster than using Visual Studio which is more expensive and a lot of time to load whereas this, you use Command Line and compile, that's it.

I actually saw -j4 fail the other day with Musashi. I guess I must have an error in my Makefile in terms of how dependencies are specified or some other subtle race condition. Certain files were not present at the right time and boom! But simply running Make again resumes where it picked off if that ever happens to you.

Let us know if you ever experience a build that crashes mysteriously as I have seen this. There are some issues in Supermodel that need to be fixed (all those compiler warnings are pointing to some real problems). Would be interesting to benchmark a gcc build against an MSVC build on a slower CPU (or just run unthrottled). Someone should also investigate link-time optimization (LTO) in gcc.

Oh and it compiles just fine in windows powershell in case anyone was wondering. I always set -j to one less than my cpu core count, in my case -j7 for a quad (8 thread) amd rizen 5.

GCC compile time in windows powershell 5.1, 48 seconds.

Thanks a lot Bart for the updated simplified method to compile Supermodel.
It worked like a breeze this time! Finally I can compile it myself

VFer97 wrote:Thanks a lot Bart for the updated simplified method to compile Supermodel.
It worked like a breeze this time! Finally I can compile it myself

Awesome This is definitely the way to go and I highly recommend doing this rather than relying on builds of unknown provenance.

Interestingly, I just tried measuring the performance impact of different optimization flags and found none. Using different permutations of -flto, -march=skylake, -O2, -O3, and -Ofast had no effect. I suspect the optimizer just can't do much given how the code is structured. The PowerPC emulator relies on indirect calls to tables of function pointers, which works against the compiler's optimizer. I do not think I've hit any sort of GPU bottleneck because I get the same results at 496x384 as at 1280x1024.

L.A. Machineguns runs at 160-175FPS during game play for me when the frame rate is uncapped.