diff --git a/.gitignore b/.gitignore
index 1c04bd5..d99755f 100644
--- a/.gitignore
+++ b/.gitignore
@@ -1,3 +1,5 @@
+/flamegraph.svg
 /luac.out
+/perf.data*
 /target
 /untracked
diff --git a/Cargo.toml b/Cargo.toml
index 8ead428..62cc3b1 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -10,3 +10,12 @@ edition = "2021"
 
 # blake3, used to hash test vectors
 blake3 = "1.5.0"
+
+[target.x86_64-unknown-linux-gnu]
+linker = "/usr/bin/clang"
+# Recommended for flamegraph
+rustflags = ["-Clink-arg=-fuse-ld=lld", "-Clink-arg=-Wl,--no-rosegment"]
+
+[profile.release]
+# Recommended for profiling, e.g. flamegraph
+debug = true
diff --git a/README.md b/README.md
index 117a989..6ed8ba2 100644
--- a/README.md
+++ b/README.md
@@ -34,3 +34,33 @@
 - [ ] Long strings
 - [ ] Using arrays internally for tables
 - [ ] Compiling Lua source code to bytecode
+
+# Benchmark
+
+9x slower than PUC Lua, 90x slower than LuaJIT.
+
+Based only on the n-body simulation from the computer language benchmarks,
+because it's easy to run.
+
+```
+time target/release/lunar_wave_vm --script test_vectors/n_body.lua -- 500000; time lua test_vectors/n_body.lua 500000; time luajit test_vectors/n_body.lua 500000;
+-0.169075164
+-0.169096567
+[src/main.rs:110] x = []
+
+real	0m10.543s
+user	0m10.535s
+sys	0m0.000s
+-0.169075164
+-0.169096567
+
+real	0m1.173s
+user	0m1.165s
+sys	0m0.004s
+-0.169075164
+-0.169096567
+
+real	0m0.117s
+user	0m0.116s
+sys	0m0.000s
+```