T0b1/wasmtime
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Optimize Func::call and its C API (#3319)

Browse Source 
* Optimize `Func::call` and its C API

This commit is an alternative to #3298 which achieves effectively the
same goal of optimizing the `Func::call` API as well as its C API
sibling of `wasmtime_func_call`. The strategy taken here is different
than #3298 though where a new API isn't created, rather a small tweak to
an existing API is done. Specifically this commit handles the major
sources of slowness with `Func::call` with:

* Looking up the type of a function, to typecheck the arguments with and
  use to guide how the results should be loaded, no longer hits the
  rwlock in the `Engine` but instead each `Func` contains its own
  `FuncType`. This can be an unnecessary allocation for funcs not used
  with `Func::call`, so this is a downside of this implementation
  relative to #3298. A mitigating factor, though, is that instance
  exports are loaded lazily into the `Store` and in theory not too many
  funcs are active in the store as `Func` objects.

* Temporary storage is amortized with a long-lived `Vec` in the `Store`
  rather than allocating a new vector on each call. This is basically
  the same strategy as #3294 only applied to different types in
  different places. Specifically `wasmtime::Store` now retains a
  `Vec<u128>` for `Func::call`, and the C API retains a `Vec<Val>` for
  calling `Func::call`.

* Finally, an API breaking change is made to `Func::call` and its type
  signature (as well as `Func::call_async`). Instead of returning
  `Box<[Val]>` as it did before this function now takes a
  `results: &mut [Val]` parameter. This allows the caller to manage the
  allocation and we can amortize-remove it in `wasmtime_func_call` by
  using space after the parameters in the `Vec<Val>` we're passing in.
  This change is naturally a breaking change and we'll want to consider
  it carefully, but mitigating factors are that most embeddings are
  likely using `TypedFunc::call` instead and this signature taking a
  mutable slice better aligns with `Func::new` which receives a mutable
  slice for the results.

Overall this change, in the benchmark of "call a nop function from the C
API" is not quite as good as #3298. It's still a bit slower, on the
order of 15ns, because there's lots of capacity checks around vectors
and the type checks are slightly less optimized than before. Overall
though this is still significantly better than today because allocations
and the rwlock to acquire the type information are both avoided. I
personally feel that this change is the best to do because it has less
of an API impact than #3298.

* Rebase issues
This commit is contained in:
Alex Crichton
2021-09-21 14:07:05 -05:00
committed by GitHub
parent 38463d11ed
commit bcf3544924
26 changed files with 399 additions and 253 deletions
Download Patch File Download Diff File Expand all files Collapse all files

