JinxRyu/ARMeilleure/Instructions/InstEmitSimdShift.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

1192 lines
38 KiB
C#

// https://github.com/intel/ARM_NEON_2_x86_SSE/blob/master/NEON_2_SSE.h
using ARMeilleure.Decoders;
using ARMeilleure.IntermediateRepresentation;
using ARMeilleure.Translation;
using System;
using System.Diagnostics;
using System.Reflection;
using static ARMeilleure.Instructions.InstEmitHelper;
using static ARMeilleure.Instructions.InstEmitSimdHelper;
using static ARMeilleure.IntermediateRepresentation.Operand.Factory;
namespace ARMeilleure.Instructions
{
using Func2I = Func<Operand, Operand, Operand>;
static partial class InstEmit
{
#region "Masks"
private static readonly long[] _masks_SliSri = new long[] // Replication masks.
{
0x0101010101010101L, 0x0001000100010001L, 0x0000000100000001L, 0x0000000000000001L
};
#endregion
public static void Rshrn_V(ArmEmitterContext context)
{
if (Optimizations.UseSsse3)
{
OpCodeSimdShImm op = (OpCodeSimdShImm)context.CurrOp;
int shift = GetImmShr(op);
long roundConst = 1L << (shift - 1);
Operand d = GetVec(op.Rd);
Operand n = GetVec(op.Rn);
Operand dLow = context.VectorZeroUpper64(d);
Operand mask = default;
switch (op.Size + 1)
{
case 1: mask = X86GetAllElements(context, (int)roundConst * 0x00010001); break;
case 2: mask = X86GetAllElements(context, (int)roundConst); break;
case 3: mask = X86GetAllElements(context, roundConst); break;
}
Intrinsic addInst = X86PaddInstruction[op.Size + 1];
Operand res = context.AddIntrinsic(addInst, n, mask);
Intrinsic srlInst = X86PsrlInstruction[op.Size + 1];
res = context.AddIntrinsic(srlInst, res, Const(shift));
Operand mask2 = X86GetAllElements(context, EvenMasks[op.Size]);
res = context.AddIntrinsic(Intrinsic.X86Pshufb, res, mask2);
Intrinsic movInst = op.RegisterSize == RegisterSize.Simd128
? Intrinsic.X86Movlhps
: Intrinsic.X86Movhlps;
res = context.AddIntrinsic(movInst, dLow, res);
context.Copy(d, res);
}
else
{
EmitVectorShrImmNarrowOpZx(context, round: true);
}
}
public static void Shl_S(ArmEmitterContext context)
{
OpCodeSimdShImm op = (OpCodeSimdShImm)context.CurrOp;
int shift = GetImmShl(op);
EmitScalarUnaryOpZx(context, (op1) => context.ShiftLeft(op1, Const(shift)));
}
public static void Shl_V(ArmEmitterContext context)
{
OpCodeSimdShImm op = (OpCodeSimdShImm)context.CurrOp;
int shift = GetImmShl(op);
if (Optimizations.UseSse2 && op.Size > 0)
{
Operand n = GetVec(op.Rn);
Intrinsic sllInst = X86PsllInstruction[op.Size];
Operand res = context.AddIntrinsic(sllInst, n, Const(shift));
if (op.RegisterSize == RegisterSize.Simd64)
{
res = context.VectorZeroUpper64(res);
}
context.Copy(GetVec(op.Rd), res);
}
else
{
EmitVectorUnaryOpZx(context, (op1) => context.ShiftLeft(op1, Const(shift)));
}
}
public static void Shll_V(ArmEmitterContext context)
{
OpCodeSimd op = (OpCodeSimd)context.CurrOp;
int shift = 8 << op.Size;
if (Optimizations.UseSse41)
{
Operand n = GetVec(op.Rn);
if (op.RegisterSize == RegisterSize.Simd128)
{
n = context.AddIntrinsic(Intrinsic.X86Psrldq, n, Const(8));
}
Intrinsic movsxInst = X86PmovsxInstruction[op.Size];
Operand res = context.AddIntrinsic(movsxInst, n);
Intrinsic sllInst = X86PsllInstruction[op.Size + 1];
res = context.AddIntrinsic(sllInst, res, Const(shift));
context.Copy(GetVec(op.Rd), res);
}
else
{
EmitVectorShImmWidenBinaryZx(context, (op1, op2) => context.ShiftLeft(op1, op2), shift);
}
}
public static void Shrn_V(ArmEmitterContext context)
{
if (Optimizations.UseSsse3)
{
