22b2cb39af
* 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)`.
850 lines
28 KiB
C#
850 lines
28 KiB
C#
using ARMeilleure.Decoders;
|
|
using ARMeilleure.IntermediateRepresentation;
|
|
using ARMeilleure.Translation;
|
|
using System.Collections.Generic;
|
|
using System.Reflection;
|
|
|
|
using static ARMeilleure.Instructions.InstEmitHelper;
|
|
using static ARMeilleure.Instructions.InstEmitSimdHelper;
|
|
using static ARMeilleure.IntermediateRepresentation.Operand.Factory;
|
|
|
|
namespace ARMeilleure.Instructions
|
|
{
|
|
static partial class InstEmit
|
|
{
|
|
#region "Masks"
|
|
private static readonly long[] _masksE0_Uzp = new long[]
|
|
{
|
|
13L << 56 | 09L << 48 | 05L << 40 | 01L << 32 | 12L << 24 | 08L << 16 | 04L << 8 | 00L << 0,
|
|
11L << 56 | 10L << 48 | 03L << 40 | 02L << 32 | 09L << 24 | 08L << 16 | 01L << 8 | 00L << 0
|
|
};
|
|
|
|
private static readonly long[] _masksE1_Uzp = new long[]
|
|
{
|
|
15L << 56 | 11L << 48 | 07L << 40 | 03L << 32 | 14L << 24 | 10L << 16 | 06L << 8 | 02L << 0,
|
|
15L << 56 | 14L << 48 | 07L << 40 | 06L << 32 | 13L << 24 | 12L << 16 | 05L << 8 | 04L << 0
|
|
};
|
|
#endregion
|
|
|
|
public static void Dup_Gp(ArmEmitterContext context)
|
|
{
|
|
OpCodeSimdIns op = (OpCodeSimdIns)context.CurrOp;
|
|
|
|
Operand n = GetIntOrZR(context, op.Rn);
|
|
|
|
if (Optimizations.UseSse2)
|
|
{
|
|
switch (op.Size)
|
|
{
|
|
case 0: n = context.ZeroExtend8 (n.Type, n); n = context.Multiply(n, Const(n.Type, 0x01010101)); break;
|
|
case 1: n = context.ZeroExtend16(n.Type, n); n = context.Multiply(n, Const(n.Type, 0x00010001)); break;
|
|
case 2: n = context.ZeroExtend32(n.Type, n); break;
|
|
}
|
|
|
|
Operand res = context.VectorInsert(context.VectorZero(), n, 0);
|
|
|
|
if (op.Size < 3)
|
|
{
|
|
if (op.RegisterSize == RegisterSize.Simd64)
|
|
{
|
|
res = context.AddIntrinsic(Intrinsic.X86Shufps, res, res, Const(0xf0));
|
|
}
|
|
else
|
|
{
|
|
res = context.AddIntrinsic(Intrinsic.X86Shufps, res, res, Const(0));
|
|
}
|
|
}
|
|
else
|
|
{
|
|
res = context.AddIntrinsic(Intrinsic.X86Movlhps, res, res);
|
|
}
|
|
|
|
context.Copy(GetVec(op.Rd), res);
|
|
}
|
|
else
|
|
{
|
|
Operand res = context.VectorZero();
|
|
|
|
int elems = op.GetBytesCount() >> op.Size;
|
|
|
|
for (int index = 0; index < elems; index++)
|
|
{
|
|
res = EmitVectorInsert(context, res, n, index, op.Size);
|
|
}
|
|
|
|
context.Copy(GetVec(op.Rd), res);
|
|
}
|
|
}
|
|
|
|
public static void Dup_S(ArmEmitterContext context)
|
|
{
|
|
OpCodeSimdIns op = (OpCodeSimdIns)context.CurrOp;
|
|
|
|
Operand ne = EmitVectorExtractZx(context, op.Rn, op.DstIndex, op.Size);
|
|
|
|
context.Copy(GetVec(op.Rd), EmitVectorInsert(context, context.VectorZero(), ne, 0, op.Size));
|
|
}
|
|
|
|
public static void Dup_V(ArmEmitterContext context)
|
|
{
|
|
