Re: Caller allocates space for callee-save registers

> our moderator writes:
>[It is my impression that with a reasonably smart compiler stack windows
>are a big lose because the compiler can usually cut the register save
>traffic to memory far below what windows do...


John Mashey (of MIPS) commented on this a few years ago, in a talk and in
response to a question I asked him about it afterward. What he said was
roughly:


1. For languages like C on a register-rich machine, careful choice of
calling conventions plus a smart compiler can essentially eliminate the
performance advantages of windows for real programs.


2. For languages like C++, with very dynamic calling patterns not readily
predictable at compile time, MIPS's studies say it would be moderately
helpful to have a few register windows... not helpful enough, in their
opinion, to justify the costs.


He also noted the existence of an undetected geological fault in
California, the "register windows fault": if you plot the locations of
RISC companies in Silicon Gulch on a map, all the windowed architectures
are on one side of a slightly crooked line, and all the non-windowed ones
are on the other side! (This situation may have changed since then, of
course...)


Also, I don't *think* he said this in so many words, but it's a simple
corollary of things he did say... Calling patterns matter a lot, and they
aren't always what you think, or what you measure from a handful of
benchmarks. The Unix kernel, for example, is a dreadful worst case for
everything from calling conventions to MMUs: it typically makes a straight
plunge down level after level of the call graph, fiddles around for an
instant near the bottom, and then zips straight back up. It's important
that machines and compilers cope reasonably well with bad cases, not just
typical ones.
--
Henry Spencer @ U of Toronto Zoology, henry@zoo.toronto.edu utzoo!henry
--

Return to the comp.compilers page.
Search the comp.compilers archives again.