OpCodeSimdShImm op = (OpCodeSimdShImm)context.CurrOp;
int shift = GetImmShr(op);
Operand d = GetVec(op.Rd);
Operand n = GetVec(op.Rn);
Operand dLow = context.VectorZeroUpper64(d);
Intrinsic srlInst = X86PsrlInstruction[op.Size + 1];
Operand nShifted = context.AddIntrinsic(srlInst, n, Const(shift));
Operand mask = X86GetAllElements(context, EvenMasks[op.Size]);
Operand res = context.AddIntrinsic(Intrinsic.X86Pshufb, nShifted, mask);
Intrinsic movInst = op.RegisterSize == RegisterSize.Simd128
? Intrinsic.X86Movlhps
: Intrinsic.X86Movhlps;
res = context.AddIntrinsic(movInst, dLow, res);
context.Copy(d, res);
}
else
{
EmitVectorShrImmNarrowOpZx(context, round: false);
}
}
public static void Sli_S(ArmEmitterContext context)
{
EmitSli(context, scalar: true);
}
public static void Sli_V(ArmEmitterContext context)
{
EmitSli(context, scalar: false);
}
public static void Sqrshl_V(ArmEmitterContext context)
{
OpCodeSimdReg op = (OpCodeSimdReg)context.CurrOp;
Operand res = context.VectorZero();
int elems = op.GetBytesCount() >> op.Size;
for (int index = 0; index < elems; index++)
{
Operand ne = EmitVectorExtractSx(context, op.Rn, index, op.Size);
Operand me = EmitVectorExtractSx(context, op.Rm, index, op.Size);
Operand e = context.Call(typeof(SoftFallback).GetMethod(nameof(SoftFallback.SignedShlRegSatQ)), ne, me, Const(1), Const(op.Size));
res = EmitVectorInsert(context, res, e, index, op.Size);
}
context.Copy(GetVec(op.Rd), res);
}
public static void Sqrshrn_S(ArmEmitterContext context)
{
EmitRoundShrImmSaturatingNarrowOp(context, ShrImmSaturatingNarrowFlags.ScalarSxSx);
}
public static void Sqrshrn_V(ArmEmitterContext context)
{
EmitRoundShrImmSaturatingNarrowOp(context, ShrImmSaturatingNarrowFlags.VectorSxSx);
}
public static void Sqrshrun_S(ArmEmitterContext context)
{
EmitRoundShrImmSaturatingNarrowOp(context, ShrImmSaturatingNarrowFlags.ScalarSxZx);
}
public static void Sqrshrun_V(ArmEmitterContext context)
{
EmitRoundShrImmSaturatingNarrowOp(context, ShrImmSaturatingNarrowFlags.VectorSxZx);
}
public static void Sqshl_V(ArmEmitterContext context)
{
OpCodeSimdReg op = (OpCodeSimdReg)context.CurrOp;
Operand res = context.VectorZero();
int elems = op.GetBytesCount() >> op.Size;
for (int index = 0; index < elems; index++)
{
Operand ne = EmitVectorExtractSx(context, op.Rn, index, op.Size);
Operand me = EmitVectorExtractSx(context, op.Rm, index, op.Size);
Operand e = context.Call(typeof(SoftFallback).GetMethod(nameof(SoftFallback.SignedShlRegSatQ)), ne, me, Const(0), Const(op.Size));
res = EmitVectorInsert(context, res, e, index, op.Size);
}
context.Copy(GetVec(op.Rd), res);
}
public static void Sqshrn_S(ArmEmitterContext context)
{
EmitShrImmSaturatingNarrowOp(context, ShrImmSaturatingNarrowFlags.ScalarSxSx);
}
public static void Sqshrn_V(ArmEmitterContext context)
{
EmitShrImmSaturatingNarrowOp(context, ShrImmSaturatingNarrowFlags.VectorSxSx);
}
public static void Sqshrun_S(ArmEmitterContext context)
{
EmitShrImmSaturatingNarrowOp(context, ShrImmSaturatingNarrowFlags.ScalarSxZx);
}
public static void Sqshrun_V(ArmEmitterContext context)
{
EmitShrImmSaturatingNarrowOp(context, ShrImmSaturatingNarrowFlags.VectorSxZx);
}
public static void Sri_S(ArmEmitterContext context)
{
EmitSri(context, scalar: true);
}
public static void Sri_V(ArmEmitterContext context)
{
EmitSri(context, scalar: false);
}
public static void Srshl_V(ArmEmitterContext context)
{
OpCodeSimdReg op = (OpCodeSimdReg)context.CurrOp;
Operand res = context.VectorZero();
