Ryujinx/ARMeilleure/IntermediateRepresentation/Operation.cs
FICTURE7 22b2cb39af
Reduce JIT GC allocations (#2515)
* Turn `MemoryOperand` into a struct

* Remove `IntrinsicOperation`

* Remove `PhiNode`

* Remove `Node`

* Turn `Operand` into a struct

* Turn `Operation` into a struct

* Clean up pool management methods

* Add `Arena` allocator

* Move `OperationHelper` to `Operation.Factory`

* Move `OperandHelper` to `Operand.Factory`

* Optimize `Operation` a bit

* Fix `Arena` initialization

* Rename `NativeList<T>` to `ArenaList<T>`

* Reduce `Operand` size from 88 to 56 bytes

* Reduce `Operation` size from 56 to 40 bytes

* Add optimistic interning of Register & Constant operands

* Optimize `RegisterUsage` pass a bit

* Optimize `RemoveUnusedNodes` pass a bit

Iterating in reverse-order allows killing dependency chains in a single
pass.

* Fix PPTC symbols

* Optimize `BasicBlock` a bit

Reduce allocations from `_successor` & `DominanceFrontiers`

* Fix `Operation` resize

* Make `Arena` expandable

Change the arena allocator to be expandable by allocating in pages, with
some of them being pooled. Currently 32 pages are pooled. An LRU removal
mechanism should probably be added to it.

Apparently MHR can allocate bitmaps large enough to exceed the 16MB
limit for the type.

* Move `Arena` & `ArenaList` to `Common`

* Remove `ThreadStaticPool` & co

* Add `PhiOperation`

* Reduce `Operand` size from 56 from 48 bytes

* Add linear-probing to `Operand` intern table

* Optimize `HybridAllocator` a bit

* Add `Allocators` class

* Tune `ArenaAllocator` sizes

* Add page removal mechanism to `ArenaAllocator`

Remove pages which have not been used for more than 5s after each reset.

I am on fence if this would be better using a Gen2 callback object like
the one in System.Buffers.ArrayPool<T>, to trim the pool. Because right
now if a large translation happens, the pages will be freed only after a
reset. This reset may not happen for a while because no new translation
is hit, but the arena base sizes are rather small.

* Fix `OOM` when allocating larger than page size in `ArenaAllocator`

Tweak resizing mechanism for Operand.Uses and Assignemnts.

* Optimize `Optimizer` a bit

* Optimize `Operand.Add<T>/Remove<T>` a bit

* Clean up `PreAllocator`

* Fix phi insertion order

Reduce codegen diffs.

* Fix code alignment

* Use new heuristics for degree of parallelism

* Suppress warnings

* Address gdkchan's feedback

Renamed `GetValue()` to `GetValueUnsafe()` to make it more clear that
`Operand.Value` should usually not be modified directly.

* Add fast path to `ArenaAllocator`

* Assembly for `ArenaAllocator.Allocate(ulong)`:

  .L0:
    mov rax, [rcx+0x18]
    lea r8, [rax+rdx]
    cmp r8, [rcx+0x10]
    ja short .L2
  .L1:
    mov rdx, [rcx+8]
    add rax, [rdx+8]
    mov [rcx+0x18], r8
    ret
  .L2:
    jmp ArenaAllocator.AllocateSlow(UInt64)

  A few variable/field had to be changed to ulong so that RyuJIT avoids
  emitting zero-extends.

* Implement a new heuristic to free pooled pages.

  If an arena is used often, it is more likely that its pages will be
  needed, so the pages are kept for longer (e.g: during PPTC rebuild or
  burst sof compilations). If is not used often, then it is more likely
  that its pages will not be needed (e.g: after PPTC rebuild or bursts
  of compilations).

