[MSP430] Update toolchain description for msp430-gcc-8.3.1.25

[atrosinenko]

This patch updates MSP430 toolchain description for current mspgcc package provided by TI. It does not try to be backward-compatible yet.

This patch expects the check for options::OPT_fexceptions to always return correct value based on -fexceptions, -fno-exceptions and the default value, that is probably not the case...

Actual object and archive file sequence is based on "deciphered" output of msp430-elf-gcc -dumpspecs and gcc/gcc/config/msp430/msp430.h source file. Still, it lacks some parts such as stack-smashing protector that seems to be supported on MSP430 according to presence of libssp.a. Of course, it omits all the sanitizer runtimes.

Provided uart-test.c contains the UART example from MSP430-Emulator documentation, both commands

/path/to/bin/clang -target msp430 --sysroot=/path/to/msp430gcc uart-test.c -v -mmcu=msp430g2553 -o uart-test
/path/to/bin/clang -target msp430 --sysroot=/path/to/msp430gcc uart-test.c -v -mmcu=msp430g2553 -o uart-test -fexceptions

produce executables that can be run on emulator by

• Uploading firmware
• dump 0xfffe
• set pc 0x<value of reset vector>
• run

Here, the --sysroot option points to TI's binary distribution:

$ cd /path/to/msp430gcc
 bin                 lib                        Revisions_MSPDebugStack.txt
 common              libexec                    rollbackBackupDirectory
 docs                msp430.dat                 share
 examples            msp430-elf                 Software_License_Agreement.pdf
'GPLv3 Notice.txt'   MSP430-GCC_manifest.html   uninstall
 include            'Release Notes.txt'         uninstall.dat
 install_scripts     Revisions_Header.txt       version.properties
$ ls include/
cc430f5123.h             msp430f5308.ld                    msp430fr2633.ld
cc430f5123.ld            msp430f5308_symbols.ld            msp430fr2633_symbols.ld
cc430f5123_symbols.ld    msp430f5309.h                     msp430fr2672.h
cc430f5125.h             msp430f5309.ld                    msp430fr2672.ld
$ ls msp430-elf/
bin  include  lib
...

[asl]

Please also updated the driver tests and submit as a separate PR.

[atrosinenko]

As I understood, OPT_fexception handling needs further rewriting as well: in clang/lib/Driver/ToolChains/Clang.cpp, there exists some non-trivial code to compute whether exceptions are enabled or not. For MSP430 backend, we have probably to use at least

  Args.hasFlag(options::OPT_fexceptions, options::OPT_fno_exceptions, false);

or better explicitly expressing the default state...

On the other hand, in GCC spec, this is described as %{fexceptions:crtend.o}%{!fexceptions:crtend_no_eh.o}, so my implementation is even correctly aligned with GCC one provided it has no special magic w.r.t. -fsomething/-fno-something handling built in.

Meanwhile, in clang/lib/Driver/ToolChains/Fuchsia.cpp, there exists nicer multilib description for roughly the same case.

[asl]

Right. But given that this is C++ stuff, please just fill an issue, we will take care of it with lower priority :)

[atrosinenko]

For the record, switching to *_no_eh.o versions by default (when no -fexceptions is explicitly given) should slightly decrease code size, so this PR seems to be not only "making it automatically discover newlib" :)

Now, I will update clang driver test, so what is the policy on supported sysroot versions? Are we planning choosing (or even autodetecting) TI's sysroot version to be used - now or in the near future?

[asl]

We need to test whatever is implemented