tidus4400/rjx-mirror
1
0
Fork 0
forked from Mirror/Ryujinx
Code Pull requests Activity

Fix Frecpe_S/V and Frsqrte_S/V (full FP emu.). Add Sse Opt. & SoftFloat Impl. for Fcmeq/ge/gt/le/lt_S/V (Reg & Zero), Faddp_S/V, Fmaxp_V, Fminp_V Inst.; add Sse Opt. for Shll_V, S/Ushll_V Inst.; improve Sse Opt. for Xtn_V Inst.. Add Tests. (#543)

Browse source 
* Update Optimizations.cs

* Update InstEmitSimdShift.cs

* Update InstEmitSimdHelper.cs

* Update InstEmitSimdArithmetic.cs

* Update InstEmitSimdMove.cs

* Update SoftFloat.cs

* Update InstEmitSimdCmp.cs

* Update CpuTestSimdShImm.cs

* Update CpuTestSimd.cs

* Update CpuTestSimdReg.cs

* Nit.

* Update SoftFloat.cs

* Update InstEmitSimdArithmetic.cs

* Update InstEmitSimdHelper.cs

* Update CpuTestSimd.cs

* Explicit some implicit casts.

* Simplify some powers; nits.

* Update OpCodeTable.cs

* Update InstEmitSimdArithmetic.cs

* Update CpuTestSimdReg.cs

* Update InstEmitSimdArithmetic.cs
This commit is contained in:
LDj3SNuD 2018-12-26 18:11:36 +01:00 committed by gdkchan
parent d8f2497f15
commit 0f5b6dfbe8
11 changed files with 1808 additions and 441 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

268
ChocolArm64/Instructions/InstEmitSimdArithmetic.cs
View file

@ -176,12 +176,119 @@ namespace ChocolArm64.Instructions
public static void Fabd_S(ILEmitterCtx context) public static void Fabd_S(ILEmitterCtx context)
{ {
EmitScalarBinaryOpF(context, () => if (Optimizations.FastFP && Optimizations.UseSse2)
{ {
context.Emit(OpCodes.Sub); OpCodeSimdReg64 op = (OpCodeSimdReg64)context.CurrOp;
EmitUnaryMathCall(context, nameof(Math.Abs)); int sizeF = op.Size & 1;
});
if (sizeF == 0)
{
Type[] typesSsv = new Type[] { typeof(float) };
Type[] typesSubAndNot = new Type[] { typeof(Vector128<float>), typeof(Vector128<float>) };
context.EmitLdc_R4(-0f);
context.EmitCall(typeof(Sse).GetMethod(nameof(Sse.SetScalarVector128), typesSsv));
context.EmitLdvec(op.Rn);
context.EmitLdvec(op.Rm);
context.EmitCall(typeof(Sse).GetMethod(nameof(Sse.SubtractScalar), typesSubAndNot));
context.EmitCall(typeof(Sse).GetMethod(nameof(Sse.AndNot), typesSubAndNot));
context.EmitStvec(op.Rd);
EmitVectorZero32_128(context, op.Rd);
}
else /* if (sizeF == 1) */
{
Type[] typesSsv = new Type[] { typeof(double) };
Type[] typesSubAndNot = new Type[] { typeof(Vector128<double>), typeof(Vector128<double>) };
context.EmitLdc_R8(-0d);
context.EmitCall(typeof(Sse2).GetMethod(nameof(Sse2.SetScalarVector128), typesSsv));
EmitLdvecWithCastToDouble(context, op.Rn);
EmitLdvecWithCastToDouble(context, op.Rm);
context.EmitCall(typeof(Sse2).GetMethod(nameof(Sse2.SubtractScalar), typesSubAndNot));
context.EmitCall(typeof(Sse2).GetMethod(nameof(Sse2.AndNot), typesSubAndNot));
EmitStvecWithCastFromDouble(context, op.Rd);
EmitVectorZeroUpper(context, op.Rd);
}
}
else
{
EmitScalarBinaryOpF(context, () =>
{
EmitSoftFloatCall(context, nameof(SoftFloat32.FPSub));
EmitUnaryMathCall(context, nameof(Math.Abs));
});
}
}
public static void Fabd_V(ILEmitterCtx context)
{
if (Optimizations.FastFP && Optimizations.UseSse2)
{
OpCodeSimdReg64 op = (OpCodeSimdReg64)context.CurrOp;
int sizeF = op.Size & 1;
if (sizeF == 0)
{
Type[] typesSav = new Type[] { typeof(float) };
Type[] typesSubAndNot = new Type[] { typeof(Vector128<float>), typeof(Vector128<float>) };
context.EmitLdc_R4(-0f);
context.EmitCall(typeof(Sse).GetMethod(nameof(Sse.SetAllVector128), typesSav));
context.EmitLdvec(op.Rn);
context.EmitLdvec(op.Rm);
context.EmitCall(typeof(Sse).GetMethod(nameof(Sse.Subtract), typesSubAndNot));
context.EmitCall(typeof(Sse).GetMethod(nameof(Sse.AndNot), typesSubAndNot));
context.EmitStvec(op.Rd);
if (op.RegisterSize == RegisterSize.Simd64)
{
EmitVectorZeroUpper(context, op.Rd);
}
}
else /* if (sizeF == 1) */
{
Type[] typesSav = new Type[] { typeof(double) };
Type[] typesSubAndNot = new Type[] { typeof(Vector128<double>), typeof(Vector128<double>) };
context.EmitLdc_R8(-0d);
context.EmitCall(typeof(Sse2).GetMethod(nameof(Sse2.SetAllVector128), typesSav));
EmitLdvecWithCastToDouble(context, op.Rn);
EmitLdvecWithCastToDouble(context, op.Rm);
context.EmitCall(typeof(Sse2).GetMethod(nameof(Sse2.Subtract), typesSubAndNot));
context.EmitCall(typeof(Sse2).GetMethod(nameof(Sse2.AndNot), typesSubAndNot));
EmitStvecWithCastFromDouble(context, op.Rd);
}
}
else
{
EmitVectorBinaryOpF(context, () =>
{
EmitSoftFloatCall(context, nameof(SoftFloat32.FPSub));
EmitUnaryMathCall(context, nameof(Math.Abs));
});
}
} }
public static void Fabs_S(ILEmitterCtx context) public static void Fabs_S(ILEmitterCtx context)
@ -321,17 +428,60 @@ namespace ChocolArm64.Instructions
int sizeF = op.Size & 1; int sizeF = op.Size & 1;
EmitVectorExtractF(context, op.Rn, 0, sizeF); if (Optimizations.FastFP && Optimizations.UseSse3)
EmitVectorExtractF(context, op.Rn, 1, sizeF); {
if (sizeF == 0)
{
Type[] typesAddH = new Type[] { typeof(Vector128<float>), typeof(Vector128<float>) };
context.Emit(OpCodes.Add); context.EmitLdvec(op.Rn);
context.Emit(OpCodes.Dup);
EmitScalarSetF(context, op.Rd, sizeF); context.EmitCall(typeof(Sse3).GetMethod(nameof(Sse3.HorizontalAdd), typesAddH));
context.EmitStvec(op.Rd);
EmitVectorZero32_128(context, op.Rd);
}
else /* if (sizeF == 1) */
{
Type[] typesAddH = new Type[] { typeof(Vector128<double>), typeof(Vector128<double>) };
EmitLdvecWithCastToDouble(context, op.Rn);
context.Emit(OpCodes.Dup);
context.EmitCall(typeof(Sse3).GetMethod(nameof(Sse3.HorizontalAdd), typesAddH));
EmitStvecWithCastFromDouble(context, op.Rd);
EmitVectorZeroUpper(context, op.Rd);
}
}
else
{
EmitVectorExtractF(context, op.Rn, 0, sizeF);
EmitVectorExtractF(context, op.Rn, 1, sizeF);
EmitSoftFloatCall(context, nameof(SoftFloat32.FPAdd));
EmitScalarSetF(context, op.Rd, sizeF);
}
} }
public static void Faddp_V(ILEmitterCtx context) public static void Faddp_V(ILEmitterCtx context)
{ {
EmitVectorPairwiseOpF(context, () => context.Emit(OpCodes.Add)); if (Optimizations.FastFP && Optimizations.UseSse
&& Optimizations.UseSse2)
{
EmitVectorPairwiseSseOrSse2OpF(context, nameof(Sse.Add));
}
else
{
EmitVectorPairwiseOpF(context, () =>
{
EmitSoftFloatCall(context, nameof(SoftFloat32.FPAdd));
});
}
} }
public static void Fdiv_S(ILEmitterCtx context) public static void Fdiv_S(ILEmitterCtx context)
@ -462,10 +612,18 @@ namespace ChocolArm64.Instructions
public static void Fmaxp_V(ILEmitterCtx context) public static void Fmaxp_V(ILEmitterCtx context)
{ {
EmitVectorPairwiseOpF(context, () => if (Optimizations.FastFP && Optimizations.UseSse
&& Optimizations.UseSse2)
{ {
EmitSoftFloatCall(context, nameof(SoftFloat32.FPMax)); EmitVectorPairwiseSseOrSse2OpF(context, nameof(Sse.Max));
}); }
else
{
EmitVectorPairwiseOpF(context, () =>
{
EmitSoftFloatCall(context, nameof(SoftFloat32.FPMax));
});
}
} }
public static void Fmin_S(ILEmitterCtx context) public static void Fmin_S(ILEmitterCtx context)
@ -518,10 +676,18 @@ namespace ChocolArm64.Instructions
public static void Fminp_V(ILEmitterCtx context) public static void Fminp_V(ILEmitterCtx context)
{ {
EmitVectorPairwiseOpF(context, () => if (Optimizations.FastFP && Optimizations.UseSse
&& Optimizations.UseSse2)
{ {
EmitSoftFloatCall(context, nameof(SoftFloat32.FPMin)); EmitVectorPairwiseSseOrSse2OpF(context, nameof(Sse.Min));
}); }
else
{
EmitVectorPairwiseOpF(context, () =>
{
EmitSoftFloatCall(context, nameof(SoftFloat32.FPMin));
});
}
} }
public static void Fmla_Se(ILEmitterCtx context) public static void Fmla_Se(ILEmitterCtx context)
@ -1085,18 +1251,42 @@ namespace ChocolArm64.Instructions
public static void Frecpe_S(ILEmitterCtx context) public static void Frecpe_S(ILEmitterCtx context)
{ {
EmitScalarUnaryOpF(context, () => OpCodeSimd64 op = (OpCodeSimd64)context.CurrOp;
int sizeF = op.Size & 1;
if (Optimizations.FastFP && Optimizations.UseSse
&& sizeF == 0)
{ {
EmitUnarySoftFloatCall(context, nameof(SoftFloat.RecipEstimate)); EmitScalarSseOrSse2OpF(context, nameof(Sse.ReciprocalScalar));
}); }
else
{
EmitScalarUnaryOpF(context, () =>
{
EmitSoftFloatCall(context, nameof(SoftFloat32.FPRecipEstimate));
});
}
} }
public static void Frecpe_V(ILEmitterCtx context) public static void Frecpe_V(ILEmitterCtx context)
{ {
EmitVectorUnaryOpF(context, () => OpCodeSimd64 op = (OpCodeSimd64)context.CurrOp;
int sizeF = op.Size & 1;
if (Optimizations.FastFP && Optimizations.UseSse
&& sizeF == 0)
{ {
EmitUnarySoftFloatCall(context, nameof(SoftFloat.RecipEstimate)); EmitVectorSseOrSse2OpF(context, nameof(Sse.Reciprocal));
}); }
else
{
EmitVectorUnaryOpF(context, () =>
{
EmitSoftFloatCall(context, nameof(SoftFloat32.FPRecipEstimate));
});
}
} }
public static void Frecps_S(ILEmitterCtx context) // Fused. public static void Frecps_S(ILEmitterCtx context) // Fused.
@ -1398,18 +1588,42 @@ namespace ChocolArm64.Instructions
public static void Frsqrte_S(ILEmitterCtx context) public static void Frsqrte_S(ILEmitterCtx context)
{ {
EmitScalarUnaryOpF(context, () => OpCodeSimd64 op = (OpCodeSimd64)context.CurrOp;
int sizeF = op.Size & 1;
if (Optimizations.FastFP && Optimizations.UseSse
&& sizeF == 0)
{ {
EmitUnarySoftFloatCall(context, nameof(SoftFloat.InvSqrtEstimate)); EmitScalarSseOrSse2OpF(context, nameof(Sse.ReciprocalSqrtScalar));
}); }
else
{
EmitScalarUnaryOpF(context, () =>
{
EmitSoftFloatCall(context, nameof(SoftFloat32.FPRSqrtEstimate));
});
}
} }
public static void Frsqrte_V(ILEmitterCtx context) public static void Frsqrte_V(ILEmitterCtx context)
{ {
EmitVectorUnaryOpF(context, () => OpCodeSimd64 op = (OpCodeSimd64)context.CurrOp;
int sizeF = op.Size & 1;
if (Optimizations.FastFP && Optimizations.UseSse
&& sizeF == 0)
{ {
EmitUnarySoftFloatCall(context, nameof(SoftFloat.InvSqrtEstimate)); EmitVectorSseOrSse2OpF(context, nameof(Sse.ReciprocalSqrt));
}); }
else
{
EmitVectorUnaryOpF(context, () =>
{
EmitSoftFloatCall(context, nameof(SoftFloat32.FPRSqrtEstimate));
});
}
} }
public static void Frsqrts_S(ILEmitterCtx context) // Fused. public static void Frsqrts_S(ILEmitterCtx context) // Fused.

