0
0
Fork 0
This repository has been archived on 2024-10-12. You can view files and clone it, but cannot push or open issues or pull requests.
ryujinx-final/ARMeilleure/Instructions/InstEmitSimdCvt.cs

1167 lines
38 KiB
C#
Raw Normal View History

Add a new JIT compiler for CPU code (#693) * Start of the ARMeilleure project * Refactoring around the old IRAdapter, now renamed to PreAllocator * Optimize the LowestBitSet method * Add CLZ support and fix CLS implementation * Add missing Equals and GetHashCode overrides on some structs, misc small tweaks * Implement the ByteSwap IR instruction, and some refactoring on the assembler * Implement the DivideUI IR instruction and fix 64-bits IDIV * Correct constant operand type on CSINC * Move division instructions implementation to InstEmitDiv * Fix destination type for the ConditionalSelect IR instruction * Implement UMULH and SMULH, with new IR instructions * Fix some issues with shift instructions * Fix constant types for BFM instructions * Fix up new tests using the new V128 struct * Update tests * Move DIV tests to a separate file * Add support for calls, and some instructions that depends on them * Start adding support for SIMD & FP types, along with some of the related ARM instructions * Fix some typos and the divide instruction with FP operands * Fix wrong method call on Clz_V * Implement ARM FP & SIMD move instructions, Saddlv_V, and misc. fixes * Implement SIMD logical instructions and more misc. fixes * Fix PSRAD x86 instruction encoding, TRN, UABD and UABDL implementations * Implement float conversion instruction, merge in LDj3SNuD fixes, and some other misc. fixes * Implement SIMD shift instruction and fix Dup_V * Add SCVTF and UCVTF (vector, fixed-point) variants to the opcode table * Fix check with tolerance on tester * Implement FP & SIMD comparison instructions, and some fixes * Update FCVT (Scalar) encoding on the table to support the Half-float variants * Support passing V128 structs, some cleanup on the register allocator, merge LDj3SNuD fixes * Use old memory access methods, made a start on SIMD memory insts support, some fixes * Fix float constant passed to functions, save and restore non-volatile XMM registers, other fixes * Fix arguments count with struct return values, other fixes * More instructions * Misc. fixes and integrate LDj3SNuD fixes * Update tests * Add a faster linear scan allocator, unwinding support on windows, and other changes * Update Ryujinx.HLE * Update Ryujinx.Graphics * Fix V128 return pointer passing, RCX is clobbered * Update Ryujinx.Tests * Update ITimeZoneService * Stop using GetFunctionPointer as that can't be called from native code, misc. fixes and tweaks * Use generic GetFunctionPointerForDelegate method and other tweaks * Some refactoring on the code generator, assert on invalid operations and use a separate enum for intrinsics * Remove some unused code on the assembler * Fix REX.W prefix regression on float conversion instructions, add some sort of profiler * Add hardware capability detection * Fix regression on Sha1h and revert Fcm** changes * Add SSE2-only paths on vector extract and insert, some refactoring on the pre-allocator * Fix silly mistake introduced on last commit on CpuId * Generate inline stack probes when the stack allocation is too large * Initial support for the System-V ABI * Support multiple destination operands * Fix SSE2 VectorInsert8 path, and other fixes * Change placement of XMM callee save and restore code to match other compilers * Rename Dest to Destination and Inst to Instruction * Fix a regression related to calls and the V128 type * Add an extra space on comments to match code style * Some refactoring * Fix vector insert FP32 SSE2 path * Port over the ARM32 instructions * Avoid memory protection races on JIT Cache * Another fix on VectorInsert FP32 (thanks to LDj3SNuD * Float operands don't need to use the same register when VEX is supported * Add a new register allocator, higher quality code for hot code (tier up), and other tweaks * Some nits, small improvements on the pre allocator * CpuThreadState is gone * Allow changing CPU emulators with a config entry * Add runtime identifiers on the ARMeilleure project * Allow switching between CPUs through a config entry (pt. 2) * Change win10-x64 to win-x64 on projects * Update the Ryujinx project to use ARMeilleure * Ensure that the selected register is valid on the hybrid allocator * Allow exiting on returns to 0 (should fix test regression) * Remove register assignments for most used variables on the hybrid allocator * Do not use fixed registers as spill temp * Add missing namespace and remove unneeded using * Address PR feedback * Fix types, etc * Enable AssumeStrictAbiCompliance by default * Ensure that Spill and Fill don't load or store any more than necessary
2019-08-08 18:56:22 +00:00
using ARMeilleure.Decoders;
using ARMeilleure.IntermediateRepresentation;
using ARMeilleure.State;
using ARMeilleure.Translation;
using System;
using System.Diagnostics;
using static ARMeilleure.Instructions.InstEmitHelper;
using static ARMeilleure.Instructions.InstEmitSimdHelper;
using static ARMeilleure.IntermediateRepresentation.OperandHelper;
namespace ARMeilleure.Instructions
{
using Func1I = Func<Operand, Operand>;
static partial class InstEmit
{
public static void Fcvt_S(ArmEmitterContext context)
{
OpCodeSimd op = (OpCodeSimd)context.CurrOp;
if (op.Size == 0 && op.Opc == 1) // Single -> Double.
