Optimize bare call overhead (#522)

* add benchmarks to test root call overhead with different amounts of parameters and results

* optimize TypedFunc::call at cost of Func::call

* remove WasmTypeList::{values, from_values} trait methods

* refactor WasmTypeList::from_untyped_values impl for a single T

* remove WasmTypeList::{Values, ValuesIter} assoc types

* rename some WasmTypeList trait methods

* apply clippy suggestions

* replace raw for loop with for_each

crates/wasmi/src/engine/mod.rs

@@ -285,23 +285,19 @@ impl EngineInner {
         Results: CallResults,
     {
         self.initialize_args(params);
-        let signature = match func.as_internal(ctx.as_context()) {
+        match func.as_internal(ctx.as_context()) {
             FuncEntityInternal::Wasm(wasm_func) => {
-                let signature = wasm_func.signature();
                 let mut frame = self.stack.call_wasm_root(wasm_func, &self.code_map)?;
                 let mut cache = InstanceCache::from(frame.instance());
                 self.execute_wasm_func(ctx.as_context_mut(), &mut frame, &mut cache)?;
-                signature
             }
             FuncEntityInternal::Host(host_func) => {
-                let signature = host_func.signature();
                 let host_func = host_func.clone();
                 self.stack
                     .call_host_root(ctx.as_context_mut(), host_func, &self.func_types)?;
-                signature
             }
         };
-        let results = self.write_results_back(signature, results);
+        let results = self.write_results_back(results);
         Ok(results)
     }
 
@@ -311,9 +307,7 @@ impl EngineInner {
         Params: CallParams,
     {
         self.stack.clear();
-        for param in params.feed_params() {
-            self.stack.values.push(param);
-        }
+        self.stack.values.extend(params.call_params());
     }
 
     /// Writes the results of the function execution back into the `results` buffer.
@@ -325,31 +319,11 @@ impl EngineInner {
     /// # Panics
     ///
     /// - If the `results` buffer length does not match the remaining amount of stack values.
-    fn write_results_back<Results>(
-        &mut self,
-        func_type: DedupFuncType,
-        results: Results,
-    ) -> <Results as CallResults>::Results
+    fn write_results_back<Results>(&mut self, results: Results) -> <Results as CallResults>::Results
     where
         Results: CallResults,
     {
-        let result_types = self.func_types.resolve_func_type(func_type).results();
-        assert_eq!(
-            self.stack.values.len(),
-            results.len_results(),
-            "expected {} values on the stack after function execution but found {}",
-            results.len_results(),
-            self.stack.values.len(),
-        );
-        assert_eq!(results.len_results(), result_types.len());
-        results.feed_results(
-            self.stack
-                .values
-                .drain()
-                .iter()
-                .zip(result_types)
-                .map(|(raw_value, value_type)| raw_value.with_type(*value_type)),
-        )
+        results.call_results(self.stack.values.drain())
     }
 
     /// Executes the top most Wasm function on the [`Stack`] until the [`Stack`] is empty.

crates/wasmi/src/engine/traits.rs

@@ -1,5 +1,6 @@
 use crate::core::Value;
 use core::{iter, slice};
+use wasmi_core::UntypedValue;
 
 /// Types implementing this trait may be used as parameters for function execution.
 ///
@@ -12,33 +13,45 @@ use core::{iter, slice};
 /// [`Engine`]: [`crate::Engine`]
 pub trait CallParams {
     /// The iterator over the parameter values.
-    type Params: Iterator<Item = Value>;
-
-    /// Returns the number of given parameter values.
-    ///
-    /// # Note
-    ///
-    /// Used by the [`Engine`] to determine how many parameters are received.
-    ///
-    /// [`Engine`]: [`crate::Engine`]
-    fn len_params(&self) -> usize;
+    type Params: ExactSizeIterator<Item = UntypedValue>;
 
     /// Feeds the parameter values from the caller.
-    fn feed_params(self) -> Self::Params;
+    fn call_params(self) -> Self::Params;
 }
 
 impl<'a> CallParams for &'a [Value] {
-    type Params = iter::Copied<slice::Iter<'a, Value>>;
+    type Params = CallParamsValueIter<'a>;
 
-    fn len_params(&self) -> usize {
-        self.len()
+    #[inline]
+    fn call_params(self) -> Self::Params {
+        CallParamsValueIter {
+            iter: self.iter().copied(),
+        }
     }
+}
+
+/// An iterator over the [`UntypedValue`] call parameters.
+#[derive(Debug)]
+pub struct CallParamsValueIter<'a> {
+    iter: iter::Copied<slice::Iter<'a, Value>>,
+}
 
