Hi Marcus:
> I could also disable the -mlittle-endian and -mbig-endian options if
> the binutils is too old, but it's questionable if getting a "I don't
> understand -mbig-endian" from gcc is more useful than getting it from
> gas. And it would be more work for the user to fix it since then they
> would have to also rebuild gcc after upgrading binutils.
Sounds reasonable to me, I am ok with keeping it as it is without
binutils change.
I tried to run with gcc testsuite on spike + pk, seems got a bunch
extra fail cases,
could you take a look for that?
========= Summary of gcc testsuite =========
| # of unexpected case / # of unique unexpected case
| gcc | g++ | gfortran |
rv64gc/ lp64/ medlow | 476 / 109 | 131 / 25 | - |
Makefile:914: recipe for target 'report-gcc-newlib' failed
I've setup test env via riscv-gnu-toolchain, you can reproduce by
following command:
$ git clone g...@github.com:riscv/riscv-gnu-toolchain.git -b big-endian
riscv-gnu-toolchain-be
$ cd riscv-gnu-toolchain-be
$ mkdir build && cd build
$ ../configure --prefix=`pwd`/install --with-arch=rv64gc
--with-multilib-generator="rv64gc-lp64--"
$ make report SIM=spike -j`nproc`
On Sat, Jan 30, 2021 at 3:54 AM Marcus Comstedt <mar...@mc.pp.se> wrote:
>
>
> Hi Kito,
>
> Kito Cheng <kito.ch...@gmail.com> writes:
>
> > You can add a check in configure.ac and config.in to detect whether
> > binutils is supported or not.
> > here is example, search HAVE_AS_MISA_SPEC in this patch:
> > https://github.com/gcc-mirror/gcc/commit/4b81528241ca682025d92558ff6aeec91dafdca8
>
> Ok, but I specifically _don't_ want to require a newer binutils for
> the little endian triplets (unless there is also some other RISC-V
> change meaning that the newer binutils is needed anyway, in which case
> this point becomes moot).
>
> I could check the binutils version in the case of big endian triplets
> only, but since older binutils can't be built for those triplets
> anyway (due to config.sub), I don't think it would do much.
>
> I could also disable the -mlittle-endian and -mbig-endian options if
> the binutils is too old, but it's questionable if getting a "I don't
> understand -mbig-endian" from gcc is more useful than getting it from
> gas. And it would be more work for the user to fix it since then they
> would have to also rebuild gcc after upgrading binutils.
>
>
> // Marcus
>
>
