Ryujinx/ARMeilleure/Instructions/InstEmitSimdLogical.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

518 lines
17 KiB
C#

using ARMeilleure.Decoders;
using ARMeilleure.IntermediateRepresentation;
using ARMeilleure.Translation;
using System;
using System.Diagnostics;
using static ARMeilleure.Instructions.InstEmitHelper;
using static ARMeilleure.Instructions.InstEmitSimdHelper;
using static ARMeilleure.IntermediateRepresentation.Operand.Factory;
namespace ARMeilleure.Instructions
{
static partial class InstEmit
{
public static void And_V(ArmEmitterContext context)
{
if (Optimizations.UseSse2)
{
OpCodeSimdReg op = (OpCodeSimdReg)context.CurrOp;
Operand n = GetVec(op.Rn);
Operand m = GetVec(op.Rm);
Operand res = context.AddIntrinsic(Intrinsic.X86Pand, n, m);
if (op.RegisterSize == RegisterSize.Simd64)
{
res = context.VectorZeroUpper64(res);
}
context.Copy(GetVec(op.Rd), res);
}
else
{
EmitVectorBinaryOpZx(context, (op1, op2) => context.BitwiseAnd(op1, op2));
}
}
public static void Bic_V(ArmEmitterContext context)
{
if (Optimizations.UseSse2)
{
OpCodeSimdReg op = (OpCodeSimdReg)context.CurrOp;
Operand n = GetVec(op.Rn);
Operand m = GetVec(op.Rm);
Operand res = context.AddIntrinsic(Intrinsic.X86Pandn, m, n);
if (op.RegisterSize == RegisterSize.Simd64)
{
res = context.VectorZeroUpper64(res);
}
context.Copy(GetVec(op.Rd), res);
}
else
{
EmitVectorBinaryOpZx(context, (op1, op2) =>
{
return context.BitwiseAnd(op1, context.BitwiseNot(op2));
});
}
}
public static void Bic_Vi(ArmEmitterContext context)
{
if (Optimizations.UseSse2)
{
OpCodeSimdImm op = (OpCodeSimdImm)context.CurrOp;
int eSize = 8 << op.Size;
Operand d = GetVec(op.Rd);
Operand imm = eSize switch {
16 => X86GetAllElements(context, (short)~op.Immediate),
32 => X86GetAllElements(context, (int)~op.Immediate),
_ => throw new InvalidOperationException($"Invalid element size {eSize}.")
};
Operand res = context.AddIntrinsic(Intrinsic.X86Pand, d, imm);
if (op.RegisterSize == RegisterSize.Simd64)
{
res = context.VectorZeroUpper64(res);
}
context.Copy(GetVec(op.Rd), res);
}
else
{
EmitVectorImmBinaryOp(context, (op1, op2) =>
{
return context.BitwiseAnd(op1, context.BitwiseNot(op2));
});
}
}
public static void Bif_V(ArmEmitterContext context)
{
EmitBifBit(context, notRm: true);
}
public static void Bit_V(ArmEmitterContext context)
{
EmitBifBit(context, notRm: false);
}
private static void EmitBifBit(ArmEmitterContext context, bool notRm)
{
OpCodeSimdReg op = (OpCodeSimdReg)context.CurrOp;
if (Optimizations.UseSse2)
{
Operand d = GetVec(op.Rd);
Operand n = GetVec(op.Rn);
Operand m = GetVec(op.Rm);
Operand res = context.AddIntrinsic(Intrinsic.X86Pxor, n, d);
if (notRm)
{
res = context.AddIntrinsic(Intrinsic.X86Pandn, m, res);
}
else
{
res = context.AddIntrinsic(Intrinsic.X86Pand, m, res);
}
res = context.AddIntrinsic(Intrinsic.X86Pxor, d, res);
if (op.RegisterSize == RegisterSize.Simd64)
{
res = context.VectorZeroUpper64(res);
}
context.Copy(d, res);
}
else
{
Operand res = context.VectorZero();
int elems = op.RegisterSize == RegisterSize.Simd128 ? 2 : 1;
for (int index = 0; index < elems; index++)
{
Operand d = EmitVectorExtractZx(context, op.Rd, index, 3);
Operand n = EmitVectorExtractZx(context, op.Rn, index, 3);
Operand m = EmitVectorExtractZx(context, op.Rm, index, 3);
