Richard Sandiford wrote:
Mark Mitchell <[EMAIL PROTECTED]> writes:
Andrew Pinski wrote:
Change powerpc-unknown-linux-gnu to powerpc64-unknown-linux-gnu so that
we also require the 64bit of PowerPC to work.
To be clear, you're suggesting that we say
"powerpc64-unknown-linux-gnu", but mean that both it's 32-bit and 64-bit
modes should work?
That makes sense to me. What about MIPS/MIPS64?
I'm glad you asked ;) Although you probably won't be.
I must admit I've always felt mips-elf to be a less-than-ideal
replacement for mips-sgi-irix6.5. The former is 32-bit only, while the
latter includes o32, n32 and n64, giving both 32-bit and 64-bit coverage.
I fully agree with removing mips-sgi-irix6.5 as a primary platform,
but I think mipsisa64-elf might be a better replacement than mips-elf.
By default, mipsisa64-elf includes libraries for {-mips32,-mips64} x
{-EL,-EB} x {-msoft-float,-mhard-float}. (And FWIW, I tend to test
all eight, and the different multilibs show up different problems.)
I take David's point about mips{,el}-linux-gnu being another alternative.
I suppose mipsisa64-elf has the advantage of being a simulator target
than anyone can test. On the other hand, mips{,el}-linux-gnu allows
proper Fortran, libstdc++ and Java testing, for instance. I can see
arguments both ways. However, by the same reasoning as above, I'd prefer
mips64{,el}-linux-gnu over mips{,el}-linux-gnu because the former includes
n32 and n64 as well as o32.
I'm not disagreeing and think that in general, it is better to make sure
there are primary or secondary targets that have lots of multilibs. These
tend to trip odd problems.
FWIW in general, *-rtems targets also usually build a larger number of
multilibs than the regular embedded targets. And since there is tasking
with POSIX threads, it is possible to run Ada tasking tests. RTEMS
supports a variety of target architectures and most of the architectures
have a board support package for a free simulator that is "good enough"
to run all the tasking tests.
RTEMS also uses newlib and bootstraps the C library as part of
building the compiler which further exercises gcc.
I am not proposing any changes for 4.3 platforms but think it would
be worth considering using RTEMS targets as an alternative to the
bare metal CPU-elf targets at some point in the future to broaden the
coverage of testing for cross-embedded environments.
At the current time, we build arm, h8300, i386, m68k, mips, powerpc,
sh, and sparc toolsets. An NIOS II port has been merged and a Blackfin
port should be merged in the near future. We have a tic4x port that we
have locked down onto an older gcc. There is a partial port to the AVR.
--joel
Richard