OpCodeSimdIns op = (OpCodeSimdIns)context.CurrOp;
|
|
|
|
if (Optimizations.UseSse2)
|
|
{
|
|
Operand res = GetVec(op.Rn);
|
|
|
|
if (op.Size == 0)
|
|
{
|
|
if (op.DstIndex != 0)
|
|
{
|
|
res = context.AddIntrinsic(Intrinsic.X86Psrldq, res, Const(op.DstIndex));
|
|
}
|
|
|
|
res = context.AddIntrinsic(Intrinsic.X86Punpcklbw, res, res);
|
|
res = context.AddIntrinsic(Intrinsic.X86Punpcklwd, res, res);
|
|
res = context.AddIntrinsic(Intrinsic.X86Shufps, res, res, Const(0));
|
|
}
|
|
else if (op.Size == 1)
|
|
{
|
|
if (op.DstIndex != 0)
|
|
{
|
|
res = context.AddIntrinsic(Intrinsic.X86Psrldq, res, Const(op.DstIndex * 2));
|
|
}
|
|
|
|
res = context.AddIntrinsic(Intrinsic.X86Punpcklwd, res, res);
|
|
res = context.AddIntrinsic(Intrinsic.X86Shufps, res, res, Const(0));
|
|
}
|
|
else if (op.Size == 2)
|
|
{
|
|
int mask = op.DstIndex * 0b01010101;
|
|
|
|
res = context.AddIntrinsic(Intrinsic.X86Shufps, res, res, Const(mask));
|
|
}
|
|
else if (op.DstIndex == 0 && op.RegisterSize != RegisterSize.Simd64)
|
|
{
|
|
res = context.AddIntrinsic(Intrinsic.X86Movlhps, res, res);
|
|
}
|
|
else if (op.DstIndex == 1)
|
|
{
|
|
res = context.AddIntrinsic(Intrinsic.X86Movhlps, res, res);
|
|
}
|
|
|
|
if (op.RegisterSize == RegisterSize.Simd64)
|
|
{
|
|
res = context.VectorZeroUpper64(res);
|
|
}
|
|
|
|
context.Copy(GetVec(op.Rd), res);
|
|
}
|
|
else
|
|
{
|
|
Operand ne = EmitVectorExtractZx(context, op.Rn, op.DstIndex, op.Size);
|
|
|
|
Operand res = context.VectorZero();
|
|
|
|
int elems = op.GetBytesCount() >> op.Size;
|
|
|
|
for (int index = 0; index < elems; index++)
|
|
{
|
|
res = EmitVectorInsert(context, res, ne, index, op.Size);
|
|
}
|
|
|
|
context.Copy(GetVec(op.Rd), res);
|
|
}
|
|
}
|
|
|
|
public static void Ext_V(ArmEmitterContext context)
|
|
{
|
|
OpCodeSimdExt op = (OpCodeSimdExt)context.CurrOp;
|
|
|
|
if (Optimizations.UseSse2)
|
|
{
|
|
Operand nShifted = GetVec(op.Rn);
|
|
|
|
if (op.RegisterSize == RegisterSize.Simd64)
|
|
{
|
|
nShifted = context.VectorZeroUpper64(nShifted);
|
|
}
|
|
|
|
nShifted = context.AddIntrinsic(Intrinsic.X86Psrldq, nShifted, Const(op.Imm4));
|
|
|
|
Operand mShifted = GetVec(op.Rm);
|
|
|
|
mShifted = context.AddIntrinsic(Intrinsic.X86Pslldq, mShifted, Const(op.GetBytesCount() - op.Imm4));
|
|
|
|
if (op.RegisterSize == RegisterSize.Simd64)
|
|
{
|
|
mShifted = context.VectorZeroUpper64(mShifted);
|
|
}
|
|
|
|
Operand res = context.AddIntrinsic(Intrinsic.X86Por, nShifted, mShifted);
|
|
|
|
context.Copy(GetVec(op.Rd), res);
|
|
}
|
|
else
|
|
{
|
|
Operand res = context.VectorZero();
|
|
|
|
int bytes = op.GetBytesCount();
|
|
|
|
int position = op.Imm4 & (bytes - 1);
|
|
|
|
for (int index = 0; index < bytes; index++)
|
|
{
|
|
int reg = op.Imm4 + index < bytes ? op.Rn : op.Rm;
|
|
|
|
Operand e = EmitVectorExtractZx(context, reg, position, 0);
