Async Future Memory Optimisation #622

Summary

We want to solve async-future memory bloat problem.

Playground

async fn combining_future(compute: impl Future<Output = ()>) {
    async {}.await;
    compute.await;
}

fn main() {
    let blob = [0u8; 65536];
    let monster = combining_future(combining_future(combining_future(combining_future(async move {
        async {}.await;
        println!("{:?}", blob)
    }))));
    println!("{}", std::mem::size_of_val(&monster));
    //~^ prints at least 1000000 bytes
}

We would like to deliver the first async-future memory packing scheme -Zpack-coroutine-layout as a nightly-only compiler flag, followed by a more aggressive memory packing scheme built on top of the foundation of the first pack-coroutine-layout support. Meanwhile, we would like to conduct two experiments in two possible approaches that could further improve the memory economy of async-futures.

• Explore the memory layout optimisation by enabling coroutine state inlining.
• Explore direct lowering of Rust async-futures, and coroutines in particular, into native LLVM coroutine intrinsics.

This issue is intended for status updates only.

For general questions or comments, please contact a goal owner or post in the Zulip topic.