-    fn feed_params(self) -> Self::Params {
-        self.iter().copied()
+impl<'a> Iterator for CallParamsValueIter<'a> {
+    type Item = UntypedValue;
+
+    #[inline]
+    fn size_hint(&self) -> (usize, Option<usize>) {
+        self.iter.size_hint()
+    }
+
+    #[inline]
+    fn next(&mut self) -> Option<Self::Item> {
+        self.iter.next().map(UntypedValue::from)
     }
 }
 
+impl ExactSizeIterator for CallParamsValueIter<'_> {}
+
 /// Types implementing this trait may be used as results for function execution.
 ///
 /// # Note
@@ -53,43 +66,22 @@ pub trait CallResults {
     /// The type of the returned results value.
     type Results;
 
-    /// Returns the number of expected result values.
-    ///
-    /// # Note
-    ///
-    /// Used by the [`Engine`] to determine how many results are expected.
-    ///
-    /// [`Engine`]: [`crate::Engine`]
-    fn len_results(&self) -> usize;
-
     /// Feeds the result values back to the caller.
     ///
     /// # Panics
     ///
     /// If the given `results` do not match the expected amount.
-    fn feed_results<T>(self, results: T) -> Self::Results
-    where
-        T: IntoIterator<Item = Value>,
-        T::IntoIter: ExactSizeIterator;
+    fn call_results(self, results: &[UntypedValue]) -> Self::Results;
 }
 
 impl<'a> CallResults for &'a mut [Value] {
     type Results = Self;
 
-    fn len_results(&self) -> usize {
-        self.len()
-    }
-
-    fn feed_results<T>(self, results: T) -> Self::Results
-    where
-        T: IntoIterator<Item = Value>,
-        T::IntoIter: ExactSizeIterator,
-    {
-        let results = results.into_iter();
-        assert_eq!(self.len_results(), results.len());
-        for (dst, src) in self.iter_mut().zip(results) {
-            *dst = src;
-        }
+    fn call_results(self, results: &[UntypedValue]) -> Self::Results {
+        assert_eq!(self.len(), results.len());
+        self.iter_mut().zip(results).for_each(|(dst, src)| {
+            *dst = src.with_type(dst.value_type());
+        });
         self
     }
 }

crates/wasmi/src/func/into_func.rs

@@ -69,8 +69,8 @@ macro_rules! impl_into_func {
             #[allow(non_snake_case)]
             fn into_func(self) -> (FuncType, HostFuncTrampoline<T>) {
                 let signature = FuncType::new(
-                    <Self::Params as WasmTypeList>::value_types(),
-                    <Self::Results as WasmTypeList>::value_types(),
+                    <Self::Params as WasmTypeList>::types(),
+                    <Self::Results as WasmTypeList>::types(),
                 );
                 let trampoline = HostFuncTrampoline::new(
                     move |caller: Caller<T>, params_results: FuncParams| -> Result<FuncResults, Trap> {
@@ -197,31 +197,32 @@ pub trait WasmTypeList: DecodeUntypedSlice + EncodeUntypedSlice + Sized {
     /// The [`ValueType`] sequence as array.
     type Types: IntoIterator<IntoIter = Self::TypesIter, Item = ValueType>
         + AsRef<[ValueType]>
-        + AsMut<[ValueType]>;
+        + AsMut<[ValueType]>
+        + Copy
+        + Clone;
 
     /// The iterator type of the sequence of [`ValueType`].
     type TypesIter: ExactSizeIterator<Item = ValueType> + DoubleEndedIterator + FusedIterator;
 
-    /// The [`Value`] sequence as array.
-    type Values: IntoIterator<IntoIter = Self::ValuesIter, Item = Value>
-        + AsRef<[Value]>
-        + AsMut<[Value]>;
+    type Values: IntoIterator<IntoIter = Self::ValuesIter, Item = UntypedValue>
+        + AsRef<[UntypedValue]>
+        + AsMut<[UntypedValue]>
+        + Copy
+        + Clone;
 
     /// The iterator type of the sequence of [`Value`].
-    type ValuesIter: ExactSizeIterator<Item = Value> + DoubleEndedIterator + FusedIterator;
+    type ValuesIter: ExactSizeIterator<Item = UntypedValue> + DoubleEndedIterator + FusedIterator;
 
