Support for RISCV-32 about selfie

[ckirsch]

This is nice. Thanks a lot! Feel free to open a PR. I will then review that and see what we can do to get this into the repo.

[andreasbaumann]

Just wanted to keep you up to date, there is now a combined branch supporting 64-bit and 32-bit
in parallel, see https://github.com/andreasbaumann/selfie/tree/risccombi.

In short:

make ARCH=RISCV64 selfie
gcc -O0 -m64 -g -Duint_t='unsigned long long' \
	-DARCHITECTURE=1 \
	-DCPUBITWIDTH=64 -DREGISTERSIZE=8 \
	-DSIZEOFUINT=8 -DSIZEOFUINTSTAR=8 \
	-DWORDSIZE=4 -DWORDSIZEINBITS=32 \
	-DINSTRUCTIONSIZE=4 -DALIGNBITS=3 \
	selfie.c -o selfie
make ARCH=RISCV32 selfie
gcc -O0 -g -m32 -D'uint_t=unsigned int' \
	-DARCHITECTURE=2 \
	-DCPUBITWIDTH=32 -DREGISTERSIZE=4 \
	-DSIZEOFUINT=4 -DSIZEOFUINTSTAR=4\
	-DWORDSIZE=4 -DWORDSIZEINBITS=32 \
	-DINSTRUCTIONSIZE=4 -DALIGNBITS=2 \
	selfie.c -o selfie

So, the basic ideas and thoughts are:

• I'm assuming that the address space is addressed with the dominant integer of the processor,
  32-bit on RISCV32, 64-bit on RISCV64.
• Provide platform constants when compiling selfie on the host, use a typedef to the one datatype
  (uint32_t or uint64_t). Once built, selfie is only able to run and compile on the same platform
  (I didn't manage to get it capable of cross-compiling).
• There is a new constant ALIGNBITS (representing the 2^3 or 2^2 when doing pointer
  arithmetic). All other constants where there before and are simply used from outside when
  building the corresponding platforms
• Another new constand ARCHITECTURE (1=RISCV64, 2=RISCV32) is used to do some setup
  correctly (in set_compile_architecture to populate the compiler with the correct global symbols).
• There is no need for a pre-processor.
• target for RISCV32 is RVE32I (with 32 registers), reducing to 16 registers poses some problems
  when allocating registers for certain functions.
• LW/LD and SW/SD are the only opcodes which differ, I made a pair LX/SX which I'm setting
  correctly depending on the target architecture (same for print_sx, etc.)
• I didn't go into the SAT-solver or formal verification, things might not be adapted there (I'm
  really not good at this kind of stuff :-) )
• String constants are padded to 4 or 8 bytes, 8 bits per char
  hard-coded (this looks a little bit like in the 'B' language. Not sure if
  a string library is overkill.
• the ELF32 and ELF64 headers are constructed in two completely different functions as
  the format is quite different.
• avoid 32-bit overflow of adress spaces on 32-bit:
  uint32_t VIRTUALMEMORYSIZE = 4294963200; // 4GB - one page of virtual memory
  decrease 2^32 by one PAGE (4k)
• the makefile has a ARCH variable to build and run the two platforms (make ARCH=RISCV32 and make ARCH=RISCV64). This requires a GNU make.
• I couldn't test spike yet as 'pk' is not really available (yet) for RISCV32
• escape.c and hello-world*.c shows trouble ahead with string constants and arithmentic on
  strings. There might be the need for string functions in the library. The registerrize (4, 8)
  should maybe be a REGISTERSIZE available via libcstar.
• symbolic.c: malloc(4) or malloc(8) => malloc(REGISTERSIZE).
• As the code differs quite heavily in a diff, there is an easy way to do some sedfu on
  the master version to get a version of selfie.c which is easier comparable to the
  risc32 or risccombi version.

# convert 64 -> 32
sed -i 's/uint64_t/uint32_t/g' 64.c
sed -i 's/SIZEOFUINT64/SIZEOFUINT32/g' 64.c
sed -i 's/SIZEOFUINT64STAR/SIZEOFUINT32STAR/g' 64.c
sed -i 's/UINT64_MAX/UINT32_MAX/g' 64.c
sed -i 's/SD/SW/g' 64.c
sed -i 's/LD/LW/g' 64.c
sed -i 's/_ld/_lw/g' 64.c
sed -i 's/_sd/_sw/g' 64.c
sed -i 's/UINT64_T/UINT32_T/g' 64.c
sed -i 's/UINT64STAR_T/UINT32STAR_T/g' 64.c
sed -i 's/INT64_MAX/INT32_MAX/g' 64.c
sed -i 's/INT64_MIN/INT32_MIN/g' 64.c
sed -i 's/64-bit/32-bit/g' 64.c
sed -i 's/SYM_UINT64/SYM_UINT32/g' 64.c
sed -i 's/F3_LD/F3_LW/g' 64.c
sed -i 's/F3_SD/F3_SW/g' 64.c

# convert 64 -> combi
sed -i 's/uint64_t/uint_t/g' selfie.c
sed -i 's/SIZEOFUINT64/SIZEOFUINT/g' selfie.c
sed -i 's/SIZEOFUINT64STAR/SIZEOFUINTSTAR/g' selfie.c
sed -i 's/UINT64_MAX/UINT_MAX/g' selfie.c
sed -i 's/SD/SX/g' selfie.c
sed -i 's/LD/LX/g' selfie.c
sed -i 's/_ld/_lx/g' selfie.c
sed -i 's/_sd/_sx/g' selfie.c
sed -i 's/UINT64_T/UINT_T/g' selfie.c
sed -i 's/UINT64STAR_T/UINTSTAR_T/g' selfie.c
sed -i 's/INT64_MAX/INT_MAX/g' selfie.c
sed -i 's/INT64_MIN/INT_MIN/g' selfie.c
sed -i 's/SYM_UINT64/SYM_UINT/g' selfie.c
sed -i 's/F3_LD/F3_LX/g' selfie.c
sed -i 's/F3_SD/F3_SX/g' selfie.c

• Also note: I didn't adapt the risc32 branch.