Lava

A compiler of C subset by USTB OWL Wheel Lab.

Lava can compile extended SysY languate to ARM assembly and LLVM IR.

Building

Lava requires:

• cmake 3.20 or later.
• C++ compiler supporting C++23 standard.

git clone https://github.com/ustb-owl/Lava
cd compiler && mkdir build && cd build
cmake .. && make -j8
./lacc -h
ninja format

Usage

Usage: lacc [OPTIONS] [input]

Positionals:
    input                      input filename

Options:
    -o, --output               set output file name
        --emit                 emit ast, ir, asm, or cfg
    -T, --dump-ast             dump AST
    -I, --dump-ir              dump LLVM IR
    -S, --dump-asm             dump assembly
    -C, --dump-cfg             dump CFG
    -O, --opt-level            optimization level (0-2)
        --no-ra                disable register allocation
        --list-passes          list middle-end transform passes
        --print-pipeline       print active middle-end pipeline
        --time-passes          print middle-end pass timings
        --disable-pass         disable one or more middle-end passes
        --run-pass             run only the listed middle-end passes
        --start-after          start the pipeline after the named pass
        --stop-after           stop the pipeline after the named pass
        --dump-ir-before       dump LLVM IR before the named pass
        --dump-ir-after        dump LLVM IR after the named pass
        --dump-ir-dir          directory for IR dumps
    -h, --help                 show this help message

Examples:

./lacc foo.sy -O2 --emit=ir
./lacc --list-passes
./lacc --print-pipeline -O2
./lacc foo.sy -O2 --disable-pass=local-gvn --emit=ir
./lacc foo.sy -O2 --stop-after=mem2reg --dump-ir-after=mem2reg

Features

• Handwritten frontend.
• Strong typed IR in SSA form.
• Can emit LLVM IR
• Optimizer based on pass and pass manager.
• Flexible phase order
• Low-level IR (LLIR) for multi-architecture machine instruction abstraction.
• Visable control flow graph.

LLVM IR generated by Lava:

define i32 @main() {
entry:
  br label %while.cond0

while.cond0: ; preds: entry, loop.body0
  %phi0 = phi i32 [ 0, %entry ], [ %t0, %loop.body0 ]
  %phi1 = phi i32 [ 3, %entry ], [ %t1, %loop.body0 ]
  %t2 = icmp slt i32 %phi0, 5
  br i1 %t2, label %loop.body0, label %inline_exit0

loop.body0: ; preds: while.cond0
  %t1 = mul i32 %phi1, 2
  %t0 = add i32 %phi0, 1
  br label %while.cond0

inline_exit0: ; preds: while.cond0
  ret i32 %phi1
}

CFG genereated by Lava:

Automated Testing Tools

See lava-test