     /// Returns an array representing the [`ValueType`] sequence of `Self`.
-    fn value_types() -> Self::Types;
+    fn types() -> Self::Types;
 
-    /// Returns an array representing the [`Value`] sequence of `self`.
+    /// Returns an array representing the [`UntypedValue`] sequence of `self`.
     fn values(self) -> Self::Values;
 
-    /// Consumes the [`Value`] iterator and creates `Self` if possible.
+    /// Consumes the [`UntypedValue`] iterator and creates `Self` if possible.
     ///
     /// Returns `None` if construction of `Self` is impossible.
-    fn from_values<T>(values: T) -> Option<Self>
-    where
-        T: Iterator<Item = Value>;
+    fn from_values(values: &[UntypedValue]) -> Option<Self>;
 }
 
 impl<T1> WasmTypeList for T1
@@ -232,31 +233,25 @@ where
 
     type Types = [ValueType; 1];
     type TypesIter = array::IntoIter<ValueType, 1>;
-    type Values = [Value; 1];
-    type ValuesIter = array::IntoIter<Value, 1>;
+    type Values = [UntypedValue; 1];
+    type ValuesIter = array::IntoIter<UntypedValue, 1>;
 
     #[inline]
-    fn value_types() -> Self::Types {
+    fn types() -> Self::Types {
         [<T1 as WasmType>::value_type()]
     }
 
     #[inline]
     fn values(self) -> Self::Values {
-        [<T1 as Into<Value>>::into(self)]
+        [<T1 as Into<UntypedValue>>::into(self)]
     }
 
-    fn from_values<T>(mut values: T) -> Option<Self>
-    where
-        T: Iterator<Item = Value>,
-    {
-        let value: T1 = values.next().and_then(Value::try_into)?;
-        if values.next().is_some() {
-            // Note: If the iterator yielded more items than
-            //       necessary we create no value from this procedure
-            //       as it is likely a bug.
-            return None;
+    #[inline]
+    fn from_values(values: &[UntypedValue]) -> Option<Self> {
+        if let [value] = *values {
+            return Some(value.into());
         }
-        Some(value)
+        None
     }
 }
 
@@ -272,37 +267,34 @@ macro_rules! impl_wasm_type_list {
 
             type Types = [ValueType; $n];
             type TypesIter = array::IntoIter<ValueType, $n>;
-            type Values = [Value; $n];
-            type ValuesIter = array::IntoIter<Value, $n>;
+            type Values = [UntypedValue; $n];
+            type ValuesIter = array::IntoIter<UntypedValue, $n>;
 
-            fn value_types() -> Self::Types {
+            #[inline]
+            fn types() -> Self::Types {
                 [$(
                     <$tuple as WasmType>::value_type()
                 ),*]
             }
 
+            #[inline]
             #[allow(non_snake_case)]
             fn values(self) -> Self::Values {
                 let ($($tuple,)*) = self;
                 [$(
-                    <$tuple as Into<Value>>::into($tuple)
+                    <$tuple as Into<UntypedValue>>::into($tuple)
                 ),*]
             }
 
-            fn from_values<T>(mut values: T) -> Option<Self>
-            where
-                T: Iterator<Item = Value>,
-            {
-                let result = ($(
-                    values.next().and_then(Value::try_into::<$tuple>)?,
-                )*);
-                if values.next().is_some() {
-                    // Note: If the iterator yielded more items than
-                    //       necessary we create no value from this procedure
-                    //       as it is likely a bug.
-                    return None
+            #[inline]
+            #[allow(non_snake_case)]
+            fn from_values(values: &[UntypedValue]) -> Option<Self> {
+                if let [$($tuple),*] = *values {
+                    return Some(
+                        ( $( Into::into($tuple), )* )
+                    )
                 }
-                Some(result)
+                None
             }
         }
     };

crates/wasmi/src/func/mod.rs

@@ -302,6 +302,10 @@ impl Func {
         if expected_outputs.len() != outputs.len() {
             return Err(FuncError::MismatchingResults { func: *self }).map_err(Into::into);
         }
+        outputs
+            .iter_mut()
+            .zip(expected_outputs.iter().copied().map(Value::default))
+            .for_each(|(output, expected_output)| *output = expected_output);
         // Note: Cloning an [`Engine`] is intentionally a cheap operation.
         ctx.as_context().store.engine().clone().execute_func(
             ctx.as_context_mut(),