if (notRm)
{
m = context.BitwiseNot(m);
}
Operand e = context.BitwiseExclusiveOr(d, n);
e = context.BitwiseAnd(e, m);
e = context.BitwiseExclusiveOr(e, d);
res = EmitVectorInsert(context, res, e, index, 3);
}
context.Copy(GetVec(op.Rd), res);
}
}
public static void Bsl_V(ArmEmitterContext context)
{
if (Optimizations.UseSse2)
{
OpCodeSimdReg op = (OpCodeSimdReg)context.CurrOp;
Operand d = GetVec(op.Rd);
Operand n = GetVec(op.Rn);
Operand m = GetVec(op.Rm);
Operand res = context.AddIntrinsic(Intrinsic.X86Pxor, n, m);
res = context.AddIntrinsic(Intrinsic.X86Pand, res, d);
res = context.AddIntrinsic(Intrinsic.X86Pxor, res, m);
if (op.RegisterSize == RegisterSize.Simd64)
{
res = context.VectorZeroUpper64(res);
}
context.Copy(d, res);
}
else
{
EmitVectorTernaryOpZx(context, (op1, op2, op3) =>
{
return context.BitwiseExclusiveOr(
context.BitwiseAnd(op1,
context.BitwiseExclusiveOr(op2, op3)), op3);
});
}
}
public static void Eor_V(ArmEmitterContext context)
{
if (Optimizations.UseSse2)
{
OpCodeSimdReg op = (OpCodeSimdReg)context.CurrOp;
Operand n = GetVec(op.Rn);
Operand m = GetVec(op.Rm);
Operand res = context.AddIntrinsic(Intrinsic.X86Pxor, n, m);
if (op.RegisterSize == RegisterSize.Simd64)
{
res = context.VectorZeroUpper64(res);
}
context.Copy(GetVec(op.Rd), res);
}
else
{
EmitVectorBinaryOpZx(context, (op1, op2) => context.BitwiseExclusiveOr(op1, op2));
}
}
public static void Not_V(ArmEmitterContext context)
{
if (Optimizations.UseSse2)
{
OpCodeSimd op = (OpCodeSimd)context.CurrOp;
Operand n = GetVec(op.Rn);
Operand mask = X86GetAllElements(context, -1L);
Operand res = context.AddIntrinsic(Intrinsic.X86Pandn, n, mask);
if (op.RegisterSize == RegisterSize.Simd64)
{
res = context.VectorZeroUpper64(res);
}
context.Copy(GetVec(op.Rd), res);
}
else
{
EmitVectorUnaryOpZx(context, (op1) => context.BitwiseNot(op1));
}
}
public static void Orn_V(ArmEmitterContext context)
{
if (Optimizations.UseSse2)
{
OpCodeSimdReg op = (OpCodeSimdReg)context.CurrOp;
Operand n = GetVec(op.Rn);
Operand m = GetVec(op.Rm);
Operand mask = X86GetAllElements(context, -1L);
Operand res = context.AddIntrinsic(Intrinsic.X86Pandn, m, mask);
res = context.AddIntrinsic(Intrinsic.X86Por, res, n);
if (op.RegisterSize == RegisterSize.Simd64)
{
res = context.VectorZeroUpper64(res);
}
context.Copy(GetVec(op.Rd), res);
}
else
{
EmitVectorBinaryOpZx(context, (op1, op2) =>
{
return context.BitwiseOr(op1, context.BitwiseNot(op2));
});
}
}
public static void Orr_V(ArmEmitterContext context)
{
if (Optimizations.UseSse2)
{
OpCodeSimdReg op = (OpCodeSimdReg)context.CurrOp;
Operand n = GetVec(op.Rn);
Operand m = GetVec(op.Rm);
Operand res = context.AddIntrinsic(Intrinsic.X86Por, n, m);
if (op.RegisterSize == RegisterSize.Simd64)
{
res = context.VectorZeroUpper64(res);
}
context.Copy(GetVec(op.Rd), res);
}
else
{
EmitVectorBinaryOpZx(context, (op1, op2) => context.BitwiseOr(op1, op2));
}
}
public static void Orr_Vi(ArmEmitterContext context)
{
if (Optimizations.UseSse2)
{
OpCodeSimdImm op = (OpCodeSimdImm)context.CurrOp;
int eSize = 8 << op.Size;
Operand d = GetVec(op.Rd);
Operand imm = eSize switch {
16 => X86GetAllElements(context, (short)op.Immediate),
32 => X86GetAllElements(context, (int)op.Immediate),
_ => throw new InvalidOperationException($"Invalid element size {eSize}.")
