leileilol wrote:Some think multithreading can magically apply to everything and some want to be so entitled for games to use 100% of their processor no matter what, like believing the hype around that TDM MinGW package to automatically parallelize all functions, tacking on -march=native and all the sse/avx flags will make everything super fast (spoiler: it won't). Given that
he can't even compile gzdoom on his own I doubt it.
Yes, it's really odd. Some people, even otherwise knowledgeable ones, sometime think that using a specific feature will bring some magic speedup. The two biggest ones I had serious clashes over with some hard to convince people:
OpenGL immediate mode is slow, moving everything to buffers will dramatically speed up your code:
Sorry, no. This works when you have stuff that contains a lot of vertices that can be output as a single large mesh. In that particular use case it indeed can being a 10-100x speedup - but not for Doom map geometry. A normal wall has 4-10 vertices, and the time saved by having the vertices ready in a buffer is minimal. It adds up when you got 10000 walls but even on a map like Frozen time it's like 44 fps vs. 48 fps roughly, i.e. 10% performance gain on NVidia and none at all on AMD because it all gets negated by their driver overhead.
Using SSE2 for floating point math is significantly better than the old x87 instruction set
I have seen benchmarks for some older systems, when SSE2 was new that it could be up to 40% faster. The problem is, these systems are long gone and in modern CPUs, say everything from the last 10 years the differences are utterly marginal, some even slightly favoring x87 speed-wise. And still, you get some people who suggest to write multiple code paths to get the most juice out of every imaginable CPU out there and of course those who accuse you of improper research. Now, what more can you do is than running some math stuff in a loop and timing it all? And it always comes back the same: Aside from minor fluctuations there's no timing differences on modern systems. Not enough for a true believer (which normally means: belief in the marketing, not belief in the facts.)