Add support for compiler-rt

[dmikushin]

Add MSP430 architecture support into compiler-rt build system and into the actual set of builtins.

[dmikushin]

Adding a new arch into compiler-rt CMake files is indeed quite easy: dmikushin/msp430-compiler-rt@18db1f7

It's not that trivial though to get the thing building. Both clang and msp430-elf-gcc currently refuse to pass entry checks, e.g.:

-- Check for working C compiler: /home/marcusmae/acctek/msp430-clang-newlib/install/bin/clang -- broken
CMake Error at /usr/share/cmake-3.10/Modules/CMakeTestCCompiler.cmake:52 (message):
  The C compiler

    "/home/marcusmae/acctek/msp430-clang-newlib/install/bin/clang"

  is not able to compile a simple test program.

  It fails with the following output:

    Change Dir: /home/marcusmae/acctek/msp430-clang-newlib/msp430-clang/src/llvm/projects/compiler-rt/CMakeFiles/CMakeTmp
    
    Run Build Command:"/usr/bin/make" "cmTC_602d7/fast"
    /usr/bin/make -f CMakeFiles/cmTC_602d7.dir/build.make CMakeFiles/cmTC_602d7.dir/build
    make[1]: Entering directory '/home/marcusmae/acctek/msp430-clang-newlib/msp430-clang/src/llvm/projects/compiler-rt/CMakeFiles/CMakeTmp'
    Building C object CMakeFiles/cmTC_602d7.dir/testCCompiler.c.o
    /home/marcusmae/acctek/msp430-clang-newlib/install/bin/clang --target=msp430-unknown-unknown    -o CMakeFiles/cmTC_602d7.dir/testCCompiler.c.o   -c /home/marcusmae/acctek/msp430-clang-newlib/msp430-clang/src/llvm/projects/compiler-rt/CMakeFiles/CMakeTmp/testCCompiler.c
    Linking C executable cmTC_602d7
    /usr/bin/cmake -E cmake_link_script CMakeFiles/cmTC_602d7.dir/link.txt --verbose=1
    /home/marcusmae/acctek/msp430-clang-newlib/install/bin/clang --target=msp430-unknown-unknown      -rdynamic CMakeFiles/cmTC_602d7.dir/testCCompiler.c.o  -o cmTC_602d7 
    clang-8: warning: argument unused during compilation: '-rdynamic' [-Wunused-command-line-argument]
    /home/marcusmae/acctek/msp430-clang-newlib/install/bin/msp430-elf-ld: cannot find crtbegin.o: No such file or directory
    /home/marcusmae/acctek/msp430-clang-newlib/install/bin/msp430-elf-ld: cannot find -lmul_none
    /home/marcusmae/acctek/msp430-clang-newlib/install/bin/msp430-elf-ld: cannot find -lgcc
    clang-8: error: msp430-elf-ld command failed with exit code 1 (use -v to see invocation)
    CMakeFiles/cmTC_602d7.dir/build.make:97: recipe for target 'cmTC_602d7' failed

We should decide which compiler we should aim for as a compiler-rt builder: clang or gcc?

[dmikushin]

Aaand the correct CMake command line to build with our clang is:

cmake -DBUILD_SHARED_LIBS=0 -DCMAKE_BUILD_TYPE=Debug -DLLVM_DEFAULT_TARGET_TRIPLE=msp430-elf -DLLVM_TARGETS_TO_BUILD=MSP430 -DCMAKE_INSTALL_PREFIX=$ROOT/install -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ -DCMAKE_TRY_COMPILE_TARGET_TYPE=STATIC_LIBRARY -DCOMPILER_RT_BAREMETAL_BUILD=TRUE -DCOMPILER_RT_OS_DIR="" ..

Two types of generic builtins are disabled:

• TLS (via COMPILER_RT_BAREMETAL_BUILD)
• UNWIND (compiler crashes)

By setting COMPILER_RT_OS_DIR to empty string, we remove useless OS subfolder and place the resulting library directly to ${INSTALL_PREFIX}/lib/libclang_rt.builtins-msp430.a.

[dmikushin]

Here is a table of all generic bultins of compiler-rt:

Atomics
__atomic_load_c
__atomic_store_c
__atomic_compare_exchange_c
__atomic_exchange_c

Bitwise
__clz{d|s|t}i2
__ctz{d|s|t}i2
__parity{d|s|t}i2	1 if number of bits is odd else 0
__popcount{d|s|t}i2	count of 1 bits
__ffs{d|s|t}i2	return position of the least significant bit set to 1
__ashl{d|t}i3
__ashr{d|t}i3
__lshrdi3
__lshrti3

Conversions
__extenddftf2
__extendhfsf2, __gnu_h2f_ieee
__extendsf{d|t}f2
__fixdf{d|s|t}i
__fixsf{d|s|t}i
__fixtf{d|s|t}i
__fixunsdf{d|s|t}i
__fixunssf{d|s|t}i
__fixunstf{d|s|t}i
__fixunsxf{d|s|t}i	convert a to a unsigned integer, rounding toward zero (negative values become zero)
__fixxf{d|t}i	convert a to a signed long long, rounding toward zero
__floatdi{d|s|t|x}f	convert a to a floating-point, rounding toward even
__floatsi{d|s|t}f
__floatti{d|s|t}f	convert a to a double, rounding toward even
__floatundi{d|s|t|x}f	convert a to a double, rounding toward even
__floatunsi{d|s|t}f
__floatunti{d|s|t|x}f

Rounding
__truncdf{h|s}f2
__truncsfhf2
__trunctf{d|s}f2

Trampoline
__gnu_unwind_frame
__gcc_personality_sj0
__gcc_personality_v0

Cache
__clear_cache	invalidate the instruction cache for the specified range

Comparison
__cmp{d|t}i2
__eqdf2
__gedf2	-1 if a < b, 0 if a == b, 1 if a > b, -1 if either a or b is NaN
__ledf2	-1 if a < b, 0 if a == b, 1 if a > b, 1 if either a or b is NaN
__gtdf2
__ltdf2
__nedf2
__unorddf2	returns 0 if both a and b are numbers; 1 if either a or b is NaN
__ucmp{d|t}i2	if (a < b) returns 0; if (a == b) returns 1; if (a > b) returns 2

Arithmetic
__absv{d|s|t}i2
__add{d|s}f3	soft-float addition with the IEEE-754 default rounding (to nearest, ties to even)
__addv{d|s|t}i3
__bswap{d|s}i2
__divd{c|f|i}3
__divmod{d|s}i4
__divs{c|f|i}3
__divt{c|f|i}3
__divxc3	the quotient of (a + ib) / (c + id)
__mod{d|s|t}i3	a % b
__mul{d|s|t}c3	(a + ib) * (c + id)
__mul{d|s|t}f3
__muldsi3	a * b
__mul{d|t}i3	a * b
__mulo{d|s|t}i4	a * b, sets overflow to 1 if a * b overflows
__mulddi3
__mulv{d|s|t}i3
__mulxc3
__negdi2	-a
__neg{d|s|t}f2	-a
__negv{d|s|t}i2	-a, aborts if -a overflows
__powi{d|s|t|x}f2	a ^ b
__sub{d|s|t}f3	flip the sign bit of b and add
__subv{d|s|t}i3	a - b, aborts if a - b overflows
__udivmod{d|s|t}i4	a / b, if rem != 0, *rem = a % b
__udiv{d|s|t}i3	a / b
__umod{d|s|t}i3	a % b