296
ChocolArm64/Instructions/InstEmitSimdCmp.cs
View file

@ -15,7 +15,7 @@ namespace ChocolArm64.Instructions
{ {
public static void Cmeq_S(ILEmitterCtx context) public static void Cmeq_S(ILEmitterCtx context)
{ {
EmitCmp(context, OpCodes.Beq_S, scalar: true); EmitCmpOp(context, OpCodes.Beq_S, scalar: true);
} }
public static void Cmeq_V(ILEmitterCtx context) public static void Cmeq_V(ILEmitterCtx context)
@ -32,28 +32,28 @@ namespace ChocolArm64.Instructions
} }
else else
{ {
EmitCmp(context, OpCodes.Beq_S, scalar: false); EmitCmpOp(context, OpCodes.Beq_S, scalar: false);
} }
} }
else else
{ {
EmitCmp(context, OpCodes.Beq_S, scalar: false); EmitCmpOp(context, OpCodes.Beq_S, scalar: false);
} }
} }
public static void Cmge_S(ILEmitterCtx context) public static void Cmge_S(ILEmitterCtx context)
{ {
EmitCmp(context, OpCodes.Bge_S, scalar: true); EmitCmpOp(context, OpCodes.Bge_S, scalar: true);
} }
public static void Cmge_V(ILEmitterCtx context) public static void Cmge_V(ILEmitterCtx context)
{ {
EmitCmp(context, OpCodes.Bge_S, scalar: false); EmitCmpOp(context, OpCodes.Bge_S, scalar: false);
} }
public static void Cmgt_S(ILEmitterCtx context) public static void Cmgt_S(ILEmitterCtx context)
{ {
EmitCmp(context, OpCodes.Bgt_S, scalar: true); EmitCmpOp(context, OpCodes.Bgt_S, scalar: true);
} }
public static void Cmgt_V(ILEmitterCtx context) public static void Cmgt_V(ILEmitterCtx context)
@ -70,63 +70,63 @@ namespace ChocolArm64.Instructions
} }
else else
{ {
EmitCmp(context, OpCodes.Bgt_S, scalar: false); EmitCmpOp(context, OpCodes.Bgt_S, scalar: false);
} }
} }
else else
{ {
EmitCmp(context, OpCodes.Bgt_S, scalar: false); EmitCmpOp(context, OpCodes.Bgt_S, scalar: false);
} }
} }
public static void Cmhi_S(ILEmitterCtx context) public static void Cmhi_S(ILEmitterCtx context)
{ {
EmitCmp(context, OpCodes.Bgt_Un_S, scalar: true); EmitCmpOp(context, OpCodes.Bgt_Un_S, scalar: true);
} }
public static void Cmhi_V(ILEmitterCtx context) public static void Cmhi_V(ILEmitterCtx context)
{ {
EmitCmp(context, OpCodes.Bgt_Un_S, scalar: false); EmitCmpOp(context, OpCodes.Bgt_Un_S, scalar: false);
} }
public static void Cmhs_S(ILEmitterCtx context) public static void Cmhs_S(ILEmitterCtx context)
{ {
EmitCmp(context, OpCodes.Bge_Un_S, scalar: true); EmitCmpOp(context, OpCodes.Bge_Un_S, scalar: true);
} }
public static void Cmhs_V(ILEmitterCtx context) public static void Cmhs_V(ILEmitterCtx context)
{ {
EmitCmp(context, OpCodes.Bge_Un_S, scalar: false); EmitCmpOp(context, OpCodes.Bge_Un_S, scalar: false);
} }
public static void Cmle_S(ILEmitterCtx context) public static void Cmle_S(ILEmitterCtx context)
{ {
EmitCmp(context, OpCodes.Ble_S, scalar: true); EmitCmpOp(context, OpCodes.Ble_S, scalar: true);
} }
public static void Cmle_V(ILEmitterCtx context) public static void Cmle_V(ILEmitterCtx context)
{ {
EmitCmp(context, OpCodes.Ble_S, scalar: false); EmitCmpOp(context, OpCodes.Ble_S, scalar: false);
} }
public static void Cmlt_S(ILEmitterCtx context) public static void Cmlt_S(ILEmitterCtx context)
{ {
EmitCmp(context, OpCodes.Blt_S, scalar: true); EmitCmpOp(context, OpCodes.Blt_S, scalar: true);
} }
public static void Cmlt_V(ILEmitterCtx context) public static void Cmlt_V(ILEmitterCtx context)
{ {
EmitCmp(context, OpCodes.Blt_S, scalar: false); EmitCmpOp(context, OpCodes.Blt_S, scalar: false);
} }
public static void Cmtst_S(ILEmitterCtx context) public static void Cmtst_S(ILEmitterCtx context)
{ {
EmitCmtst(context, scalar: true); EmitCmtstOp(context, scalar: true);
} }
public static void Cmtst_V(ILEmitterCtx context) public static void Cmtst_V(ILEmitterCtx context)
{ {
EmitCmtst(context, scalar: false); EmitCmtstOp(context, scalar: false);
} }
public static void Fccmp_S(ILEmitterCtx context) public static void Fccmp_S(ILEmitterCtx context)
@ -145,7 +145,7 @@ namespace ChocolArm64.Instructions
context.MarkLabel(lblTrue); context.MarkLabel(lblTrue);
EmitFcmpE(context, signalNaNs: false); EmitFcmpOrFcmpe(context, signalNaNs: false);
context.MarkLabel(lblEnd); context.MarkLabel(lblEnd);
} }
@ -166,120 +166,152 @@ namespace ChocolArm64.Instructions
context.MarkLabel(lblTrue); context.MarkLabel(lblTrue);
EmitFcmpE(context, signalNaNs: true); EmitFcmpOrFcmpe(context, signalNaNs: true);
context.MarkLabel(lblEnd); context.MarkLabel(lblEnd);
} }
public static void Fcmeq_S(ILEmitterCtx context) public static void Fcmeq_S(ILEmitterCtx context)
{ {
if (context.CurrOp is OpCodeSimdReg64 && Optimizations.UseSse if (Optimizations.FastFP && Optimizations.UseSse
&& Optimizations.UseSse2) && Optimizations.UseSse2)
{ {
EmitScalarSseOrSse2OpF(context, nameof(Sse.CompareEqualScalar)); EmitCmpSseOrSse2OpF(context, nameof(Sse.CompareEqualScalar), scalar: true);
} }
else else
{ {
EmitScalarFcmp(context, OpCodes.Beq_S); EmitCmpOpF(context, nameof(SoftFloat32.FPCompareEQ), scalar: true);
} }
} }
public static void Fcmeq_V(ILEmitterCtx context) public static void Fcmeq_V(ILEmitterCtx context)
{ {
if (context.CurrOp is OpCodeSimdReg64 && Optimizations.UseSse if (Optimizations.FastFP && Optimizations.UseSse
&& Optimizations.UseSse2) && Optimizations.UseSse2)
{ {
EmitVectorSseOrSse2OpF(context, nameof(Sse.CompareEqual)); EmitCmpSseOrSse2OpF(context, nameof(Sse.CompareEqual), scalar: false);
} }
else else
{ {
EmitVectorFcmp(context, OpCodes.Beq_S); EmitCmpOpF(context, nameof(SoftFloat32.FPCompareEQ), scalar: false);
} }
} }
public static void Fcmge_S(ILEmitterCtx context) public static void Fcmge_S(ILEmitterCtx context)
{ {
if (context.CurrOp is OpCodeSimdReg64 && Optimizations.UseSse if (Optimizations.FastFP && Optimizations.UseSse
&& Optimizations.UseSse2) && Optimizations.UseSse2)
{ {
EmitScalarSseOrSse2OpF(context, nameof(Sse.CompareGreaterThanOrEqualScalar)); EmitCmpSseOrSse2OpF(context, nameof(Sse.CompareGreaterThanOrEqualScalar), scalar: true);
} }
else else
{ {
EmitScalarFcmp(context, OpCodes.Bge_S); EmitCmpOpF(context, nameof(SoftFloat32.FPCompareGE), scalar: true);
} }
} }
public static void Fcmge_V(ILEmitterCtx context) public static void Fcmge_V(ILEmitterCtx context)
{ {
if (context.CurrOp is OpCodeSimdReg64 && Optimizations.UseSse if (Optimizations.FastFP && Optimizations.UseSse
&& Optimizations.UseSse2) && Optimizations.UseSse2)
{ {
EmitVectorSseOrSse2OpF(context, nameof(Sse.CompareGreaterThanOrEqual)); EmitCmpSseOrSse2OpF(context, nameof(Sse.CompareGreaterThanOrEqual), scalar: false);
} }
else else
{ {
EmitVectorFcmp(context, OpCodes.Bge_S); EmitCmpOpF(context, nameof(SoftFloat32.FPCompareGE), scalar: false);
} }
} }
public static void Fcmgt_S(ILEmitterCtx context) public static void Fcmgt_S(ILEmitterCtx context)
{ {
if (context.CurrOp is OpCodeSimdReg64 && Optimizations.UseSse if (Optimizations.FastFP && Optimizations.UseSse
&& Optimizations.UseSse2) && Optimizations.UseSse2)
{ {
EmitScalarSseOrSse2OpF(context, nameof(Sse.CompareGreaterThanScalar)); EmitCmpSseOrSse2OpF(context, nameof(Sse.CompareGreaterThanScalar), scalar: true);
} }
else else
{ {
EmitScalarFcmp(context, OpCodes.Bgt_S); EmitCmpOpF(context, nameof(SoftFloat32.FPCompareGT), scalar: true);
} }
} }
public static void Fcmgt_V(ILEmitterCtx context) public static void Fcmgt_V(ILEmitterCtx context)
{ {
if (context.CurrOp is OpCodeSimdReg64 && Optimizations.UseSse if (Optimizations.FastFP && Optimizations.UseSse
&& Optimizations.UseSse2) && Optimizations.UseSse2)
{ {
EmitVectorSseOrSse2OpF(context, nameof(Sse.CompareGreaterThan)); EmitCmpSseOrSse2OpF(context, nameof(Sse.CompareGreaterThan), scalar: false);
} }
else else
{ {
EmitVectorFcmp(context, OpCodes.Bgt_S); EmitCmpOpF(context, nameof(SoftFloat32.FPCompareGT), scalar: false);
} }
} }
public static void Fcmle_S(ILEmitterCtx context) public static void Fcmle_S(ILEmitterCtx context)
{ {
EmitScalarFcmp(context, OpCodes.Ble_S); if (Optimizations.FastFP && Optimizations.UseSse
&& Optimizations.UseSse2)
{
EmitCmpSseOrSse2OpF(context, nameof(Sse.CompareGreaterThanOrEqualScalar), scalar: true, isLeOrLt: true);
}
else
{
EmitCmpOpF(context, nameof(SoftFloat32.FPCompareLE), scalar: true);
}
} }
public static void Fcmle_V(ILEmitterCtx context) public static void Fcmle_V(ILEmitterCtx context)
{ {
EmitVectorFcmp(context, OpCodes.Ble_S); if (Optimizations.FastFP && Optimizations.UseSse
&& Optimizations.UseSse2)
{
EmitCmpSseOrSse2OpF(context, nameof(Sse.CompareGreaterThanOrEqual), scalar: false, isLeOrLt: true);
}
else
{
EmitCmpOpF(context, nameof(SoftFloat32.FPCompareLE), scalar: false);
}
} }
public static void Fcmlt_S(ILEmitterCtx context) public static void Fcmlt_S(ILEmitterCtx context)
{ {
EmitScalarFcmp(context, OpCodes.Blt_S); if (Optimizations.FastFP && Optimizations.UseSse
&& Optimizations.UseSse2)
{
EmitCmpSseOrSse2OpF(context, nameof(Sse.CompareGreaterThanScalar), scalar: true, isLeOrLt: true);
}
else
{
EmitCmpOpF(context, nameof(SoftFloat32.FPCompareLT), scalar: true);
}
} }
public static void Fcmlt_V(ILEmitterCtx context) public static void Fcmlt_V(ILEmitterCtx context)
{ {
EmitVectorFcmp(context, OpCodes.Blt_S); if (Optimizations.FastFP && Optimizations.UseSse
&& Optimizations.UseSse2)
{
EmitCmpSseOrSse2OpF(context, nameof(Sse.CompareGreaterThan), scalar: false, isLeOrLt: true);
}
else
{
EmitCmpOpF(context, nameof(SoftFloat32.FPCompareLT), scalar: false);
}
} }
public static void Fcmp_S(ILEmitterCtx context) public static void Fcmp_S(ILEmitterCtx context)
{ {
EmitFcmpE(context, signalNaNs: false); EmitFcmpOrFcmpe(context, signalNaNs: false);
} }
public static void Fcmpe_S(ILEmitterCtx context) public static void Fcmpe_S(ILEmitterCtx context)
{ {
EmitFcmpE(context, signalNaNs: true); EmitFcmpOrFcmpe(context, signalNaNs: true);
} }
private static void EmitFcmpE(ILEmitterCtx context, bool signalNaNs) private static void EmitFcmpOrFcmpe(ILEmitterCtx context, bool signalNaNs)
{ {
OpCodeSimdReg64 op = (OpCodeSimdReg64)context.CurrOp; OpCodeSimdReg64 op = (OpCodeSimdReg64)context.CurrOp;
@ -430,7 +462,7 @@ namespace ChocolArm64.Instructions
{ {
context.EmitLdc_R4(0f); context.EmitLdc_R4(0f);
} }
else // if (op.Size == 1) else /* if (op.Size == 1) */
{ {
context.EmitLdc_R8(0d); context.EmitLdc_R8(0d);
} }
@ -448,7 +480,7 @@ namespace ChocolArm64.Instructions
} }
} }
private static void EmitCmp(ILEmitterCtx context, OpCode ilOp, bool scalar) private static void EmitCmpOp(ILEmitterCtx context, OpCode ilOp, bool scalar)
{ {
OpCodeSimd64 op = (OpCodeSimd64)context.CurrOp; OpCodeSimd64 op = (OpCodeSimd64)context.CurrOp;
@ -492,7 +524,7 @@ namespace ChocolArm64.Instructions
} }
} }
private static void EmitCmtst(ILEmitterCtx context, bool scalar) private static void EmitCmtstOp(ILEmitterCtx context, bool scalar)
{ {
OpCodeSimdReg64 op = (OpCodeSimdReg64)context.CurrOp; OpCodeSimdReg64 op = (OpCodeSimdReg64)context.CurrOp;
@ -532,84 +564,134 @@ namespace ChocolArm64.Instructions
} }
} }
private static void EmitScalarFcmp(ILEmitterCtx context, OpCode ilOp) private static void EmitCmpOpF(ILEmitterCtx context, string name, bool scalar)
{
EmitFcmp(context, ilOp, 0, scalar: true);
}
private static void EmitVectorFcmp(ILEmitterCtx context, OpCode ilOp)
{ {
OpCodeSimd64 op = (OpCodeSimd64)context.CurrOp; OpCodeSimd64 op = (OpCodeSimd64)context.CurrOp;
int sizeF = op.Size & 1; int sizeF = op.Size & 1;
int bytes = op.GetBitsCount() >> 3; int bytes = op.GetBitsCount() >> 3;
int elems = bytes >> sizeF + 2; int elems = !scalar ? bytes >> sizeF + 2 : 1;
for (int index = 0; index < elems; index++) for (int index = 0; index < elems; index++)
{ {
EmitFcmp(context, ilOp, index, scalar: false); EmitVectorExtractF(context, op.Rn, index, sizeF);
if (op is OpCodeSimdReg64 binOp)
{
EmitVectorExtractF(context, binOp.Rm, index, sizeF);
}
else
{
if (sizeF == 0)
{
context.EmitLdc_R4(0f);
}
else /* if (sizeF == 1) */
{
context.EmitLdc_R8(0d);
}
}
EmitSoftFloatCall(context, name);
EmitVectorInsertF(context, op.Rd, index, sizeF);
} }
if (op.RegisterSize == RegisterSize.Simd64) if (!scalar)
{ {
EmitVectorZeroUpper(context, op.Rd); if (op.RegisterSize == RegisterSize.Simd64)
{
EmitVectorZeroUpper(context, op.Rd);
}
}
else
{
if (sizeF == 0)
{
EmitVectorZero32_128(context, op.Rd);
}
else /* if (sizeF == 1) */
{
EmitVectorZeroUpper(context, op.Rd);
}
} }
} }
private static void EmitFcmp(ILEmitterCtx context, OpCode ilOp, int index, bool scalar) private static void EmitCmpSseOrSse2OpF(ILEmitterCtx context, string name, bool scalar, bool isLeOrLt = false)
{ {
OpCodeSimd64 op = (OpCodeSimd64)context.CurrOp; OpCodeSimd64 op = (OpCodeSimd64)context.CurrOp;
int sizeF = op.Size & 1; int sizeF = op.Size & 1;
ulong szMask = ulong.MaxValue >> (64 - (32 << sizeF)); if (sizeF == 0)
EmitVectorExtractF(context, op.Rn, index, sizeF);
if (op is OpCodeSimdReg64 binOp)
{ {
EmitVectorExtractF(context, binOp.Rm, index, sizeF); Type[] types = new Type[] { typeof(Vector128<float>), typeof(Vector128<float>) };
}
else if (sizeF == 0) if (!isLeOrLt)
{ {
context.EmitLdc_R4(0f); context.EmitLdvec(op.Rn);
}
if (op is OpCodeSimdReg64 binOp)
{
context.EmitLdvec(binOp.Rm);
}
else
{
VectorHelper.EmitCall(context, nameof(VectorHelper.VectorSingleZero));
}
if (isLeOrLt)
{
context.EmitLdvec(op.Rn);
}
context.EmitCall(typeof(Sse).GetMethod(name, types));
context.EmitStvec(op.Rd);
if (scalar)
{
EmitVectorZero32_128(context, op.Rd);
}
else if (op.RegisterSize == RegisterSize.Simd64)
{
EmitVectorZeroUpper(context, op.Rd);
}
} }
else /* if (sizeF == 1) */ else /* if (sizeF == 1) */
{ {
context.EmitLdc_R8(0d); Type[] types = new Type[] { typeof(Vector128<double>), typeof(Vector128<double>) };
if (!isLeOrLt)
{
EmitLdvecWithCastToDouble(context, op.Rn);
}
if (op is OpCodeSimdReg64 binOp)
{
EmitLdvecWithCastToDouble(context, binOp.Rm);
}
else
{
VectorHelper.EmitCall(context, nameof(VectorHelper.VectorDoubleZero));
}
if (isLeOrLt)
{
EmitLdvecWithCastToDouble(context, op.Rn);
}
context.EmitCall(typeof(Sse2).GetMethod(name, types));
EmitStvecWithCastFromDouble(context, op.Rd);
if (scalar)
{
EmitVectorZeroUpper(context, op.Rd);
}
} }
ILLabel lblTrue = new ILLabel();
ILLabel lblEnd = new ILLabel();
context.Emit(ilOp, lblTrue);
if (scalar)
{
EmitVectorZeroAll(context, op.Rd);
}
else
{
EmitVectorInsert(context, op.Rd, index, sizeF + 2, 0);
}
context.Emit(OpCodes.Br_S, lblEnd);
context.MarkLabel(lblTrue);
if (scalar)
{
EmitVectorInsert(context, op.Rd, index, 3, (long)szMask);
EmitVectorZeroUpper(context, op.Rd);
}
else
{
EmitVectorInsert(context, op.Rd, index, sizeF + 2, (long)szMask);
}
context.MarkLabel(lblEnd);
} }
} }
} }