{
if (Optimizations.UseSse2)
{
Operand n = GetVec(op.Rn);
Operand res = context.AddIntrinsic(Intrinsic.X86Cvtss2sd, context.VectorZero(), n);
context.Copy(GetVec(op.Rd), res);
}
else
{
Operand ne = context.VectorExtract(OperandType.FP32, GetVec(op.Rn), 0);
Operand res = context.ConvertToFP(OperandType.FP64, ne);
context.Copy(GetVec(op.Rd), context.VectorInsert(context.VectorZero(), res, 0));
}
}
else if (op.Size == 1 && op.Opc == 0) // Double -> Single.
{
if (Optimizations.UseSse2)
{
Operand n = GetVec(op.Rn);
Operand res = context.AddIntrinsic(Intrinsic.X86Cvtsd2ss, context.VectorZero(), n);
context.Copy(GetVec(op.Rd), res);
}
else
{
Operand ne = context.VectorExtract(OperandType.FP64, GetVec(op.Rn), 0);
Operand res = context.ConvertToFP(OperandType.FP32, ne);
context.Copy(GetVec(op.Rd), context.VectorInsert(context.VectorZero(), res, 0));
}
}
else if (op.Size == 0 && op.Opc == 3) // Single -> Half.
{
Operand ne = context.VectorExtract(OperandType.FP32, GetVec(op.Rn), 0);
Delegate dlg = new _U16_F32(SoftFloat32_16.FPConvert);
Operand res = context.Call(dlg, ne);
res = context.ZeroExtend16(OperandType.I64, res);
context.Copy(GetVec(op.Rd), EmitVectorInsert(context, context.VectorZero(), res, 0, 1));
}
else if (op.Size == 3 && op.Opc == 0) // Half -> Single.
{
Operand ne = EmitVectorExtractZx(context, op.Rn, 0, 1);
Delegate dlg = new _F32_U16(SoftFloat16_32.FPConvert);
Operand res = context.Call(dlg, ne);
context.Copy(GetVec(op.Rd), context.VectorInsert(context.VectorZero(), res, 0));
}
else if (op.Size == 1 && op.Opc == 3) // Double -> Half.
{
throw new NotImplementedException("Double-precision to half-precision.");
}
else if (op.Size == 3 && op.Opc == 1) // Double -> Half.
{
throw new NotImplementedException("Half-precision to double-precision.");
}
else // Invalid encoding.
{
Debug.Assert(false, $"type == {op.Size} && opc == {op.Opc}");
}
}
public static void Fcvtas_Gp(ArmEmitterContext context)
{
EmitFcvt_s_Gp(context, (op1) => EmitRoundMathCall(context, MidpointRounding.AwayFromZero, op1));
}
public static void Fcvtau_Gp(ArmEmitterContext context)
{
EmitFcvt_u_Gp(context, (op1) => EmitRoundMathCall(context, MidpointRounding.AwayFromZero, op1));
}
public static void Fcvtl_V(ArmEmitterContext context)
{
OpCodeSimd op = (OpCodeSimd)context.CurrOp;
int sizeF = op.Size & 1;
if (Optimizations.UseSse2 && sizeF == 1)
{
Operand n = GetVec(op.Rn);
Operand res;
if (op.RegisterSize == RegisterSize.Simd128)
{
res = context.AddIntrinsic(Intrinsic.X86Movhlps, n, n);
}
else
{
res = n;
}
res = context.AddIntrinsic(Intrinsic.X86Cvtps2pd, res);
context.Copy(GetVec(op.Rd), res);
}
else
{
Operand res = context.VectorZero();
int elems = 4 >> sizeF;
int part = op.RegisterSize == RegisterSize.Simd128 ? elems : 0;
for (int index = 0; index < elems; index++)
{
if (sizeF == 0)
{
Operand ne = EmitVectorExtractZx(context, op.Rn, part + index, 1);
Delegate dlg = new _F32_U16(SoftFloat16_32.FPConvert);
Operand e = context.Call(dlg, ne);
res = context.VectorInsert(res, e, index);
}
else /* if (sizeF == 1) */
{
Operand ne = context.VectorExtract(OperandType.FP32, GetVec(op.Rn), part + index);
Operand e = context.ConvertToFP(OperandType.FP64, ne);
res = context.VectorInsert(res, e, index);
}
}
context.Copy(GetVec(op.Rd), res);
}
}
public static void Fcvtms_Gp(ArmEmitterContext context)
{
EmitFcvt_s_Gp(context, (op1) => EmitUnaryMathCall(context, MathF.Floor, Math.Floor, op1));
}
public static void Fcvtmu_Gp(ArmEmitterContext context)
{
EmitFcvt_u_Gp(context, (op1) => EmitUnaryMathCall(context, MathF.Floor, Math.Floor, op1));
}
public static void Fcvtn_V(ArmEmitterContext context)
{
OpCodeSimd op = (OpCodeSimd)context.CurrOp;
int sizeF = op.Size & 1;
if (Optimizations.UseSse2 && sizeF == 1)
{
Operand d = GetVec(op.Rd);
Operand n = GetVec(op.Rn);
Operand res = context.AddIntrinsic(Intrinsic.X86Movlhps, d, context.VectorZero());
