1744978 - Implement jit support for Records and Tuples

Status: UNCONFIRMED

(Core :: JavaScript Engine, enhancement, P3)

Description

[Nicolò Ribaudo]

The --enable-record-tuple build flag currently disables JIT. We should implement jit support for the new Record, Tuple and Box primitives.

Comment 1

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 1. GC

Depends on D134842

Comment 2

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 1. Wrap test code in function definitions in order to trigger JIT

Depends on D135771

Comment 3

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 17. Expose ArrayNativeSort

This will be used by JIT for records (since record keys need to be sorted).

Depends on D135772

Comment 4

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 4. Add Record to friend classes for ObjectElements

This will be useful when implementing JIT for records.

Depends on D135773

Comment 5

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 5. Add VMFunction wrappers for record/tuple JIT

Add wrappers for NativeAddProperty, GrowElements, ArrayNativeSort, Int32ToAtom

Depends on D135774

Comment 6

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 6. Allow Record and Tuple opcodes to have ICs

Depends on D135775

Comment 7

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 2. Add methods to support JIT for Object.is on Records/Tuples

Depends on D135776

Comment 8

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 3. Make EXTENDED_PRIMITIVE_VALUE_SLOT public

Depends on D135777

Comment 9

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 5. Add Int32ToAtom as an ABI function

Depends on D135778

Comment 10

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 10. Add R/T opcodes to BaselineCodeGen

Depends on D135779

Comment 11

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 20. Add ExtendedPrimOperandId type to CacheIR.h

Depends on D135780

Comment 12

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 41. Add Record and Tuple MIR types

Depends on D135781

Comment 13

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 13. Add GuardToPrimitive CacheIR op

Depends on D135782

Comment 14

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 14. Add AtomizeString and AtomizeInt CacheIR ops

These auxiliary ops will be used to implement baseline codegen for
the AddRecordProperty opcode.

Depends on D135783

Comment 15

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 10. Add MacroAssembler operations to be used in implementing FinishRecord

Depends on D135784

Comment 16

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 16. Add CacheIR kinds for record/tuple opcodes

Depends on D135785

Comment 17

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 11. Add MacroAssembler type testing operations for records/tuples

Depends on D135786

Comment 18

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 21. Add GuardToExtendedPrimitive CacheIR op

Depends on D135787

Comment 19

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 12. Refactor emitStoreDenseElement

This code will be used for code generation for AddTupleElement and AddRecordProperty.

Depends on D135788

Comment 20

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 9. Take objectFlags into account when determining if shape has changed

This arises when setting object flags on records/tuples; if the record or
tuple becomes non-extensible, the shape should change (otherwise, the initial
default shape is mutated).

Depends on D135789

Comment 21

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 21. Baseline JIT compilation for InitRecord opcode

Depends on D135790

Comment 22

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 22. Baseline JIT compilation for FinishRecord opcode

Depends on D135791

Comment 23

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 23. Baseline JIT compilation for AddRecordProperty opcode

Depends on D135792

Comment 24

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 30. Baseline JIT compilation for InitTuple opcode

Depends on D135793

Comment 25

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 31. Baseline JIT compilation for FinishTuple opcode

Depends on D135795

Comment 26

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 32. Baseline JIT compilation for AddTupleElement opcode

Depends on D135796

Comment 27

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 22. Add CompareExtendedPrimitive CacheIR op

Depends on D135797

Comment 28

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 24. Add TypeOfExtendedPrimitive CacheIR op

Depends on D135798

Comment 29

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 29. Generate CacheIR stubs for tuple element accesses

Depends on D135799

Comment 30

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 34. Handle ExtendedPrimitives in CacheIR case for typeOf

Depends on D135800

Comment 31

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 23. Handle ExtendedPrimitives in CacheIR case for Object.is()

Depends on D135801

Comment 32

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 25. Add methods for manipulating ExtendedPrimitive properties

Depends on D135802

Comment 33

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 27. Handle property accesses on Records/Tuples properly in CacheIR

Depends on D135803

Comment 34

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 35. Implement JIT for IsTuple intrinsic

Depends on D135804

Comment 35

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 36. Implement JIT for ThisTupleValue intrinsic

Depends on D135805

Comment 36

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 37. Implement JIT for TupleLength intrinsic

Depends on D135806

Comment 37

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 33. Increase iterations for record tests

Depends on D135777

Comment 38

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 6. Add GrowElements VM call

This will be used by the JIT implementation for AddRecordProperty and AddTupleElement.

