From: Paul M. <le...@li...> - 2003-06-15 17:05:24
|
On Sun, Jun 15, 2003 at 01:52:10PM +0100, Adrian McMenamin wrote: > I am using gcc 3.0.4 which is broken (won't build shared in userland) and= =20 > while gcc 3.3 is meant to support sh4 without patching, I userstand that = it=20 > too is buggy. >=20 3.3 and 3.4 are still both broken to some degree.. the lack of -mno-implici= t-fp is the primary concern. Though using -m4-nofpu and implicitly setting __SH4= __ does work to some extent (well, and undefining __sh3__, since we're not at = all interested in the sh3 ABI). I've also run into both register and stack corruption with 3.4 when using things like __builtin_return_address(). Presumably most of these things should be fixed before the branch is frozen. > Therefore I want to use 3.2 (assuming that works - I think it does). But = where=20 > are the patches? Or should I be using a different version? >=20 3.2 has been working fine for me for quite awhile. 3.2.2 is also rumored to= work. I've posted the 3.2 patch at: =09 http://www.stampede.org/~lethal/gcc-3.2-sh.diff Most of this is derived from the uClibc toolchain patches. It also looks like Kojima-san has a 3.2.2 patch at: http://dodo.nurs.or.jp/~kkojima/gnu-on-sh/gcc-3.2.2-tmp.diff.gz Though I have no idea how reliable it is. Optionally, Sugioka-san has prebuilt 3.2.3 rpms at: http://www.sh-linux.org/rpm-index-2003/index.html Again, I'm not sure how well these work, as I've not tested them myself. |