110
ChocolArm64/Instructions/InstEmitSimdHelper.cs
View file

@ -322,26 +322,6 @@ namespace ChocolArm64.Instructions
context.EmitCall(mthdInfo); context.EmitCall(mthdInfo);
} }
public static void EmitUnarySoftFloatCall(ILEmitterCtx context, string name)
{
IOpCodeSimd64 op = (IOpCodeSimd64)context.CurrOp;
int sizeF = op.Size & 1;
MethodInfo mthdInfo;
if (sizeF == 0)
{
mthdInfo = typeof(SoftFloat).GetMethod(name, new Type[] { typeof(float) });
}
else /* if (sizeF == 1) */
{
mthdInfo = typeof(SoftFloat).GetMethod(name, new Type[] { typeof(double) });
}
context.EmitCall(mthdInfo);
}
public static void EmitSoftFloatCall(ILEmitterCtx context, string name) public static void EmitSoftFloatCall(ILEmitterCtx context, string name)
{ {
IOpCodeSimd64 op = (IOpCodeSimd64)context.CurrOp; IOpCodeSimd64 op = (IOpCodeSimd64)context.CurrOp;
@ -909,6 +889,96 @@ namespace ChocolArm64.Instructions
} }
} }
public static void EmitVectorPairwiseSseOrSse2OpF(ILEmitterCtx context, string name)
{
OpCodeSimdReg64 op = (OpCodeSimdReg64)context.CurrOp;
int sizeF = op.Size & 1;
if (sizeF == 0)
{
if (op.RegisterSize == RegisterSize.Simd64)
{
Type[] types = new Type[] { typeof(Vector128<float>), typeof(Vector128<float>) };
context.EmitLdvec(op.Rn);
context.EmitLdvec(op.Rm);
context.EmitCall(typeof(Sse).GetMethod(nameof(Sse.UnpackLow), types));
context.Emit(OpCodes.Dup);
context.EmitStvectmp();
VectorHelper.EmitCall(context, nameof(VectorHelper.VectorSingleZero));
context.EmitCall(typeof(Sse).GetMethod(nameof(Sse.MoveLowToHigh), types));
VectorHelper.EmitCall(context, nameof(VectorHelper.VectorSingleZero));
context.EmitLdvectmp();
context.EmitCall(typeof(Sse).GetMethod(nameof(Sse.MoveHighToLow), types));
context.EmitCall(typeof(Sse).GetMethod(name, types));
context.EmitStvec(op.Rd);
}
else /* if (op.RegisterSize == RegisterSize.Simd128) */
{
Type[] typesSfl = new Type[] { typeof(Vector128<float>), typeof(Vector128<float>), typeof(byte) };
Type[] types = new Type[] { typeof(Vector128<float>), typeof(Vector128<float>) };
context.EmitLdvec(op.Rn);
context.Emit(OpCodes.Dup);
context.EmitStvectmp();
context.EmitLdvec(op.Rm);
context.Emit(OpCodes.Dup);
context.EmitStvectmp2();
context.EmitLdc_I4(2 << 6 | 0 << 4 | 2 << 2 | 0 << 0);
context.EmitCall(typeof(Sse).GetMethod(nameof(Sse.Shuffle), typesSfl));
context.EmitLdvectmp();
context.EmitLdvectmp2();
context.EmitLdc_I4(3 << 6 | 1 << 4 | 3 << 2 | 1 << 0);
context.EmitCall(typeof(Sse).GetMethod(nameof(Sse.Shuffle), typesSfl));
context.EmitCall(typeof(Sse).GetMethod(name, types));
context.EmitStvec(op.Rd);
}
}
else /* if (sizeF == 1) */
{
Type[] types = new Type[] { typeof(Vector128<double>), typeof(Vector128<double>) };
EmitLdvecWithCastToDouble(context, op.Rn);
context.Emit(OpCodes.Dup);
context.EmitStvectmp();
EmitLdvecWithCastToDouble(context, op.Rm);
context.Emit(OpCodes.Dup);
context.EmitStvectmp2();
context.EmitCall(typeof(Sse2).GetMethod(nameof(Sse2.UnpackLow), types));
context.EmitLdvectmp();
context.EmitLdvectmp2();
context.EmitCall(typeof(Sse2).GetMethod(nameof(Sse2.UnpackHigh), types));
context.EmitCall(typeof(Sse2).GetMethod(name, types));
EmitStvecWithCastFromDouble(context, op.Rd);
}
}
[Flags] [Flags]
public enum SaturatingFlags public enum SaturatingFlags
{ {

80
ChocolArm64/Instructions/InstEmitSimdMove.cs
View file

@ -377,75 +377,47 @@ namespace ChocolArm64.Instructions
{ {
OpCodeSimd64 op = (OpCodeSimd64)context.CurrOp; OpCodeSimd64 op = (OpCodeSimd64)context.CurrOp;
int elems = 8 >> op.Size; if (Optimizations.UseSsse3)
int part = op.RegisterSize == RegisterSize.Simd128 ? elems : 0;
if (Optimizations.UseSse41 && op.Size < 2)
{ {
void EmitZeroVector() long[] masks = new long[]
{ {
switch (op.Size) 14L << 56 | 12L << 48 | 10L << 40 | 08L << 32 | 06L << 24 | 04L << 16 | 02L << 8 | 00L << 0,
{ 13L << 56 | 12L << 48 | 09L << 40 | 08L << 32 | 05L << 24 | 04L << 16 | 01L << 8 | 00L << 0,
case 0: VectorHelper.EmitCall(context, nameof(VectorHelper.VectorInt16Zero)); break; 11L << 56 | 10L << 48 | 09L << 40 | 08L << 32 | 03L << 24 | 02L << 16 | 01L << 8 | 00L << 0
case 1: VectorHelper.EmitCall(context, nameof(VectorHelper.VectorInt32Zero)); break; };
}
}
//For XTN, first operand is source, second operand is 0. Type[] typesMov = new Type[] { typeof(Vector128<float>), typeof(Vector128<float>) };
//For XTN2, first operand is 0, second operand is source. Type[] typesSfl = new Type[] { typeof(Vector128<sbyte>), typeof(Vector128<sbyte>) };
if (part != 0) Type[] typesSve = new Type[] { typeof(long), typeof(long) };
{
EmitZeroVector();
}
EmitLdvecWithSignedCast(context, op.Rn, op.Size + 1); string nameMov = op.RegisterSize == RegisterSize.Simd128
? nameof(Sse.MoveLowToHigh)
: nameof(Sse.MoveHighToLow);
//Set mask to discard the upper half of the wide elements. context.EmitLdvec(op.Rd);
switch (op.Size) VectorHelper.EmitCall(context, nameof(VectorHelper.VectorSingleZero));
{
case 0: context.EmitLdc_I4(0x00ff); break;
case 1: context.EmitLdc_I4(0x0000ffff); break;
}
Type wideType = IntTypesPerSizeLog2[op.Size + 1]; context.EmitCall(typeof(Sse).GetMethod(nameof(Sse.MoveLowToHigh), typesMov));
context.EmitCall(typeof(Sse2).GetMethod(nameof(Sse2.SetAllVector128), new Type[] { wideType })); EmitLdvecWithSignedCast(context, op.Rn, 0);
wideType = VectorIntTypesPerSizeLog2[op.Size + 1]; context.EmitLdc_I8(masks[op.Size]);
context.Emit(OpCodes.Dup);
Type[] wideTypes = new Type[] { wideType, wideType }; context.EmitCall(typeof(Sse2).GetMethod(nameof(Sse2.SetVector128), typesSve));
context.EmitCall(typeof(Sse2).GetMethod(nameof(Sse2.And), wideTypes)); context.EmitCall(typeof(Ssse3).GetMethod(nameof(Ssse3.Shuffle), typesSfl));
if (part == 0) context.EmitCall(typeof(Sse).GetMethod(nameMov, typesMov));
{
EmitZeroVector();
}
//Pack values with signed saturation, the signed saturation shouldn't context.EmitStvec(op.Rd);
//saturate anything since the upper bits were masked off.
Type sseType = op.Size == 0 ? typeof(Sse2) : typeof(Sse41);
context.EmitCall(sseType.GetMethod(nameof(Sse2.PackUnsignedSaturate), wideTypes));
if (part != 0)
{
//For XTN2, we additionally need to discard the upper bits
//of the target register and OR the result with it.
EmitVectorZeroUpper(context, op.Rd);
EmitLdvecWithUnsignedCast(context, op.Rd, op.Size);
Type narrowType = VectorUIntTypesPerSizeLog2[op.Size];
context.EmitCall(typeof(Sse2).GetMethod(nameof(Sse2.Or), new Type[] { narrowType, narrowType }));
}
EmitStvecWithUnsignedCast(context, op.Rd, op.Size);
} }
else else
{ {
int elems = 8 >> op.Size;
int part = op.RegisterSize == RegisterSize.Simd128 ? elems : 0;
if (part != 0) if (part != 0)
{ {
context.EmitLdvec(op.Rd); context.EmitLdvec(op.Rd);