Depends on D135779

Comment 39

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 7. Add ShrinkCapacityToInitializedLength VM call

This will be used by the JIT implementation for FinishRecord and FinishTuple.

Depends on D152987

Comment 40

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 8. Add NativeAddProperty VM call

This will be used by the JIT implementation for AddRecordProperty.

Depends on D152988

Comment 41

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 14. Add ThrowNonPrimitiveRecordValueError() VM function

This will be called by the JIT implementation of AddRecordProperty if a
value is non-primitive.

Depends on D135783

Comment 42

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 15. Add unboxing for Record values to MacroAssembler

Depends on D152990

Comment 43

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 16. Add SetNonExtensible() VM function

This function is used to implement FinishRecord.

Depends on D152991

Comment 44

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 18. Add FinishSortedKeysArray() VM function

This is used to implement FinishRecord.

Depends on D135773

Comment 45

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 19. Add IdToRecordKey() VM function

This is used to implement AddRecordProperty.

Depends on D152993

Comment 46

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 26. Don't generate CacheIR for isPackedArray() calls on extended primitives

Depends on D135803

Comment 47

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 26.5. Don't generate CacheIR for map/set operations on extended primitives

Depends on D152995

Comment 48

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 28. Add createUninitializedRecord() method to MacroAssembler

Depends on D135804

Comment 49

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 29. Implement records in baseline JIT compiler

Depends on D152997

Comment 50

[Tim Chevalier]

Attachment: WIP: Bug 1744978: BaselineFrameReg to FramePointer

Depends on D135797

Comment 51

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 38. Root a variable in Object::TryEnumerableOwnPropertiesNative()

Depends on D135807

Comment 52

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 39. Add type assertions to visitToHashableString()

Depends on D153000

Comment 53

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 40. Allow SetArrayLength and SetInitializedLength to use a non-constant index

This is helpful in implementing record initialization (in the presence of spread operations, record length isn't always statically known.)

Depends on D153001

Comment 54

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 42. Allow Record type in MUnbox instruction

Depends on D135782

Comment 55

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 43. Handle record and tuple types in TypeOf instruction

Depends on D153003

Comment 56

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 44. Add RecordPolicy type

Depends on D153004

Comment 57

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 45. Add RecordToObject instruction to MIR

This instruction is a no-op at runtime, but allows us to re-use
MIR instructions with Object operands for Records and Tuples.

Depends on D153005

Comment 58

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 46. Factor out EmitLoadAndUnbox and GetElements

Depends on D153006

Comment 59

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 47. Implement InitRecord in Warp

Depends on D153007

Comment 60

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 48. Add GuardIsNotSymbol to LIR/MIR

This is used to implement AddRecordProperty to ensure that record keys
are not symbols.

Depends on D153008

Comment 61

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 49. Add SetNonExtensible and SetNonwritableArrayLength to LIR

These are used to implement FinishRecord.

Depends on D153009

Comment 62

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 50. Add CallEnsureElements to LIR

This is used in implementing AddRecordProperty and AddTupleElement.

Depends on D153010

Comment 63

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 51. Add CallNativeAddProperty to MIR and LIR

This is used in implementing AddRecordProperty.

Depends on D153011

Comment 64

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 52. Add CallSort to MIR

This is used to implement FinishRecord.