int elems = op.GetBytesCount() >> op.Size;
for (int index = 0; index < elems; index++)
{
Operand ne = EmitVectorExtractSx(context, op.Rn, index, op.Size);
Operand me = EmitVectorExtractSx(context, op.Rm, index, op.Size);
Operand e = context.Call(typeof(SoftFallback).GetMethod(nameof(SoftFallback.SignedShlReg)), ne, me, Const(1), Const(op.Size));
res = EmitVectorInsert(context, res, e, index, op.Size);
}
context.Copy(GetVec(op.Rd), res);
}
public static void Srshr_S(ArmEmitterContext context)
{
EmitScalarShrImmOpSx(context, ShrImmFlags.Round);
}
public static void Srshr_V(ArmEmitterContext context)
{
OpCodeSimdShImm op = (OpCodeSimdShImm)context.CurrOp;
if (Optimizations.UseSse2 && op.Size > 0 && op.Size < 3)
{
int shift = GetImmShr(op);
int eSize = 8 << op.Size;
Operand n = GetVec(op.Rn);
Intrinsic sllInst = X86PsllInstruction[op.Size];
Operand res = context.AddIntrinsic(sllInst, n, Const(eSize - shift));
Intrinsic srlInst = X86PsrlInstruction[op.Size];
res = context.AddIntrinsic(srlInst, res, Const(eSize - 1));
Intrinsic sraInst = X86PsraInstruction[op.Size];
Operand nSra = context.AddIntrinsic(sraInst, n, Const(shift));
Intrinsic addInst = X86PaddInstruction[op.Size];
res = context.AddIntrinsic(addInst, res, nSra);
if (op.RegisterSize == RegisterSize.Simd64)
{
res = context.VectorZeroUpper64(res);
}
context.Copy(GetVec(op.Rd), res);
}
else
{
EmitVectorShrImmOpSx(context, ShrImmFlags.Round);
}
}
public static void Srsra_S(ArmEmitterContext context)
{
EmitScalarShrImmOpSx(context, ShrImmFlags.Round | ShrImmFlags.Accumulate);
}
public static void Srsra_V(ArmEmitterContext context)
{
OpCodeSimdShImm op = (OpCodeSimdShImm)context.CurrOp;
if (Optimizations.UseSse2 && op.Size > 0 && op.Size < 3)
{
int shift = GetImmShr(op);
int eSize = 8 << op.Size;
Operand d = GetVec(op.Rd);
Operand n = GetVec(op.Rn);
Intrinsic sllInst = X86PsllInstruction[op.Size];
Operand res = context.AddIntrinsic(sllInst, n, Const(eSize - shift));
Intrinsic srlInst = X86PsrlInstruction[op.Size];
res = context.AddIntrinsic(srlInst, res, Const(eSize - 1));
Intrinsic sraInst = X86PsraInstruction[op.Size];
Operand nSra = context.AddIntrinsic(sraInst, n, Const(shift));
Intrinsic addInst = X86PaddInstruction[op.Size];
res = context.AddIntrinsic(addInst, res, nSra);
res = context.AddIntrinsic(addInst, res, d);
if (op.RegisterSize == RegisterSize.Simd64)
{
res = context.VectorZeroUpper64(res);
}
context.Copy(d, res);
}
else
{
EmitVectorShrImmOpSx(context, ShrImmFlags.Round | ShrImmFlags.Accumulate);
}
}
public static void Sshl_S(ArmEmitterContext context)
{
EmitSshlOrUshl(context, signed: true, scalar: true);
}
public static void Sshl_V(ArmEmitterContext context)
{
EmitSshlOrUshl(context, signed: true, scalar: false);
}
public static void Sshll_V(ArmEmitterContext context)
{
OpCodeSimdShImm op = (OpCodeSimdShImm)context.CurrOp;
int shift = GetImmShl(op);
if (Optimizations.UseSse41)
{
Operand n = GetVec(op.Rn);
if (op.RegisterSize == RegisterSize.Simd128)
{
n = context.AddIntrinsic(Intrinsic.X86Psrldq, n, Const(8));
}
Intrinsic movsxInst = X86PmovsxInstruction[op.Size];
Operand res = context.AddIntrinsic(movsxInst, n);
if (shift != 0)
{
Intrinsic sllInst = X86PsllInstruction[op.Size + 1];
res = context.AddIntrinsic(sllInst, res, Const(shift));
}
context.Copy(GetVec(op.Rd), res);
}
else
{
EmitVectorShImmWidenBinarySx(context, (op1, op2) => context.ShiftLeft(op1, op2), shift);
}
}
public static void Sshr_S(ArmEmitterContext context)
{
EmitShrImmOp(context, ShrImmFlags.ScalarSx);
