Mx-Compiler

updated on 2022/12/5

documents

• commands
• semantic notes
• semantic checklist
• IR design
• raw IR checklist
• codegen notes
• raw codegen checklist

my test commands

IR test in debug/.

test.ll(generated by my compiler from test.mx) and builtin.ll are both x86-64 target.

• sh ir-test.sh
• sh ir-draw.sh

IR auto test in src/autotestspace.

The testIR flag in Compiler.java should be set to true.

• python3 ir-auto-test.py

codegen test in debug/.

• sh asm-riscv-gen.sh to generate test.s in risv32 target.
• sh codegen-test.sh use test.s, builtin.s and ravel to execute the program.

llvm commands

LLVM IR generation

• to .ll
  • clang -S -emit-llvm test.c
  • clang -S -emit-llvm test.c --target=riscv32
• to .bc
  • clang -c -emit-llvm test.c

link LLVM IR (to bitcode)

• llvm-link test.ll builtin.ll -o linked.bc

compile LLVM IR(.ll or .bc) to executable

• clang test.ll -o test
• clang test.bc -o test

execute program in LLVM IR format

• lli test.ll
• lli test.bc

compile LLVM IR to assembly

• llc test.ll -o test.s
• llc test.bc -o test.s

code generation (from .c to .s)

• clang -S test.c
• clang -S test.c --target=riscv32 -march=rv32im i means basic integer instructions, m means multiply and divide, which we will use in code-generation phase.

code generation (from .ll to .s)

• llc test.ll -o test.s -march=riscv32 -mattr=+m m tells the compiler not to use __mul but to use mul instruction.