Vm.run_func() FuncTypeMismatch using wasmedge_bindgen

[KeeganJin]

Hi all,
I transformed the .wasm file of the wasi-nn example into a lib file and compiled it into .wasm file. It basically just changed the function name main() of the pytorch example into pytorch_infer(), turn it into lib.rs and compiled it into wasm, so we can potentially use wasmedge_bindgen to get the inference result from the host.

//lib.rs
#[wasmedge_bindgen]
pub fn pytorch_infer() {
    let args: Vec<String> = env::args().collect();
    let model_bin_name: &str = &args[1];
    let image_name: &str = &args[2];
     ----
   // no return yet
    )

But when I call the wasm-lib from the host by using
vm.run_func(Some("extern"), "pytorch_infer", params!())?; to call the wasm lib
(the rest part of the code is the same as the main.rs in wasi-nn example )

//main.rs
fn main(){
    infer()
}
fn infer(){
    let dir_mapping = &String::from(".:.");
    let wasm_file = &String::from("ml_pytorch_lib.wasm");
    ----


    vm.run_func(Some("extern"), "pytorch_infer", params!())?;
}

I got the error

load plugin
[2023-09-19 20:20:28.849] [error] execution failed: function signature mismatch, Code: 0x83
[2023-09-19 20:20:28.849] [error]     Mismatched function type. Expected: FuncType {params{i32 , i32} returns{i32}} , Got: FuncType {params{} returns{i32}}
Error: Core(Execution(FuncTypeMismatch))

So I am wondering how to make wasmedge_bindgen work in this case and how can the host function get the inference result from the wasmedge VM?

[apepkuss]

@KeeganJin If you'd like to use wasmedge_bindgen in your project, you should use VmDocker instead of Vm. This example shows how to use VmDocker. It can help you. Thanks!

[KeeganJin]

Hi @apepkuss , thanks for your reply. I checked the documents and transformed Vm into VmDocker by doing

    vm.wasi_module_mut()
        // .ok_or("Not found wasi module")?
        .expect("Not found wasi module")
        .initialize(
            Some(vec![wasm_file, model_bin, image_file]),
            None,
            Some(vec![dir_mapping]),
        );

    let vm = VmDock::new(vm);

// call the pytorch_infer in ml_pytorch_lib
    vm.run_func("pytorch_infer", params!())?;

And got an error saying that

Error: Vm(NotFoundActiveModule)

Thanks. :>

[apepkuss]

From the error message, you didn't register the wasm module, which holds the pytorch_infer function, into the vm. You can use Module::from_file and Vm::register_module to register the wasm file into the vm. After that, you're able to call run_func method. For example,

use wasmedge_sdk::Module;

// load wasm module that holds the `pytorch_infer` func
let module = Module::from_file(None, "/path/to/your/some.wasm")?;

// register the module as an active module into the vm
let vm = vm.register_module(None, module)?;
     
vm.wasi_module_mut()
        .expect("Not found wasi module")
        .initialize(
            Some(vec![wasm_file, model_bin, image_file]),
            None,
            Some(vec![dir_mapping]),
        );

let vm = VmDock::new(vm);
vm.run_func("pytorch_infer", params!())?;

In addition, I drafted a new example for the wasi_nn-ggml plugin. It contains two parts: one is a wasm lib containing the infer function, the other an app that loads the wasm lib and invoke the infer func. Even though it does not use wasmedge_bindgen, the example should help you design your code.

[KeeganJin]

Hi @apepkuss ,

thanks for your reply.
In the previous code, I registered the WASM file by doing

   // build the Vm
    let mut vm = VmBuilder::new()
        .with_config(config)
        .with_plugin_wasi_nn()
        .build()?
        .register_module(Some("extern"), module)?;

The complete code is as shown below, thanks:

#[cfg(all(target_os = "linux", target_arch = "x86_64"))]
use wasmedge_sdk::{
    config::{CommonConfigOptions, ConfigBuilder, HostRegistrationConfigOptions},
    params,
    dock::{VmDock},
    plugin::PluginManager,
    Module, VmBuilder,
};

struct InferenceResult(usize, f32);

fn main() -> Result<(), Box<dyn std::error::Error>> {
    #[cfg(all(target_os = "linux", target_arch = "x86_64"))]
    infer()?;
    println!("first inference");

    // println!("second inference");
    // infer()?;

    Ok(())
}


#[cfg(all(target_os = "linux", target_arch = "x86_64"))]
fn infer() -> Result<(), Box<dyn std::error::Error>> {

    //works in the Terminal
    let dir_mapping = &String::from(".:.");
    // let wasm_file = &String::from("wasmedge-wasinn-example-mobilenet-image.wasm");
    // let wasm_file = &String::from("ml_pytorch_lib.wasm");
    let wasm_file = &String::from("ml_pytorch.wasm");

    let model_bin = &String::from("mobilenet.pt");
    let image_file = &String::from("input.jpg");
    // let args: Vec<String> = vec![dir_mapping,wasm_file,model_bin,image_file];


    println!("load plugin");
    PluginManager::load(None)?;

    let config = ConfigBuilder::new(CommonConfigOptions::default())
        .with_host_registration_config(HostRegistrationConfigOptions::default().wasi(true))
        .build()?;
    assert!(config.wasi_enabled());
    // assert!(config.wasi_nn_enabled());

    // load wasm module from file
    let module = Module::from_file(Some(&config), wasm_file)?;

   // build a Vm
    let mut vm = VmBuilder::new()
        .with_config(config)
        .with_plugin_wasi_nn()
        .build()?
        .register_module(Some("extern"), module)?;


    // init wasi module
    vm.wasi_module_mut()
        // .ok_or("Not found wasi module")?
        .expect("Not found wasi module")
        .initialize(
            Some(vec![wasm_file, model_bin, image_file]),
            None,
            Some(vec![dir_mapping]),
        );
    
    // vm.run_func(Some("extern"), "_start", params!())?;

    let vm = VmDock::new(vm);

    // call the call_infer in ml_pytorch_lib.wasm
    vm.run_func("call_infer", params!())?;


    // vm.run_func(Some("extern"), "_start", params!())?;

    Ok(())
}

[apepkuss]

As your code shows, you register module as a named module. It is named "extern". However, you invoke the call_infer from an active module that does not exist.

Fix: replace .register_module(Some("extern"), module)?; with .register_module(None, module)?;, then try again.

...

#[cfg(all(target_os = "linux", target_arch = "x86_64"))]
fn infer() -> Result<(), Box<dyn std::error::Error>> {

    ...

    // load wasm module from file
    let module = Module::from_file(Some(&config), wasm_file)?;

   // build a Vm
    let mut vm = VmBuilder::new()
        .with_config(config)
        .with_plugin_wasi_nn()
        .build()?
        .register_module(Some("extern"), module)?;   <=========== register a named module, called "extern"



   ...

    let vm = VmDock::new(vm);

    // call the call_infer in ml_pytorch_lib.wasm
    vm.run_func("call_infer", params!())?;           <============ call "call_infer" from an active module that doesn't exist


    // vm.run_func(Some("extern"), "_start", params!())?;

    Ok(())
}

[KeeganJin]

Thanks a lot! That fixes that issue!

[apepkuss]

If the issue have been solved, please close it. Thanks a lot!