98
ChocolArm64/Instructions/InstEmitSimdShift.cs
View file

@ -22,9 +22,11 @@ namespace ChocolArm64.Instructions
{ {
OpCodeSimdShImm64 op = (OpCodeSimdShImm64)context.CurrOp; OpCodeSimdShImm64 op = (OpCodeSimdShImm64)context.CurrOp;
int shift = GetImmShl(op);
EmitScalarUnaryOpZx(context, () => EmitScalarUnaryOpZx(context, () =>
{ {
context.EmitLdc_I4(GetImmShl(op)); context.EmitLdc_I4(shift);
context.Emit(OpCodes.Shl); context.Emit(OpCodes.Shl);
}); });
@ -34,13 +36,15 @@ namespace ChocolArm64.Instructions
{ {
OpCodeSimdShImm64 op = (OpCodeSimdShImm64)context.CurrOp; OpCodeSimdShImm64 op = (OpCodeSimdShImm64)context.CurrOp;
int shift = GetImmShl(op);
if (Optimizations.UseSse2 && op.Size > 0) if (Optimizations.UseSse2 && op.Size > 0)
{ {
Type[] typesSll = new Type[] { VectorUIntTypesPerSizeLog2[op.Size], typeof(byte) }; Type[] typesSll = new Type[] { VectorUIntTypesPerSizeLog2[op.Size], typeof(byte) };
EmitLdvecWithUnsignedCast(context, op.Rn, op.Size); EmitLdvecWithUnsignedCast(context, op.Rn, op.Size);
context.EmitLdc_I4(GetImmShl(op)); context.EmitLdc_I4(shift);
context.EmitCall(typeof(Sse2).GetMethod(nameof(Sse2.ShiftLeftLogical), typesSll)); context.EmitCall(typeof(Sse2).GetMethod(nameof(Sse2.ShiftLeftLogical), typesSll));
EmitStvecWithUnsignedCast(context, op.Rd, op.Size); EmitStvecWithUnsignedCast(context, op.Rd, op.Size);
@ -54,7 +58,7 @@ namespace ChocolArm64.Instructions
{ {
EmitVectorUnaryOpZx(context, () => EmitVectorUnaryOpZx(context, () =>
{ {
context.EmitLdc_I4(GetImmShl(op)); context.EmitLdc_I4(shift);
context.Emit(OpCodes.Shl); context.Emit(OpCodes.Shl);
}); });
@ -67,7 +71,33 @@ namespace ChocolArm64.Instructions
int shift = 8 << op.Size; int shift = 8 << op.Size;
EmitVectorShImmWidenBinaryZx(context, () => context.Emit(OpCodes.Shl), shift); if (Optimizations.UseSse41)
{
Type[] typesSll = new Type[] { VectorUIntTypesPerSizeLog2[op.Size + 1], typeof(byte) };
Type[] typesCvt = new Type[] { VectorUIntTypesPerSizeLog2[op.Size] };
string[] namesCvt = new string[] { nameof(Sse41.ConvertToVector128Int16),
nameof(Sse41.ConvertToVector128Int32),
nameof(Sse41.ConvertToVector128Int64) };
int numBytes = op.RegisterSize == RegisterSize.Simd128 ? 8 : 0;
EmitLdvecWithUnsignedCast(context, op.Rn, op.Size);
context.EmitLdc_I4(numBytes);
context.EmitCall(typeof(Sse2).GetMethod(nameof(Sse2.ShiftRightLogical128BitLane), typesSll));
context.EmitCall(typeof(Sse41).GetMethod(namesCvt[op.Size], typesCvt));
context.EmitLdc_I4(shift);
context.EmitCall(typeof(Sse2).GetMethod(nameof(Sse2.ShiftLeftLogical), typesSll));
EmitStvecWithUnsignedCast(context, op.Rd, op.Size + 1);
}
else
{
EmitVectorShImmWidenBinaryZx(context, () => context.Emit(OpCodes.Shl), shift);
}
} }
public static void Shrn_V(ILEmitterCtx context) public static void Shrn_V(ILEmitterCtx context)
@ -362,7 +392,35 @@ namespace ChocolArm64.Instructions
{ {
OpCodeSimdShImm64 op = (OpCodeSimdShImm64)context.CurrOp; OpCodeSimdShImm64 op = (OpCodeSimdShImm64)context.CurrOp;
EmitVectorShImmWidenBinarySx(context, () => context.Emit(OpCodes.Shl), GetImmShl(op)); int shift = GetImmShl(op);
if (Optimizations.UseSse41)
{
Type[] typesSll = new Type[] { VectorIntTypesPerSizeLog2[op.Size + 1], typeof(byte) };
Type[] typesCvt = new Type[] { VectorIntTypesPerSizeLog2[op.Size] };
string[] namesCvt = new string[] { nameof(Sse41.ConvertToVector128Int16),
nameof(Sse41.ConvertToVector128Int32),
nameof(Sse41.ConvertToVector128Int64) };
int numBytes = op.RegisterSize == RegisterSize.Simd128 ? 8 : 0;
EmitLdvecWithSignedCast(context, op.Rn, op.Size);
context.EmitLdc_I4(numBytes);
context.EmitCall(typeof(Sse2).GetMethod(nameof(Sse2.ShiftRightLogical128BitLane), typesSll));
context.EmitCall(typeof(Sse41).GetMethod(namesCvt[op.Size], typesCvt));
context.EmitLdc_I4(shift);
context.EmitCall(typeof(Sse2).GetMethod(nameof(Sse2.ShiftLeftLogical), typesSll));
EmitStvecWithSignedCast(context, op.Rd, op.Size + 1);
}
else
{
EmitVectorShImmWidenBinarySx(context, () => context.Emit(OpCodes.Shl), shift);
}
} }
public static void Sshr_S(ILEmitterCtx context) public static void Sshr_S(ILEmitterCtx context)
@ -663,7 +721,35 @@ namespace ChocolArm64.Instructions
{ {
OpCodeSimdShImm64 op = (OpCodeSimdShImm64)context.CurrOp; OpCodeSimdShImm64 op = (OpCodeSimdShImm64)context.CurrOp;
EmitVectorShImmWidenBinaryZx(context, () => context.Emit(OpCodes.Shl), GetImmShl(op)); int shift = GetImmShl(op);
if (Optimizations.UseSse41)
{
Type[] typesSll = new Type[] { VectorUIntTypesPerSizeLog2[op.Size + 1], typeof(byte) };
Type[] typesCvt = new Type[] { VectorUIntTypesPerSizeLog2[op.Size] };
string[] namesCvt = new string[] { nameof(Sse41.ConvertToVector128Int16),
nameof(Sse41.ConvertToVector128Int32),
nameof(Sse41.ConvertToVector128Int64) };
int numBytes = op.RegisterSize == RegisterSize.Simd128 ? 8 : 0;
EmitLdvecWithUnsignedCast(context, op.Rn, op.Size);
context.EmitLdc_I4(numBytes);
context.EmitCall(typeof(Sse2).GetMethod(nameof(Sse2.ShiftRightLogical128BitLane), typesSll));
context.EmitCall(typeof(Sse41).GetMethod(namesCvt[op.Size], typesCvt));
context.EmitLdc_I4(shift);
context.EmitCall(typeof(Sse2).GetMethod(nameof(Sse2.ShiftLeftLogical), typesSll));
EmitStvecWithUnsignedCast(context, op.Rd, op.Size + 1);
}
else
{
EmitVectorShImmWidenBinaryZx(context, () => context.Emit(OpCodes.Shl), shift);
}
} }
public static void Ushr_S(ILEmitterCtx context) public static void Ushr_S(ILEmitterCtx context)

