Brian Ward

I wonder if it’s slower to compile position independent code? It’s also possible some of the precompiled header logic in CmdStan didn’t make it into our makefile, though that stuff...

If you compile with autodiff hessians you’re also including all of stan/math/fwd, which definitely adds a lot of files for the preprocessor to handle

I can take a closer look tomorrow, with that ruled out my questions would be: Are there any additional commands used in our compilation compared to CmdStan (I think not,...

I'm not able to re-create this with a warmed-up (meaning I've built at least one model previously) `cmdstan` and `bridgestan`: CmdStan output ```shell $ /usr/bin/time make examples/bernoulli/bernoulli --- Translating Stan...

Yeah, that is the “precompiled header” file. I’m not sure why it would be one place but not the other, since the math library is the same in both

I think I have the precompiled header disabled in my cmdstan installation. It's a decent speedup, but if we do support it I would make the opposite choice of CmdStan...

I believe @rok-cesnovar knows more about pre-compiled headers/helped set them up for CmdStan. The doc for them always has lots of [scary warnings](https://gcc.gnu.org/onlinedocs/gcc/Precompiled-Headers.html), like > Some other command-line options starting...

For me the precompiled header takes compiling a model from 17s to 7s

Could we do something similar to the `_threads` trick and hack the compiler version number into the filename? How does the compiler know to use `model_header_threads.hpp.gch` when the C++ file...

Not "including the world" is a project we've been talking about in the math library since I joined the project (and probably before). A fair number of things in math...