}
public static void Sshr_V(ArmEmitterContext context)
{
OpCodeSimdShImm op = (OpCodeSimdShImm)context.CurrOp;
if (Optimizations.UseSse2 && op.Size > 0 && op.Size < 3)
{
int shift = GetImmShr(op);
Operand n = GetVec(op.Rn);
Intrinsic sraInst = X86PsraInstruction[op.Size];
Operand res = context.AddIntrinsic(sraInst, n, Const(shift));
if (op.RegisterSize == RegisterSize.Simd64)
{
res = context.VectorZeroUpper64(res);
}
context.Copy(GetVec(op.Rd), res);
}
else
{
EmitShrImmOp(context, ShrImmFlags.VectorSx);
}
}
public static void Ssra_S(ArmEmitterContext context)
{
EmitScalarShrImmOpSx(context, ShrImmFlags.Accumulate);
}
public static void Ssra_V(ArmEmitterContext context)
{
OpCodeSimdShImm op = (OpCodeSimdShImm)context.CurrOp;
if (Optimizations.UseSse2 && op.Size > 0 && op.Size < 3)
{
int shift = GetImmShr(op);
Operand d = GetVec(op.Rd);
Operand n = GetVec(op.Rn);
Intrinsic sraInst = X86PsraInstruction[op.Size];
Operand res = context.AddIntrinsic(sraInst, n, Const(shift));
Intrinsic addInst = X86PaddInstruction[op.Size];
res = context.AddIntrinsic(addInst, res, d);
if (op.RegisterSize == RegisterSize.Simd64)
{
res = context.VectorZeroUpper64(res);
}
context.Copy(d, res);
}
else
{
EmitVectorShrImmOpSx(context, ShrImmFlags.Accumulate);
}
}
public static void Uqrshl_V(ArmEmitterContext context)
{
OpCodeSimdReg op = (OpCodeSimdReg)context.CurrOp;
Operand res = context.VectorZero();
int elems = op.GetBytesCount() >> op.Size;
for (int index = 0; index < elems; index++)
{
Operand ne = EmitVectorExtractZx(context, op.Rn, index, op.Size);
Operand me = EmitVectorExtractZx(context, op.Rm, index, op.Size);
Operand e = context.Call(typeof(SoftFallback).GetMethod(nameof(SoftFallback.UnsignedShlRegSatQ)), ne, me, Const(1), Const(op.Size));
res = EmitVectorInsert(context, res, e, index, op.Size);
}
context.Copy(GetVec(op.Rd), res);
}
public static void Uqrshrn_S(ArmEmitterContext context)
{
EmitRoundShrImmSaturatingNarrowOp(context, ShrImmSaturatingNarrowFlags.ScalarZxZx);
}
public static void Uqrshrn_V(ArmEmitterContext context)
{
EmitRoundShrImmSaturatingNarrowOp(context, ShrImmSaturatingNarrowFlags.VectorZxZx);
}
public static void Uqshl_V(ArmEmitterContext context)
{
OpCodeSimdReg op = (OpCodeSimdReg)context.CurrOp;
Operand res = context.VectorZero();
int elems = op.GetBytesCount() >> op.Size;
for (int index = 0; index < elems; index++)
{
Operand ne = EmitVectorExtractZx(context, op.Rn, index, op.Size);
Operand me = EmitVectorExtractZx(context, op.Rm, index, op.Size);
Operand e = context.Call(typeof(SoftFallback).GetMethod(nameof(SoftFallback.UnsignedShlRegSatQ)), ne, me, Const(0), Const(op.Size));
res = EmitVectorInsert(context, res, e, index, op.Size);
}
context.Copy(GetVec(op.Rd), res);
}
public static void Uqshrn_S(ArmEmitterContext context)
{
EmitShrImmSaturatingNarrowOp(context, ShrImmSaturatingNarrowFlags.ScalarZxZx);
}
public static void Uqshrn_V(ArmEmitterContext context)
{
EmitShrImmSaturatingNarrowOp(context, ShrImmSaturatingNarrowFlags.VectorZxZx);
}
public static void Urshl_V(ArmEmitterContext context)
{
OpCodeSimdReg op = (OpCodeSimdReg)context.CurrOp;
Operand res = context.VectorZero();
int elems = op.GetBytesCount() >> op.Size;
for (int index = 0; index < elems; index++)
{
Operand ne = EmitVectorExtractZx(context, op.Rn, index, op.Size);
Operand me = EmitVectorExtractZx(context, op.Rm, index, op.Size);
Operand e = context.Call(typeof(SoftFallback).GetMethod(nameof(SoftFallback.UnsignedShlReg)), ne, me, Const(1), Const(op.Size));
