On 20090131.1659+0100, Vincent Sanders <vince@kyllikki.org> wrote:
[This]patch is a bit more involved. The patch addresses a gcc
regression in the 4.3 series (specifically this patch is against 4.3.2
which does *not* have a lot of other issues which affect kernel building)

GCC bug tracker has this issue as
#38453 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38453
#32044 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32044

comment 65 of #32044 has the fix being applied to gcc trunk as revision #142719

The attached patch is a backport to gcc 4.3.2 which allows this
version to be used to generate correct output for various ARM kernel
build (and indeed is teh correct answer in general).

/trunk/patches/gcc/4.3.2/360-fix-expensive-optimize.patch | 207 207 0 0 +++++++++++++++++++++
1 file changed, 207 insertions(+)

http://mirror.sh-linux.org/rpm-2003/SRPMS/gcc-3.2.3-3.src.rpm contains the following patches:

gcc-20030210-sh-linux-1.patch

gcc-3.2.3-libffi-1.patch

gcc-3.2.3-sh-linux-dwarf2-1.patch (*not* applied by the spec file, it's in there by accident)

gcc-3.2.3-libffi-1.patch was needed just to build, I think.

After that was applied, sh4 gcc seemed to compile fine, but c++ programs

failed to execute because libstdc++.so.5 was built without version

info.  This was caused directly by libstdc++-v3/configure setting

SYMVER_MAP=config/linker-map.dummy because it sees that 

no libgcc_s.so was generated; configure says

  checking for shared libgcc... no.

Applying gcc-20030210-sh-linux-1.patch in hopes it makes those problems go away.