PGO testing on refactor branch?

Discussion
#1

Hey folks, I’m doing some refactoring that theoretically shouldn’t impact performance (and doesn’t, on my system) but could cause compilers to optimize more poorly. Anyone who has a setup for running PGO builds and/or cares about performance, could you run some tests on my vmi-refactor branch?

I’ve pulled a few files out of the pl-wam.c compilation unit and moved a couple functions into pl-inline.h. Outcomes should be identical to current master builds, but there might be differences if the compiler suddenly finds itself unable to inline a hot function.

Thanks!

#2

There is no measurable impact :slight_smile: Some of the stuff should probably be moved to the inline version though. Think about the “GENERIC INDIRECT OPERATIONS” functions (which should also avoid GET_LD). This doesn’t show in the default benchmark test as there are no string or bignum manipulations and I doubt there are even floats in there (our indirect types).

If you clean this up it might be wise to split it into a file doing the heap allocations and one doing the stack allocations. Don’t ask me how these ended up in the same place :slight_smile: If you create files you need to add the file header. Without I get complaints from the Linux package managers …