Remove some allocations in CodeMemory (#3253)

* Remove some allocations in `CodeMemory`

This commit removes the `FinishedFunctions` type as well as allocations
associated with trampolines when allocating inside of a `CodeMemory`.
The main goal of this commit is to improve the time spent in
`CodeMemory` where currently today a good portion of time is spent
simply parsing symbol names and trying to extract function indices from
them. Instead this commit implements a new strategy (different from #3236)
where compilation records offset/length information for all
functions/trampolines so this doesn't need to be re-learned from the
object file later.

A consequence of this commit is that this offset information will be
decoded/encoded through `bincode` unconditionally, but we can also
optimize that later if necessary as well.

Internally this involved quite a bit of refactoring since the previous
map for `FinishedFunctions` was relatively heavily relied upon.

* comments
This commit is contained in:
Alex Crichton
2021-08-30 10:35:17 -05:00
committed by GitHub
parent c73be1f13a
commit a237e73b5a
26 changed files with 271 additions and 288 deletions

View File

@@ -2,8 +2,8 @@
//! module.
use crate::{
DefinedFuncIndex, FilePos, FunctionBodyData, ModuleTranslation, PrimaryMap, StackMap, Tunables,
TypeTables, WasmError, WasmFuncType,
DefinedFuncIndex, FilePos, FunctionBodyData, ModuleTranslation, PrimaryMap, SignatureIndex,
StackMap, Tunables, TypeTables, WasmError, WasmFuncType,
};
use anyhow::Result;
use object::write::Object;
@@ -22,6 +22,25 @@ use thiserror::Error;
pub struct FunctionInfo {
pub start_srcloc: FilePos,
pub stack_maps: Vec<StackMapInformation>,
/// Offset in the text section of where this function starts.
pub start: u64,
/// The size of the compiled function, in bytes.
pub length: u32,
}
/// Information about a compiled trampoline which the host can call to enter
/// wasm.
#[derive(Serialize, Deserialize, Clone)]
#[allow(missing_docs)]
pub struct Trampoline {
/// The signature this trampoline is for
pub signature: SignatureIndex,
/// Offset in the text section of where this function starts.
pub start: u64,
/// The size of the compiled function, in bytes.
pub length: u32,
}
/// The offset within a function of a GC safepoint, and its associated stack
@@ -154,19 +173,21 @@ pub trait Compiler: Send + Sync {
funcs: PrimaryMap<DefinedFuncIndex, Box<dyn Any + Send>>,
emit_dwarf: bool,
obj: &mut Object,
) -> Result<PrimaryMap<DefinedFuncIndex, FunctionInfo>>;
) -> Result<(PrimaryMap<DefinedFuncIndex, FunctionInfo>, Vec<Trampoline>)>;
/// Inserts two functions for host-to-wasm and wasm-to-host trampolines into
/// the `obj` provided.
///
/// This will configure the same sections as `emit_obj`, but will likely be
/// much smaller.
/// much smaller. The two returned `Trampoline` structures describe where to
/// find the host-to-wasm and wasm-to-host trampolines in the text section,
/// respectively.
fn emit_trampoline_obj(
&self,
ty: &WasmFuncType,
host_fn: usize,
obj: &mut Object,
) -> Result<()>;
) -> Result<(Trampoline, Trampoline)>;
/// Creates a new `Object` file which is used to build the results of a
/// compilation into.

View File

@@ -43,6 +43,7 @@ pub use crate::stack_map::StackMap;
pub use crate::trap_encoding::*;
pub use crate::tunables::Tunables;
pub use crate::vmoffsets::*;
pub use object;
// Reexport all of these type-level since they're quite commonly used and it's
// much easier to refer to everything through one crate rather than importing

View File

@@ -3,7 +3,7 @@
use crate::{EntityRef, ModuleTranslation, PrimaryMap, Tunables, WASM_PAGE_SIZE};
use indexmap::IndexMap;
use serde::{Deserialize, Serialize};
use std::collections::{BTreeMap, BTreeSet};
use std::collections::BTreeMap;
use std::convert::TryFrom;
use std::ops::Range;
use wasmtime_types::*;
@@ -422,10 +422,6 @@ pub struct Module {
/// The type of each nested wasm module this module contains.
pub modules: PrimaryMap<ModuleIndex, ModuleTypeIndex>,
/// The set of defined functions within this module which are located in
/// element segments.
pub possibly_exported_funcs: BTreeSet<DefinedFuncIndex>,
}
/// Initialization routines for creating an instance, encompassing imports,