Operand nInt = context.AddIntrinsic(Intrinsic.X86Cvtpd2ps, n);
nInt = context.AddIntrinsic(Intrinsic.X86Movlhps, nInt, nInt);
Intrinsic movInst = op.RegisterSize == RegisterSize.Simd128
? Intrinsic.X86Movlhps
: Intrinsic.X86Movhlps;
res = context.AddIntrinsic(movInst, res, nInt);
context.Copy(GetVec(op.Rd), res);
}
else
{
OperandType type = sizeF == 0 ? OperandType.FP32 : OperandType.FP64;
int elems = 4 >> sizeF;
int part = op.RegisterSize == RegisterSize.Simd128 ? elems : 0;
Operand res = part == 0 ? context.VectorZero() : context.Copy(GetVec(op.Rd));
for (int index = 0; index < elems; index++)
{
Operand ne = context.VectorExtract(type, GetVec(op.Rn), 0);
if (sizeF == 0)
{
Delegate dlg = new _U16_F32(SoftFloat32_16.FPConvert);
Operand e = context.Call(dlg, ne);
e = context.ZeroExtend16(OperandType.I64, e);
res = EmitVectorInsert(context, res, e, part + index, 1);
}
else /* if (sizeF == 1) */
{
Operand e = context.ConvertToFP(OperandType.FP32, ne);
res = context.VectorInsert(res, e, part + index);
}
}
context.Copy(GetVec(op.Rd), res);
}
}
public static void Fcvtns_S(ArmEmitterContext context)
{
if (Optimizations.UseSse41)
{
EmitSse41Fcvts(context, FPRoundingMode.ToNearest, scalar: true);
}
else
{
EmitFcvtn(context, signed: true, scalar: true);
}
}
public static void Fcvtns_V(ArmEmitterContext context)
{
if (Optimizations.UseSse41)
{
EmitSse41Fcvts(context, FPRoundingMode.ToNearest, scalar: false);
}
else
{
EmitFcvtn(context, signed: true, scalar: false);
}
}
public static void Fcvtnu_S(ArmEmitterContext context)
{
if (Optimizations.UseSse41)
{
EmitSse41Fcvtu(context, FPRoundingMode.ToNearest, scalar: true);
}
else
{
EmitFcvtn(context, signed: false, scalar: true);
}
}
public static void Fcvtnu_V(ArmEmitterContext context)
{
if (Optimizations.UseSse41)
{
EmitSse41Fcvtu(context, FPRoundingMode.ToNearest, scalar: false);
}
else
{
EmitFcvtn(context, signed: false, scalar: false);
}
}
public static void Fcvtps_Gp(ArmEmitterContext context)
{
EmitFcvt_s_Gp(context, (op1) => EmitUnaryMathCall(context, MathF.Ceiling, Math.Ceiling, op1));
}
public static void Fcvtpu_Gp(ArmEmitterContext context)
{
EmitFcvt_u_Gp(context, (op1) => EmitUnaryMathCall(context, MathF.Ceiling, Math.Ceiling, op1));
}
public static void Fcvtzs_Gp(ArmEmitterContext context)
{
EmitFcvt_s_Gp(context, (op1) => op1);
}
public static void Fcvtzs_Gp_Fixed(ArmEmitterContext context)
{
EmitFcvtzs_Gp_Fixed(context);
}
public static void Fcvtzs_S(ArmEmitterContext context)
{
if (Optimizations.UseSse41)
{
EmitSse41Fcvts(context, FPRoundingMode.TowardsZero, scalar: true);
}
else
{
EmitFcvtz(context, signed: true, scalar: true);
}
}
public static void Fcvtzs_V(ArmEmitterContext context)
{
if (Optimizations.UseSse41)
{
EmitSse41Fcvts(context, FPRoundingMode.TowardsZero, scalar: false);
}
else
{
EmitFcvtz(context, signed: true, scalar: false);
}
}
public static void Fcvtzs_V_Fixed(ArmEmitterContext context)
{
if (Optimizations.UseSse41)
{
EmitSse41Fcvts(context, FPRoundingMode.TowardsZero, scalar: false);
}
else
{
EmitFcvtz(context, signed: true, scalar: false);
}
}
public static void Fcvtzu_Gp(ArmEmitterContext context)
{
EmitFcvt_u_Gp(context, (op1) => op1);
}
public static void Fcvtzu_Gp_Fixed(ArmEmitterContext context)
{
EmitFcvtzu_Gp_Fixed(context);
}
public static void Fcvtzu_S(ArmEmitterContext context)
{
if (Optimizations.UseSse41)
{
EmitSse41Fcvtu(context, FPRoundingMode.TowardsZero, scalar: true);
}
else
{
EmitFcvtz(context, signed: false, scalar: true);
}
}
public static void Fcvtzu_V(ArmEmitterContext context)
{
if (Optimizations.UseSse41)
{
EmitSse41Fcvtu(context, FPRoundingMode.TowardsZero, scalar: false);
}
else
{
EmitFcvtz(context, signed: false, scalar: false);
}
}
public static void Fcvtzu_V_Fixed(ArmEmitterContext context)
{
if (Optimizations.UseSse41)
{