721
ChocolArm64/Instructions/SoftFloat.cs
View file

@ -9,191 +9,72 @@ namespace ChocolArm64.Instructions
{ {
static SoftFloat() static SoftFloat()
{ {
RecipEstimateTable = BuildRecipEstimateTable(); RecipEstimateTable = BuildRecipEstimateTable();
InvSqrtEstimateTable = BuildInvSqrtEstimateTable(); RecipSqrtEstimateTable = BuildRecipSqrtEstimateTable();
} }
private static readonly byte[] RecipEstimateTable; internal static readonly byte[] RecipEstimateTable;
private static readonly byte[] InvSqrtEstimateTable; internal static readonly byte[] RecipSqrtEstimateTable;
private static byte[] BuildRecipEstimateTable() private static byte[] BuildRecipEstimateTable()
{ {
byte[] table = new byte[256]; byte[] tbl = new byte[256];
for (ulong index = 0; index < 256; index++)
for (int idx = 0; idx < 256; idx++)
{ {
ulong a = index | 0x100; uint src = (uint)idx + 256u;
a = (a << 1) + 1; Debug.Assert(256u <= src && src < 512u);
ulong b = 0x80000 / a;
b = (b + 1) >> 1;
table[index] = (byte)(b & 0xFF); src = (src << 1) + 1u;
uint aux = (1u << 19) / src;
uint dst = (aux + 1u) >> 1;
Debug.Assert(256u <= dst && dst < 512u);
tbl[idx] = (byte)(dst - 256u);
} }
return table;
return tbl;
} }
private static byte[] BuildInvSqrtEstimateTable() private static byte[] BuildRecipSqrtEstimateTable()
{ {
byte[] table = new byte[512]; byte[] tbl = new byte[384];
for (ulong index = 128; index < 512; index++)
for (int idx = 0; idx < 384; idx++)
{ {
ulong a = index; uint src = (uint)idx + 128u;
if (a < 256)
Debug.Assert(128u <= src && src < 512u);
if (src < 256u)
{ {
a = (a << 1) + 1; src = (src << 1) + 1u;
} }
else else
{ {
a = (a | 1) << 1; src = (src >> 1) << 1;
src = (src + 1u) << 1;
} }
ulong b = 256; uint aux = 512u;
while (a * (b + 1) * (b + 1) < (1ul << 28))
while (src * (aux + 1u) * (aux + 1u) < (1u << 28))
{ {
b++; aux = aux + 1u;
}
b = (b + 1) >> 1;
table[index] = (byte)(b & 0xFF);
}
return table;
}
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public static float RecipEstimate(float x)
{
return (float)RecipEstimate((double)x);
}
public static double RecipEstimate(double x)
{
ulong xBits = (ulong)BitConverter.DoubleToInt64Bits(x);
ulong xSign = xBits & 0x8000000000000000;
ulong xExp = (xBits >> 52) & 0x7FF;
ulong scaled = xBits & ((1ul << 52) - 1);
if (xExp >= 2045)
{
if (xExp == 0x7ff && scaled != 0)
{
// NaN
return BitConverter.Int64BitsToDouble((long)(xBits | 0x0008000000000000));
} }
// Infinity, or Out of range -> Zero uint dst = (aux + 1u) >> 1;
return BitConverter.Int64BitsToDouble((long)xSign);
Debug.Assert(256u <= dst && dst < 512u);
tbl[idx] = (byte)(dst - 256u);
} }
if (xExp == 0) return tbl;
{
if (scaled == 0)
{
// Zero -> Infinity
return BitConverter.Int64BitsToDouble((long)(xSign | 0x7FF0000000000000));
}
// Denormal
if ((scaled & (1ul << 51)) == 0)
{
xExp = ~0ul;
scaled <<= 2;
}
else
{
scaled <<= 1;
}
}
scaled >>= 44;
scaled &= 0xFF;
ulong resultExp = (2045 - xExp) & 0x7FF;
ulong estimate = (ulong)RecipEstimateTable[scaled];
ulong fraction = estimate << 44;
if (resultExp == 0)
{
fraction >>= 1;
fraction |= 1ul << 51;
}
else if (resultExp == 0x7FF)
{
resultExp = 0;
fraction >>= 2;
fraction |= 1ul << 50;
}
ulong result = xSign | (resultExp << 52) | fraction;
return BitConverter.Int64BitsToDouble((long)result);
}
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public static float InvSqrtEstimate(float x)
{
return (float)InvSqrtEstimate((double)x);
}
public static double InvSqrtEstimate(double x)
{
ulong xBits = (ulong)BitConverter.DoubleToInt64Bits(x);
ulong xSign = xBits & 0x8000000000000000;
long xExp = (long)((xBits >> 52) & 0x7FF);
ulong scaled = xBits & ((1ul << 52) - 1);
if (xExp == 0x7FF && scaled != 0)
{
// NaN
return BitConverter.Int64BitsToDouble((long)(xBits | 0x0008000000000000));
}
if (xExp == 0)
{
if (scaled == 0)
{
// Zero -> Infinity
return BitConverter.Int64BitsToDouble((long)(xSign | 0x7FF0000000000000));
}
// Denormal
while ((scaled & (1 << 51)) == 0)
{
scaled <<= 1;
xExp--;
}
scaled <<= 1;
}
if (xSign != 0)
{
// Negative -> NaN
return BitConverter.Int64BitsToDouble((long)0x7FF8000000000000);
}
if (xExp == 0x7ff && scaled == 0)
{
// Infinity -> Zero
return BitConverter.Int64BitsToDouble((long)xSign);
}
if (((ulong)xExp & 1) == 1)
{
scaled >>= 45;
scaled &= 0xFF;
scaled |= 0x80;
}
else
{
scaled >>= 44;
scaled &= 0xFF;
scaled |= 0x100;
}
ulong resultExp = ((ulong)(3068 - xExp) / 2) & 0x7FF;
ulong estimate = (ulong)InvSqrtEstimateTable[scaled];
ulong fraction = estimate << 44;
ulong result = xSign | (resultExp << 52) | fraction;
return BitConverter.Int64BitsToDouble((long)result);
} }
} }
@ -395,12 +276,12 @@ namespace ChocolArm64.Instructions
{ {
intMant++; intMant++;
if (intMant == (uint)Math.Pow(2d, f)) if (intMant == 1u << f)
{ {
biasedExp = 1u; biasedExp = 1u;
} }
if (intMant == (uint)Math.Pow(2d, f + 1)) if (intMant == 1u << (f + 1))
{ {
biasedExp++; biasedExp++;
intMant >>= 1; intMant >>= 1;
@ -409,7 +290,7 @@ namespace ChocolArm64.Instructions
float result; float result;
if (biasedExp >= (uint)Math.Pow(2d, e) - 1u) if (biasedExp >= (1u << e) - 1u)
{ {
result = overflowToInf ? FPInfinity(sign) : FPMaxNormal(sign); result = overflowToInf ? FPInfinity(sign) : FPMaxNormal(sign);
@ -666,12 +547,12 @@ namespace ChocolArm64.Instructions
{ {
intMant++; intMant++;
if (intMant == (uint)Math.Pow(2d, f)) if (intMant == 1u << f)
{ {
biasedExp = 1u; biasedExp = 1u;
} }
if (intMant == (uint)Math.Pow(2d, f + 1)) if (intMant == 1u << (f + 1))
{ {
biasedExp++; biasedExp++;
intMant >>= 1; intMant >>= 1;
@ -682,7 +563,7 @@ namespace ChocolArm64.Instructions
if (!state.GetFpcrFlag(Fpcr.Ahp)) if (!state.GetFpcrFlag(Fpcr.Ahp))
{ {
if (biasedExp >= (uint)Math.Pow(2d, e) - 1u) if (biasedExp >= (1u << e) - 1u)
{ {
resultBits = overflowToInf ? FPInfinity(sign) : FPMaxNormal(sign); resultBits = overflowToInf ? FPInfinity(sign) : FPMaxNormal(sign);
@ -697,7 +578,7 @@ namespace ChocolArm64.Instructions
} }
else else
{ {
if (biasedExp >= (uint)Math.Pow(2d, e)) if (biasedExp >= 1u << e)
{ {
resultBits = (ushort)((sign ? 1u : 0u) << 15 | 0x7FFFu); resultBits = (ushort)((sign ? 1u : 0u) << 15 | 0x7FFFu);
@ -826,6 +707,94 @@ namespace ChocolArm64.Instructions
return result; return result;
} }
public static float FPCompareEQ(float value1, float value2, CpuThreadState state)
{
Debug.WriteLineIf(state.Fpcr != 0, $"SoftFloat32.FPCompareEQ: state.Fpcr = 0x{state.Fpcr:X8}");
value1 = value1.FPUnpack(out FpType type1, out _, out _, state);
value2 = value2.FPUnpack(out FpType type2, out _, out _, state);
float result;
if (type1 == FpType.SNaN || type1 == FpType.QNaN || type2 == FpType.SNaN || type2 == FpType.QNaN)
{
result = ZerosOrOnes(false);
if (type1 == FpType.SNaN || type2 == FpType.SNaN)
{
FPProcessException(FpExc.InvalidOp, state);
}
}
else
{
result = ZerosOrOnes(value1 == value2);
}
return result;
}
public static float FPCompareGE(float value1, float value2, CpuThreadState state)
{
Debug.WriteLineIf(state.Fpcr != 0, $"SoftFloat32.FPCompareGE: state.Fpcr = 0x{state.Fpcr:X8}");
value1 = value1.FPUnpack(out FpType type1, out _, out _, state);
value2 = value2.FPUnpack(out FpType type2, out _, out _, state);
float result;
if (type1 == FpType.SNaN || type1 == FpType.QNaN || type2 == FpType.SNaN || type2 == FpType.QNaN)
{
result = ZerosOrOnes(false);
FPProcessException(FpExc.InvalidOp, state);
}
else
{
result = ZerosOrOnes(value1 >= value2);
}
return result;
}
public static float FPCompareGT(float value1, float value2, CpuThreadState state)
{
Debug.WriteLineIf(state.Fpcr != 0, $"SoftFloat32.FPCompareGT: state.Fpcr = 0x{state.Fpcr:X8}");
value1 = value1.FPUnpack(out FpType type1, out _, out _, state);
value2 = value2.FPUnpack(out FpType type2, out _, out _, state);
float result;
if (type1 == FpType.SNaN || type1 == FpType.QNaN || type2 == FpType.SNaN || type2 == FpType.QNaN)
{
result = ZerosOrOnes(false);
FPProcessException(FpExc.InvalidOp, state);
}
else
{
result = ZerosOrOnes(value1 > value2);
}
return result;
}
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public static float FPCompareLE(float value1, float value2, CpuThreadState state)
{
Debug.WriteLineIf(state.Fpcr != 0, $"SoftFloat32.FPCompareLE: state.Fpcr = 0x{state.Fpcr:X8}");
return FPCompareGE(value2, value1, state);
}
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public static float FPCompareLT(float value1, float value2, CpuThreadState state)
{
Debug.WriteLineIf(state.Fpcr != 0, $"SoftFloat32.FPCompareLT: state.Fpcr = 0x{state.Fpcr:X8}");
return FPCompareGT(value2, value1, state);
}
public static float FPDiv(float value1, float value2, CpuThreadState state) public static float FPDiv(float value1, float value2, CpuThreadState state)
{ {
Debug.WriteLineIf(state.Fpcr != 0, $"SoftFloat32.FPDiv: state.Fpcr = 0x{state.Fpcr:X8}"); Debug.WriteLineIf(state.Fpcr != 0, $"SoftFloat32.FPDiv: state.Fpcr = 0x{state.Fpcr:X8}");
@ -1188,6 +1157,95 @@ namespace ChocolArm64.Instructions
return result; return result;
} }
public static float FPRecipEstimate(float value, CpuThreadState state)
{
Debug.WriteLineIf(state.Fpcr != 0, $"SoftFloat32.FPRecipEstimate: state.Fpcr = 0x{state.Fpcr:X8}");
value.FPUnpack(out FpType type, out bool sign, out uint op, state);
float result;
if (type == FpType.SNaN || type == FpType.QNaN)
{
result = FPProcessNaN(type, op, state);
}
else if (type == FpType.Infinity)
{
result = FPZero(sign);
}
else if (type == FpType.Zero)
{
result = FPInfinity(sign);
FPProcessException(FpExc.DivideByZero, state);
}
else if (MathF.Abs(value) < MathF.Pow(2f, -128))
{
bool overflowToInf;
switch (state.FPRoundingMode())
{
default:
case RoundMode.ToNearest: overflowToInf = true; break;
case RoundMode.TowardsPlusInfinity: overflowToInf = !sign; break;
case RoundMode.TowardsMinusInfinity: overflowToInf = sign; break;
case RoundMode.TowardsZero: overflowToInf = false; break;
}
result = overflowToInf ? FPInfinity(sign) : FPMaxNormal(sign);
FPProcessException(FpExc.Overflow, state);
FPProcessException(FpExc.Inexact, state);
}
else if (state.GetFpcrFlag(Fpcr.Fz) && (MathF.Abs(value) >= MathF.Pow(2f, 126)))
{
result = FPZero(sign);
state.SetFpsrFlag(Fpsr.Ufc);
}
else
{
ulong fraction = (ulong)(op & 0x007FFFFFu) << 29;
uint exp = (op & 0x7F800000u) >> 23;
if (exp == 0u)
{
if ((fraction & 0x0008000000000000ul) == 0ul)
{
fraction = (fraction & 0x0003FFFFFFFFFFFFul) << 2;
exp -= 1u;
}
else
{
fraction = (fraction & 0x0007FFFFFFFFFFFFul) << 1;
}
}
uint scaled = (uint)(((fraction & 0x000FF00000000000ul) | 0x0010000000000000ul) >> 44);
uint resultExp = 253u - exp;
uint estimate = (uint)SoftFloat.RecipEstimateTable[scaled - 256u] + 256u;
fraction = (ulong)(estimate & 0xFFu) << 44;
if (resultExp == 0u)
{
fraction = ((fraction & 0x000FFFFFFFFFFFFEul) | 0x0010000000000000ul) >> 1;
}
else if (resultExp + 1u == 0u)
{
fraction = ((fraction & 0x000FFFFFFFFFFFFCul) | 0x0010000000000000ul) >> 2;
resultExp = 0u;
}
result = BitConverter.Int32BitsToSingle(
(int)((sign ? 1u : 0u) << 31 | (resultExp & 0xFFu) << 23 | (uint)(fraction >> 29) & 0x007FFFFFu));
}
return result;
}
public static float FPRecipStepFused(float value1, float value2, CpuThreadState state) public static float FPRecipStepFused(float value1, float value2, CpuThreadState state)
{ {
Debug.WriteLineIf(state.Fpcr != 0, $"SoftFloat32.FPRecipStepFused: state.Fpcr = 0x{state.Fpcr:X8}"); Debug.WriteLineIf(state.Fpcr != 0, $"SoftFloat32.FPRecipStepFused: state.Fpcr = 0x{state.Fpcr:X8}");
@ -1255,6 +1313,71 @@ namespace ChocolArm64.Instructions
return result; return result;
} }
public static float FPRSqrtEstimate(float value, CpuThreadState state)
{
Debug.WriteLineIf(state.Fpcr != 0, $"SoftFloat32.FPRSqrtEstimate: state.Fpcr = 0x{state.Fpcr:X8}");
value.FPUnpack(out FpType type, out bool sign, out uint op, state);
float result;
if (type == FpType.SNaN || type == FpType.QNaN)
{
result = FPProcessNaN(type, op, state);
}
else if (type == FpType.Zero)
{
result = FPInfinity(sign);
FPProcessException(FpExc.DivideByZero, state);
}
else if (sign)
{
result = FPDefaultNaN();
FPProcessException(FpExc.InvalidOp, state);
}
else if (type == FpType.Infinity)
{
result = FPZero(false);
}
else
{
ulong fraction = (ulong)(op & 0x007FFFFFu) << 29;
uint exp = (op & 0x7F800000u) >> 23;
if (exp == 0u)
{
while ((fraction & 0x0008000000000000ul) == 0ul)
{
fraction = (fraction & 0x0007FFFFFFFFFFFFul) << 1;
exp -= 1u;
}
fraction = (fraction & 0x0007FFFFFFFFFFFFul) << 1;
}
uint scaled;
if ((exp & 1u) == 0u)
{
scaled = (uint)(((fraction & 0x000FF00000000000ul) | 0x0010000000000000ul) >> 44);
}
else
{
scaled = (uint)(((fraction & 0x000FE00000000000ul) | 0x0010000000000000ul) >> 45);
}
uint resultExp = (380u - exp) >> 1;
uint estimate = (uint)SoftFloat.RecipSqrtEstimateTable[scaled - 128u] + 256u;
result = BitConverter.Int32BitsToSingle((int)((resultExp & 0xFFu) << 23 | (estimate & 0xFFu) << 15));
}
return result;
}
public static float FPRSqrtStepFused(float value1, float value2, CpuThreadState state) public static float FPRSqrtStepFused(float value1, float value2, CpuThreadState state)
{ {
Debug.WriteLineIf(state.Fpcr != 0, $"SoftFloat32.FPRSqrtStepFused: state.Fpcr = 0x{state.Fpcr:X8}"); Debug.WriteLineIf(state.Fpcr != 0, $"SoftFloat32.FPRSqrtStepFused: state.Fpcr = 0x{state.Fpcr:X8}");
@ -1402,6 +1525,11 @@ namespace ChocolArm64.Instructions
return sign ? -0f : +0f; return sign ? -0f : +0f;
} }
private static float FPMaxNormal(bool sign)
{
return sign ? float.MinValue : float.MaxValue;
}
private static float FPTwo(bool sign) private static float FPTwo(bool sign)
{ {
return sign ? -2f : +2f; return sign ? -2f : +2f;
@ -1417,6 +1545,11 @@ namespace ChocolArm64.Instructions
return -value; return -value;
} }
private static float ZerosOrOnes(bool zeros)
{
return BitConverter.Int32BitsToSingle(!zeros ? 0 : -1);
}
private static float FPUnpack( private static float FPUnpack(
this float value, this float value,
out FpType type, out FpType type,
@ -1658,6 +1791,94 @@ namespace ChocolArm64.Instructions
return result; return result;
} }
public static double FPCompareEQ(double value1, double value2, CpuThreadState state)
{
Debug.WriteLineIf(state.Fpcr != 0, $"SoftFloat64.FPCompareEQ: state.Fpcr = 0x{state.Fpcr:X8}");
value1 = value1.FPUnpack(out FpType type1, out _, out _, state);
value2 = value2.FPUnpack(out FpType type2, out _, out _, state);
double result;
if (type1 == FpType.SNaN || type1 == FpType.QNaN || type2 == FpType.SNaN || type2 == FpType.QNaN)
{
result = ZerosOrOnes(false);
if (type1 == FpType.SNaN || type2 == FpType.SNaN)
{
FPProcessException(FpExc.InvalidOp, state);
}
}
else
{
result = ZerosOrOnes(value1 == value2);
}
return result;
}
public static double FPCompareGE(double value1, double value2, CpuThreadState state)
{
Debug.WriteLineIf(state.Fpcr != 0, $"SoftFloat64.FPCompareGE: state.Fpcr = 0x{state.Fpcr:X8}");
value1 = value1.FPUnpack(out FpType type1, out _, out _, state);
value2 = value2.FPUnpack(out FpType type2, out _, out _, state);
double result;
if (type1 == FpType.SNaN || type1 == FpType.QNaN || type2 == FpType.SNaN || type2 == FpType.QNaN)
{
result = ZerosOrOnes(false);
FPProcessException(FpExc.InvalidOp, state);
}
else
{
result = ZerosOrOnes(value1 >= value2);
}
return result;
}
public static double FPCompareGT(double value1, double value2, CpuThreadState state)
{
Debug.WriteLineIf(state.Fpcr != 0, $"SoftFloat64.FPCompareGT: state.Fpcr = 0x{state.Fpcr:X8}");
value1 = value1.FPUnpack(out FpType type1, out _, out _, state);
value2 = value2.FPUnpack(out FpType type2, out _, out _, state);
double result;
if (type1 == FpType.SNaN || type1 == FpType.QNaN || type2 == FpType.SNaN || type2 == FpType.QNaN)
{
result = ZerosOrOnes(false);
FPProcessException(FpExc.InvalidOp, state);
}
else
{
result = ZerosOrOnes(value1 > value2);
}
return result;
}
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public static double FPCompareLE(double value1, double value2, CpuThreadState state)
{
Debug.WriteLineIf(state.Fpcr != 0, $"SoftFloat64.FPCompareLE: state.Fpcr = 0x{state.Fpcr:X8}");
return FPCompareGE(value2, value1, state);
}
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public static double FPCompareLT(double value1, double value2, CpuThreadState state)
{
Debug.WriteLineIf(state.Fpcr != 0, $"SoftFloat64.FPCompareLT: state.Fpcr = 0x{state.Fpcr:X8}");
return FPCompareGT(value2, value1, state);
}
public static double FPDiv(double value1, double value2, CpuThreadState state) public static double FPDiv(double value1, double value2, CpuThreadState state)
{ {
Debug.WriteLineIf(state.Fpcr != 0, $"SoftFloat64.FPDiv: state.Fpcr = 0x{state.Fpcr:X8}"); Debug.WriteLineIf(state.Fpcr != 0, $"SoftFloat64.FPDiv: state.Fpcr = 0x{state.Fpcr:X8}");
@ -2020,6 +2241,95 @@ namespace ChocolArm64.Instructions
return result; return result;
} }
public static double FPRecipEstimate(double value, CpuThreadState state)
{
Debug.WriteLineIf(state.Fpcr != 0, $"SoftFloat64.FPRecipEstimate: state.Fpcr = 0x{state.Fpcr:X8}");
value.FPUnpack(out FpType type, out bool sign, out ulong op, state);
double result;
if (type == FpType.SNaN || type == FpType.QNaN)
{
result = FPProcessNaN(type, op, state);
}
else if (type == FpType.Infinity)
{
result = FPZero(sign);
}
else if (type == FpType.Zero)
{
result = FPInfinity(sign);
FPProcessException(FpExc.DivideByZero, state);
}
else if (Math.Abs(value) < Math.Pow(2d, -1024))
{
bool overflowToInf;
switch (state.FPRoundingMode())
{
default:
case RoundMode.ToNearest: overflowToInf = true; break;
case RoundMode.TowardsPlusInfinity: overflowToInf = !sign; break;
case RoundMode.TowardsMinusInfinity: overflowToInf = sign; break;
case RoundMode.TowardsZero: overflowToInf = false; break;
}
result = overflowToInf ? FPInfinity(sign) : FPMaxNormal(sign);
FPProcessException(FpExc.Overflow, state);
FPProcessException(FpExc.Inexact, state);
}
else if (state.GetFpcrFlag(Fpcr.Fz) && (Math.Abs(value) >= Math.Pow(2d, 1022)))
{
result = FPZero(sign);
state.SetFpsrFlag(Fpsr.Ufc);
}
else
{
ulong fraction = op & 0x000FFFFFFFFFFFFFul;
uint exp = (uint)((op & 0x7FF0000000000000ul) >> 52);
if (exp == 0u)
{
if ((fraction & 0x0008000000000000ul) == 0ul)
{
fraction = (fraction & 0x0003FFFFFFFFFFFFul) << 2;
exp -= 1u;
}
else
{
fraction = (fraction & 0x0007FFFFFFFFFFFFul) << 1;
}
}
uint scaled = (uint)(((fraction & 0x000FF00000000000ul) | 0x0010000000000000ul) >> 44);
uint resultExp = 2045u - exp;
uint estimate = (uint)SoftFloat.RecipEstimateTable[scaled - 256u] + 256u;
fraction = (ulong)(estimate & 0xFFu) << 44;
if (resultExp == 0u)
{
fraction = ((fraction & 0x000FFFFFFFFFFFFEul) | 0x0010000000000000ul) >> 1;
}
else if (resultExp + 1u == 0u)
{
fraction = ((fraction & 0x000FFFFFFFFFFFFCul) | 0x0010000000000000ul) >> 2;
resultExp = 0u;
}
result = BitConverter.Int64BitsToDouble(
(long)((sign ? 1ul : 0ul) << 63 | (resultExp & 0x7FFul) << 52 | (fraction & 0x000FFFFFFFFFFFFFul)));
}
return result;
}
public static double FPRecipStepFused(double value1, double value2, CpuThreadState state) public static double FPRecipStepFused(double value1, double value2, CpuThreadState state)
{ {
Debug.WriteLineIf(state.Fpcr != 0, $"SoftFloat64.FPRecipStepFused: state.Fpcr = 0x{state.Fpcr:X8}"); Debug.WriteLineIf(state.Fpcr != 0, $"SoftFloat64.FPRecipStepFused: state.Fpcr = 0x{state.Fpcr:X8}");
@ -2087,6 +2397,71 @@ namespace ChocolArm64.Instructions
return result; return result;
} }
public static double FPRSqrtEstimate(double value, CpuThreadState state)
{
Debug.WriteLineIf(state.Fpcr != 0, $"SoftFloat64.FPRSqrtEstimate: state.Fpcr = 0x{state.Fpcr:X8}");
value.FPUnpack(out FpType type, out bool sign, out ulong op, state);
double result;
if (type == FpType.SNaN || type == FpType.QNaN)
{
result = FPProcessNaN(type, op, state);
}
else if (type == FpType.Zero)
{
result = FPInfinity(sign);
FPProcessException(FpExc.DivideByZero, state);
}
else if (sign)
{
result = FPDefaultNaN();
FPProcessException(FpExc.InvalidOp, state);
}
else if (type == FpType.Infinity)
{
result = FPZero(false);
}
else
{
ulong fraction = op & 0x000FFFFFFFFFFFFFul;
uint exp = (uint)((op & 0x7FF0000000000000ul) >> 52);
if (exp == 0u)
{
while ((fraction & 0x0008000000000000ul) == 0ul)
{
fraction = (fraction & 0x0007FFFFFFFFFFFFul) << 1;
exp -= 1u;
}
fraction = (fraction & 0x0007FFFFFFFFFFFFul) << 1;
}
uint scaled;
if ((exp & 1u) == 0u)
{
scaled = (uint)(((fraction & 0x000FF00000000000ul) | 0x0010000000000000ul) >> 44);
}
else
{
scaled = (uint)(((fraction & 0x000FE00000000000ul) | 0x0010000000000000ul) >> 45);
}
uint resultExp = (3068u - exp) >> 1;
uint estimate = (uint)SoftFloat.RecipSqrtEstimateTable[scaled - 128u] + 256u;
result = BitConverter.Int64BitsToDouble((long)((resultExp & 0x7FFul) << 52 | (estimate & 0xFFul) << 44));
}
return result;
}
public static double FPRSqrtStepFused(double value1, double value2, CpuThreadState state) public static double FPRSqrtStepFused(double value1, double value2, CpuThreadState state)
{ {
Debug.WriteLineIf(state.Fpcr != 0, $"SoftFloat64.FPRSqrtStepFused: state.Fpcr = 0x{state.Fpcr:X8}"); Debug.WriteLineIf(state.Fpcr != 0, $"SoftFloat64.FPRSqrtStepFused: state.Fpcr = 0x{state.Fpcr:X8}");
@ -2234,6 +2609,11 @@ namespace ChocolArm64.Instructions
return sign ? -0d : +0d; return sign ? -0d : +0d;
} }
private static double FPMaxNormal(bool sign)
{
return sign ? double.MinValue : double.MaxValue;
}
private static double FPTwo(bool sign) private static double FPTwo(bool sign)
{ {
return sign ? -2d : +2d; return sign ? -2d : +2d;
@ -2249,6 +2629,11 @@ namespace ChocolArm64.Instructions
return -value; return -value;
} }
private static double ZerosOrOnes(bool zeros)
{
return BitConverter.Int64BitsToDouble(!zeros ? 0L : -1L);
}
private static double FPUnpack( private static double FPUnpack(
this double value, this double value,
out FpType type, out FpType type,