|
|
|
|
position = (position + 1) & (bytes - 1);
|
|
|
|
res = EmitVectorInsert(context, res, e, index, 0);
|
|
}
|
|
|
|
context.Copy(GetVec(op.Rd), res);
|
|
}
|
|
}
|
|
|
|
public static void Fcsel_S(ArmEmitterContext context)
|
|
{
|
|
OpCodeSimdFcond op = (OpCodeSimdFcond)context.CurrOp;
|
|
|
|
Operand lblTrue = Label();
|
|
Operand lblEnd = Label();
|
|
|
|
Operand isTrue = InstEmitFlowHelper.GetCondTrue(context, op.Cond);
|
|
|
|
context.BranchIfTrue(lblTrue, isTrue);
|
|
|
|
OperandType type = op.Size == 0 ? OperandType.FP32 : OperandType.FP64;
|
|
|
|
Operand me = context.VectorExtract(type, GetVec(op.Rm), 0);
|
|
|
|
context.Copy(GetVec(op.Rd), context.VectorInsert(context.VectorZero(), me, 0));
|
|
|
|
context.Branch(lblEnd);
|
|
|
|
context.MarkLabel(lblTrue);
|
|
|
|
Operand ne = context.VectorExtract(type, GetVec(op.Rn), 0);
|
|
|
|
context.Copy(GetVec(op.Rd), context.VectorInsert(context.VectorZero(), ne, 0));
|
|
|
|
context.MarkLabel(lblEnd);
|
|
}
|
|
|
|
public static void Fmov_Ftoi(ArmEmitterContext context)
|
|
{
|
|
OpCodeSimd op = (OpCodeSimd)context.CurrOp;
|
|
|
|
Operand ne = EmitVectorExtractZx(context, op.Rn, 0, op.Size + 2);
|
|
|
|
SetIntOrZR(context, op.Rd, ne);
|
|
}
|
|
|
|
public static void Fmov_Ftoi1(ArmEmitterContext context)
|
|
{
|
|
OpCodeSimd op = (OpCodeSimd)context.CurrOp;
|
|
|
|
Operand ne = EmitVectorExtractZx(context, op.Rn, 1, 3);
|
|
|
|
SetIntOrZR(context, op.Rd, ne);
|
|
}
|
|
|
|
public static void Fmov_Itof(ArmEmitterContext context)
|
|
{
|
|
OpCodeSimd op = (OpCodeSimd)context.CurrOp;
|
|
|
|
Operand n = GetIntOrZR(context, op.Rn);
|
|
|
|
context.Copy(GetVec(op.Rd), EmitVectorInsert(context, context.VectorZero(), n, 0, op.Size + 2));
|
|
}
|
|
|
|
public static void Fmov_Itof1(ArmEmitterContext context)
|
|
{
|
|
OpCodeSimd op = (OpCodeSimd)context.CurrOp;
|
|
|
|
Operand d = GetVec(op.Rd);
|
|
Operand n = GetIntOrZR(context, op.Rn);
|
|
|
|
context.Copy(d, EmitVectorInsert(context, d, n, 1, 3));
|
|
}
|
|
|
|
public static void Fmov_S(ArmEmitterContext context)
|
|
{
|
|
OpCodeSimd op = (OpCodeSimd)context.CurrOp;
|
|
|
|
OperandType type = op.Size == 0 ? OperandType.FP32 : OperandType.FP64;
|
|
|
|
Operand ne = context.VectorExtract(type, GetVec(op.Rn), 0);
|
|
|
|
context.Copy(GetVec(op.Rd), context.VectorInsert(context.VectorZero(), ne, 0));
|
|
}
|
|
|
|
public static void Fmov_Si(ArmEmitterContext context)
|
|
{
|
|
OpCodeSimdFmov op = (OpCodeSimdFmov)context.CurrOp;
|
|
|
|
if (Optimizations.UseSse2)
|
|
{
|
|
if (op.Size == 0)
|
|
{
|
|
context.Copy(GetVec(op.Rd), X86GetScalar(context, (int)op.Immediate));
|
|
}
|
|
else
|
|
{
|
|
context.Copy(GetVec(op.Rd), X86GetScalar(context, op.Immediate));
|
|
}
|
|
}
|
|
else
|
|
{
|
|
Operand e = Const(op.Immediate);
|
|
|
|
Operand res = context.VectorZero();
