Benchmark of AMD64 PUSH vs SUB+MOV

Cargo.toml

[package]
name = "push"
version = "0.1.0"
authors = ["A non-e-moose"]

[dependencies]

[dev-dependencies]
criterion = "0.2.5"
memmap = "0.7.0"

[[bench]]
name = "push"
harness = false

push.rs

#[macro_use]
extern crate criterion;
extern crate memmap;

use criterion::Criterion;
use memmap::MmapMut;

fn push(c: &mut criterion::Criterion) {
    let code = vec![0x50, // push %rax
                    0x58, // pop %rax
                    0xc3]; // ret

    let mut m = MmapMut::map_anon(code.len()).unwrap();
    for (i, c) in code.iter().enumerate() {
        m[i] = *c;
    }

    let m = m.make_exec().unwrap();
    let f = unsafe { std::mem::transmute::<*const u8, fn() -> ()>(m.as_ptr()) };
    c.bench_function("push", move |b| b.iter(|| f()));
}

fn mov(c: &mut criterion::Criterion) {
    let code = vec![0x48, 0x83, 0xec, 0x08, // sub $8, %rsp
                    0x48, 0x89, 0x04, 0x24, // mov %rax, (%rsp)
                    0x48, 0x8b, 0x04, 0x24, // mov (%rsp), %rax
                    0x48, 0x83, 0xc4, 0x08, // add $8, %rsp
                    0xc3]; // ret

    let mut m = MmapMut::map_anon(code.len()).unwrap();
    for (i, c) in code.iter().enumerate() {
        m[i] = *c;
    }

    let m = m.make_exec().unwrap();
    let f = unsafe { std::mem::transmute::<*const u8, fn() -> ()>(m.as_ptr()) };
    c.bench_function("mov", move |b| b.iter(|| f()));
}

criterion_group!(benches, push, mov);
criterion_main!(benches);