EmitSse41Fcvtu(context, FPRoundingMode.TowardsZero, scalar: false);
}
else
{
EmitFcvtz(context, signed: false, scalar: false);
}
}
public static void Scvtf_Gp(ArmEmitterContext context)
{
OpCodeSimdCvt op = (OpCodeSimdCvt)context.CurrOp;
Operand res = GetIntOrZR(context, op.Rn);
if (op.RegisterSize == RegisterSize.Int32)
{
res = context.SignExtend32(OperandType.I64, res);
}
res = EmitFPConvert(context, res, op.Size, signed: true);
context.Copy(GetVec(op.Rd), context.VectorInsert(context.VectorZero(), res, 0));
}
public static void Scvtf_Gp_Fixed(ArmEmitterContext context)
{
OpCodeSimdCvt op = (OpCodeSimdCvt)context.CurrOp;
Operand res = GetIntOrZR(context, op.Rn);
if (op.RegisterSize == RegisterSize.Int32)
{
res = context.SignExtend32(OperandType.I64, res);
}
res = EmitFPConvert(context, res, op.Size, signed: true);
res = EmitI2fFBitsMul(context, res, op.FBits);
context.Copy(GetVec(op.Rd), context.VectorInsert(context.VectorZero(), res, 0));
}
public static void Scvtf_S(ArmEmitterContext context)
{
OpCodeSimd op = (OpCodeSimd)context.CurrOp;
int sizeF = op.Size & 1;
if (Optimizations.UseSse2 && sizeF == 0)
{
EmitSse2Scvtf(context, scalar: true);
}
else
{
Operand res = EmitVectorLongExtract(context, op.Rn, 0, sizeF + 2);
res = EmitFPConvert(context, res, op.Size, signed: true);
context.Copy(GetVec(op.Rd), context.VectorInsert(context.VectorZero(), res, 0));
}
}
public static void Scvtf_V(ArmEmitterContext context)
{
OpCodeSimd op = (OpCodeSimd)context.CurrOp;
int sizeF = op.Size & 1;
if (Optimizations.UseSse2 && sizeF == 0)
{
EmitSse2Scvtf(context, scalar: false);
}
else
{
EmitVectorCvtf(context, signed: true);
}
}
public static void Scvtf_V_Fixed(ArmEmitterContext context)
{
OpCodeSimd op = (OpCodeSimd)context.CurrOp;
// sizeF == ((OpCodeSimdShImm64)op).Size - 2
int sizeF = op.Size & 1;
if (Optimizations.UseSse2 && sizeF == 0)
{
EmitSse2Scvtf(context, scalar: false);
}
else
{
EmitVectorCvtf(context, signed: true);
}
}
public static void Ucvtf_Gp(ArmEmitterContext context)
{
OpCodeSimdCvt op = (OpCodeSimdCvt)context.CurrOp;
Operand res = GetIntOrZR(context, op.Rn);
res = EmitFPConvert(context, res, op.Size, signed: false);
context.Copy(GetVec(op.Rd), context.VectorInsert(context.VectorZero(), res, 0));
}
public static void Ucvtf_Gp_Fixed(ArmEmitterContext context)
{
OpCodeSimdCvt op = (OpCodeSimdCvt)context.CurrOp;
Operand res = GetIntOrZR(context, op.Rn);
res = EmitFPConvert(context, res, op.Size, signed: false);
res = EmitI2fFBitsMul(context, res, op.FBits);
context.Copy(GetVec(op.Rd), context.VectorInsert(context.VectorZero(), res, 0));
}
public static void Ucvtf_S(ArmEmitterContext context)
{
OpCodeSimd op = (OpCodeSimd)context.CurrOp;
int sizeF = op.Size & 1;
if (Optimizations.UseSse2 && sizeF == 0)
{
EmitSse2Ucvtf(context, scalar: true);
}
else
{
Operand ne = EmitVectorLongExtract(context, op.Rn, 0, sizeF + 2);
Operand res = EmitFPConvert(context, ne, sizeF, signed: false);
context.Copy(GetVec(op.Rd), context.VectorInsert(context.VectorZero(), res, 0));
}
}
public static void Ucvtf_V(ArmEmitterContext context)
{
OpCodeSimd op = (OpCodeSimd)context.CurrOp;
int sizeF = op.Size & 1;
if (Optimizations.UseSse2 && sizeF == 0)
{
EmitSse2Ucvtf(context, scalar: false);
}
else
{
EmitVectorCvtf(context, signed: false);
}
}
public static void Ucvtf_V_Fixed(ArmEmitterContext context)
{
OpCodeSimd op = (OpCodeSimd)context.CurrOp;
// sizeF == ((OpCodeSimdShImm)op).Size - 2
int sizeF = op.Size & 1;
if (Optimizations.UseSse2 && sizeF == 0)
{
EmitSse2Ucvtf(context, scalar: false);
}
else
{
EmitVectorCvtf(context, signed: false);
}
}
private static void EmitFcvtn(ArmEmitterContext context, bool signed, bool scalar)
{
OpCodeSimd op = (OpCodeSimd)context.CurrOp;
Operand res = context.VectorZero();
Operand n = GetVec(op.Rn);
int sizeF = op.Size & 1;
int sizeI = sizeF + 2;
OperandType type = sizeF == 0 ? OperandType.FP32 : OperandType.FP64;