};
Operand res = context.AddIntrinsic(Intrinsic.X86Por, d, imm);
if (op.RegisterSize == RegisterSize.Simd64)
{
res = context.VectorZeroUpper64(res);
}
context.Copy(GetVec(op.Rd), res);
}
else
{
EmitVectorImmBinaryOp(context, (op1, op2) => context.BitwiseOr(op1, op2));
}
}
public static void Rbit_V(ArmEmitterContext context)
{
OpCodeSimd op = (OpCodeSimd)context.CurrOp;
Operand res = context.VectorZero();
int elems = op.RegisterSize == RegisterSize.Simd128 ? 16 : 8;
for (int index = 0; index < elems; index++)
{
Operand ne = EmitVectorExtractZx(context, op.Rn, index, 0);
Operand de = EmitReverseBits8Op(context, ne);
res = EmitVectorInsert(context, res, de, index, 0);
}
context.Copy(GetVec(op.Rd), res);
}
private static Operand EmitReverseBits8Op(ArmEmitterContext context, Operand op)
{
Debug.Assert(op.Type == OperandType.I64);
Operand val = context.BitwiseOr(context.ShiftRightUI(context.BitwiseAnd(op, Const(0xaaul)), Const(1)),
context.ShiftLeft (context.BitwiseAnd(op, Const(0x55ul)), Const(1)));
val = context.BitwiseOr(context.ShiftRightUI(context.BitwiseAnd(val, Const(0xccul)), Const(2)),
context.ShiftLeft (context.BitwiseAnd(val, Const(0x33ul)), Const(2)));
return context.BitwiseOr(context.ShiftRightUI(val, Const(4)),
context.ShiftLeft (context.BitwiseAnd(val, Const(0x0ful)), Const(4)));
}
public static void Rev16_V(ArmEmitterContext context)
{
if (Optimizations.UseSsse3)
{
OpCodeSimd op = (OpCodeSimd)context.CurrOp;
Operand n = GetVec(op.Rn);
const long maskE0 = 06L << 56 | 07L << 48 | 04L << 40 | 05L << 32 | 02L << 24 | 03L << 16 | 00L << 8 | 01L << 0;
const long maskE1 = 14L << 56 | 15L << 48 | 12L << 40 | 13L << 32 | 10L << 24 | 11L << 16 | 08L << 8 | 09L << 0;
Operand mask = X86GetScalar(context, maskE0);
mask = EmitVectorInsert(context, mask, Const(maskE1), 1, 3);
Operand res = context.AddIntrinsic(Intrinsic.X86Pshufb, n, mask);
if (op.RegisterSize == RegisterSize.Simd64)
{
res = context.VectorZeroUpper64(res);
}
context.Copy(GetVec(op.Rd), res);
}
else
{
EmitRev_V(context, containerSize: 1);
}
}
public static void Rev32_V(ArmEmitterContext context)
{
if (Optimizations.UseSsse3)
{
OpCodeSimd op = (OpCodeSimd)context.CurrOp;
Operand n = GetVec(op.Rn);
Operand mask;
if (op.Size == 0)
{
const long maskE0 = 04L << 56 | 05L << 48 | 06L << 40 | 07L << 32 | 00L << 24 | 01L << 16 | 02L << 8 | 03L << 0;
const long maskE1 = 12L << 56 | 13L << 48 | 14L << 40 | 15L << 32 | 08L << 24 | 09L << 16 | 10L << 8 | 11L << 0;
mask = X86GetScalar(context, maskE0);
mask = EmitVectorInsert(context, mask, Const(maskE1), 1, 3);
}