Depends on D153012

Comment 65

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 53. Implement AddRecordProperty in Warp

Depends on D153013

Comment 66

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 54. Implement FinishRecord in Warp

Depends on D153014

Comment 67

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 55. Implement AddRecordSpread in Warp

Depends on D153015

Comment 68

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 56. Don't automatically disable JIT if record-tuple is enabled

Depends on D153016

Comment 69

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 57. Add missing includes

Include mozilla/Maybe.h in TupleObject.cpp
Include vm/JSContext-inl.h in RecordTupleShared.cpp

This fixes build failures on build-macosx64-hybrid/plain

Depends on D153017

Comment 70

[Tim Chevalier]

Attachment: WIP: Bug 1744978 - 58. Revise help message for --enable-record-tuple build option

Depends on D153018

Comment 71

[Tim Chevalier]

Attachment: WIP: Bug 1744978: Allow GetProperty() to operate on ExtendedPrimitives

Depends on D153019

Comment 72

[Tim Chevalier]

Attachment: WIP: Bug 1744978: Adding jit-test tests for explainer text in the proposal repo.

From the README.md: https://github.com/tc39/proposal-record-tuple#readme
From the cookbook: https://tc39.es/proposal-record-tuple/cookbook/
And possibly wrong examples in Details.md (not updated since 2020): https://github.com/tc39/proposal-record-tuple/blob/main/details.md

Depends on D153020

Comment 73

[Tim Chevalier]

Attachment: WIP: Bug 1744978: Don't run tests that use Reflect.parse when not in the shell

Depends on D153021

Comment 74

[Tim Chevalier]

Attachment: WIP: Bug 1744978: Add isConstructor() to shell.js for Tuple tests

Depends on D153022

Comment 75

[Tim Chevalier]

Attachment: WIP: Bug 1744978: tmp

Depends on D153023

Comment 76

[Tim Chevalier]

Attachment: WIP: Bug 1744978: tmp

Depends on D153024

Comment 77

[Tim Chevalier]

Attachment: WIP: Bug 1744978: tmp

Depends on D153025

Comment 78

[Tim Chevalier]

Attachment: WIP: Bug 1744978: Reduced register usage; FinishTuple/AddTupleElement WIP

Depends on D153026

Comment 79

[Tim Chevalier]

Attachment: WIP: Bug 1744978: tmp

Depends on D153027

Comment 80

[Tim Chevalier]

Attachment: WIP: Bug 1744978: Fixed bug in CacheIRCompiler::emitCompareExtendedPrimitiveResult()

The bug was that it was calling prepare() twice, which meant that in
one of the paths, the stack didn't get popped all the way, meaning we
used some random garbage as the return address

Record tests pass now except for json

Depends on D153028

Comment 81

[Tim Chevalier]

Attachment: WIP: Bug 1744978: Developer console fixes

This needs to be cleaned up.
Records still render as just "Object { }", but that's the best I can do
for now without knowing the code better, and it doesn't crash

Main issue I fixed was that there were two versions of createValueGrip()
in the console and both needed to be changed to return something non-null
for Record/Tuple

Also needs tests

Depends on D153029

Comment 82

[Tim Chevalier]

Attachment: WIP: Bug 1744978: Set fixed elements flag correctly in BaselineCacheIRCompiler for InitTuple

Depends on D153030

Comment 83

[Tim Chevalier]

Attachment: WIP: Bug 1744978: Record tests

Depends on D153031

Comment 84

[Tim Chevalier]

Attachment: WIP: Bug 1744978: Fix calling conventions for calls to IC and fallback for AddTupleElement

Depends on D153032

Comment 85

[Tim Chevalier]

Attachment: WIP: Bug 1744978: Handle tuple elements correctly in CacheIR

The code was incorrectly coercing ExtPrimOperandIds into ObjOperandIds,
causing useRegister(...) to use the wrong type when unboxing Tuples,
creating invalid memory addresses

Depends on D153033

Comment 86

[Tim Chevalier]

Attachment: WIP: Bug 1744978: Cache both length and tuple in CacheIR for InitTuple

Depends on D153034

Comment 87

[Tim Chevalier]

Attachment: WIP: Bug 1744978: Don't use pre-barriers for writes in BaselineCacheIR AddTupleElement

Depends on D153035

Comment 88

[Tim Chevalier]

Attachment: WIP: Bug 1744978: Don't use a custom writer for guardToTuple

Because guardToTuple may return something that's different from its input
(it might unbox a TupleObject), it can't use a custom writer.