1
ChocolArm64/OpCodeTable.cs
View file

@ -222,6 +222,7 @@ namespace ChocolArm64
SetA64("0x101110001xxxxx000111xxxxxxxxxx", InstEmit.Eor_V, typeof(OpCodeSimdReg64)); SetA64("0x101110001xxxxx000111xxxxxxxxxx", InstEmit.Eor_V, typeof(OpCodeSimdReg64));
SetA64("0>101110000xxxxx0<xxx0xxxxxxxxxx", InstEmit.Ext_V, typeof(OpCodeSimdExt64)); SetA64("0>101110000xxxxx0<xxx0xxxxxxxxxx", InstEmit.Ext_V, typeof(OpCodeSimdExt64));
SetA64("011111101x1xxxxx110101xxxxxxxxxx", InstEmit.Fabd_S, typeof(OpCodeSimdReg64)); SetA64("011111101x1xxxxx110101xxxxxxxxxx", InstEmit.Fabd_S, typeof(OpCodeSimdReg64));
SetA64("0>1011101<1xxxxx110101xxxxxxxxxx", InstEmit.Fabd_V, typeof(OpCodeSimdReg64));
SetA64("000111100x100000110000xxxxxxxxxx", InstEmit.Fabs_S, typeof(OpCodeSimd64)); SetA64("000111100x100000110000xxxxxxxxxx", InstEmit.Fabs_S, typeof(OpCodeSimd64));
SetA64("0>0011101<100000111110xxxxxxxxxx", InstEmit.Fabs_V, typeof(OpCodeSimd64)); SetA64("0>0011101<100000111110xxxxxxxxxx", InstEmit.Fabs_V, typeof(OpCodeSimd64));
SetA64("000111100x1xxxxx001010xxxxxxxxxx", InstEmit.Fadd_S, typeof(OpCodeSimdReg64)); SetA64("000111100x1xxxxx001010xxxxxxxxxx", InstEmit.Fadd_S, typeof(OpCodeSimdReg64));

2
ChocolArm64/Optimizations.cs
View file

@ -8,12 +8,14 @@ public static class Optimizations
private static bool _useSseIfAvailable = true; private static bool _useSseIfAvailable = true;
private static bool _useSse2IfAvailable = true; private static bool _useSse2IfAvailable = true;
private static bool _useSse3IfAvailable = true;
private static bool _useSsse3IfAvailable = true; private static bool _useSsse3IfAvailable = true;
private static bool _useSse41IfAvailable = true; private static bool _useSse41IfAvailable = true;
private static bool _useSse42IfAvailable = true; private static bool _useSse42IfAvailable = true;
internal static bool UseSse = (_useAllSseIfAvailable && _useSseIfAvailable) && Sse.IsSupported; internal static bool UseSse = (_useAllSseIfAvailable && _useSseIfAvailable) && Sse.IsSupported;
internal static bool UseSse2 = (_useAllSseIfAvailable && _useSse2IfAvailable) && Sse2.IsSupported; internal static bool UseSse2 = (_useAllSseIfAvailable && _useSse2IfAvailable) && Sse2.IsSupported;
internal static bool UseSse3 = (_useAllSseIfAvailable && _useSse3IfAvailable) && Sse3.IsSupported;
internal static bool UseSsse3 = (_useAllSseIfAvailable && _useSsse3IfAvailable) && Ssse3.IsSupported; internal static bool UseSsse3 = (_useAllSseIfAvailable && _useSsse3IfAvailable) && Ssse3.IsSupported;
internal static bool UseSse41 = (_useAllSseIfAvailable && _useSse41IfAvailable) && Sse41.IsSupported; internal static bool UseSse41 = (_useAllSseIfAvailable && _useSse41IfAvailable) && Sse41.IsSupported;
internal static bool UseSse42 = (_useAllSseIfAvailable && _useSse42IfAvailable) && Sse42.IsSupported; internal static bool UseSse42 = (_useAllSseIfAvailable && _useSse42IfAvailable) && Sse42.IsSupported;

