JinxRyu/ARMeilleure/CodeGen/Optimizations/Optimizer.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

266 lines
No EOL
8.5 KiB
C#

using ARMeilleure.IntermediateRepresentation;
using ARMeilleure.Translation;
using System;
using System.Diagnostics;
using static ARMeilleure.IntermediateRepresentation.Operand.Factory;
namespace ARMeilleure.CodeGen.Optimizations
{
static class Optimizer
{
public static void RunPass(ControlFlowGraph cfg)
{
// Scratch buffer used to store uses.
Span<Operation> buffer = default;
bool modified;
do
{
modified = false;
for (BasicBlock block = cfg.Blocks.Last; block != null; block = block.ListPrevious)
{
Operation node;
Operation prevNode;
for (node = block.Operations.Last; node != default; node = prevNode)
{
prevNode = node.ListPrevious;
if (IsUnused(node))
{
RemoveNode(block, node);
modified = true;
continue;
}
else if (node.Instruction == Instruction.Phi)
{
continue;
}
ConstantFolding.RunPass(node);
Simplification.RunPass(node);
if (DestIsLocalVar(node))
{
if (IsPropagableCompare(node))
{
modified |= PropagateCompare(ref buffer, node);
if (modified && IsUnused(node))
{
RemoveNode(block, node);
}
}
else if (IsPropagableCopy(node))
{
PropagateCopy(ref buffer, node);
RemoveNode(block, node);
modified = true;
}
}
}
}
}
while (modified);
}
public static void RemoveUnusedNodes(ControlFlowGraph cfg)
{
bool modified;
do
{
modified = false;
for (BasicBlock block = cfg.Blocks.Last; block != null; block = block.ListPrevious)
{
Operation node;
Operation prevNode;
for (node = block.Operations.Last; node != default; node = prevNode)
{
prevNode = node.ListPrevious;
if (IsUnused(node))
{
RemoveNode(block, node);
modified = true;
}
}
}
}
while (modified);
}
private static Span<Operation> GetUses(ref Span<Operation> buffer, Operand operand)
{
ReadOnlySpan<Operation> uses = operand.Uses;
if (buffer.Length < uses.Length)
{
buffer = Allocators.Default.AllocateSpan<Operation>((uint)uses.Length);
}
uses.CopyTo(buffer);
return buffer.Slice(0, uses.Length);
}
private static bool PropagateCompare(ref Span<Operation> buffer, Operation compOp)
{
// Try to propagate Compare operations into their BranchIf uses, when these BranchIf uses are in the form
// of:
//
// - BranchIf %x, 0x0, Equal ;; i.e BranchIfFalse %x
// - BranchIf %x, 0x0, NotEqual ;; i.e BranchIfTrue %x
//
// The commutative property of Equal and NotEqual is taken into consideration as well.
//
// For example:
//
// %x = Compare %a, %b, comp
// BranchIf %x, 0x0, NotEqual
//
// =>
//
// BranchIf %a, %b, comp
static bool IsZeroBranch(Operation operation, out Comparison compType)
{
compType = Comparison.Equal;
if (operation.Instruction != Instruction.BranchIf)
{
return false;
}
Operand src1 = operation.GetSource(0);
Operand src2 = operation.GetSource(1);
Operand comp = operation.GetSource(2);
compType = (Comparison)comp.AsInt32();
return (src1.Kind == OperandKind.Constant && src1.Value == 0) ||
(src2.Kind == OperandKind.Constant && src2.Value == 0);
}
bool modified = false;
Operand dest = compOp.Destination;
Operand src1 = compOp.GetSource(0);
Operand src2 = compOp.GetSource(1);
Operand comp = compOp.GetSource(2);
Comparison compType = (Comparison)comp.AsInt32();
Span<Operation> uses = GetUses(ref buffer, dest);
foreach (Operation use in uses)
{
// If operation is a BranchIf and has a constant value 0 in its RHS or LHS source operands.
if (IsZeroBranch(use, out Comparison otherCompType))
{
Comparison propCompType;
if (otherCompType == Comparison.NotEqual)
{
propCompType = compType;
}
else if (otherCompType == Comparison.Equal)
{
propCompType = compType.Invert();
}
else
{
continue;
}
use.SetSource(0, src1);
use.SetSource(1, src2);
use.SetSource(2, Const((int)propCompType));
modified = true;
}
}
return modified;
}
private static void PropagateCopy(ref Span<Operation> buffer, Operation copyOp)
{
// Propagate copy source operand to all uses of the destination operand.
Operand dest = copyOp.Destination;
Operand source = copyOp.GetSource(0);
Span<Operation> uses = GetUses(ref buffer, dest);
foreach (Operation use in uses)
{
for (int index = 0; index < use.SourcesCount; index++)
{
if (use.GetSource(index) == dest)
{
use.SetSource(index, source);
}
}
}
}
private static void RemoveNode(BasicBlock block, Operation node)
{
// Remove a node from the nodes list, and also remove itself
// from all the use lists on the operands that this node uses.
block.Operations.Remove(node);
for (int index = 0; index < node.SourcesCount; index++)
{
node.SetSource(index, default);
}
Debug.Assert(node.Destination == default || node.Destination.UsesCount == 0);
node.Destination = default;
}
private static bool IsUnused(Operation node)
{
return DestIsLocalVar(node) && node.Destination.UsesCount == 0 && !HasSideEffects(node);
}
private static bool DestIsLocalVar(Operation node)
{
return node.Destination != default && node.Destination.Kind == OperandKind.LocalVariable;
}
private static bool HasSideEffects(Operation node)
{
return node.Instruction == Instruction.Call
|| node.Instruction == Instruction.Tailcall
|| node.Instruction == Instruction.CompareAndSwap
|| node.Instruction == Instruction.CompareAndSwap16
|| node.Instruction == Instruction.CompareAndSwap8;
}
private static bool IsPropagableCompare(Operation operation)
{
return operation.Instruction == Instruction.Compare;
}
private static bool IsPropagableCopy(Operation operation)
{
if (operation.Instruction != Instruction.Copy)
{
return false;
}
return operation.Destination.Type == operation.GetSource(0).Type;
}
}
}