Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

feat: use second table to visit members #2913

Open
wants to merge 2 commits into
base: main
Choose a base branch
from
Open

feat: use second table to visit members #2913

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
e1ae559
feat: use second table to visit members
HerrCai0907 Mar 24, 2025
1533149
bugfix
HerrCai0907 Mar 25, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
69 changes: 58 additions & 11 deletions src/builtins.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -99,6 +99,7 @@ import {

import {
CommonFlags,
CommonNames,
Feature,
featureToString,
TypeinfoFlags
Expand Down Expand Up @@ -10841,16 +10842,12 @@ function ensureVisitMembersOf(compiler: Compiler, instance: Class): void {
}
}

/** Compiles the `__visit_members` function. */
export function compileVisitMembers(compiler: Compiler): void {
function compileVisitMembersWithSwitchCase(compiler: Compiler): ExpressionRef {
let program = compiler.program;
let module = compiler.module;
let usizeType = program.options.usizeType;
let sizeTypeRef = usizeType.toRef();
let managedClasses = program.managedClasses;
let visitInstance = assert(program.visitInstance);
compiler.compileFunction(visitInstance, true); // is lazy, make sure it is compiled

// Prepare a mapping of class names to visitor calls. Each name corresponds to
// the respective sequential (0..N) class id.
let names = new Array<string>();
Expand Down Expand Up @@ -10907,16 +10904,66 @@ export function compileVisitMembers(compiler: Compiler): void {
current,
cases[names.length - 1]
], TypeRef.None);
return module.flatten([current, module.unreachable()]);
}

function compileVisitMembersWithCallIndirect(compiler: Compiler): ExpressionRef {
let program = compiler.program;
let module = compiler.module;
let sizeTypeRef = program.options.usizeType.toRef();
let managedClasses = program.managedClasses;
let objectInstance = program.objectInstance;
let tableFunc = new Array<string>();
let managedClassKeys = Map_keys(managedClasses), l = managedClassKeys.length;
for (let i = 0; i < l; ++i) {
let instanceId = managedClassKeys[i];
let instance = assert(managedClasses.get(instanceId));
if (instance.isPointerfree) {
// optimize for non pointer object
ensureVisitMembersOf(compiler, objectInstance);
tableFunc[i] = `${objectInstance.internalName}~visit`;
} else {
ensureVisitMembersOf(compiler, instance);
tableFunc[i] = `${instance.internalName}~visit`;
}
}
module.addFunctionTable(CommonNames.VisitorTable, l, l, tableFunc, module.i32(0));
return module.call_indirect(CommonNames.VisitorTable,
// load<u32>(changetype<usize>(this) - 8)
module.load(4, false,
sizeTypeRef == TypeRef.I64
? module.binary(BinaryOp.SubI64,
module.local_get(0, sizeTypeRef),
module.i64(8)
)
: module.binary(BinaryOp.SubI32,
module.local_get(0, sizeTypeRef),
module.i32(8) // rtId is at -8
),
TypeRef.I32, 0
), [
module.local_get(0, sizeTypeRef), // this
module.local_get(1, TypeRef.I32) // cookie
], createType([sizeTypeRef, TypeRef.I32]), TypeRef.None, false);
}
/** Compiles the `__visit_members` function. */
export function compileVisitMembers(compiler: Compiler): void {
let program = compiler.program;
let module = compiler.module;
let usizeType = program.options.usizeType;
let sizeTypeRef = usizeType.toRef();
let visitInstance = assert(program.visitInstance);
compiler.compileFunction(visitInstance, true); // is lazy, make sure it is compiled
const current = compiler.options.hasFeature(Feature.ReferenceTypes)
? compileVisitMembersWithCallIndirect(compiler)
: compileVisitMembersWithSwitchCase(compiler);
Comment on lines +10957 to +10959
Copy link
Member

@MaxGraey MaxGraey Mar 24, 2025 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

What do you think to use compileVisitMembersWithCallIndirect only for large amount of members (for example >= 16) for optimized for speed builds and always use it (without a threshold) for optimized for size builds?

Copy link
Member Author

@HerrCai0907 HerrCai0907 Mar 25, 2025 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think we cannot get any benefit from old implement in runtime which supports new features. So maybe emit call_indirect is better in each cases (more clear, easier to further opt, ...)