View File

@@ -10,7 +10,7 @@ use crate::{
};
use cranelift_entity::packed_option::ReservedValue;
use std::borrow::Cow;
use std::collections::{hash_map::Entry, HashMap};
use std::collections::{hash_map::Entry, HashMap, HashSet};
use std::convert::{TryFrom, TryInto};
use std::mem;
use std::path::PathBuf;
@@ -61,6 +61,23 @@ pub struct ModuleTranslation<'data> {
/// References to the function bodies.
pub function_body_inputs: PrimaryMap<DefinedFuncIndex, FunctionBodyData<'data>>,
/// The set of defined functions within this module which are "possibly
/// exported" which means that the host can possibly call them. This
/// includes functions such as:
///
/// * Exported functions
/// * Functions in element segments
/// * Functions via `ref.func` instructions
///
/// This set is used to determine the set of type signatures that need
/// trampolines for the host to call into.
pub escaped_funcs: HashSet<DefinedFuncIndex>,
/// A list of type signatures which are considered exported from this
/// module, or those that can possibly be called. This list is sorted, and
/// trampolines for each of these signatures are required.
pub exported_signatures: Vec<SignatureIndex>,
/// DWARF debug information, if enabled, parsed from the module.
pub debuginfo: DebugInfoData<'data>,
@@ -221,6 +238,22 @@ impl<'data> ModuleEnvironment<'data> {
Payload::End => {
validator.end()?;
// With the `escaped_funcs` set of functions finished
// we can calculate the set of signatures that are exported as
// the set of exported functions' signatures.
self.result.exported_signatures = self
.result
.module
.functions
.iter()
.filter_map(|(i, sig)| match self.result.module.defined_func_index(i) {
Some(i) if !self.result.escaped_funcs.contains(&i) => None,
_ => Some(*sig),
})
.collect();
self.result.exported_signatures.sort_unstable();
self.result.exported_signatures.dedup();
self.result.creation_artifacts.shrink_to_fit();
self.result.creation_modules.shrink_to_fit();
@@ -417,7 +450,7 @@ impl<'data> ModuleEnvironment<'data> {
Operator::RefNull { ty: _ } => GlobalInit::RefNullConst,
Operator::RefFunc { function_index } => {
let index = FuncIndex::from_u32(function_index);
self.flag_func_possibly_exported(index);
self.flag_func_escaped(index);
GlobalInit::RefFunc(index)
}
Operator::GlobalGet { global_index } => {
@@ -446,7 +479,7 @@ impl<'data> ModuleEnvironment<'data> {
let entity = match kind {
ExternalKind::Function => {
let index = FuncIndex::from_u32(index);
self.flag_func_possibly_exported(index);
self.flag_func_escaped(index);
EntityIndex::Function(index)
}
ExternalKind::Table => EntityIndex::Table(TableIndex::from_u32(index)),
@@ -471,7 +504,7 @@ impl<'data> ModuleEnvironment<'data> {
validator.start_section(func, &range)?;
let func_index = FuncIndex::from_u32(func);
self.flag_func_possibly_exported(func_index);
self.flag_func_escaped(func_index);
debug_assert!(self.result.module.start_func.is_none());
self.result.module.start_func = Some(func_index);
}
@@ -497,7 +530,7 @@ impl<'data> ModuleEnvironment<'data> {
elements.push(match item? {
ElementItem::Func(f) => {
let f = FuncIndex::from_u32(f);
self.flag_func_possibly_exported(f);
self.flag_func_escaped(f);
f
}
ElementItem::Null(_ty) => FuncIndex::reserved_value(),
@@ -1113,9 +1146,9 @@ and for re-adding support for interface types you can see this issue:
.push(ModuleSignature { imports, exports })
}
fn flag_func_possibly_exported(&mut self, func: FuncIndex) {
fn flag_func_escaped(&mut self, func: FuncIndex) {
if let Some(idx) = self.result.module.defined_func_index(func) {
self.result.module.possibly_exported_funcs.insert(idx);
self.result.escaped_funcs.insert(idx);
}
}