Notes of MiniDecaf

These are some notes when doing the experiment MiniDecaf.

1. Basis

Steps for compiling:

1. Lexical (relating to words) analysis (词法分析) → tokens
2. Syntactic Analysis (according to syntax) / Parsing (句法分析) → Abstract Syntax Tree
   • frontend/ast/tree.py: defines the nodes of AST
   • frontend/parser/ply_parser.py: define the rule of parser to generate AST nodes
3. Semantic (relating to meaning) Analysis (go through the AST, checking the type correctness, etc.)
   • frontend/typecheck/namer.py: the namer phase; resolve all symbols defined in the AST and store them in symbol tables
4. Generate intermediate code (e.g. three-address code) for code generation and optimization
   • utils/tac/tacop.py: define operations in TAC
   • frontend/tacgen/tacgen.py: generate TAC
5. Generate object code
   • utils/riscv.py: Convert TAC into RISCV code

2. Stage 1

2.1 Step 2

In step 2, we need only modify some codes related to TAC generation.

1. frontend/tacgen/tacgen.py :
   • expr.setattr("val", ...) sets the value of child to the temporary value of this node.
   • expr.op which is node.XXXopneeds to be projected to tacop.XXXop
2. utils/tac/tacinstr.py :
   • Complement the __str__ attribute of related TAC operations.

Overflow example: -(~2147483647).

2.2 Step 3

Modify frontend/tacgen/tacgen.py for all binary operations.

#include <stdio.h>

int main() {
  int a = 0x80000000;
  int b = -1;
  printf("%d\n", a / b);
  return 0;
}

❯ gcc -O0 test.c -o test
❯ ./test
[1]    72874 floating point exception  ./test
# same as divided by zero

❯ riscv64-unknown-elf-gcc -O0 test.c -march=rv32im -mabi=ilp32 -o test
❯ spike --isa=RV32G /usr/local/bin/pk test
bbl loader
-2147483648

After enabling optimization:

• gcc on macOS gives random values.
• qemu of RISCV gives -2147483648.

---

Last update: October 17, 2021

Authors: Co1lin