|
|
|
|
res = EmitVectorInsert(context, res, e, 0, op.Size + 2);
|
|
|
|
context.Copy(GetVec(op.Rd), res);
|
|
}
|
|
}
|
|
|
|
public static void Fmov_Vi(ArmEmitterContext context)
|
|
{
|
|
OpCodeSimdImm op = (OpCodeSimdImm)context.CurrOp;
|
|
|
|
if (Optimizations.UseSse2)
|
|
{
|
|
if (op.RegisterSize == RegisterSize.Simd128)
|
|
{
|
|
context.Copy(GetVec(op.Rd), X86GetAllElements(context, op.Immediate));
|
|
}
|
|
else
|
|
{
|
|
context.Copy(GetVec(op.Rd), X86GetScalar(context, op.Immediate));
|
|
}
|
|
}
|
|
else
|
|
{
|
|
Operand e = Const(op.Immediate);
|
|
|
|
Operand res = context.VectorZero();
|
|
|
|
int elems = op.RegisterSize == RegisterSize.Simd128 ? 2 : 1;
|
|
|
|
for (int index = 0; index < elems; index++)
|
|
{
|
|
res = EmitVectorInsert(context, res, e, index, 3);
|
|
}
|
|
|
|
context.Copy(GetVec(op.Rd), res);
|
|
}
|
|
}
|
|
|
|
public static void Ins_Gp(ArmEmitterContext context)
|
|
{
|
|
OpCodeSimdIns op = (OpCodeSimdIns)context.CurrOp;
|
|
|
|
Operand d = GetVec(op.Rd);
|
|
Operand n = GetIntOrZR(context, op.Rn);
|
|
|
|
context.Copy(d, EmitVectorInsert(context, d, n, op.DstIndex, op.Size));
|
|
}
|
|
|
|
public static void Ins_V(ArmEmitterContext context)
|
|
{
|
|
OpCodeSimdIns op = (OpCodeSimdIns)context.CurrOp;
|
|
|
|
Operand d = GetVec(op.Rd);
|
|
Operand ne = EmitVectorExtractZx(context, op.Rn, op.SrcIndex, op.Size);
|
|
|
|
context.Copy(d, EmitVectorInsert(context, d, ne, op.DstIndex, op.Size));
|
|
}
|
|
|
|
public static void Movi_V(ArmEmitterContext context)
|
|
{
|
|
if (Optimizations.UseSse2)
|
|
{
|
|
EmitSse2VectorMoviMvniOp(context, not: false);
|
|
}
|
|
else
|
|
{
|
|
EmitVectorImmUnaryOp(context, (op1) => op1);
|
|
}
|
|
}
|
|
|
|
public static void Mvni_V(ArmEmitterContext context)
|
|
{
|
|
if (Optimizations.UseSse2)
|
|
{
|
|
EmitSse2VectorMoviMvniOp(context, not: true);
|
|
}
|
|
else
|
|
{
|
|
EmitVectorImmUnaryOp(context, (op1) => context.BitwiseNot(op1));
|
|
}
|
|
}
|
|
|
|
public static void Smov_S(ArmEmitterContext context)
|
|
{
|
|
OpCodeSimdIns op = (OpCodeSimdIns)context.CurrOp;
|
|
|
|
Operand ne = EmitVectorExtractSx(context, op.Rn, op.DstIndex, op.Size);
|
|
|
|
if (op.RegisterSize == RegisterSize.Simd64)
|
|
{
|
|
ne = context.ZeroExtend32(OperandType.I64, ne);
|
|
}
|
|
|
|
SetIntOrZR(context, op.Rd, ne);
|
|
}
|
|
|
|
public static void Tbl_V(ArmEmitterContext context)
|
|
{
|
|
EmitTableVectorLookup(context, isTbl: true);
|
|
}
|
|
|
|
public static void Tbx_V(ArmEmitterContext context)
|
|
{
|
|
EmitTableVectorLookup(context, isTbl: false);
|
|
}
|
|
|
|
public static void Trn1_V(ArmEmitterContext context)
|
|
{
|
|
EmitVectorTranspose(context, part: 0);
|
|
}
|
|
|
|
public static void Trn2_V(ArmEmitterContext context)
|
|
{
|
|
EmitVectorTranspose(context, part: 1);
|
|
}
|
|
|
|