res = EmitVectorInsert(context, res, e, index, op.Size);
}
context.Copy(GetVec(op.Rd), res);
}
public static void Urshr_S(ArmEmitterContext context)
{
EmitScalarShrImmOpZx(context, ShrImmFlags.Round);
}
public static void Urshr_V(ArmEmitterContext context)
{
OpCodeSimdShImm op = (OpCodeSimdShImm)context.CurrOp;
if (Optimizations.UseSse2 && op.Size > 0)
{
int shift = GetImmShr(op);
int eSize = 8 << op.Size;
Operand n = GetVec(op.Rn);
Intrinsic sllInst = X86PsllInstruction[op.Size];
Operand res = context.AddIntrinsic(sllInst, n, Const(eSize - shift));
Intrinsic srlInst = X86PsrlInstruction[op.Size];
res = context.AddIntrinsic(srlInst, res, Const(eSize - 1));
Operand nSrl = context.AddIntrinsic(srlInst, n, Const(shift));
Intrinsic addInst = X86PaddInstruction[op.Size];
res = context.AddIntrinsic(addInst, res, nSrl);
if (op.RegisterSize == RegisterSize.Simd64)
{
res = context.VectorZeroUpper64(res);
}
context.Copy(GetVec(op.Rd), res);
}
else
{
EmitVectorShrImmOpZx(context, ShrImmFlags.Round);
}
}
public static void Ursra_S(ArmEmitterContext context)
{
EmitScalarShrImmOpZx(context, ShrImmFlags.Round | ShrImmFlags.Accumulate);
}
public static void Ursra_V(ArmEmitterContext context)
{
OpCodeSimdShImm op = (OpCodeSimdShImm)context.CurrOp;
if (Optimizations.UseSse2 && op.Size > 0)
{
int shift = GetImmShr(op);
int eSize = 8 << op.Size;
Operand d = GetVec(op.Rd);
Operand n = GetVec(op.Rn);
Intrinsic sllInst = X86PsllInstruction[op.Size];
Operand res = context.AddIntrinsic(sllInst, n, Const(eSize - shift));
Intrinsic srlInst = X86PsrlInstruction[op.Size];
res = context.AddIntrinsic(srlInst, res, Const(eSize - 1));
Operand nSrl = context.AddIntrinsic(srlInst, n, Const(shift));
Intrinsic addInst = X86PaddInstruction[op.Size];
res = context.AddIntrinsic(addInst, res, nSrl);
res = context.AddIntrinsic(addInst, res, d);
if (op.RegisterSize == RegisterSize.Simd64)
{
res = context.VectorZeroUpper64(res);
}
context.Copy(d, res);
}
else
{
EmitVectorShrImmOpZx(context, ShrImmFlags.Round | ShrImmFlags.Accumulate);
}
}
public static void Ushl_S(ArmEmitterContext context)
{
EmitSshlOrUshl(context, signed: false, scalar: true);
}
public static void Ushl_V(ArmEmitterContext context)
{
EmitSshlOrUshl(context, signed: false, scalar: false);
}
public static void Ushll_V(ArmEmitterContext context)
{
OpCodeSimdShImm op = (OpCodeSimdShImm)context.CurrOp;
int shift = GetImmShl(op);
if (Optimizations.UseSse41)
{
Operand n = GetVec(op.Rn);
if (op.RegisterSize == RegisterSize.Simd128)
{
n = context.AddIntrinsic(Intrinsic.X86Psrldq, n, Const(8));
}
Intrinsic movzxInst = X86PmovzxInstruction[op.Size];
Operand res = context.AddIntrinsic(movzxInst, n);
if (shift != 0)
{
Intrinsic sllInst = X86PsllInstruction[op.Size + 1];
res = context.AddIntrinsic(sllInst, res, Const(shift));
}
context.Copy(GetVec(op.Rd), res);
}
else
{
EmitVectorShImmWidenBinaryZx(context, (op1, op2) => context.ShiftLeft(op1, op2), shift);
}
}
public static void Ushr_S(ArmEmitterContext context)
{
EmitShrImmOp(context, ShrImmFlags.ScalarZx);
}
public static void Ushr_V(ArmEmitterContext context)
{
OpCodeSimdShImm op = (OpCodeSimdShImm)context.CurrOp;
if (Optimizations.UseSse2 && op.Size > 0)
{
int shift = GetImmShr(op);
Operand n = GetVec(op.Rn);
Intrinsic srlInst = X86PsrlInstruction[op.Size];
Operand res = context.AddIntrinsic(srlInst, n, Const(shift));
if (op.RegisterSize == RegisterSize.Simd64)
{
res = context.VectorZeroUpper64(res);
}
context.Copy(GetVec(op.Rd), res);
}
