Path: utzoo!utgpu!jarvis.csri.toronto.edu!rutgers!usc!brutus.cs.uiuc.edu!wuarchive!uwm.edu!ux1.cso.uiuc.edu!tank!ncar!boulder!grunwald
From: grunwald@foobar.colorado.edu (Dirk Grunwald)
Newsgroups: gnu.g++.lib.bug
Subject: Re: lack of inline functions in libg++
Message-ID: <13193@boulder.Colorado.EDU>
Date: 26 Oct 89 16:08:23 GMT
References: <8910261030.AA01450@g.oswego.edu>
Sender: news@boulder.Colorado.EDU
Reply-To: grunwald@foobar.colorado.edu
Distribution: gnu
Organization: University of Colorado at Boulder
Lines: 18
In-reply-to: dl@G.OSWEGO.EDU's message of 26 Oct 89 10:30:30 GMT


I think that a lot of the effort spent on ``cleaning up inlining'' &
the like would be better spent on global optimization, although that
process is much harder.

I'm thinking of things along the lines of the compilation to symbolic
intermediate (rtl?), and later ``linking'' of the intermediate form
followed by a global procedure integration & register allocation pass.

Examples of this include the MIPS compilers (using ucode as
intermediate) and the work at DEC-WRL for the titan (mahler
intermediate code).

This allows much better procedure integration and you'll improve
register allocation in leaf procedures. Also, it would make G++ much
nicer because your header files would only include class declerations
- there would be not need to include inline procedure bodies in the
header files.