[PATCH] GAS/MIPS: Add `-mfix-r5900' option for the R5900 short loop errata

[Maciej W. Rozycki]

Hi Fredrik,

> >  What target was that with though?
> 
> It was "mipsr5900el-unknown-linux-gnu".

 Please mention that when making submissions.

 I pushed your change through testing with my usual set of MIPS targets 
and there were no issues, so things are looking good.  Sometimes tweaks 
are needed to otherwise reasonable test cases because of subtle 
differences between various MIPS targets, for example such as these with 
section alignment between Linux and bare metal ELF targets causing output 
not to match the patterns expected.

 You could include a bare metal ELF target in your testing, e.g. 
`mipsr5900el-elf' would be an obvious choice in your case, to avoid 
surprises.

> >  But are these actual regressions or preexisting failures?
> 
> These are preexisting failures.

 There's no need to mention preexisting failures (unless you're fixing 
them).  Also I seem to remember now our discussion on these CP0 register 
disassembly issues.

> Updating to HEAD this morning yields
> two additional ones:
> 
> FAIL: MIPS MIPS64 MIPS-3D ASE instructions (-mips3d flag)
> FAIL: MIPS MIPS64 MDMX ASE instructions (-mdmx flag)

 Weird.

> As mentioned a while ago, there is also a preexisting build error with
> the GDB part of Binutils:

 GDB is actually a separate project, although it shares the repository.  
In the old days of the `src' CVS repo even more projects shared the tree, 
and you'd normally check out just the CVS module you were interested in.  
You can't do this with GIT, so instead you can omit non-binutils parts 
from the build by adding `--disable-gdb --disable-libdecnumber 
--disable-readline --disable-sim' to the `configure' invocation (you can 
actually name arbitrary top-level subdirectories to exclude this way).

 Likewise `--disable-binutils --disable-gas --disable-gold --disable-gprof 
--disable-ld' will build the GDB parts only of the repository.

> I just filed a bug report for that here:
> 
> https://sourceware.org/bugzilla/show_bug.cgi?id=23835

 Thanks!

  Maciej