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

As far as I know, switch-case is usually faster than dynamic dispatching. Rust even has a whole package for this, https://crates.io/crates/enum_dispatch. Or have you already tried to evaluate the performance, and it turned out to be the same as jump table / switch-case?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It is not switch case vs dynamic dispatching. It is switch case based dynamic dispatching vs table lookup based dynamic dispatching. The major task is dispatching to actually visitor function.
old implement is

switch (obj.rtid) {
  case 0:
    call fn0;
  case 1:
   call fn1;
  ...
}

new implement is

jump_table = [fn0, fn1, ...]
call_indirect jump_table[obj.rtid];

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I understand. The problem is that call_indirect is not a very cheap operation even without wasm vm. In Wasm engines bounds check operation + long indirect jump is usually performed. In switch-case approach it is a very fast deterministic jump table. So I would first of all test the performance and how much it regresses if you always use call_indirect

Copy link
Member

@MaxGraey MaxGraey Mar 25, 2025 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I can't reproduce benchmark. I bootstraped compiler with this PR (with enabled reference-types feature) and main branch .wasm files. Optimize via wasm-opt as you suggested but when I try run js bench file it can't find __visit_members. All builds in debug btw.

    ins.instance.exports["__visit_members"](ptr, 0);
                                           ^

TypeError: ins.instance.exports.__visit_members is not a function

However rest of exports is present:

...
  __new: [Function: 50],
  __pin: [Function: 2389],
  __unpin: [Function: 2390],
  __collect: [Function: 2391],
  __rtti_base: Global [WebAssembly.Global] {},
  memory: Memory [WebAssembly.Memory] {},
  __setArgumentsLength: [Function: 2805],
  _initialize: [Function: 2806],
  setTarget: [Function: 5291],
  setRuntime: [Function: 5292],
  setNoAssert: [Function: 5293],
  setExportMemory: [Function: 5294],
  
  ...

And as I remember __visit_members is not exported from module and that's expected. So I'm confused...

Copy link
Member

@MaxGraey MaxGraey Mar 25, 2025 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Can you prepare a worked benchmark and share it with GitHub gist for example?

Copy link
Member

@MaxGraey MaxGraey Mar 25, 2025 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

When we tall about performance, we should not keep eyes in switch and call_indirect itself.
Here is diff between old and new.
old

It's typical codegen for any switch statement in wasm. Btw size of compiler (.wasm)

- with switch-case:    1,824,783 bytes (main)
- with indirect_call:  1,822,204 bytes (this pr)

So with indirect call we just shrink ~0.14% for non-optimized build.
For an optimized build, it will be even less

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Sorry, I forgot one step, in bootstrap wat, manual export __visit_members.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

So with indirect call we just shrink ~0.14% for non-optimized build.

It is not a module level change, it is a function level change. which means we save 2.5kB wasm code by optimizing one function.

// Add the function, executing an unreachable if breaking to 'invalid'
module.addFunction(BuiltinNames.visit_members,
createType([ sizeTypeRef, TypeRef.I32 ]), // this, cookie
module.addFunction(
BuiltinNames.visit_members,
createType([sizeTypeRef, TypeRef.I32]), // this, cookie
TypeRef.None, // => void
null,
module.flatten([
current,
module.unreachable()
])
current,
);
}

Expand Down
1 change: 1 addition & 0 deletions src/common.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -267,6 +267,7 @@ export namespace CommonNames {
// memory & table
export const DefaultMemory = "0";
export const DefaultTable = "0";
export const VisitorTable = "~/lib/rt/visitor";
}