else
{
EmitShrImmOp(context, ShrImmFlags.VectorZx);
}
}
public static void Usra_S(ArmEmitterContext context)
{
EmitScalarShrImmOpZx(context, ShrImmFlags.Accumulate);
}
public static void Usra_V(ArmEmitterContext context)
{
OpCodeSimdShImm op = (OpCodeSimdShImm)context.CurrOp;
if (Optimizations.UseSse2 && op.Size > 0)
{
int shift = GetImmShr(op);
Operand d = GetVec(op.Rd);
Operand n = GetVec(op.Rn);
Intrinsic srlInst = X86PsrlInstruction[op.Size];
Operand res = context.AddIntrinsic(srlInst, n, Const(shift));
Intrinsic addInst = X86PaddInstruction[op.Size];
res = context.AddIntrinsic(addInst, res, d);
if (op.RegisterSize == RegisterSize.Simd64)
{
res = context.VectorZeroUpper64(res);
}
context.Copy(d, res);
}
else
{
EmitVectorShrImmOpZx(context, ShrImmFlags.Accumulate);
}
}
[Flags]
private enum ShrImmFlags
{
Scalar = 1 << 0,
Signed = 1 << 1,
Round = 1 << 2,
Accumulate = 1 << 3,
ScalarSx = Scalar | Signed,
ScalarZx = Scalar,
VectorSx = Signed,
VectorZx = 0
}
private static void EmitScalarShrImmOpSx(ArmEmitterContext context, ShrImmFlags flags)
{
EmitShrImmOp(context, ShrImmFlags.ScalarSx | flags);
}
private static void EmitScalarShrImmOpZx(ArmEmitterContext context, ShrImmFlags flags)
{
EmitShrImmOp(context, ShrImmFlags.ScalarZx | flags);
}
private static void EmitVectorShrImmOpSx(ArmEmitterContext context, ShrImmFlags flags)
{
EmitShrImmOp(context, ShrImmFlags.VectorSx | flags);
}
private static void EmitVectorShrImmOpZx(ArmEmitterContext context, ShrImmFlags flags)
{
EmitShrImmOp(context, ShrImmFlags.VectorZx | flags);
}
private static void EmitShrImmOp(ArmEmitterContext context, ShrImmFlags flags)
{
OpCodeSimdShImm op = (OpCodeSimdShImm)context.CurrOp;
Operand res = context.VectorZero();
bool scalar = (flags & ShrImmFlags.Scalar) != 0;
bool signed = (flags & ShrImmFlags.Signed) != 0;
bool round = (flags & ShrImmFlags.Round) != 0;
bool accumulate = (flags & ShrImmFlags.Accumulate) != 0;
int shift = GetImmShr(op);
long roundConst = 1L << (shift - 1);
int elems = !scalar ? op.GetBytesCount() >> op.Size : 1;
for (int index = 0; index < elems; index++)
{
Operand e = EmitVectorExtract(context, op.Rn, index, op.Size, signed);
if (op.Size <= 2)
{
if (round)
{
e = context.Add(e, Const(roundConst));
}
e = signed ? context.ShiftRightSI(e, Const(shift)) : context.ShiftRightUI(e, Const(shift));
}
else /* if (op.Size == 3) */
{
e = EmitShrImm64(context, e, signed, round ? roundConst : 0L, shift);
}
if (accumulate)
{
Operand de = EmitVectorExtract(context, op.Rd, index, op.Size, signed);
e = context.Add(e, de);
}
res = EmitVectorInsert(context, res, e, index, op.Size);
}
context.Copy(GetVec(op.Rd), res);
}
private static Operand EmitShlRegOp(ArmEmitterContext context, Operand op, Operand shiftLsB, int size, bool signed)
{
Debug.Assert(op.Type == OperandType.I64);
Debug.Assert(shiftLsB.Type == OperandType.I32);
Debug.Assert((uint)size < 4u);
Operand negShiftLsB = context.Negate(shiftLsB);
Operand isInRange = context.BitwiseAnd(
context.ICompareLess(shiftLsB, Const(8 << size)),
context.ICompareLess(negShiftLsB, Const(8 << size)));
Operand isPositive = context.ICompareGreaterOrEqual(shiftLsB, Const(0));
Operand shl = context.ShiftLeft(op, shiftLsB);
Operand sarOrShr = signed
? context.ShiftRightSI(op, negShiftLsB)
: context.ShiftRightUI(op, negShiftLsB);
Operand res = context.ConditionalSelect(isPositive, shl, sarOrShr);
if (signed)
{
Operand isPositive2 = context.ICompareGreaterOrEqual(op, Const(0L));