public static void Umov_S(ArmEmitterContext context)
|
|
{
|
|
OpCodeSimdIns op = (OpCodeSimdIns)context.CurrOp;
|
|
|
|
Operand ne = EmitVectorExtractZx(context, op.Rn, op.DstIndex, op.Size);
|
|
|
|
SetIntOrZR(context, op.Rd, ne);
|
|
}
|
|
|
|
public static void Uzp1_V(ArmEmitterContext context)
|
|
{
|
|
EmitVectorUnzip(context, part: 0);
|
|
}
|
|
|
|
public static void Uzp2_V(ArmEmitterContext context)
|
|
{
|
|
EmitVectorUnzip(context, part: 1);
|
|
}
|
|
|
|
public static void Xtn_V(ArmEmitterContext context)
|
|
{
|
|
OpCodeSimd op = (OpCodeSimd)context.CurrOp;
|
|
|
|
if (Optimizations.UseSsse3)
|
|
{
|
|
Operand d = GetVec(op.Rd);
|
|
|
|
Operand res = context.VectorZeroUpper64(d);
|
|
|
|
Operand mask = X86GetAllElements(context, EvenMasks[op.Size]);
|
|
|
|
Operand res2 = context.AddIntrinsic(Intrinsic.X86Pshufb, GetVec(op.Rn), mask);
|
|
|
|
Intrinsic movInst = op.RegisterSize == RegisterSize.Simd128
|
|
? Intrinsic.X86Movlhps
|
|
: Intrinsic.X86Movhlps;
|
|
|
|
res = context.AddIntrinsic(movInst, res, res2);
|
|
|
|
context.Copy(d, res);
|
|
}
|
|
else
|
|
{
|
|
int elems = 8 >> op.Size;
|
|
|
|
int part = op.RegisterSize == RegisterSize.Simd128 ? elems : 0;
|
|
|
|
Operand d = GetVec(op.Rd);
|
|
|
|
Operand res = part == 0 ? context.VectorZero() : context.Copy(d);
|
|
|
|
for (int index = 0; index < elems; index++)
|
|
{
|
|
Operand ne = EmitVectorExtractZx(context, op.Rn, index, op.Size + 1);
|
|
|
|
res = EmitVectorInsert(context, res, ne, part + index, op.Size);
|
|
}
|
|
|
|
context.Copy(d, res);
|
|
}
|
|
}
|
|
|
|
public static void Zip1_V(ArmEmitterContext context)
|
|
{
|
|
EmitVectorZip(context, part: 0);
|
|
}
|
|
|
|
public static void Zip2_V(ArmEmitterContext context)
|
|
{
|
|
EmitVectorZip(context, part: 1);
|
|
}
|
|
|
|
private static void EmitSse2VectorMoviMvniOp(ArmEmitterContext context, bool not)
|
|
{
|
|
OpCodeSimdImm op = (OpCodeSimdImm)context.CurrOp;
|
|
|
|
long imm = op.Immediate;
|
|
|
|
switch (op.Size)
|
|
{
|
|
case 0: imm *= 0x01010101; break;
|
|
case 1: imm *= 0x00010001; break;
|
|
}
|
|
|
|
if (not)
|
|
{
|
|
imm = ~imm;
|
|
}
|
|
|
|
Operand mask;
|
|
|
|
if (op.Size < 3)
|
|
{
|
|
mask = X86GetAllElements(context, (int)imm);
|
|
}
|
|
else
|
|
{
|
|
mask = X86GetAllElements(context, imm);
|
|
}
|
|
|
|
if (op.RegisterSize == RegisterSize.Simd64)
|
|
{
|
|
mask = context.VectorZeroUpper64(mask);
|
|
}
|
|
|
|
context.Copy(GetVec(op.Rd), mask);
|
|
}
|
|
|
|
private static void EmitTableVectorLookup(ArmEmitterContext context, bool isTbl)
|
|
{
|
|
OpCodeSimdTbl op = (OpCodeSimdTbl)context.CurrOp;
|
|
|
|
if (Optimizations.UseSsse3)
|
|
{
|
|
Operand d = GetVec(op.Rd);
|
|
Operand m = GetVec(op.Rm);
|
|
|
|
Operand res;
|
|
|
|
Operand mask = X86GetAllElements(context, 0x0F0F0F0F0F0F0F0FL);
|
|
|
|
// Fast path for single register table.