64
crates/c-api/src/func.rs
View File

@@ -3,7 +3,6 @@ use crate::{
wasm_extern_t, wasm_functype_t, wasm_store_t, wasm_val_t, wasm_val_vec_t, wasmtime_error_t,
wasmtime_extern_t, wasmtime_val_t, wasmtime_val_union, CStoreContext, CStoreContextMut,
};
use anyhow::anyhow;
use std::ffi::c_void;
use std::mem::{self, MaybeUninit};
use std::panic::{self, AssertUnwindSafe};
@@ -109,6 +108,22 @@ pub unsafe extern "C" fn wasm_func_new_with_env(
})
}
/// Places the `args` into `dst` and additionally reserves space in `dst` for `results_size`
/// returns. The params/results slices are then returned separately.
fn translate_args<'a>(
dst: &'a mut Vec<Val>,
args: impl ExactSizeIterator<Item = Val>,
results_size: usize,
) -> (&'a [Val], &'a mut [Val]) {
debug_assert!(dst.is_empty());
let num_args = args.len();
dst.reserve(args.len() + results_size);
dst.extend(args);
dst.extend((0..results_size).map(|_| Val::null()));
let (a, b) = dst.split_at_mut(num_args);
(a, b)
}
#[no_mangle]
pub unsafe extern "C" fn wasm_func_call(
func: &mut wasm_func_t,
@@ -118,23 +133,20 @@ pub unsafe extern "C" fn wasm_func_call(
let f = func.func();
let results = (*results).as_uninit_slice();
let args = (*args).as_slice();
if results.len() != f.ty(func.ext.store.context()).results().len() {
return Box::into_raw(Box::new(wasm_trap_t::new(
anyhow!("wrong number of results provided").into(),
)));
}
let params = args.iter().map(|i| i.val()).collect::<Vec<_>>();
let mut dst = Vec::new();
let (wt_params, wt_results) =
translate_args(&mut dst, args.iter().map(|i| i.val()), results.len());
// We're calling arbitrary code here most of the time, and we in general
// want to try to insulate callers against bugs in wasmtime/wasi/etc if we
// can. As a result we catch panics here and transform them to traps to
// allow the caller to have any insulation possible against Rust panics.
let result = panic::catch_unwind(AssertUnwindSafe(|| {
f.call(func.ext.store.context_mut(), &params)
f.call(func.ext.store.context_mut(), wt_params, wt_results)
}));
match result {
Ok(Ok(out)) => {
for (slot, val) in results.iter_mut().zip(out.into_vec().into_iter()) {
Ok(Ok(())) => {
for (slot, val) in results.iter_mut().zip(wt_results.iter().cloned()) {
crate::initialize(slot, wasm_val_t::from_val(val));
}
ptr::null_mut()
@@ -261,7 +273,7 @@ pub(crate) unsafe fn c_callback_to_rust_fn(
#[no_mangle]
pub unsafe extern "C" fn wasmtime_func_call(
store: CStoreContextMut<'_>,
mut store: CStoreContextMut<'_>,
func: &Func,
args: *const wasmtime_val_t,
nargs: usize,
@@ -269,27 +281,31 @@ pub unsafe extern "C" fn wasmtime_func_call(
nresults: usize,
trap_ret: &mut *mut wasm_trap_t,
) -> Option<Box<wasmtime_error_t>> {
if nresults != func.ty(&store).results().len() {
return Some(Box::new(wasmtime_error_t::from(anyhow!(
"wrong number of results provided"
))));
}
let params = crate::slice_from_raw_parts(args, nargs)
.iter()
.map(|i| i.to_val())
.collect::<Vec<_>>();
let mut store = store.as_context_mut();
let mut params = mem::take(&mut store.data_mut().wasm_val_storage);
let (wt_params, wt_results) = translate_args(
&mut params,
crate::slice_from_raw_parts(args, nargs)
.iter()
.map(|i| i.to_val()),
nresults,
);
// We're calling arbitrary code here most of the time, and we in general
// want to try to insulate callers against bugs in wasmtime/wasi/etc if we
// can. As a result we catch panics here and transform them to traps to
// allow the caller to have any insulation possible against Rust panics.
let result = panic::catch_unwind(AssertUnwindSafe(|| func.call(store, &params)));
let result = panic::catch_unwind(AssertUnwindSafe(|| {
func.call(&mut store, wt_params, wt_results)
}));
match result {
Ok(Ok(out)) => {
Ok(Ok(())) => {
let results = crate::slice_from_raw_parts_mut(results, nresults);
for (slot, val) in results.iter_mut().zip(out.into_vec().into_iter()) {
crate::initialize(slot, wasmtime_val_t::from_val(val));
for (slot, val) in results.iter_mut().zip(wt_results.iter()) {
crate::initialize(slot, wasmtime_val_t::from_val(val.clone()));
}
params.truncate(0);
store.data_mut().wasm_val_storage = params;
None
}
Ok(Err(trap)) => match trap.downcast::<Trap>() {

9
crates/c-api/src/store.rs
View File

@@ -2,7 +2,9 @@ use crate::{wasm_engine_t, wasmtime_error_t, wasmtime_val_t, ForeignData};
use std::cell::UnsafeCell;
use std::ffi::c_void;
use std::sync::Arc;
use wasmtime::{AsContext, AsContextMut, InterruptHandle, Store, StoreContext, StoreContextMut};
use wasmtime::{
AsContext, AsContextMut, InterruptHandle, Store, StoreContext, StoreContextMut, Val,
};
/// This representation of a `Store` is used to implement the `wasm.h` API.
///
@@ -71,6 +73,10 @@ pub struct StoreData {
/// Temporary storage for usage during a wasm->host call to store values
/// in a slice we pass to the C API.
pub hostcall_val_storage: Vec<wasmtime_val_t>,
/// Temporary storage for usage during host->wasm calls, same as above but
/// for a different direction.
pub wasm_val_storage: Vec<Val>,
}
#[no_mangle]
@@ -90,6 +96,7 @@ pub extern "C" fn wasmtime_store_new(
#[cfg(feature = "wasi")]
wasi: None,
hostcall_val_storage: Vec::new(),
wasm_val_storage: Vec::new(),
},
),
})