Operand res2 = context.ConditionalSelect(isPositive2, Const(0L), Const(-1L));
res2 = context.ConditionalSelect(isPositive, Const(0L), res2);
return context.ConditionalSelect(isInRange, res, res2);
}
else
{
return context.ConditionalSelect(isInRange, res, Const(0UL));
}
}
private static void EmitVectorShrImmNarrowOpZx(ArmEmitterContext context, bool round)
{
OpCodeSimdShImm op = (OpCodeSimdShImm)context.CurrOp;
int shift = GetImmShr(op);
long roundConst = 1L << (shift - 1);
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 e = EmitVectorExtractZx(context, op.Rn, index, op.Size + 1);
if (round)
{
e = context.Add(e, Const(roundConst));
}
e = context.ShiftRightUI(e, Const(shift));
res = EmitVectorInsert(context, res, e, part + index, op.Size);
}
context.Copy(d, res);
}
[Flags]
private enum ShrImmSaturatingNarrowFlags
{
Scalar = 1 << 0,
SignedSrc = 1 << 1,
SignedDst = 1 << 2,
Round = 1 << 3,
ScalarSxSx = Scalar | SignedSrc | SignedDst,
ScalarSxZx = Scalar | SignedSrc,
ScalarZxZx = Scalar,
VectorSxSx = SignedSrc | SignedDst,
VectorSxZx = SignedSrc,
VectorZxZx = 0
}
private static void EmitRoundShrImmSaturatingNarrowOp(ArmEmitterContext context, ShrImmSaturatingNarrowFlags flags)
{
EmitShrImmSaturatingNarrowOp(context, ShrImmSaturatingNarrowFlags.Round | flags);
}
private static void EmitShrImmSaturatingNarrowOp(ArmEmitterContext context, ShrImmSaturatingNarrowFlags flags)
{
OpCodeSimdShImm op = (OpCodeSimdShImm)context.CurrOp;
bool scalar = (flags & ShrImmSaturatingNarrowFlags.Scalar) != 0;
bool signedSrc = (flags & ShrImmSaturatingNarrowFlags.SignedSrc) != 0;
bool signedDst = (flags & ShrImmSaturatingNarrowFlags.SignedDst) != 0;
bool round = (flags & ShrImmSaturatingNarrowFlags.Round) != 0;
int shift = GetImmShr(op);
long roundConst = 1L << (shift - 1);
int elems = !scalar ? 8 >> op.Size : 1;
int part = !scalar && (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 e = EmitVectorExtract(context, op.Rn, index, op.Size + 1, signedSrc);
if (op.Size <= 1 || !round)
{
if (round)
{
e = context.Add(e, Const(roundConst));
}
e = signedSrc ? context.ShiftRightSI(e, Const(shift)) : context.ShiftRightUI(e, Const(shift));
}
else /* if (op.Size == 2 && round) */
{
e = EmitShrImm64(context, e, signedSrc, roundConst, shift); // shift <= 32
}
e = EmitSatQ(context, e, op.Size, signedSrc, signedDst);
res = EmitVectorInsert(context, res, e, part + index, op.Size);
}
context.Copy(d, res);
}
// dst64 = (Int(src64, signed) + roundConst) >> shift;
private static Operand EmitShrImm64(
ArmEmitterContext context,
Operand value,
bool signed,
long roundConst,
int shift)
{
MethodInfo info = signed
? typeof(SoftFallback).GetMethod(nameof(SoftFallback.SignedShrImm64))
: typeof(SoftFallback).GetMethod(nameof(SoftFallback.UnsignedShrImm64));
return context.Call(info, value, Const(roundConst), Const(shift));
}
private static void EmitVectorShImmWidenBinarySx(ArmEmitterContext context, Func2I emit, int imm)
{
EmitVectorShImmWidenBinaryOp(context, emit, imm, signed: true);
}
private static void EmitVectorShImmWidenBinaryZx(ArmEmitterContext context, Func2I emit, int imm)
{
EmitVectorShImmWidenBinaryOp(context, emit, imm, signed: false);
}
private static void EmitVectorShImmWidenBinaryOp(ArmEmitterContext context, Func2I emit, int imm, bool signed)
{
OpCodeSimd op = (OpCodeSimd)context.CurrOp;
Operand res = context.VectorZero();
int elems = 8 >> op.Size;
int part = op.RegisterSize == RegisterSize.Simd128 ? elems : 0;
for (int index = 0; index < elems; index++)
{