409
Ryujinx.Tests/Cpu/CpuTestSimd.cs
View file

@ -256,6 +256,112 @@ namespace Ryujinx.Tests.Cpu
#endregion #endregion
#region "ValueSource (Opcodes)" #region "ValueSource (Opcodes)"
private static uint[] _F_Abs_Neg_Recpx_Sqrt_S_S_()
{
return new uint[]
{
0x1E20C020u, // FABS S0, S1
0x1E214020u, // FNEG S0, S1
0x5EA1F820u, // FRECPX S0, S1
0x1E21C020u // FSQRT S0, S1
};
}
private static uint[] _F_Abs_Neg_Recpx_Sqrt_S_D_()
{
return new uint[]
{
0x1E60C020u, // FABS D0, D1
0x1E614020u, // FNEG D0, D1
0x5EE1F820u, // FRECPX D0, D1
0x1E61C020u // FSQRT D0, D1
};
}
private static uint[] _F_Abs_Neg_Sqrt_V_2S_4S_()
{
return new uint[]
{
0x0EA0F800u, // FABS V0.2S, V0.2S
0x2EA0F800u, // FNEG V0.2S, V0.2S
0x2EA1F800u // FSQRT V0.2S, V0.2S
};
}
private static uint[] _F_Abs_Neg_Sqrt_V_2D_()
{
return new uint[]
{
0x4EE0F800u, // FABS V0.2D, V0.2D
0x6EE0F800u, // FNEG V0.2D, V0.2D
0x6EE1F800u // FSQRT V0.2D, V0.2D
};
}
private static uint[] _F_Add_P_S_2SS_()
{
return new uint[]
{
0x7E30D820u // FADDP S0, V1.2S
};
}
private static uint[] _F_Add_P_S_2DD_()
{
return new uint[]
{
0x7E70D820u // FADDP D0, V1.2D
};
}
private static uint[] _F_Cm_EqGeGtLeLt_S_S_()
{
return new uint[]
{
0x5EA0D820u, // FCMEQ S0, S1, #0.0
0x7EA0C820u, // FCMGE S0, S1, #0.0
0x5EA0C820u, // FCMGT S0, S1, #0.0
0x7EA0D820u, // FCMLE S0, S1, #0.0
0x5EA0E820u // FCMLT S0, S1, #0.0
};
}
private static uint[] _F_Cm_EqGeGtLeLt_S_D_()
{
return new uint[]
{
0x5EE0D820u, // FCMEQ D0, D1, #0.0
0x7EE0C820u, // FCMGE D0, D1, #0.0
0x5EE0C820u, // FCMGT D0, D1, #0.0
0x7EE0D820u, // FCMLE D0, D1, #0.0
0x5EE0E820u // FCMLT D0, D1, #0.0
};
}
private static uint[] _F_Cm_EqGeGtLeLt_V_2S_4S_()
{
return new uint[]
{
0x0EA0D800u, // FCMEQ V0.2S, V0.2S, #0.0
0x2EA0C800u, // FCMGE V0.2S, V0.2S, #0.0
0x0EA0C800u, // FCMGT V0.2S, V0.2S, #0.0
0x2EA0D800u, // FCMLE V0.2S, V0.2S, #0.0
0x0EA0E800u // FCMLT V0.2S, V0.2S, #0.0
};
}
private static uint[] _F_Cm_EqGeGtLeLt_V_2D_()
{
return new uint[]
{
0x4EE0D800u, // FCMEQ V0.2D, V0.2D, #0.0
0x6EE0C800u, // FCMGE V0.2D, V0.2D, #0.0
0x4EE0C800u, // FCMGT V0.2D, V0.2D, #0.0
0x6EE0D800u, // FCMLE V0.2D, V0.2D, #0.0
0x4EE0E800u // FCMLT V0.2D, V0.2D, #0.0
};
}
private static uint[] _F_Cmp_Cmpe_S_S_() private static uint[] _F_Cmp_Cmpe_S_S_()
{ {
return new uint[] return new uint[]
@ -366,45 +472,39 @@ namespace Ryujinx.Tests.Cpu
}; };
} }
private static uint[] _F_Abs_Neg_Recpx_Sqrt_S_S_() private static uint[] _F_Recpe_Rsqrte_S_S_()
{ {
return new uint[] return new uint[]
{ {
0x1E20C020u, // FABS S0, S1 0x5EA1D820u, // FRECPE S0, S1
0x1E214020u, // FNEG S0, S1 0x7EA1D820u // FRSQRTE S0, S1
0x5EA1F820u, // FRECPX S0, S1
0x1E21C020u // FSQRT S0, S1
}; };
} }
private static uint[] _F_Abs_Neg_Recpx_Sqrt_S_D_() private static uint[] _F_Recpe_Rsqrte_S_D_()
{ {
return new uint[] return new uint[]
{ {
0x1E60C020u, // FABS D0, D1 0x5EE1D820u, // FRECPE D0, D1
0x1E614020u, // FNEG D0, D1 0x7EE1D820u // FRSQRTE D0, D1
0x5EE1F820u, // FRECPX D0, D1
0x1E61C020u // FSQRT D0, D1
}; };
} }
private static uint[] _F_Abs_Neg_Sqrt_V_2S_4S_() private static uint[] _F_Recpe_Rsqrte_V_2S_4S_()
{ {
return new uint[] return new uint[]
{ {
0x0EA0F800u, // FABS V0.2S, V0.2S 0x0EA1D800u, // FRECPE V0.2S, V0.2S
0x2EA0F800u, // FNEG V0.2S, V0.2S 0x2EA1D800u // FRSQRTE V0.2S, V0.2S
0x2EA1F800u // FSQRT V0.2S, V0.2S
}; };
} }
private static uint[] _F_Abs_Neg_Sqrt_V_2D_() private static uint[] _F_Recpe_Rsqrte_V_2D_()
{ {
return new uint[] return new uint[]
{ {
0x4EE0F800u, // FABS V0.2D, V0.2D 0x4EE1D800u, // FRECPE V0.2D, V0.2D
0x6EE0F800u, // FNEG V0.2D, V0.2D 0x6EE1D800u // FRSQRTE V0.2D, V0.2D
0x6EE1F800u // FSQRT V0.2D, V0.2D
}; };
} }
@ -963,6 +1063,202 @@ namespace Ryujinx.Tests.Cpu
CompareAgainstUnicorn(); CompareAgainstUnicorn();
} }
[Test, Pairwise] [Explicit]
public void F_Abs_Neg_Recpx_Sqrt_S_S([ValueSource("_F_Abs_Neg_Recpx_Sqrt_S_S_")] uint opcodes,
[ValueSource("_1S_F_")] ulong a)
{
ulong z = TestContext.CurrentContext.Random.NextULong();
Vector128<float> v0 = MakeVectorE0E1(z, z);
Vector128<float> v1 = MakeVectorE0(a);
int rnd = (int)TestContext.CurrentContext.Random.NextUInt();
int fpcr = rnd & (1 << (int)Fpcr.Fz);
fpcr |= rnd & (1 << (int)Fpcr.Dn);
SingleOpcode(opcodes, v0: v0, v1: v1, fpcr: fpcr);
CompareAgainstUnicorn(fpsrMask: Fpsr.Ioc | Fpsr.Idc);
}
[Test, Pairwise] [Explicit]
public void F_Abs_Neg_Recpx_Sqrt_S_D([ValueSource("_F_Abs_Neg_Recpx_Sqrt_S_D_")] uint opcodes,
[ValueSource("_1D_F_")] ulong a)
{
ulong z = TestContext.CurrentContext.Random.NextULong();
Vector128<float> v0 = MakeVectorE1(z);
Vector128<float> v1 = MakeVectorE0(a);
int rnd = (int)TestContext.CurrentContext.Random.NextUInt();
int fpcr = rnd & (1 << (int)Fpcr.Fz);
fpcr |= rnd & (1 << (int)Fpcr.Dn);
SingleOpcode(opcodes, v0: v0, v1: v1, fpcr: fpcr);
CompareAgainstUnicorn(fpsrMask: Fpsr.Ioc | Fpsr.Idc);
}
[Test, Pairwise] [Explicit]
public void F_Abs_Neg_Sqrt_V_2S_4S([ValueSource("_F_Abs_Neg_Sqrt_V_2S_4S_")] uint opcodes,
[Values(0u)] uint rd,
[Values(1u, 0u)] uint rn,
[ValueSource("_2S_F_")] ulong z,
[ValueSource("_2S_F_")] ulong a,
[Values(0b0u, 0b1u)] uint q) // <2S, 4S>
{
opcodes |= ((rn & 31) << 5) | ((rd & 31) << 0);
opcodes |= ((q & 1) << 30);
Vector128<float> v0 = MakeVectorE0E1(z, z);
Vector128<float> v1 = MakeVectorE0E1(a, a * q);
int rnd = (int)TestContext.CurrentContext.Random.NextUInt();
int fpcr = rnd & (1 << (int)Fpcr.Fz);
fpcr |= rnd & (1 << (int)Fpcr.Dn);
SingleOpcode(opcodes, v0: v0, v1: v1, fpcr: fpcr);
CompareAgainstUnicorn(fpsrMask: Fpsr.Ioc | Fpsr.Idc);
}
[Test, Pairwise] [Explicit]
public void F_Abs_Neg_Sqrt_V_2D([ValueSource("_F_Abs_Neg_Sqrt_V_2D_")] uint opcodes,
[Values(0u)] uint rd,
[Values(1u, 0u)] uint rn,
[ValueSource("_1D_F_")] ulong z,
[ValueSource("_1D_F_")] ulong a)
{
opcodes |= ((rn & 31) << 5) | ((rd & 31) << 0);
Vector128<float> v0 = MakeVectorE0E1(z, z);
Vector128<float> v1 = MakeVectorE0E1(a, a);
int rnd = (int)TestContext.CurrentContext.Random.NextUInt();
int fpcr = rnd & (1 << (int)Fpcr.Fz);
fpcr |= rnd & (1 << (int)Fpcr.Dn);
SingleOpcode(opcodes, v0: v0, v1: v1, fpcr: fpcr);
CompareAgainstUnicorn(fpsrMask: Fpsr.Ioc | Fpsr.Idc);
}
[Test, Pairwise] [Explicit]
public void F_Add_P_S_2SS([ValueSource("_F_Add_P_S_2SS_")] uint opcodes,
[ValueSource("_2S_F_")] ulong a)
{
ulong z = TestContext.CurrentContext.Random.NextULong();
Vector128<float> v0 = MakeVectorE0E1(z, z);
Vector128<float> v1 = MakeVectorE0(a);
int rnd = (int)TestContext.CurrentContext.Random.NextUInt();
int fpcr = rnd & (1 << (int)Fpcr.Fz);
fpcr |= rnd & (1 << (int)Fpcr.Dn);
SingleOpcode(opcodes, v0: v0, v1: v1, fpcr: fpcr);
CompareAgainstUnicorn(fpsrMask: Fpsr.Ioc | Fpsr.Idc);
}
[Test, Pairwise] [Explicit]
public void F_Add_P_S_2DD([ValueSource("_F_Add_P_S_2DD_")] uint opcodes,
[ValueSource("_1D_F_")] ulong a)
{
ulong z = TestContext.CurrentContext.Random.NextULong();
Vector128<float> v0 = MakeVectorE1(z);
Vector128<float> v1 = MakeVectorE0E1(a, a);
int rnd = (int)TestContext.CurrentContext.Random.NextUInt();
int fpcr = rnd & (1 << (int)Fpcr.Fz);
fpcr |= rnd & (1 << (int)Fpcr.Dn);
SingleOpcode(opcodes, v0: v0, v1: v1, fpcr: fpcr);
CompareAgainstUnicorn(fpsrMask: Fpsr.Ioc | Fpsr.Idc);
}
[Test, Pairwise] [Explicit]
public void F_Cm_EqGeGtLeLt_S_S([ValueSource("_F_Cm_EqGeGtLeLt_S_S_")] uint opcodes,
[ValueSource("_1S_F_")] ulong a)
{
ulong z = TestContext.CurrentContext.Random.NextULong();
Vector128<float> v0 = MakeVectorE0E1(z, z);
Vector128<float> v1 = MakeVectorE0(a);
int rnd = (int)TestContext.CurrentContext.Random.NextUInt();
int fpcr = rnd & (1 << (int)Fpcr.Fz);
SingleOpcode(opcodes, v0: v0, v1: v1, fpcr: fpcr);
CompareAgainstUnicorn(fpsrMask: Fpsr.Ioc | Fpsr.Idc);
}
[Test, Pairwise] [Explicit]
public void F_Cm_EqGeGtLeLt_S_D([ValueSource("_F_Cm_EqGeGtLeLt_S_D_")] uint opcodes,
[ValueSource("_1D_F_")] ulong a)
{
ulong z = TestContext.CurrentContext.Random.NextULong();
Vector128<float> v0 = MakeVectorE1(z);
Vector128<float> v1 = MakeVectorE0(a);
int rnd = (int)TestContext.CurrentContext.Random.NextUInt();
int fpcr = rnd & (1 << (int)Fpcr.Fz);
SingleOpcode(opcodes, v0: v0, v1: v1, fpcr: fpcr);
CompareAgainstUnicorn(fpsrMask: Fpsr.Ioc | Fpsr.Idc);
}
[Test, Pairwise] [Explicit]
public void F_Cm_EqGeGtLeLt_V_2S_4S([ValueSource("_F_Cm_EqGeGtLeLt_V_2S_4S_")] uint opcodes,
[Values(0u)] uint rd,
[Values(1u, 0u)] uint rn,
[ValueSource("_2S_F_")] ulong z,
[ValueSource("_2S_F_")] ulong a,
[Values(0b0u, 0b1u)] uint q) // <2S, 4S>
{
opcodes |= ((rn & 31) << 5) | ((rd & 31) << 0);
opcodes |= ((q & 1) << 30);
Vector128<float> v0 = MakeVectorE0E1(z, z);
Vector128<float> v1 = MakeVectorE0E1(a, a * q);
int rnd = (int)TestContext.CurrentContext.Random.NextUInt();
int fpcr = rnd & (1 << (int)Fpcr.Fz);
SingleOpcode(opcodes, v0: v0, v1: v1, fpcr: fpcr);
CompareAgainstUnicorn(fpsrMask: Fpsr.Ioc | Fpsr.Idc);
}
[Test, Pairwise] [Explicit]
public void F_Cm_EqGeGtLeLt_V_2D([ValueSource("_F_Cm_EqGeGtLeLt_V_2D_")] uint opcodes,
[Values(0u)] uint rd,
[Values(1u, 0u)] uint rn,
[ValueSource("_1D_F_")] ulong z,
[ValueSource("_1D_F_")] ulong a)