int elems = !scalar ? op.GetBytesCount() >> sizeI : 1;
for (int index = 0; index < elems; index++)
{
Operand ne = context.VectorExtract(type, n, index);
Operand e = EmitRoundMathCall(context, MidpointRounding.ToEven, ne);
if (sizeF == 0)
{
Delegate dlg = signed
? (Delegate)new _S32_F32(SoftFallback.SatF32ToS32)
: (Delegate)new _U32_F32(SoftFallback.SatF32ToU32);
e = context.Call(dlg, e);
e = context.ZeroExtend32(OperandType.I64, e);
}
else /* if (sizeF == 1) */
{
Delegate dlg = signed
? (Delegate)new _S64_F64(SoftFallback.SatF64ToS64)
: (Delegate)new _U64_F64(SoftFallback.SatF64ToU64);
e = context.Call(dlg, e);
}
res = EmitVectorInsert(context, res, e, index, sizeI);
}
context.Copy(GetVec(op.Rd), res);
}
private static void EmitFcvtz(ArmEmitterContext context, bool signed, bool scalar)
{
OpCodeSimd op = (OpCodeSimd)context.CurrOp;
Operand res = context.VectorZero();
Operand n = GetVec(op.Rn);
int sizeF = op.Size & 1;
int sizeI = sizeF + 2;
OperandType type = sizeF == 0 ? OperandType.FP32 : OperandType.FP64;
int fBits = GetFBits(context);
int elems = !scalar ? op.GetBytesCount() >> sizeI : 1;
for (int index = 0; index < elems; index++)
{
Operand ne = context.VectorExtract(type, n, index);
Operand e = EmitF2iFBitsMul(context, ne, fBits);
if (sizeF == 0)
{
Delegate dlg = signed
? (Delegate)new _S32_F32(SoftFallback.SatF32ToS32)
: (Delegate)new _U32_F32(SoftFallback.SatF32ToU32);
e = context.Call(dlg, e);
e = context.ZeroExtend32(OperandType.I64, e);
}
else /* if (sizeF == 1) */
{
Delegate dlg = signed
? (Delegate)new _S64_F64(SoftFallback.SatF64ToS64)
: (Delegate)new _U64_F64(SoftFallback.SatF64ToU64);
e = context.Call(dlg, e);
}
res = EmitVectorInsert(context, res, e, index, sizeI);
}
context.Copy(GetVec(op.Rd), res);
}
private static void EmitFcvt_s_Gp(ArmEmitterContext context, Func1I emit)
{
EmitFcvt___Gp(context, emit, signed: true);
}
private static void EmitFcvt_u_Gp(ArmEmitterContext context, Func1I emit)
{
EmitFcvt___Gp(context, emit, signed: false);
}
private static void EmitFcvt___Gp(ArmEmitterContext context, Func1I emit, bool signed)
{
OpCodeSimdCvt op = (OpCodeSimdCvt)context.CurrOp;
OperandType type = op.Size == 0 ? OperandType.FP32 : OperandType.FP64;
Operand ne = context.VectorExtract(type, GetVec(op.Rn), 0);
Operand res = signed
? EmitScalarFcvts(context, emit(ne), 0)
: EmitScalarFcvtu(context, emit(ne), 0);
SetIntOrZR(context, op.Rd, res);
}
private static void EmitFcvtzs_Gp_Fixed(ArmEmitterContext context)
{
EmitFcvtz__Gp_Fixed(context, signed: true);
}
private static void EmitFcvtzu_Gp_Fixed(ArmEmitterContext context)
{
EmitFcvtz__Gp_Fixed(context, signed: false);
}
private static void EmitFcvtz__Gp_Fixed(ArmEmitterContext context, bool signed)
{
OpCodeSimdCvt op = (OpCodeSimdCvt)context.CurrOp;
OperandType type = op.Size == 0 ? OperandType.FP32 : OperandType.FP64;
Operand ne = context.VectorExtract(type, GetVec(op.Rn), 0);
Operand res = signed
? EmitScalarFcvts(context, ne, op.FBits)
: EmitScalarFcvtu(context, ne, op.FBits);
SetIntOrZR(context, op.Rd, res);
}
private static void EmitVectorCvtf(ArmEmitterContext context, bool signed)
{
OpCodeSimd op = (OpCodeSimd)context.CurrOp;
Operand res = context.VectorZero();
int sizeF = op.Size & 1;
int sizeI = sizeF + 2;
int fBits = GetFBits(context);
int elems = op.GetBytesCount() >> sizeI;
for (int index = 0; index < elems; index++)
{
Operand ne = EmitVectorLongExtract(context, op.Rn, index, sizeI);
Operand e = EmitFPConvert(context, ne, sizeF, signed);
e = EmitI2fFBitsMul(context, e, fBits);
res = context.VectorInsert(res, e, index);
}
context.Copy(GetVec(op.Rd), res);
}
private static int GetFBits(ArmEmitterContext context)
{
if (context.CurrOp is OpCodeSimdShImm op)
{
return GetImmShr(op);
}
return 0;
}
private static Operand EmitFPConvert(ArmEmitterContext context, Operand value, int size, bool signed)