|
|
{
|
|
Operand n = GetVec(op.Rn);
|
|
|
|
Operand mMask = context.AddIntrinsic(Intrinsic.X86Pcmpgtb, m, mask);
|
|
mMask = context.AddIntrinsic(Intrinsic.X86Por, mMask, m);
|
|
|
|
res = context.AddIntrinsic(Intrinsic.X86Pshufb, n, mMask);
|
|
}
|
|
|
|
for (int index = 1; index < op.Size; index++)
|
|
{
|
|
Operand ni = GetVec((op.Rn + index) & 0x1F);
|
|
|
|
Operand idxMask = X86GetAllElements(context, 0x1010101010101010L * index);
|
|
|
|
Operand mSubMask = context.AddIntrinsic(Intrinsic.X86Psubb, m, idxMask);
|
|
|
|
Operand mMask = context.AddIntrinsic(Intrinsic.X86Pcmpgtb, mSubMask, mask);
|
|
mMask = context.AddIntrinsic(Intrinsic.X86Por, mMask, mSubMask);
|
|
|
|
Operand res2 = context.AddIntrinsic(Intrinsic.X86Pshufb, ni, mMask);
|
|
|
|
res = context.AddIntrinsic(Intrinsic.X86Por, res, res2);
|
|
}
|
|
|
|
if (!isTbl)
|
|
{
|
|
Operand idxMask = X86GetAllElements(context, (0x1010101010101010L * op.Size) - 0x0101010101010101L);
|
|
Operand zeroMask = context.VectorZero();
|
|
|
|
Operand mPosMask = context.AddIntrinsic(Intrinsic.X86Pcmpgtb, m, idxMask);
|
|
Operand mNegMask = context.AddIntrinsic(Intrinsic.X86Pcmpgtb, zeroMask, m);
|
|
|
|
Operand mMask = context.AddIntrinsic(Intrinsic.X86Por, mPosMask, mNegMask);
|
|
|
|
Operand dMask = context.AddIntrinsic(Intrinsic.X86Pand, d, mMask);
|
|
|
|
res = context.AddIntrinsic(Intrinsic.X86Por, res, dMask);
|
|
}
|
|
|
|
if (op.RegisterSize == RegisterSize.Simd64)
|
|
{
|
|
res = context.VectorZeroUpper64(res);
|
|
}
|
|
|
|
context.Copy(d, res);
|
|
}
|
|
else
|
|
{
|
|
Operand d = GetVec(op.Rd);
|
|
|
|
List<Operand> args = new List<Operand>();
|
|
|
|
if (!isTbl)
|
|
{
|
|
args.Add(d);
|
|
}
|
|
|
|
args.Add(GetVec(op.Rm));
|
|
|
|
args.Add(Const(op.RegisterSize == RegisterSize.Simd64 ? 8 : 16));
|
|
|
|
for (int index = 0; index < op.Size; index++)
|
|
{
|
|
args.Add(GetVec((op.Rn + index) & 0x1F));
|
|
}
|
|
|
|
MethodInfo info = null;
|
|
|
|
if (isTbl)
|
|
{
|
|
switch (op.Size)
|
|
{
|
|
case 1: info = typeof(SoftFallback).GetMethod(nameof(SoftFallback.Tbl1)); break;
|
|
case 2: info = typeof(SoftFallback).GetMethod(nameof(SoftFallback.Tbl2)); break;
|
|
case 3: info = typeof(SoftFallback).GetMethod(nameof(SoftFallback.Tbl3)); break;
|
|
case 4: info = typeof(SoftFallback).GetMethod(nameof(SoftFallback.Tbl4)); break;
|
|
}
|
|
}
|
|
else
|
|
{
|
|
switch (op.Size)
|
|
{
|
|
case 1: info = typeof(SoftFallback).GetMethod(nameof(SoftFallback.Tbx1)); break;
|
|
case 2: info = typeof(SoftFallback).GetMethod(nameof(SoftFallback.Tbx2)); break;
|
|
case 3: info = typeof(SoftFallback).GetMethod(nameof(SoftFallback.Tbx3)); break;
|
|
case 4: info = typeof(SoftFallback).GetMethod(nameof(SoftFallback.Tbx4)); break;
