Re: [PATCH] Trivial inliner cleanups

[Richard Guenther <rguenth at tat dot physik dot uni-tuebingen dot de>]

Mark Mitchell wrote:
> Richard Guenther wrote:
>
> > On Tue, 1 Mar 2005, Mark Mitchell wrote:
> >
> >
> > > Richard Guenther wrote:
> > >
> > > > This patch does some trivial cleanups to the inliner and related
> > > > code, namely exposing a local define to tree-inline.h and using
> > > > it in more places instead of 10; removing an unused define and
> > > > marking a local helper function inline (it's uses are such that
> > > > it will collapse to almost a NOP at most call-sites).
> > > >
> > > > Bootstrapped on x86_64-unknown-linux-gnu.
> > > >
> > > > Ok for mainline?
> > >
> > >
> > > No, because ...
> > >
> > >
> > > >      if (size < 0 || size > MOVE_MAX_PIECES * MOVE_RATIO)
> > > > !       *count += 10;
> > >
> > >
> > > ... that 10 isn't INSNS_PER_CALL; it's an estimate of how expensive the
> > > copy will be.  It should be INSNS_PER_<something else>.
> >
> >
> >
> > I thought that was, "for moves greater than MOVE_MAX_PIECES * MOVE_RATIO
> > we use memcpy", which is a call?
>
>
> You may be right!  However, if so, there's definitely a comment missing. 
> :-)  Does anyone else know where this "10" came from?

It was introduced to c-common.c rev. 1.427 by Jan - the "10" was there
from the beginning.  I might even suggest that the only left "10" for
DIV and MOD may also be accounted for a call, but those were added later.

Richard.