MCJOHN974 / bench

Cranelift bench

In this repo I want to measure performance difference between code compiled via:

1. Rust -> aarch64
2. Rust -> Cranelift IR -> aarch64
3. Rust -> Wasm -> Cranelift IR -> aarch64

Workloads

Currently matrix multiplication and monte-carlo pi calculation only. I think with such results it is enought.

How to reproduce

Below are steps to reproduce each of the methods

Direct compilation using rustc

cargo run --release > results/rustc.txt

Using rustc_codegen_cranelift

# Download and unpack compiler from here https://github.com/bjorn3/rustc_codegen_cranelift/releases/tag/dev
wget https://github.com/rust-lang/rustc_codegen_cranelift/releases/download/dev/cg_clif-x86_64-apple-darwin.tar.xz
tar -xf cg_clif-x86_64-apple-darwin.tar.xz

# Compile the project with rustc_codegen_cranelift and run the benchmark
dist/cargo-clif run --release > results/rustc_cranelift.txt

Using Rustc WASM target

# Init cargo for compiling to wasm
rustup target add wasm32-wasi

# Compile Rust to wasm
cargo build --target wasm32-wasi --release

# Install wasmtime
brew install wasmtime

# Precompile wasm via cranelift
wasmtime compile target/wasm32-wasi/release/cranelift-bench.wasm -o compiled-wasm

# Run precompiled wasm:
wasmtime compiled-wasm --allow-precompiled > results/wasm_cranelift.txt

Results

Apple M2

	rustc	rustc_cranelift	wasm_cranelift
monte-carlo-pi	8.9s	562.8s	13.7s
matrix-multiplication	4.7s	16.3s	4.7s

Intel i7-6700K CPU

	rustc	rustc_cranelift	wasm_cranelift
monte-carlo-pi	3.1s	550s	20.1s
matrix-multiplication	7.3s	31.7s	9.2s