On Sun, Mar 17, 2002 at 10:06:46PM -0800, Andy Ross wrote: > David Megginson wrote: > > The biggest surprise was that inlining methods made things slower, not > > faster, in most cases (there were a couple of exceptions). That may > > be a quirk of G++'s code generation, but it's probably worth > > considering -- I had inlined much of the infrastructure to try to > > speed things up, but then put it back out of line again piece by > > piece. > > It's probably not a quirk. Inlining actually helps very little except > for VERY small functions. It used to be that a function call was slow > -- you had to spill a bunch of registers and a return value onto the > stack, and then clean them up later. But modern processors can, for > the most part, stick all that mess into into a few extra pipeline > stages. > > And all the extra code takes up extra cache lines, that are critical > to performance on modern memories. So naively inlined code can > actually be slower. > This was discussed on lkml a while ago, and the consensus was that optimising for cache usage was more important than almost anything else on a modern chip - using -Os rather than -O2/3 can be better, at least for performance critical code, to minimise the cache footprint.
As for inline functions, they're not as much of a performance improvement as you might think, because modern compilers are very good at minimising the cost of function calls. This was also discussed on lkml . . . Cutting and pasting from the lkml thread: ------------------------------------------------------------ The first patch should be against Documentation/CodingStyle. What are we trying to achieve here? What are the guidelines for when-to and when-to-not? I'd say: - If a function has a single call site and is static then it is always correct to inline. - If a function is very small (20-30 bytes) then inlining is correct even if it has many call sites. - If a function is less-small, and has only one or two *commonly called* call sites, then inlining is OK. - If a function is a leaf function, then it is more inlinable than a function which makes another function call. ------------------------------------------------------------ Not sure how this translates to C++ inlining (I'm a C-only man at present), but it probably isn't too different. Simon -- PGP public key Id 0x144A991C, or ftp://bg77.anu.edu.au/pub/himi/himi.asc (crappy) Homepage: http://bg77.anu.edu.au doe #237 (see http://www.lemuria.org/DeCSS) My DeCSS mirror: ftp://bg77.anu.edu.au/pub/mirrors/css/
msg04155/pgp00000.pgp
Description: PGP signature