else /* if (op.Size == 1) */
{
const long maskE0 = 05L << 56 | 04L << 48 | 07L << 40 | 06L << 32 | 01L << 24 | 00L << 16 | 03L << 8 | 02L << 0;
const long maskE1 = 13L << 56 | 12L << 48 | 15L << 40 | 14L << 32 | 09L << 24 | 08L << 16 | 11L << 8 | 10L << 0;
mask = X86GetScalar(context, maskE0);
mask = EmitVectorInsert(context, mask, Const(maskE1), 1, 3);
}
Operand res = context.AddIntrinsic(Intrinsic.X86Pshufb, n, mask);
if (op.RegisterSize == RegisterSize.Simd64)
{
res = context.VectorZeroUpper64(res);
}
context.Copy(GetVec(op.Rd), res);
}
else
{
EmitRev_V(context, containerSize: 2);
}
}
public static void Rev64_V(ArmEmitterContext context)
{
if (Optimizations.UseSsse3)
{
OpCodeSimd op = (OpCodeSimd)context.CurrOp;
Operand n = GetVec(op.Rn);
Operand mask;
if (op.Size == 0)
{
const long maskE0 = 00L << 56 | 01L << 48 | 02L << 40 | 03L << 32 | 04L << 24 | 05L << 16 | 06L << 8 | 07L << 0;
const long maskE1 = 08L << 56 | 09L << 48 | 10L << 40 | 11L << 32 | 12L << 24 | 13L << 16 | 14L << 8 | 15L << 0;
mask = X86GetScalar(context, maskE0);
mask = EmitVectorInsert(context, mask, Const(maskE1), 1, 3);
}
else if (op.Size == 1)
{
const long maskE0 = 01L << 56 | 00L << 48 | 03L << 40 | 02L << 32 | 05L << 24 | 04L << 16 | 07L << 8 | 06L << 0;
const long maskE1 = 09L << 56 | 08L << 48 | 11L << 40 | 10L << 32 | 13L << 24 | 12L << 16 | 15L << 8 | 14L << 0;
mask = X86GetScalar(context, maskE0);
mask = EmitVectorInsert(context, mask, Const(maskE1), 1, 3);
}
else /* if (op.Size == 2) */
{
const long maskE0 = 03L << 56 | 02L << 48 | 01L << 40 | 00L << 32 | 07L << 24 | 06L << 16 | 05L << 8 | 04L << 0;
const long maskE1 = 11L << 56 | 10L << 48 | 09L << 40 | 08L << 32 | 15L << 24 | 14L << 16 | 13L << 8 | 12L << 0;
mask = X86GetScalar(context, maskE0);
mask = EmitVectorInsert(context, mask, Const(maskE1), 1, 3);
}
Operand res = context.AddIntrinsic(Intrinsic.X86Pshufb, n, mask);
if (op.RegisterSize == RegisterSize.Simd64)
{
res = context.VectorZeroUpper64(res);
}
context.Copy(GetVec(op.Rd), res);
}
else
{
EmitRev_V(context, containerSize: 3);
}
}
private static void EmitRev_V(ArmEmitterContext context, int containerSize)
{
OpCodeSimd op = (OpCodeSimd)context.CurrOp;
Operand res = context.VectorZero();
int elems = op.GetBytesCount() >> op.Size;
int containerMask = (1 << (containerSize - op.Size)) - 1;
for (int index = 0; index < elems; index++)
{
int revIndex = index ^ containerMask;
Operand ne = EmitVectorExtractZx(context, op.Rn, revIndex, op.Size);
res = EmitVectorInsert(context, res, ne, index, op.Size);
}
context.Copy(GetVec(op.Rd), res);
}
}
}