This is the mail archive of the
mailing list for the libc-ports project.
Re: [patch, mips] Improved memset for MIPS
- From: "Joseph S. Myers" <joseph at codesourcery dot com>
- To: Steve Ellcey <sellcey at mips dot com>
- Cc: <libc-ports at sourceware dot org>
- Date: Fri, 6 Sep 2013 16:59:33 +0000
- Subject: Re: [patch, mips] Improved memset for MIPS
- Authentication-results: sourceware.org; auth=none
- References: <93a232b5-9d0b-4a27-bbb5-16e3ae7c4b89 at BAMAIL02 dot ba dot imgtec dot org> <Pine dot LNX dot 4 dot 64 dot 1309061430150 dot 5886 at digraph dot polyomino dot org dot uk> <1378483039 dot 5770 dot 302 dot camel at ubuntu-sellcey> <Pine dot LNX dot 4 dot 64 dot 1309061603380 dot 8532 at digraph dot polyomino dot org dot uk> <1378486241 dot 5770 dot 327 dot camel at ubuntu-sellcey>
On Fri, 6 Sep 2013, Steve Ellcey wrote:
> I have found that --prefix=/usr is more of a problem then a help when
> building general cross compiler toolchains. Using a prefix of /usr
> triggers various special case code in
> ports/sysdeps/unix/sysv/linux/mips/configure to put things in lib32 and
> lib64 and I don't actually want any of that so I use a prefix
> of /usr/fake instead of /usr.
Not using --prefix=/usr runs into ABI testsuite problems with bug 14664.
> The "undefined reference to `__libc_global_ctors'" has just shown up
> again in a parallel build, but it seems to go away when I rebuild. I am
> still trying to understand what is going on with this.
Since it seems to be about parallel builds and linkobj/libc.so, try with
<https://sourceware.org/ml/libc-alpha/2013-08/msg00597.html>? (Which, if
it works OK on master for a while, should probably be backported to 2.18
> > The expectation is that the glibc testsuite is the normal way to test
> > patches before submission, and string function patches like this need it
> > to be run for all six relevant ABI variants.
> I think some flexibility here would be better. There is no floating
> point code in this routine so running all three ABI's in both hard and
> soft float modes seems like overkill to me. Testing in big and little
> endian modes would seem more likely to turn up problems then testing in
> hard and soft float.
That's why I said six rather than twelve ABI variants; floating-point
variants aren't relevant here, but the other variants are. You could
argue for full testing for three variants that cover all of o32, n32 and
n64 and both BE and LE, plus just the string/ directory tests for the
other three variants, but I think that would be the minimum.
Joseph S. Myers