// shared
Expand Down
98 changes: 12 additions & 86 deletions tests/compiler/bindings/esm.debug.wat
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -84,8 +84,10 @@
(data $20 (i32.const 1052) "<\00\00\00\00\00\00\00\00\00\00\00\02\00\00\00*\00\00\00O\00b\00j\00e\00c\00t\00 \00a\00l\00r\00e\00a\00d\00y\00 \00p\00i\00n\00n\00e\00d\00\00\00")
(data $21 (i32.const 1116) "<\00\00\00\00\00\00\00\00\00\00\00\02\00\00\00(\00\00\00O\00b\00j\00e\00c\00t\00 \00i\00s\00 \00n\00o\00t\00 \00p\00i\00n\00n\00e\00d\00\00\00\00\00")
(data $22 (i32.const 1184) "\10\00\00\00 \00\00\00 \00\00\00 \00\00\00\00\00\00\00\00\00\00\00\81\08\00\00\01\19\00\00\01\02\00\00$\t\00\00\a4\00\00\00$\n\00\00\02\t\00\00\02A\00\00\00\00\00\00A\00\00\00 \00\00\00")
(table $~/lib/rt/visitor 16 16 funcref)
(table $0 2 2 funcref)
(elem $0 (i32.const 1) $start:bindings/esm~anonymous|0)
(elem $~/lib/rt/visitor (table $~/lib/rt/visitor) (i32.const 0) func $~lib/object/Object~visit $~lib/object/Object~visit $~lib/object/Object~visit $~lib/arraybuffer/ArrayBufferView~visit $~lib/function/Function<%28%29=>void>~visit $~lib/typedarray/Int16Array~visit $~lib/typedarray/Float32Array~visit $~lib/typedarray/Uint64Array~visit $~lib/object/Object~visit $~lib/object/Object~visit $~lib/object/Object~visit $~lib/array/Array<i32>~visit $~lib/array/Array<~lib/string/String>~visit $bindings/esm/PlainObject~visit $~lib/typedarray/Uint8Array~visit $~lib/object/Object~visit)
(elem $0 (table $0) (i32.const 1) func $start:bindings/esm~anonymous|0)
(export "plainGlobal" (global $bindings/esm/plainGlobal))
(export "plainMutableGlobal" (global $bindings/esm/plainMutableGlobal))
(export "stringGlobal" (global $bindings/esm/stringGlobal))
Expand Down Expand Up @@ -2867,6 +2869,8 @@
local.get $0
call $~lib/rt/itcms/__visit
)
(func $~lib/object/Object~visit (param $0 i32) (param $1 i32)
)
(func $~lib/arraybuffer/ArrayBufferView~visit (param $0 i32) (param $1 i32)
(local $2 i32)
local.get $0
Expand All @@ -2881,8 +2885,6 @@
call $~lib/rt/itcms/__visit
end
)
(func $~lib/object/Object~visit (param $0 i32) (param $1 i32)
)
(func $~lib/function/Function<%28%29=>void>#get:_env (param $this i32) (result i32)
local.get $this
i32.load offset=4
Expand Down Expand Up @@ -2970,89 +2972,13 @@
call $~lib/arraybuffer/ArrayBufferView~visit
)
(func $~lib/rt/__visit_members (param $0 i32) (param $1 i32)
block $invalid
block $bindings/esm/NonPlainObject
block $~lib/typedarray/Uint8Array
block $bindings/esm/PlainObject
block $~lib/array/Array<~lib/string/String>
block $~lib/array/Array<i32>
block $~lib/staticarray/StaticArray<i64>
block $~lib/staticarray/StaticArray<u16>
block $~lib/staticarray/StaticArray<i32>
block $~lib/typedarray/Uint64Array
block $~lib/typedarray/Float32Array
block $~lib/typedarray/Int16Array
block $~lib/function/Function<%28%29=>void>
block $~lib/arraybuffer/ArrayBufferView
block $~lib/string/String
block $~lib/arraybuffer/ArrayBuffer
block $~lib/object/Object
local.get $0
i32.const 8
i32.sub
i32.load
br_table $~lib/object/Object $~lib/arraybuffer/ArrayBuffer $~lib/string/String $~lib/arraybuffer/ArrayBufferView $~lib/function/Function<%28%29=>void> $~lib/typedarray/Int16Array $~lib/typedarray/Float32Array $~lib/typedarray/Uint64Array $~lib/staticarray/StaticArray<i32> $~lib/staticarray/StaticArray<u16> $~lib/staticarray/StaticArray<i64> $~lib/array/Array<i32> $~lib/array/Array<~lib/string/String> $bindings/esm/PlainObject $~lib/typedarray/Uint8Array $bindings/esm/NonPlainObject $invalid
end
return
end
return
end
return
end
local.get $0
local.get $1
call $~lib/arraybuffer/ArrayBufferView~visit
return
end
local.get $0
local.get $1
call $~lib/function/Function<%28%29=>void>~visit
return
end
local.get $0
local.get $1
call $~lib/typedarray/Int16Array~visit
return
end
local.get $0
local.get $1
call $~lib/typedarray/Float32Array~visit
return
end
local.get $0
local.get $1
call $~lib/typedarray/Uint64Array~visit
return
end
return
end
return
end
return
end
local.get $0
local.get $1
call $~lib/array/Array<i32>~visit
return
end
local.get $0
local.get $1
call $~lib/array/Array<~lib/string/String>~visit
return
end
local.get $0
local.get $1
call $bindings/esm/PlainObject~visit
return
end
local.get $0
local.get $1
call $~lib/typedarray/Uint8Array~visit
return
end
return
end
unreachable
local.get $0
local.get $1
local.get $0
i32.const 8
i32.sub
i32.load
call_indirect $~/lib/rt/visitor (type $0)
)
(func $~setArgumentsLength (param $0 i32)
local.get $0
Expand Down
Loading