{
Debug.Assert(value.Type == OperandType.I32 || value.Type == OperandType.I64);
Debug.Assert((uint)size < 2);
OperandType type = size == 0 ? OperandType.FP32
: OperandType.FP64;
if (signed)
{
return context.ConvertToFP(type, value);
}
else
{
return context.ConvertToFPUI(type, value);
}
}
private static Operand EmitScalarFcvts(ArmEmitterContext context, Operand value, int fBits)
{
Debug.Assert(value.Type == OperandType.FP32 || value.Type == OperandType.FP64);
value = EmitF2iFBitsMul(context, value, fBits);
if (context.CurrOp.RegisterSize == RegisterSize.Int32)
{
Delegate dlg = value.Type == OperandType.FP32
? (Delegate)new _S32_F32(SoftFallback.SatF32ToS32)
: (Delegate)new _S32_F64(SoftFallback.SatF64ToS32);
return context.Call(dlg, value);
}
else
{
Delegate dlg = value.Type == OperandType.FP32
? (Delegate)new _S64_F32(SoftFallback.SatF32ToS64)
: (Delegate)new _S64_F64(SoftFallback.SatF64ToS64);
return context.Call(dlg, value);
}
}
private static Operand EmitScalarFcvtu(ArmEmitterContext context, Operand value, int fBits)
{
Debug.Assert(value.Type == OperandType.FP32 || value.Type == OperandType.FP64);
value = EmitF2iFBitsMul(context, value, fBits);
if (context.CurrOp.RegisterSize == RegisterSize.Int32)
{
Delegate dlg = value.Type == OperandType.FP32
? (Delegate)new _U32_F32(SoftFallback.SatF32ToU32)
: (Delegate)new _U32_F64(SoftFallback.SatF64ToU32);
return context.Call(dlg, value);
}
else
{
Delegate dlg = value.Type == OperandType.FP32
? (Delegate)new _U64_F32(SoftFallback.SatF32ToU64)
: (Delegate)new _U64_F64(SoftFallback.SatF64ToU64);
return context.Call(dlg, value);
}
}
private static Operand EmitF2iFBitsMul(ArmEmitterContext context, Operand value, int fBits)
{
Debug.Assert(value.Type == OperandType.FP32 || value.Type == OperandType.FP64);
if (fBits == 0)
{
return value;
}
if (value.Type == OperandType.FP32)
{
return context.Multiply(value, ConstF(MathF.Pow(2f, fBits)));
}
else /* if (value.Type == OperandType.FP64) */
{
return context.Multiply(value, ConstF(Math.Pow(2d, fBits)));
}
}
private static Operand EmitI2fFBitsMul(ArmEmitterContext context, Operand value, int fBits)
{
Debug.Assert(value.Type == OperandType.FP32 || value.Type == OperandType.FP64);
if (fBits == 0)
{
return value;
}
if (value.Type == OperandType.FP32)
{
return context.Multiply(value, ConstF(1f / MathF.Pow(2f, fBits)));
}
else /* if (value.Type == OperandType.FP64) */
{
return context.Multiply(value, ConstF(1d / Math.Pow(2d, fBits)));
}
}
private static void EmitSse41Fcvts(ArmEmitterContext context, FPRoundingMode roundMode, bool scalar)
{
OpCodeSimd op = (OpCodeSimd)context.CurrOp;
Operand n = GetVec(op.Rn);
const int cmpGreaterThanOrEqual = 5;
const int cmpOrdered = 7;
// sizeF == ((OpCodeSimdShImm64)op).Size - 2
int sizeF = op.Size & 1;
if (sizeF == 0)
{
Operand nMask = context.AddIntrinsic(Intrinsic.X86Cmpps, n, n, Const(cmpOrdered));
Operand nScaled = context.AddIntrinsic(Intrinsic.X86Pand, nMask, n);
if (op is OpCodeSimdShImm fixedOp)
{
int fBits = GetImmShr(fixedOp);
// BitConverter.Int32BitsToSingle(fpScaled) == MathF.Pow(2f, fBits)
int fpScaled = 0x3F800000 + fBits * 0x800000;
Operand scale = X86GetAllElements(context, fpScaled);
nScaled = context.AddIntrinsic(Intrinsic.X86Mulps, nScaled, scale);
}
Operand nRnd = context.AddIntrinsic(Intrinsic.X86Roundps, nScaled, Const(X86GetRoundControl(roundMode)));
Operand nInt = context.AddIntrinsic(Intrinsic.X86Cvtps2dq, nRnd);
Operand mask = X86GetAllElements(context, 0x4F000000); // 2.14748365E9f (2147483648)
Operand mask2 = context.AddIntrinsic(Intrinsic.X86Cmpps, nRnd, mask, Const(cmpGreaterThanOrEqual));
Operand res = context.AddIntrinsic(Intrinsic.X86Pxor, nInt, mask2);
if (scalar)
{
res = context.VectorZeroUpper96(res);
}
else if (op.RegisterSize == RegisterSize.Simd64)
{
res = context.VectorZeroUpper64(res);
}
