bfjit

brainfuck interpreter + x86_64 jit

the goal: accuracy first, then speed

platform

• x86_64 only
• windows + linux

run

hello:

cargo run --release -- samples/hello.bf

mandelbrot:

cargo run --release -- samples/mandelbrot.bf

modes:

cargo run --release -- samples/hello.bf --interp
cargo run --release -- samples/hello.bf --jit

options:

cargo run --release -- samples/mandelbrot.bf --tape 30000
cargo run --release -- samples/mandelbrot.bf --guard
cargo run --release -- samples/mandelbrot.bf --noguard
cargo run --release -- samples/mandelbrot.bf --dump-ir
cargo run --release -- samples/mandelbrot.bf --dump-asm

benches

fast:

cargo bench --bench bench

heavy (mandelbrot):

$env:BFJIT_HEAVY = "1"
cargo bench --bench bench -- mandelbrot

fuzz

requires nightly:

cargo +nightly fuzz run ir_equiv
cargo +nightly fuzz run bf_equiv

correctness

• interp is the oracle
• fuzz targets compare interp vs jit (tape + output)
• mandelbrot output matches byte-for-byte (interp vs jit)

numbers

mandelbrot on my box:

• interp: ~5.6s
• jit: ~0.85s
• jit compile: ~85us