Re: gcc3 "millicode" problems with sh3 (toolchain/22452)
"Valeriy E. Ushakov" <[email protected]> Wed, 3 Sep 2003 22:23:56 +0400
| Newsgroups | gmane.os.netbsd.ports.sh3,gmane.os.netbsd.devel.toolchain |
|---|---|
| Message-ID | <[email protected]> |
On Wed, Sep 03, 2003 at 10:22:55 -0700, [email protected] wrote: > > # gcc2 > > $ /usr/nb/tools/bin/shle--netbsdelf-gcc -dumpspecs | grep -e -lgcc > > %{!symbolic:%{!shared:-lgcc}%{shared:-lgcc_pic}} > > > > # gcc3 > > $ /usr/nb3/tools/bin/shle--netbsdelf-gcc -dumpspecs | grep -e -lgcc > > %{!shared:%{!symbolic: -lgcc}} > > I'm not sure that it would be right to do it, either. > > I believe in the new gcc world (not necessarily the new netbsd gcc > world; i've not been paying attention to netbsd's gcc integration > work), the right thing is to link with the GCC shared library. > > That has the effect of getting you one copy of the exception handling > functions and data structures, which is important (and works much > better than having independent copies of them 8-). That's what worries me too, but I don't know libgcc internals at all (the part that's responsible for the exception handling, &c). FWIW, Kaz Kojima (linux/sh gcc person) told me that's what linux do, that's all I know. > I'm not sure that linking libgcc into shlibs directly will do the > right thing w.r.t. this issue. Ok, then we do need a separate "millicode" library to be linked into shlibs (apparently hppa needs something like that as well). SY, Uwe -- [email protected] | Zu Grunde kommen http://www.ptc.spbu.ru/~uwe/ | Ist zu Grunde gehen