context.Copy(GetVec(op.Rd), res);
}
else /* if (sizeF == 1) */
{
Operand nMask = context.AddIntrinsic(Intrinsic.X86Cmppd, n, n, Const(cmpOrdered));
Operand nScaled = context.AddIntrinsic(Intrinsic.X86Pand, nMask, n);
if (op is OpCodeSimdShImm fixedOp)
{
int fBits = GetImmShr(fixedOp);
// BitConverter.Int64BitsToDouble(fpScaled) == Math.Pow(2d, fBits)
long fpScaled = 0x3FF0000000000000L + fBits * 0x10000000000000L;
Operand scale = X86GetAllElements(context, fpScaled);
nScaled = context.AddIntrinsic(Intrinsic.X86Mulpd, nScaled, scale);
}
Operand nRnd = context.AddIntrinsic(Intrinsic.X86Roundpd, nScaled, Const(X86GetRoundControl(roundMode)));
Operand high;
if (!scalar)
{
high = context.AddIntrinsic(Intrinsic.X86Unpckhpd, nRnd, nRnd);
high = context.AddIntrinsicLong(Intrinsic.X86Cvtsd2si, high);
}
else
{
high = Const(0L);
}
Operand low = context.AddIntrinsicLong(Intrinsic.X86Cvtsd2si, nRnd);
Operand nInt = EmitVectorLongCreate(context, low, high);
Operand mask = X86GetAllElements(context, 0x43E0000000000000L); // 9.2233720368547760E18d (9223372036854775808)
Operand mask2 = context.AddIntrinsic(Intrinsic.X86Cmppd, nRnd, mask, Const(cmpGreaterThanOrEqual));
Operand res = context.AddIntrinsic(Intrinsic.X86Pxor, nInt, mask2);
if (scalar)
{
res = context.VectorZeroUpper64(res);
}
context.Copy(GetVec(op.Rd), res);
}
}
private static void EmitSse41Fcvtu(ArmEmitterContext context, FPRoundingMode roundMode, bool scalar)
{
OpCodeSimd op = (OpCodeSimd)context.CurrOp;
Operand n = GetVec(op.Rn);
const int cmpGreaterThanOrEqual = 5;
const int cmpGreaterThan = 6;
const int cmpOrdered = 7;
// sizeF == ((OpCodeSimdShImm)op).Size - 2
int sizeF = op.Size & 1;
if (sizeF == 0)
{
Operand nMask = context.AddIntrinsic(Intrinsic.X86Cmpps, n, n, Const(cmpOrdered));
Operand nScaled = context.AddIntrinsic(Intrinsic.X86Pand, nMask, n);
if (op is OpCodeSimdShImm fixedOp)
{
int fBits = GetImmShr(fixedOp);
// BitConverter.Int32BitsToSingle(fpScaled) == MathF.Pow(2f, fBits)
int fpScaled = 0x3F800000 + fBits * 0x800000;
Operand scale = X86GetAllElements(context, fpScaled);
nScaled = context.AddIntrinsic(Intrinsic.X86Mulps, nScaled, scale);
}
Operand nRnd = context.AddIntrinsic(Intrinsic.X86Roundps, nScaled, Const(X86GetRoundControl(roundMode)));
Operand nRndMask = context.AddIntrinsic(Intrinsic.X86Cmpps, nRnd, context.VectorZero(), Const(cmpGreaterThan));
Operand nRndMasked = context.AddIntrinsic(Intrinsic.X86Pand, nRnd, nRndMask);
Operand nInt = context.AddIntrinsic(Intrinsic.X86Cvtps2dq, nRndMasked);
Operand mask = X86GetAllElements(context, 0x4F000000); // 2.14748365E9f (2147483648)
Operand res = context.AddIntrinsic(Intrinsic.X86Subps, nRndMasked, mask);
Operand mask2 = context.AddIntrinsic(Intrinsic.X86Cmpps, res, context.VectorZero(), Const(cmpGreaterThan));
Operand resMasked = context.AddIntrinsic(Intrinsic.X86Pand, res, mask2);
res = context.AddIntrinsic(Intrinsic.X86Cvtps2dq, resMasked);
Operand mask3 = context.AddIntrinsic(Intrinsic.X86Cmpps, resMasked, mask, Const(cmpGreaterThanOrEqual));
res = context.AddIntrinsic(Intrinsic.X86Pxor, res, mask3);
res = context.AddIntrinsic(Intrinsic.X86Paddd, res, nInt);
if (scalar)
{
res = context.VectorZeroUpper96(res);
}
else if (op.RegisterSize == RegisterSize.Simd64)
{
res = context.VectorZeroUpper64(res);
}
context.Copy(GetVec(op.Rd), res);
}
else /* if (sizeF == 1) */
{
Operand nMask = context.AddIntrinsic(Intrinsic.X86Cmppd, n, n, Const(cmpOrdered));
Operand nScaled = context.AddIntrinsic(Intrinsic.X86Pand, nMask, n);
if (op is OpCodeSimdShImm fixedOp)
{
int fBits = GetImmShr(fixedOp);
// BitConverter.Int64BitsToDouble(fpScaled) == Math.Pow(2d, fBits)
long fpScaled = 0x3FF0000000000000L + fBits * 0x10000000000000L;
Operand scale = X86GetAllElements(context, fpScaled);
nScaled = context.AddIntrinsic(Intrinsic.X86Mulpd, nScaled, scale);
}