|
|
}
|
|
}
|
|
|
|
context.Copy(d, context.Call(info, args.ToArray()));
|
|
}
|
|
}
|
|
|
|
private static void EmitVectorTranspose(ArmEmitterContext context, int part)
|
|
{
|
|
OpCodeSimdReg op = (OpCodeSimdReg)context.CurrOp;
|
|
|
|
if (Optimizations.UseSsse3)
|
|
{
|
|
Operand mask = default;
|
|
|
|
if (op.Size < 3)
|
|
{
|
|
long maskE0 = EvenMasks[op.Size];
|
|
long maskE1 = OddMasks [op.Size];
|
|
|
|
mask = X86GetScalar(context, maskE0);
|
|
|
|
mask = EmitVectorInsert(context, mask, Const(maskE1), 1, 3);
|
|
}
|
|
|
|
Operand n = GetVec(op.Rn);
|
|
|
|
if (op.Size < 3)
|
|
{
|
|
n = context.AddIntrinsic(Intrinsic.X86Pshufb, n, mask);
|
|
}
|
|
|
|
Operand m = GetVec(op.Rm);
|
|
|
|
if (op.Size < 3)
|
|
{
|
|
m = context.AddIntrinsic(Intrinsic.X86Pshufb, m, mask);
|
|
}
|
|
|
|
Intrinsic punpckInst = part == 0
|
|
? X86PunpcklInstruction[op.Size]
|
|
: X86PunpckhInstruction[op.Size];
|
|
|
|
Operand res = context.AddIntrinsic(punpckInst, n, m);
|
|
|
|
if (op.RegisterSize == RegisterSize.Simd64)
|
|
{
|
|
res = context.VectorZeroUpper64(res);
|
|
}
|
|
|
|
context.Copy(GetVec(op.Rd), res);
|
|
}
|
|
else
|
|
{
|
|
Operand res = context.VectorZero();
|
|
|
|
int pairs = op.GetPairsCount() >> op.Size;
|
|
|
|
for (int index = 0; index < pairs; index++)
|
|
{
|
|
int pairIndex = index << 1;
|
|
|
|
Operand ne = EmitVectorExtractZx(context, op.Rn, pairIndex + part, op.Size);
|
|
Operand me = EmitVectorExtractZx(context, op.Rm, pairIndex + part, op.Size);
|
|
|
|
res = EmitVectorInsert(context, res, ne, pairIndex, op.Size);
|
|
res = EmitVectorInsert(context, res, me, pairIndex + 1, op.Size);
|
|
}
|
|
|
|
context.Copy(GetVec(op.Rd), res);
|
|
}
|
|
}
|
|
|
|
private static void EmitVectorUnzip(ArmEmitterContext context, int part)
|
|
{
|
|
OpCodeSimdReg op = (OpCodeSimdReg)context.CurrOp;
|
|
|
|
if (Optimizations.UseSsse3)
|
|
{
|
|
if (op.RegisterSize == RegisterSize.Simd128)
|
|
{
|
|
Operand mask = default;
|
|
|
|
if (op.Size < 3)
|
|
{
|
|
long maskE0 = EvenMasks[op.Size];
|
|
long maskE1 = OddMasks [op.Size];
|
|
|
|
mask = X86GetScalar(context, maskE0);
|
|
|
|
mask = EmitVectorInsert(context, mask, Const(maskE1), 1, 3);
|
|
}
|
|
|
|
Operand n = GetVec(op.Rn);
|
|
|
|
if (op.Size < 3)
|
|
{
|
|
n = context.AddIntrinsic(Intrinsic.X86Pshufb, n, mask);
|
|
}
|
|
|
|
Operand m = GetVec(op.Rm);
|
|
|
|
if (op.Size < 3)
|
|
{
|
|
m = context.AddIntrinsic(Intrinsic.X86Pshufb, m, mask);
|
|
}
|
|
|
|
Intrinsic punpckInst = part == 0
|
|
? Intrinsic.X86Punpcklqdq
|
|
: Intrinsic.X86Punpckhqdq;
|
|
|
|
Operand res = context.AddIntrinsic(punpckInst, n, m);
|
|
|
|
context.Copy(GetVec(op.Rd), res);
|
|
}
|
|
else
|
|
{
|
|
Operand n = GetVec(op.Rn);