Operand ne = EmitVectorExtract(context, op.Rn, part + index, op.Size, signed);
res = EmitVectorInsert(context, res, emit(ne, Const(imm)), index, op.Size + 1);
}
context.Copy(GetVec(op.Rd), res);
}
private static void EmitSli(ArmEmitterContext context, bool scalar)
{
OpCodeSimdShImm op = (OpCodeSimdShImm)context.CurrOp;
int shift = GetImmShl(op);
ulong mask = shift != 0 ? ulong.MaxValue >> (64 - shift) : 0UL;
if (Optimizations.UseSse2 && op.Size > 0)
{
Operand d = GetVec(op.Rd);
Operand n = GetVec(op.Rn);
Intrinsic sllInst = X86PsllInstruction[op.Size];
Operand nShifted = context.AddIntrinsic(sllInst, n, Const(shift));
Operand dMask = X86GetAllElements(context, (long)mask * _masks_SliSri[op.Size]);
Operand dMasked = context.AddIntrinsic(Intrinsic.X86Pand, d, dMask);
Operand res = context.AddIntrinsic(Intrinsic.X86Por, nShifted, dMasked);
if ((op.RegisterSize == RegisterSize.Simd64) || scalar)
{
res = context.VectorZeroUpper64(res);
}
context.Copy(d, res);
}
else
{
Operand res = context.VectorZero();
int elems = !scalar ? op.GetBytesCount() >> op.Size : 1;
for (int index = 0; index < elems; index++)
{
Operand ne = EmitVectorExtractZx(context, op.Rn, index, op.Size);
Operand neShifted = context.ShiftLeft(ne, Const(shift));
Operand de = EmitVectorExtractZx(context, op.Rd, index, op.Size);
Operand deMasked = context.BitwiseAnd(de, Const(mask));
Operand e = context.BitwiseOr(neShifted, deMasked);
res = EmitVectorInsert(context, res, e, index, op.Size);
}
context.Copy(GetVec(op.Rd), res);
}
}
private static void EmitSri(ArmEmitterContext context, bool scalar)
{
OpCodeSimdShImm op = (OpCodeSimdShImm)context.CurrOp;
int shift = GetImmShr(op);
int eSize = 8 << op.Size;
ulong mask = (ulong.MaxValue << (eSize - shift)) & (ulong.MaxValue >> (64 - eSize));
if (Optimizations.UseSse2 && op.Size > 0)
{
Operand d = GetVec(op.Rd);
Operand n = GetVec(op.Rn);
Intrinsic srlInst = X86PsrlInstruction[op.Size];
Operand nShifted = context.AddIntrinsic(srlInst, n, Const(shift));
Operand dMask = X86GetAllElements(context, (long)mask * _masks_SliSri[op.Size]);
Operand dMasked = context.AddIntrinsic(Intrinsic.X86Pand, d, dMask);
Operand res = context.AddIntrinsic(Intrinsic.X86Por, nShifted, dMasked);
if ((op.RegisterSize == RegisterSize.Simd64) || scalar)
{
res = context.VectorZeroUpper64(res);
}
context.Copy(d, res);
}
else
{
Operand res = context.VectorZero();
int elems = !scalar ? op.GetBytesCount() >> op.Size : 1;
for (int index = 0; index < elems; index++)
{
Operand ne = EmitVectorExtractZx(context, op.Rn, index, op.Size);
Operand neShifted = shift != 64 ? context.ShiftRightUI(ne, Const(shift)) : Const(0UL);
Operand de = EmitVectorExtractZx(context, op.Rd, index, op.Size);
Operand deMasked = context.BitwiseAnd(de, Const(mask));
Operand e = context.BitwiseOr(neShifted, deMasked);
res = EmitVectorInsert(context, res, e, index, op.Size);
}
context.Copy(GetVec(op.Rd), res);
}
}
private static void EmitSshlOrUshl(ArmEmitterContext context, bool signed, bool scalar)
{
OpCodeSimdReg op = (OpCodeSimdReg)context.CurrOp;
Operand res = context.VectorZero();
int elems = !scalar ? op.GetBytesCount() >> op.Size : 1;
for (int index = 0; index < elems; index++)
{
Operand ne = EmitVectorExtract (context, op.Rn, index, op.Size, signed);
Operand me = EmitVectorExtractSx(context, op.Rm, index << op.Size, 0);
Operand e = EmitShlRegOp(context, ne, context.ConvertI64ToI32(me), op.Size, signed);
res = EmitVectorInsert(context, res, e, index, op.Size);
}
context.Copy(GetVec(op.Rd), res);
}
}
}