NetNews Usenet Archive 1992 #18/NN_1992_18.iso/spool/comp/os/linux/8419

Internet Message Format | 1992-08-16 | 1.8 KB

Path: sparky!uunet!cis.ohio-state.edu!zaphod.mps.ohio-state.edu!sdd.hp.com!mips!carbon!news.cs.indiana.edu!umn.edu!msus1.msus.edu!msus1.msus.edu!nop
Newsgroups: comp.os.linux
Subject: Re: shared libs - can everyone be happy with this?
Message-ID: <[e-mail]>
From: [e-mail] (Jay A. Carlson)
Date: 16 Aug 92 22:00:27
References: <[e-mail]> <[e-mail]>
Organization: Mankato State University
Nntp-Posting-Host: theory.cs.mankato.msus.edu
In-reply-to: [e-mail]'s message of Sat, 15 Aug 92 04:24:20 GMTLines: 24
Lines: 24

In article <[e-mail]> [e-mail] (Hongjiu Lu) writes:

   I am not sure if we are thinking the same thing. In up-coming gcc 2.2.2d,
   which I am planning to release this Sunday. There is only one version of the
   shared images, which is used by both jump table and good, old shared lib.
   If you use -jump in LDFLAGS, your executables are linked with jump table. The
   good, old shared lib is used by default, which uses the same shared images
   but bypasses the jump table.

Could the jumptables be set up to look where they were called from and
then patch their target into the instruction from whence they were
called?  There then would only be overhead on the first call from each
point.

This would lose when functions were called through registers.  I don't
know how much gcc tries to do this and how much user programs make use
of libraries calls through variables.  I certainly use int (*foo)();
often in my programs but never when it points to library calls.

For such calls, the overhead of identifying the instruction as neither
a direct jump or subroutine call would be incurred each time.

Jay Carlson
[e-mail]