cvs commit: src/sys/kern kern_sig.c

[Marcel Moolenaar]

On Mon, Mar 07, 2005 at 01:47:42PM -0700, Scott Long wrote:
> David O'Brien wrote:
> >On Thu, Mar 03, 2005 at 07:06:24AM -0700, Scott Long wrote:
> >
> >>a bit.  Also, there is talk about increasing the default kstack size due
> >>to all of the extra inlining that the compiler does with the -O2 option
> >
> >
> >I'd love more details on the extra inlining people are seeing with -O2.
> >(i.e. specifics)  -O2 is not supose to do extra function inlining.  That
> >is suppose to be a -O3 thing.
> >
> >From the GCC manual:
> >
> >-O3 Optimize yet more.
> >    -O3 turns on all optimizations specified by -O2 and also turns on the
> >    -finline-functions, -fweb and -frename-registers options.
> >
> >-O2 Optimize even more.
> >    GCC performs nearly all supported optimizations that do not involve a
> >    space-speed tradeoff. The compiler does not perform loop unrolling or
> >    function inlining when you specify -O2. As compared to -O, this
> >    option increases both compilation time and the performance of the
> >    generated code.
> >    ..snip..
> >
> >The -O2 options that affect size are:
> >    -falign-functions  -falign-jumps  -falign-loops -falign-labels
> >    -freorder-blocks -fprefetch-loop-arrays
> 
> Talk to Bill Paul about the massive amount of inlining that is happening
> in the ieee80211 ioctl code, and the no_inline directives he had to use
> to kludge around it.

gcc(1) does inlining of functions labeled with inline at -O1.
Functions not labeled with inline will be inlined at -O2.
Both cases are subject to preconditions (i.e. complexity of
the function being inlined).

In other words: the gcc(1) manual is inaccurate.

-- 
 Marcel Moolenaar	  USPA: A-39004		 marcel at xcllnt.net