Again, that's a strawman. I'm just looking for suggestions about what
we might to do -- or even feedback that there's no need to do anything.
This isn't really suitable for an automated tester, but what I used to do was
keep around some .i files of some version of some compiler files (I think it
was reload.i, combine.i, and loop.i) along with .s files for some
architectures and every so often make new .s files and compare them to see if
something got better or worse. Of course, with modern machines, manually
trying to decide if assembler code is better or worse isn't too practical,
but it did detect if some optimizations (e.g., strength reduction) got turned
off. It also forced something that all too few people do: look at generated
code on a routine basis with an eye to performance.
As I said, I don't see a way to automate this, but even now feel that this
sort of thing would have some value.
