Hi there. I've been playing w/dosbox for a couple years now but haven't touched the code. At any rate, too slow for me, should be faster. By trade I'm an embedded/realtime software engineer. Is there any profiling/benchmarking stuff setup for dosbox already? I'd like to know where the biggest problem areas are in the games I'm trying to run.
nothing officially.
but there is a patch at sf.net tackling some parts of the code.
Water flows down the stream
How to ask questions the smart way!
Depending on the platform, for example linux, there's tons of built in profiling. If you're running windows, you'll have to set everything up(which reminds me, I really need to setup a linux box). I've been using gprof in mingw from windows. If you havn't played with gprof just google it, though as an imbedded programmer I'd imagine you're familier. I've also been using the Pentium RDTSC for cycle measuring. For that I added some functions for calling and formating the results. Calling RDTSC is very simple and many examples of it can be found.
Your one-stop linux profiling source:
http://oprofile.sourceforge.net/news/
I've never seen a better profiling solution. Profiles optimized executables, no recompilation neccessary, low overhead, even profiles kernel code (that way I found out my sound driver chews quite some amount of CPU).
Yep, oprofile for x86 is system wide RDTSC measuring and there's also a patch for it to do call graphs like gprof so under linux you get what I've done in windows but automagically for everything.
Thanks, I don't think I've ever seen oprofile before. Playing with it now. I was playing around with some DOS benchmarks, but I didn't find anything comprehensive -- basically they all just come up with some number that says 'this is how fast your system is', and that's just not very useful. They all seem to have names that all sound the same, too, so it's hard to tell if I've tried them all or not. Is there any particular benchmark(s) that folks on here use, or is it just checking to see if one particular game runs faster or no?
The easiest low-tech solution is "turn up cycles until sound starts skipping". Unfortunately, it is rather inaccurate. Unless you compare two different versions side-by-side, it's difficult to get accurate results. It's enough for estimations which may be 10% off, and that's useful since common wisdom says that 10% improvement is the smallest that humans actually notice.
