Comment by jcelerier
1 year ago
> so we want a way to tell them to not worry about the ABI and avoid wasting time on preserving registers across the call
that's what -fvisibility=internal already does, no?
1 year ago
> so we want a way to tell them to not worry about the ABI and avoid wasting time on preserving registers across the call
that's what -fvisibility=internal already does, no?
That’s what static could do (if the function’s address is not taken, or given sufficiently powerful dataflow analysis), but C and C++ compilers don’t take advantage of that. Getting that out of -fvisibility=hidden -flto would also be possible, but requires even more nonexistent compiler smarts. (From a quick web search, I can't figure out what internal visibility brings over hidden.)
(Granted, it’s not like this is completely impossible—I seem to remember GHC and MLton can invent custom calling conventions for Haskell and SML respectively. But the popular C or C++ compilers can’t.)
Really? I've always assumed that static gives C/C++ compilers the right to disregard calling conventions altogether. Now that I think about it, this assumption might be unjustified. Was there any strong reason for compilers to keep the calling convention?
I think that "static register" should make the compiler allowed to disregard calling conventions (if optimizations are enabled), but neither "static" nor "register" alone should do so. (In a discussion on an IRC some time ago, I had suggested what "register" alone should do in my opinion, which is something else than this.)
I mean, it does give them that right as far as I know [with, again, the caveat of function pointers—you can’t pass a non-ABI-compliant function to qsort(), however static it is—and an additional one of inline assembly], it’s just that they choose not to exercise it. Why? Not sure. Maybe it’s too difficult (register allocation and spilling is NP-complete and fairly expensive). Maybe it’s too difficult when your compiler starts out as a function-at-a-time one and builds on that. The popular ones don’t do that, is my point. (Would be interesting to check what SDCC does, by the way, on register-starved micros such tricks could make sense.)