Depends on D153036

Comment 89

[Tim Chevalier]

Attachment: WIP: Bug 1744978: remove debugging printfs

Depends on D153037

Comment 90

[Tim Chevalier]

Attachment: WIP: Bug 1744978: fix clang-format errors

Depends on D153038

Comment 91

[Tim Chevalier]

Attachment: WIP: Bug 1744978: fix ifdefs

Depends on D153039

Comment 92

[Tim Chevalier]

Attachment: WIP: Bug 1744978: fix MacroAssembler-arm64 syntax error

Depends on D153040

Comment 93

[Tim Chevalier]

Attachment: WIP: Bug 1744978: Clean up equality.js

Depends on D153041

Comment 94

[Tim Chevalier]

Attachment: WIP: Bug 1744978: emitCompareExtendedPrimitiveResult bug fix + cleanup

Depends on D153042

Comment 95

[Tim Chevalier]

Attachment: WIP: Bug 1744978: add tests for empty tuples and extend tuple proposal examples to induce Warp under --ion-eager

Depends on D153043

Comment 96

[Tim Chevalier]

Attachment: WIP: Bug 1744978: Removed inline caches for FinishTuple.

Depends on D153044

Comment 97

[Tim Chevalier]

Attachment: WIP: Bug 1744978: Uncommenting tests after Tim's last fixes from our meeting. (commits are rearranged) Also some typesetting and superficial refactoring.

Uncommented tests in records-proposal-examples.js,
and some in record-tuple-cookbook.js which I had forgotten to uncomment
last time. Some tests still fail.

Depends on D153045

Comment 98

[Tim Chevalier]

Attachment: WIP: Bug 1744978: Uncommenting tests after Tim's last fixes from our meeting. (commits are rearranged) Also some typesetting and superficial refactoring.

Uncommented tests in records-proposal-examples.js,
and some in record-tuple-cookbook.js which I had forgotten to uncomment
last time.

FIXME: Errors with JSON.stringify

Depends on D153046

Comment 99

[Tim Chevalier]

Attachment: WIP: Bug 1744978: WIP: Adding Warp support for empty tuples

Implements InitTuple and FinishTuple in Warp.

Depends on D153047

Comment 100

[Tim Chevalier]

Attachment: WIP: Bug 1744978: Warp implementation for empty tuples.

Finishes WIP of previous commit.
Extends jit-test/tests/record-tuple/basic-tuples.js to induce warp (hopefully).

Depends on D153048

Comment 101

[Tim Chevalier]

Attachment: WIP: Bug 1744978: fixes for fallibleUnboxRecord

Depends on D153049

Comment 102

[Tim Chevalier]

Attachment: WIP: Bug 1744978: Save/restore `val` in emitAddTupleElementResult()

This was resulting in a garbage value being written into a tuple with
an array spread on arm32

Depends on D153050

Comment 103

[Tim Chevalier]

Attachment: WIP: Bug 1744978: fix stack alignment bug in emitFinishTuple, arm32

Depends on D153051

Comment 104

[Tim Chevalier]

Attachment: WIP: Bug 1744978: Temporarily disable basic-tuples test

Depends on D153052

Comment 105

[Tim Chevalier]

Attachment: WIP: Bug 1744978: Remove spurious pre-barrier for sorted keys elements in Warp AddRecordProperty

Depends on D153053

Comment 106

[Tim Chevalier]

Attachment: WIP: Bug 1744978: Make CallAddRecordSpread MIR op take two Values

Changed CallAddRecordSpread to take both the record and the object as
Values, to use fewer registers

Depends on D153054

Comment 107

[Tim Chevalier]

Attachment: WIP: Bug 1744978: fix rebase error

Depends on D153055

Comment 108

[Tim Chevalier]

Attachment: WIP: Bug 1744978: fix arm64 compilation error

Depends on D153056

Comment 109

[Tim Chevalier]

Attachment: WIP: Bug 1744978: Update for upstream

Depends on D153057

Comment 110

[BugBot [:suhaib / :marco/ :calixte]]

The bug assignee is inactive on Bugzilla, so the assignee is being reset.