* Address riperiperi's feedback

* Use `EqualityComparer<T>` in `IntrusiveList<T>`

Avoids a potential GC hole in `Equals(T, T)`.
2021-08-17 15:08:34 -03:00

376 lines
10 KiB
C#

using System;
using System.Diagnostics;
using System.Runtime.CompilerServices;
namespace ARMeilleure.IntermediateRepresentation
{
unsafe struct Operation : IEquatable<Operation>, IIntrusiveListNode<Operation>
{
internal struct Data
{
public ushort Instruction;
public ushort Intrinsic;
public ushort SourcesCount;
public ushort DestinationsCount;
public Operation ListPrevious;
public Operation ListNext;
public Operand* Destinations;
public Operand* Sources;
}
private Data* _data;
public Instruction Instruction
{
get => (Instruction)_data->Instruction;
private set => _data->Instruction = (ushort)value;
}
public Intrinsic Intrinsic
{
get => (Intrinsic)_data->Intrinsic;
private set => _data->Intrinsic = (ushort)value;
}
public Operation ListPrevious
{
get => _data->ListPrevious;
set => _data->ListPrevious = value;
}
public Operation ListNext
{
get => _data->ListNext;
set => _data->ListNext = value;
}
public Operand Destination
{
get => _data->DestinationsCount != 0 ? GetDestination(0) : default;
set => SetDestination(value);
}
public int DestinationsCount => _data->DestinationsCount;
public int SourcesCount => _data->SourcesCount;
private Span<Operand> Destinations => new(_data->Destinations, _data->DestinationsCount);
private Span<Operand> Sources => new(_data->Sources, _data->SourcesCount);
public PhiOperation AsPhi()
{
Debug.Assert(Instruction == Instruction.Phi);
return new PhiOperation(this);
}
public Operand GetDestination(int index)
{
return Destinations[index];
}
public Operand GetSource(int index)
{
return Sources[index];
}
public void SetDestination(int index, Operand dest)
{
ref Operand curDest = ref Destinations[index];
RemoveAssignment(curDest);
AddAssignment(dest);
curDest = dest;
}
public void SetSource(int index, Operand src)
{
ref Operand curSrc = ref Sources[index];
RemoveUse(curSrc);
AddUse(src);
curSrc = src;
}
private void RemoveOldDestinations()
{
for (int i = 0; i < _data->DestinationsCount; i++)
{
RemoveAssignment(_data->Destinations[i]);
}
}
public void SetDestination(Operand dest)
{
RemoveOldDestinations();
if (dest == default)
{
_data->DestinationsCount = 0;
}
else
{
EnsureCapacity(ref _data->Destinations, ref _data->DestinationsCount, 1);
_data->Destinations[0] = dest;
AddAssignment(dest);
}
}
public void SetDestinations(Operand[] dests)
{
RemoveOldDestinations();
EnsureCapacity(ref _data->Destinations, ref _data->DestinationsCount, dests.Length);
for (int index = 0; index < dests.Length; index++)
{
Operand newOp = dests[index];
_data->Destinations[index] = newOp;
AddAssignment(newOp);
}
}
private void RemoveOldSources()
{
for (int index = 0; index < _data->SourcesCount; index++)
{
RemoveUse(_data->Sources[index]);
}
}
public void SetSource(Operand src)
{
RemoveOldSources();
if (src == default)
{
_data->SourcesCount = 0;
}
else
{
EnsureCapacity(ref _data->Sources, ref _data->SourcesCount, 1);
_data->Sources[0] = src;
AddUse(src);
}
}
public void SetSources(Operand[] srcs)
{
RemoveOldSources();
EnsureCapacity(ref _data->Sources, ref _data->SourcesCount, srcs.Length);
for (int index = 0; index < srcs.Length; index++)
{
Operand newOp = srcs[index];
_data->Sources[index] = newOp;
AddUse(newOp);
}
}
public void TurnIntoCopy(Operand source)
{
Instruction = Instruction.Copy;
SetSource(source);
}
private void AddAssignment(Operand op)
{
if (op != default)
{
op.AddAssignment(this);
}
}
private void RemoveAssignment(Operand op)
{
if (op != default)
{
op.RemoveAssignment(this);
}
}
private void AddUse(Operand op)
{
if (op != default)
