Yeap, -O3 is mostly voodoo. Berger has some measurements.

Spoiler: He found your username has a bigger effect on performance than most compiler flags:-)

It doesn’t matter because for the most part, you’re not spending most of your time in Wine code, you’re spending your time in the application’s code.

So you might see some improvements if you’re running something that’s more of a Win32 API performance benchmark than an application benchmark.

But even then, -O2 is already pretty good, and -O3 is a bit of a gamble whether it’ll be faster or slower. Sometimes O3 will pull the big guns and the real world dataset is always small and it wastes time in setup.

-march=native might give you some nice speedups but it depends entirely on your CPU. If you have the latest CPU with all the features, it may take advantage of it.

But it’s still all mostly things like using SIMD instructions to crunch through numbers faster. The Windows API doesn’t do a whole lot of that in the first place, but there’s also a few ABI boundaries you can’t change. It needs to receive and output data in the expected format in the expected registers. It can only optimize things that happen internally in Wine and are never observed by the application it’s running, further reducing the amount of optimization potential.

So, it’s kind of a waste of time. It’s a lot more worth it for like video codecs and image processing

IMO there’s no point in optimising without being able to measure the results.

A profiler should be able to tell you how much time is spent running the code you’re attempting to optimise. It might turn out that the wine runtime code is not a significant factor in performance.