Wow, good stuff. Skimming through to apply my own intuition: 6.99% ssgEntity::cull_test 5.28% ssgBranch::cull 4.85% ssgVtxTable::draw_geometry 3.42% FGHitList::IntersectLeaf 3.28% FGHitList::IntersectBranch 2.85% ssgVtxTable::getNumVertices 2.71% sgFrustum::contains 1.85% sgdPointInTriangle 1.28% sgdClosestPointToLineDistSquared 1.28% ssgSimpleState::apply
These guys are all directly related to the rendering process, and are where we should expect to spend our time. The only odd one I see is ssgVtxTable::getNumVertices. I don't care how many times that's called, how does returning an integer figure in global performance? Maybe it's a profiler interaction. 8.84% sgdMakeNormal This one is notable only because it's so expensive. Sometimes, you just have to normalize something before passing it to OpenGL, and you have to eat the cost. As far as I know, all of our geometry is fixed and we don't use funny scalings. So most stuff shouldn't require per-frame normalization. This looks to me like a good spot to concentrate on; I'll bet most or all of these calls can be eliminated with a little work. 17.40% ssgMakeMipMaps 1.57% ssgSGIHeader::ssgSGIHeader[in-charge] 1.43% ssgSGIHeader::getRow These are clearly related to texture loading. How long did you run your profiler for? If it was just for a few seconds, then these might be showing the time spent in initialization and not in the routine frame-to-frame running of the program. Certainly, parsing and generating textures is going to be expensive at startup. 2.14% calc_magvar This has me stumped too. First, why on earth is it being called so much? I'd think once per frame would be more than sufficient. Second, this is (or should be) an rather cheap computation. A few hundred CPU cycles at most. Weird. Andy -- Andrew J. Ross NextBus Information Systems Senior Software Engineer Emeryville, CA [EMAIL PROTECTED] http://www.nextbus.com "Men go crazy in conflagrations. They only get better one by one." - Sting (misquoted) _______________________________________________ Flightgear-devel mailing list [EMAIL PROTECTED] http://mail.flightgear.org/mailman/listinfo/flightgear-devel
