Stream: git-wasmtime

Topic: wasmtime / issue #2952 make ResourceLimiter operate on St...

Wasmtime GitHub notifications bot (May 29 2021 at 00:53):

github-actions[bot] commented on issue #2952:

Subscribe to Label Action

cc @peterhuene

<details>
This issue or pull request has been labeled: "cranelift", "wasmtime:c-api"

Thus the following users have been cc'd because of the following labels:

• peterhuene: wasmtime:c-api

To subscribe or unsubscribe from this label, edit the <code>.github/subscribe-to-label.json</code> configuration file.

Learn more.
</details>

Wasmtime GitHub notifications bot (Jun 03 2021 at 23:16):

github-actions[bot] commented on issue #2952:

Subscribe to Label Action

cc @fitzgen, @peterhuene

<details>
This issue or pull request has been labeled: "fuzzing", "wasmtime:api"

Thus the following users have been cc'd because of the following labels:

• fitzgen: fuzzing
• peterhuene: wasmtime:api

To subscribe or unsubscribe from this label, edit the <code>.github/subscribe-to-label.json</code> configuration file.

Learn more.
</details>

Wasmtime GitHub notifications bot (Jun 07 2021 at 14:52):

alexcrichton commented on issue #2952:

Locally I've got two benchmarks for changes like this:

<details>

<summary>host -> wasm overhead</summary>

fn host_to_wasm_overhead() -> anyhow::Result<()> {
    use wasmtime::*;
    let engine = Engine::default();
    let module = Module::new(
        &engine,
        r#"
            (module
                (func (export "run") (param i32 i32) (result i32)
                    local.get 0
                    local.get 1
                    i32.add
                )
            )
        "#,
    )?;
    let mut store = Store::new(&engine, ());
    let instance = Instance::new(&mut store, &module, &[])?;
    let run = instance.get_typed_func::<(i32, i32), i32, _>(&mut store, "run")?;
    let now = std::time::Instant::now();
    let total = 100_000_000;
    for _ in 0..total {
        run.call(&mut store, (1, 1))?;
    }
    let time = now.elapsed();
    println!(
        "use Time: {:?}, each:{} ns/op",
        time,
        time.as_nanos() / (total as u128)
    );
    Ok(())
}

</details>

<details>

<summary>wasm -> host overhead</summary>

fn wasm_to_host() -> anyhow::Result<()> {
    use wasmtime::*;
    let engine = Engine::default();
    let module = Module::new(
        &engine,
        r#"
            (module
                (import "" "add" (func $add (param i32 i32) (result i32)))

                (func (export "run") (param i32)
                    loop
                        i32.const 1
                        i32.const 2
                        call 0
                        drop

                        local.get 0
                        i32.const -1
                        i32.add
                        local.tee 0
                        br_if 0
                    end
                ))
        "#,
    )?;
    let mut store = Store::new(&engine, ());
    let func = Func::wrap(&mut store, |a: u32, b: u32| a + b);
    let instance = Instance::new(&mut store, &module, &[func.into()])?;
    let run = instance.get_typed_func::<i32, (), _>(&mut store, "run")?;
    let now = std::time::Instant::now();
    let total = 1_000_000_000;
    run.call(&mut store, total)?;
    let time = now.elapsed();
    println!(
        "use Time: {:?}, each:{} ns/op",
        time,
        time.as_nanos() / (total as u128)
    );
    Ok(())
}

</details>

The numbers I got for this PR on my local macOS machine are:

	before	after
host -> wasm	29ns	30ns
wasm -> host	1ns	2ns

So it looks like adding these unconditional branches has pretty negligible overhead. If it really matters in some use case we could always add a compile-time option to disable them, but otherwise I don't think there's too much impact.

One thing that's required, though, is to add #[inline] to a few more methods. I noticed this when profiling and gathered the numbers with some local changes to this PR. The reason is that StoreInner methods which previously had a generic in scope are now concrete on StoreInnermost so they require #[inline] to get cross-crate inlined. The ones I added inline too are:

• StoreInnermost::signal_handler
• StoreInnermost::vminterrupts
• StoreInnermost::async_cx
• StoreInnermost::default_callee
• StoreInnermost::async_support
• StoreInnermost::engine
• StoreInnermost::data
• StoreInnermost::data_mut

Wasmtime GitHub notifications bot (Jun 08 2021 at 14:37):

alexcrichton commented on issue #2952:

:+1:

Last updated: Nov 22 2024 at 16:03 UTC