|
|
Operand m = GetVec(op.Rm);
|
|
|
|
Intrinsic punpcklInst = X86PunpcklInstruction[op.Size];
|
|
|
|
Operand res = context.AddIntrinsic(punpcklInst, n, m);
|
|
|
|
if (op.Size < 2)
|
|
{
|
|
long maskE0 = _masksE0_Uzp[op.Size];
|
|
long maskE1 = _masksE1_Uzp[op.Size];
|
|
|
|
Operand mask = X86GetScalar(context, maskE0);
|
|
|
|
mask = EmitVectorInsert(context, mask, Const(maskE1), 1, 3);
|
|
|
|
res = context.AddIntrinsic(Intrinsic.X86Pshufb, res, mask);
|
|
}
|
|
|
|
Intrinsic punpckInst = part == 0
|
|
? Intrinsic.X86Punpcklqdq
|
|
: Intrinsic.X86Punpckhqdq;
|
|
|
|
res = context.AddIntrinsic(punpckInst, res, context.VectorZero());
|
|
|
|
context.Copy(GetVec(op.Rd), res);
|
|
}
|
|
}
|
|
else
|
|
{
|
|
Operand res = context.VectorZero();
|
|
|
|
int pairs = op.GetPairsCount() >> op.Size;
|
|
|
|
for (int index = 0; index < pairs; index++)
|
|
{
|
|
int idx = index << 1;
|
|
|
|
Operand ne = EmitVectorExtractZx(context, op.Rn, idx + part, op.Size);
|
|
Operand me = EmitVectorExtractZx(context, op.Rm, idx + part, op.Size);
|
|
|
|
res = EmitVectorInsert(context, res, ne, index, op.Size);
|
|
res = EmitVectorInsert(context, res, me, pairs + index, op.Size);
|
|
}
|
|
|
|
context.Copy(GetVec(op.Rd), res);
|
|
}
|
|
}
|
|
|
|
private static void EmitVectorZip(ArmEmitterContext context, int part)
|
|
{
|
|
OpCodeSimdReg op = (OpCodeSimdReg)context.CurrOp;
|
|
|
|
if (Optimizations.UseSse2)
|
|
{
|
|
Operand n = GetVec(op.Rn);
|
|
Operand m = GetVec(op.Rm);
|
|
|
|
if (op.RegisterSize == RegisterSize.Simd128)
|
|
{
|
|
Intrinsic punpckInst = part == 0
|
|
? X86PunpcklInstruction[op.Size]
|
|
: X86PunpckhInstruction[op.Size];
|
|
|
|
Operand res = context.AddIntrinsic(punpckInst, n, m);
|
|
|
|
context.Copy(GetVec(op.Rd), res);
|
|
}
|
|
else
|
|
{
|
|
Operand res = context.AddIntrinsic(X86PunpcklInstruction[op.Size], n, m);
|
|
|
|
Intrinsic punpckInst = part == 0
|
|
? Intrinsic.X86Punpcklqdq
|
|
: Intrinsic.X86Punpckhqdq;
|
|
|
|
res = context.AddIntrinsic(punpckInst, res, context.VectorZero());
|
|
|
|
context.Copy(GetVec(op.Rd), res);
|
|
}
|
|
}
|
|
else
|
|
{
|
|
Operand res = context.VectorZero();
|
|
|
|
int pairs = op.GetPairsCount() >> op.Size;
|
|
|
|
int baseIndex = part != 0 ? pairs : 0;
|
|
|
|
for (int index = 0; index < pairs; index++)
|
|
{
|
|
int pairIndex = index << 1;
|
|
|
|
Operand ne = EmitVectorExtractZx(context, op.Rn, baseIndex + index, op.Size);
|
|
Operand me = EmitVectorExtractZx(context, op.Rm, baseIndex + index, op.Size);
|
|
|
|
res = EmitVectorInsert(context, res, ne, pairIndex, op.Size);
|
|
res = EmitVectorInsert(context, res, me, pairIndex + 1, op.Size);
|
|
}
|
|
|
|
context.Copy(GetVec(op.Rd), res);
|
|
}
|
|
}
|
|
}
|
|
}
|