If anybody was wondering about program guided optimization... \ VOGONS

If anybody was wondering about program guided optimization...

Topic actions

First post, by frobme

Posted on 2010-01-13, 23:31

frobme Offline

Rank Member

Rank: Member
Posts: 137
Joined: 2007-01-09, 09:03

Every couple of years I check this out with Dosbox to see if there are any major efficiency gains. Every couple of years the answer is "not really" =).

For what it's worth, using the latest SVN trunk, building with PGO and instrumenting over 4-5 game sessions (various games) only results in about a 4% FPS increase for my tests. It's hard to be exhaustive with the example scenarios, but that's a pretty good slug of data, so I dont think there are substantial gains to be made here and it's not really worth the effort (this was under Visual Studio 2008 btw, my mingw/GCC install went stale and I need to fiddle with it).

Anyway, that's not bad news at all, it means that Dosbox is structured well enough to deal with all the obvious gains already. Just thought I would mention if anybody else was curious.

-Frob

Reply 1 of 2, by wd

Posted on 2010-01-13, 23:41

wd Offline

Rank DOSBox Author

Rank: DOSBox Author
Posts: 10813
Joined: 2003-12-03, 21:23

it means that Dosbox is structured well enough to deal with all the obvious gains already.

Wrong, means that the recompiler (which is out of scope of pgo or profiling) is eating most of the cpu time 😉

Reply 2 of 2, by frobme

Posted on 2010-01-14, 05:39

frobme Offline

Rank Member

Rank: Member
Posts: 137
Joined: 2007-01-09, 09:03

Fair enough, although I made sure to stress video by doing things I don't normally, like 3x expansion and scalars =). But the central idea remains - red herring, don't need to focus here.

Now focusing on the recompiler might reveal some gains apparently!

-Frob

Go to top of page Go to top of page

Back to DOSBox Development