{
op.AddUse(this);
}
}
private void RemoveUse(Operand op)
{
if (op != default)
{
op.RemoveUse(this);
}
}
public bool Equals(Operation operation)
{
return operation._data == _data;
}
public override bool Equals(object obj)
{
return obj is Operation operation && Equals(operation);
}
public override int GetHashCode()
{
return HashCode.Combine((IntPtr)_data);
}
public static bool operator ==(Operation a, Operation b)
{
return a.Equals(b);
}
public static bool operator !=(Operation a, Operation b)
{
return !a.Equals(b);
}
[MethodImpl(MethodImplOptions.AggressiveInlining)]
private static void EnsureCapacity(ref Operand* list, ref ushort capacity, int newCapacity)
{
if (newCapacity > ushort.MaxValue)
{
ThrowOverflow(newCapacity);
}
// We only need to allocate a new buffer if we're increasing the size.
else if (newCapacity > capacity)
{
list = Allocators.References.Allocate<Operand>((uint)newCapacity);
}
capacity = (ushort)newCapacity;
}
private static void ThrowOverflow(int count) =>
throw new OverflowException($"Exceeded maximum size for Source or Destinations. Required {count}.");
public static class Factory
{
private static Operation Make(Instruction inst, int destCount, int srcCount)
{
Data* data = Allocators.Operations.Allocate<Data>();
*data = default;
Operation result = new();
result._data = data;
result.Instruction = inst;
EnsureCapacity(ref result._data->Destinations, ref result._data->DestinationsCount, destCount);
EnsureCapacity(ref result._data->Sources, ref result._data->SourcesCount, srcCount);
result.Destinations.Clear();
result.Sources.Clear();
return result;
}
public static Operation Operation(Instruction inst, Operand dest)
{
Operation result = Make(inst, 0, 0);
result.SetDestination(dest);
return result;
}
public static Operation Operation(Instruction inst, Operand dest, Operand src0)
{
Operation result = Make(inst, 0, 1);
result.SetDestination(dest);
result.SetSource(0, src0);
return result;
}
public static Operation Operation(Instruction inst, Operand dest, Operand src0, Operand src1)
{
Operation result = Make(inst, 0, 2);
result.SetDestination(dest);
result.SetSource(0, src0);
result.SetSource(1, src1);
return result;
}
public static Operation Operation(Instruction inst, Operand dest, Operand src0, Operand src1, Operand src2)
{
Operation result = Make(inst, 0, 3);
result.SetDestination(dest);
result.SetSource(0, src0);
result.SetSource(1, src1);
result.SetSource(2, src2);
return result;
}
public static Operation Operation(Instruction inst, Operand dest, int srcCount)
{
Operation result = Make(inst, 0, srcCount);
result.SetDestination(dest);
return result;
}
public static Operation Operation(Instruction inst, Operand dest, Operand[] srcs)
{
Operation result = Make(inst, 0, srcs.Length);
result.SetDestination(dest);
for (int index = 0; index < srcs.Length; index++)
{
result.SetSource(index, srcs[index]);
}
return result;
}
public static Operation Operation(Intrinsic intrin, Operand dest, params Operand[] srcs)
{
Operation result = Make(Instruction.Extended, 0, srcs.Length);
result.Intrinsic = intrin;
result.SetDestination(dest);
for (int index = 0; index < srcs.Length; index++)
{
result.SetSource(index, srcs[index]);
}
return result;
}
public static Operation Operation(Instruction inst, Operand[] dests, Operand[] srcs)
{
Operation result = Make(inst, dests.Length, srcs.Length);
for (int index = 0; index < dests.Length; index++)
{
result.SetDestination(index, dests[index]);
}
for (int index = 0; index < srcs.Length; index++)
{
result.SetSource(index, srcs[index]);
}
return result;
}
public static Operation PhiOperation(Operand dest, int srcCount)
{
return Operation(Instruction.Phi, dest, srcCount * 2);
}
}
}
}