Operand nRnd = context.AddIntrinsic(Intrinsic.X86Roundpd, nScaled, Const(X86GetRoundControl(roundMode)));
Operand nRndMask = context.AddIntrinsic(Intrinsic.X86Cmppd, nRnd, context.VectorZero(), Const(cmpGreaterThan));
Operand nRndMasked = context.AddIntrinsic(Intrinsic.X86Pand, nRnd, nRndMask);
Operand high;
if (!scalar)
{
high = context.AddIntrinsic(Intrinsic.X86Unpckhpd, nRndMasked, nRndMasked);
high = context.AddIntrinsicLong(Intrinsic.X86Cvtsd2si, high);
}
else
{
high = Const(0L);
}
Operand low = context.AddIntrinsicLong(Intrinsic.X86Cvtsd2si, nRndMasked);
Operand nInt = EmitVectorLongCreate(context, low, high);
Operand mask = X86GetAllElements(context, 0x43E0000000000000L); // 9.2233720368547760E18d (9223372036854775808)
Operand res = context.AddIntrinsic(Intrinsic.X86Subpd, nRndMasked, mask);
Operand mask2 = context.AddIntrinsic(Intrinsic.X86Cmppd, res, context.VectorZero(), Const(cmpGreaterThan));
Operand resMasked = context.AddIntrinsic(Intrinsic.X86Pand, res, mask2);
if (!scalar)
{
high = context.AddIntrinsic(Intrinsic.X86Unpckhpd, resMasked, resMasked);
high = context.AddIntrinsicLong(Intrinsic.X86Cvtsd2si, high);
}
low = context.AddIntrinsicLong(Intrinsic.X86Cvtsd2si, resMasked);
res = EmitVectorLongCreate(context, low, high);
Operand mask3 = context.AddIntrinsic(Intrinsic.X86Cmppd, resMasked, mask, Const(cmpGreaterThanOrEqual));
res = context.AddIntrinsic(Intrinsic.X86Pxor, res, mask3);
res = context.AddIntrinsic(Intrinsic.X86Paddq, res, nInt);
if (scalar)
{
res = context.VectorZeroUpper64(res);
}
context.Copy(GetVec(op.Rd), res);
}
}
private static void EmitSse2Scvtf(ArmEmitterContext context, bool scalar)
{
OpCodeSimd op = (OpCodeSimd)context.CurrOp;
Operand n = GetVec(op.Rn);
Operand res = context.AddIntrinsic(Intrinsic.X86Cvtdq2ps, n);
if (op is OpCodeSimdShImm fixedOp)
{
int fBits = GetImmShr(fixedOp);
// BitConverter.Int32BitsToSingle(fpScaled) == 1f / MathF.Pow(2f, fBits)
int fpScaled = 0x3F800000 - fBits * 0x800000;
Operand scale = X86GetAllElements(context, fpScaled);
res = context.AddIntrinsic(Intrinsic.X86Mulps, res, scale);
}
if (scalar)
{
res = context.VectorZeroUpper96(res);
}
else if (op.RegisterSize == RegisterSize.Simd64)
{
res = context.VectorZeroUpper64(res);
}
context.Copy(GetVec(op.Rd), res);
}
private static void EmitSse2Ucvtf(ArmEmitterContext context, bool scalar)
{
OpCodeSimd op = (OpCodeSimd)context.CurrOp;
Operand n = GetVec(op.Rn);
Operand res = context.AddIntrinsic(Intrinsic.X86Psrld, n, Const(16));
res = context.AddIntrinsic(Intrinsic.X86Cvtdq2ps, res);
Operand mask = X86GetAllElements(context, 0x47800000); // 65536.0f (1 << 16)
res = context.AddIntrinsic(Intrinsic.X86Mulps, res, mask);
Operand res2 = context.AddIntrinsic(Intrinsic.X86Pslld, n, Const(16));
res2 = context.AddIntrinsic(Intrinsic.X86Psrld, res2, Const(16));
res2 = context.AddIntrinsic(Intrinsic.X86Cvtdq2ps, res2);
res = context.AddIntrinsic(Intrinsic.X86Addps, res, res2);
if (op is OpCodeSimdShImm fixedOp)
{
int fBits = GetImmShr(fixedOp);
// BitConverter.Int32BitsToSingle(fpScaled) == 1f / MathF.Pow(2f, fBits)
int fpScaled = 0x3F800000 - fBits * 0x800000;
Operand scale = X86GetAllElements(context, fpScaled);
res = context.AddIntrinsic(Intrinsic.X86Mulps, res, scale);
}
if (scalar)
{
res = context.VectorZeroUpper96(res);
}
else if (op.RegisterSize == RegisterSize.Simd64)
{
res = context.VectorZeroUpper64(res);
}
context.Copy(GetVec(op.Rd), res);
}
private static Operand EmitVectorLongExtract(ArmEmitterContext context, int reg, int index, int size)
{
OperandType type = size == 3 ? OperandType.I64 : OperandType.I32;
return context.VectorExtract(type, GetVec(reg), index);
}
private static Operand EmitVectorLongCreate(ArmEmitterContext context, Operand low, Operand high)
{
Operand vector = context.VectorCreateScalar(low);
vector = context.VectorInsert(vector, high, 1);
return vector;
}
}
}