{
opcodes |= ((rn & 31) << 5) | ((rd & 31) << 0);
Vector128<float> v0 = MakeVectorE0E1(z, z);
Vector128<float> v1 = MakeVectorE0E1(a, a);
int rnd = (int)TestContext.CurrentContext.Random.NextUInt();
int fpcr = rnd & (1 << (int)Fpcr.Fz);
SingleOpcode(opcodes, v0: v0, v1: v1, fpcr: fpcr);
CompareAgainstUnicorn(fpsrMask: Fpsr.Ioc | Fpsr.Idc);
}
[Test, Pairwise] [Explicit] [Test, Pairwise] [Explicit]
public void F_Cmp_Cmpe_S_S([ValueSource("_F_Cmp_Cmpe_S_S_")] uint opcodes, public void F_Cmp_Cmpe_S_S([ValueSource("_F_Cmp_Cmpe_S_S_")] uint opcodes,
[ValueSource("_1S_F_")] ulong a) [ValueSource("_1S_F_")] ulong a)
@ -1089,13 +1385,13 @@ namespace Ryujinx.Tests.Cpu
[Values(1u, 0u)] uint rn, [Values(1u, 0u)] uint rn,
[ValueSource("_4H_F_")] ulong z, [ValueSource("_4H_F_")] ulong z,
[ValueSource("_4H_F_")] ulong a, [ValueSource("_4H_F_")] ulong a,
[Values(0b0u, 0b1u)] uint q, // <4H, 8H> [Values(0b0u, 0b1u)] uint q, // <4H4S, 8H4S>
[Values(RMode.Rn)] RMode rMode) [Values(RMode.Rn)] RMode rMode)
{ {
opcodes |= ((rn & 31) << 5) | ((rd & 31) << 0); opcodes |= ((rn & 31) << 5) | ((rd & 31) << 0);
opcodes |= ((q & 1) << 30); opcodes |= ((q & 1) << 30);
Vector128<float> v0 = MakeVectorE0E1(q == 0u ? z : 0ul, q == 1u ? z : 0ul); Vector128<float> v0 = MakeVectorE0E1(z, z);
Vector128<float> v1 = MakeVectorE0E1(q == 0u ? a : 0ul, q == 1u ? a : 0ul); Vector128<float> v1 = MakeVectorE0E1(q == 0u ? a : 0ul, q == 1u ? a : 0ul);
int rnd = (int)TestContext.CurrentContext.Random.NextUInt(); int rnd = (int)TestContext.CurrentContext.Random.NextUInt();
@ -1116,12 +1412,12 @@ namespace Ryujinx.Tests.Cpu
[Values(1u, 0u)] uint rn, [Values(1u, 0u)] uint rn,
[ValueSource("_2S_F_")] ulong z, [ValueSource("_2S_F_")] ulong z,
[ValueSource("_2S_F_")] ulong a, [ValueSource("_2S_F_")] ulong a,
[Values(0b0u, 0b1u)] uint q) // <2S, 4S> [Values(0b0u, 0b1u)] uint q) // <2S2D, 4S2D>
{ {
opcodes |= ((rn & 31) << 5) | ((rd & 31) << 0); opcodes |= ((rn & 31) << 5) | ((rd & 31) << 0);
opcodes |= ((q & 1) << 30); opcodes |= ((q & 1) << 30);
Vector128<float> v0 = MakeVectorE0E1(q == 0u ? z : 0ul, q == 1u ? z : 0ul); Vector128<float> v0 = MakeVectorE0E1(z, z);
Vector128<float> v1 = MakeVectorE0E1(q == 0u ? a : 0ul, q == 1u ? a : 0ul); Vector128<float> v1 = MakeVectorE0E1(q == 0u ? a : 0ul, q == 1u ? a : 0ul);
SingleOpcode(opcodes, v0: v0, v1: v1); SingleOpcode(opcodes, v0: v0, v1: v1);
@ -1135,7 +1431,7 @@ namespace Ryujinx.Tests.Cpu
[Values(1u, 0u)] uint rn, [Values(1u, 0u)] uint rn,
[ValueSource("_2S_F_")] ulong z, [ValueSource("_2S_F_")] ulong z,
[ValueSource("_2S_F_")] ulong a, [ValueSource("_2S_F_")] ulong a,
[Values(0b0u, 0b1u)] uint q, // <4H, 8H> [Values(0b0u, 0b1u)] uint q, // <4S4H, 4S8H>
[Values(RMode.Rn)] RMode rMode) [Values(RMode.Rn)] RMode rMode)
{ {
opcodes |= ((rn & 31) << 5) | ((rd & 31) << 0); opcodes |= ((rn & 31) << 5) | ((rd & 31) << 0);
@ -1162,7 +1458,7 @@ namespace Ryujinx.Tests.Cpu
[Values(1u, 0u)] uint rn, [Values(1u, 0u)] uint rn,
[ValueSource("_1D_F_")] ulong z, [ValueSource("_1D_F_")] ulong z,
[ValueSource("_1D_F_")] ulong a, [ValueSource("_1D_F_")] ulong a,
[Values(0b0u, 0b1u)] uint q) // <2S, 4S> [Values(0b0u, 0b1u)] uint q) // <2D2S, 2D4S>
{ {
opcodes |= ((rn & 31) << 5) | ((rd & 31) << 0); opcodes |= ((rn & 31) << 5) | ((rd & 31) << 0);
opcodes |= ((q & 1) << 30); opcodes |= ((q & 1) << 30);
@ -1176,48 +1472,53 @@ namespace Ryujinx.Tests.Cpu
} }
[Test, Pairwise] [Explicit] [Test, Pairwise] [Explicit]
public void F_Abs_Neg_Recpx_Sqrt_S_S([ValueSource("_F_Abs_Neg_Recpx_Sqrt_S_S_")] uint opcodes, public void F_Recpe_Rsqrte_S_S([ValueSource("_F_Recpe_Rsqrte_S_S_")] uint opcodes,
[ValueSource("_1S_F_")] ulong a) [ValueSource("_1S_F_")] ulong a,
[Values(RMode.Rn)] RMode rMode)
{ {
ulong z = TestContext.CurrentContext.Random.NextULong(); ulong z = TestContext.CurrentContext.Random.NextULong();
Vector128<float> v0 = MakeVectorE0E1(z, z); Vector128<float> v0 = MakeVectorE0E1(z, z);
Vector128<float> v1 = MakeVectorE0E1(a, z); Vector128<float> v1 = MakeVectorE0(a);
int rnd = (int)TestContext.CurrentContext.Random.NextUInt(); int rnd = (int)TestContext.CurrentContext.Random.NextUInt();
int fpcr = rnd & (1 << (int)Fpcr.Fz); int fpcr = (int)rMode << (int)Fpcr.RMode;
fpcr |= rnd & (1 << (int)Fpcr.Fz);
fpcr |= rnd & (1 << (int)Fpcr.Dn); fpcr |= rnd & (1 << (int)Fpcr.Dn);
SingleOpcode(opcodes, v0: v0, v1: v1, fpcr: fpcr); SingleOpcode(opcodes, v0: v0, v1: v1, fpcr: fpcr);
CompareAgainstUnicorn(fpsrMask: Fpsr.Ioc | Fpsr.Idc); CompareAgainstUnicorn(fpsrMask: Fpsr.Ioc | Fpsr.Dzc | Fpsr.Ofc | Fpsr.Ufc | Fpsr.Ixc | Fpsr.Idc);
} }
[Test, Pairwise] [Explicit] [Test, Pairwise] [Explicit]
public void F_Abs_Neg_Recpx_Sqrt_S_D([ValueSource("_F_Abs_Neg_Recpx_Sqrt_S_D_")] uint opcodes, public void F_Recpe_Rsqrte_S_D([ValueSource("_F_Recpe_Rsqrte_S_D_")] uint opcodes,
[ValueSource("_1D_F_")] ulong a) [ValueSource("_1D_F_")] ulong a,
[Values(RMode.Rn)] RMode rMode)
{ {
ulong z = TestContext.CurrentContext.Random.NextULong(); ulong z = TestContext.CurrentContext.Random.NextULong();
Vector128<float> v0 = MakeVectorE1(z); Vector128<float> v0 = MakeVectorE1(z);
Vector128<float> v1 = MakeVectorE0E1(a, z); Vector128<float> v1 = MakeVectorE0(a);
int rnd = (int)TestContext.CurrentContext.Random.NextUInt(); int rnd = (int)TestContext.CurrentContext.Random.NextUInt();
int fpcr = rnd & (1 << (int)Fpcr.Fz); int fpcr = (int)rMode << (int)Fpcr.RMode;
fpcr |= rnd & (1 << (int)Fpcr.Fz);
fpcr |= rnd & (1 << (int)Fpcr.Dn); fpcr |= rnd & (1 << (int)Fpcr.Dn);
SingleOpcode(opcodes, v0: v0, v1: v1, fpcr: fpcr); SingleOpcode(opcodes, v0: v0, v1: v1, fpcr: fpcr);
CompareAgainstUnicorn(fpsrMask: Fpsr.Ioc | Fpsr.Idc); CompareAgainstUnicorn(fpsrMask: Fpsr.Ioc | Fpsr.Dzc | Fpsr.Ofc | Fpsr.Ufc | Fpsr.Ixc | Fpsr.Idc);
} }
[Test, Pairwise] [Explicit] [Test, Pairwise] [Explicit]
public void F_Abs_Neg_Sqrt_V_2S_4S([ValueSource("_F_Abs_Neg_Sqrt_V_2S_4S_")] uint opcodes, public void F_Recpe_Rsqrte_V_2S_4S([ValueSource("_F_Recpe_Rsqrte_V_2S_4S_")] uint opcodes,
[Values(0u)] uint rd, [Values(0u)] uint rd,
[Values(1u, 0u)] uint rn, [Values(1u, 0u)] uint rn,
[ValueSource("_2S_F_")] ulong z, [ValueSource("_2S_F_")] ulong z,
[ValueSource("_2S_F_")] ulong a, [ValueSource("_2S_F_")] ulong a,
[Values(0b0u, 0b1u)] uint q) // <2S, 4S> [Values(0b0u, 0b1u)] uint q, // <2S, 4S>
[Values(RMode.Rn)] RMode rMode)
{ {
opcodes |= ((rn & 31) << 5) | ((rd & 31) << 0); opcodes |= ((rn & 31) << 5) | ((rd & 31) << 0);
opcodes |= ((q & 1) << 30); opcodes |= ((q & 1) << 30);
@ -1227,20 +1528,22 @@ namespace Ryujinx.Tests.Cpu
int rnd = (int)TestContext.CurrentContext.Random.NextUInt(); int rnd = (int)TestContext.CurrentContext.Random.NextUInt();
int fpcr = rnd & (1 << (int)Fpcr.Fz); int fpcr = (int)rMode << (int)Fpcr.RMode;
fpcr |= rnd & (1 << (int)Fpcr.Fz);
fpcr |= rnd & (1 << (int)Fpcr.Dn); fpcr |= rnd & (1 << (int)Fpcr.Dn);
SingleOpcode(opcodes, v0: v0, v1: v1, fpcr: fpcr); SingleOpcode(opcodes, v0: v0, v1: v1, fpcr: fpcr);
CompareAgainstUnicorn(fpsrMask: Fpsr.Ioc | Fpsr.Idc); CompareAgainstUnicorn(fpsrMask: Fpsr.Ioc | Fpsr.Dzc | Fpsr.Ofc | Fpsr.Ufc | Fpsr.Ixc | Fpsr.Idc);
} }
[Test, Pairwise] [Explicit] [Test, Pairwise] [Explicit]
public void F_Abs_Neg_Sqrt_V_2D([ValueSource("_F_Abs_Neg_Sqrt_V_2D_")] uint opcodes, public void F_Recpe_Rsqrte_V_2D([ValueSource("_F_Recpe_Rsqrte_V_2D_")] uint opcodes,
[Values(0u)] uint rd, [Values(0u)] uint rd,
[Values(1u, 0u)] uint rn, [Values(1u, 0u)] uint rn,
[ValueSource("_1D_F_")] ulong z, [ValueSource("_1D_F_")] ulong z,
[ValueSource("_1D_F_")] ulong a) [ValueSource("_1D_F_")] ulong a,
[Values(RMode.Rn)] RMode rMode)
{ {
opcodes |= ((rn & 31) << 5) | ((rd & 31) << 0); opcodes |= ((rn & 31) << 5) | ((rd & 31) << 0);
@ -1249,12 +1552,13 @@ namespace Ryujinx.Tests.Cpu
int rnd = (int)TestContext.CurrentContext.Random.NextUInt(); int rnd = (int)TestContext.CurrentContext.Random.NextUInt();
int fpcr = rnd & (1 << (int)Fpcr.Fz); int fpcr = (int)rMode << (int)Fpcr.RMode;
fpcr |= rnd & (1 << (int)Fpcr.Fz);
fpcr |= rnd & (1 << (int)Fpcr.Dn); fpcr |= rnd & (1 << (int)Fpcr.Dn);
SingleOpcode(opcodes, v0: v0, v1: v1, fpcr: fpcr); SingleOpcode(opcodes, v0: v0, v1: v1, fpcr: fpcr);
CompareAgainstUnicorn(fpsrMask: Fpsr.Ioc | Fpsr.Idc); CompareAgainstUnicorn(fpsrMask: Fpsr.Ioc | Fpsr.Dzc | Fpsr.Ofc | Fpsr.Ufc | Fpsr.Ixc | Fpsr.Idc);
} }
[Test, Pairwise, Description("NEG <V><d>, <V><n>")] [Test, Pairwise, Description("NEG <V><d>, <V><n>")]
@ -1662,6 +1966,27 @@ namespace Ryujinx.Tests.Cpu
CompareAgainstUnicorn(); CompareAgainstUnicorn();
} }
[Test, Pairwise, Description("SHLL{2} <Vd>.<Ta>, <Vn>.<Tb>, #<shift>")]
public void Shll_V([Values(0u)] uint rd,
[Values(1u, 0u)] uint rn,
[ValueSource("_8B4H2S_")] [Random(RndCnt)] ulong z,
[ValueSource("_8B4H2S_")] [Random(RndCnt)] ulong a,
[Values(0b00u, 0b01u, 0b10u)] uint size, // <shift: 8, 16, 32>
[Values(0b0u, 0b1u)] uint q)
{
uint opcode = 0x2E213800; // SHLL V0.8H, V0.8B, #8
opcode |= ((rn & 31) << 5) | ((rd & 31) << 0);
opcode |= ((size & 3) << 22);
opcode |= ((q & 1) << 30);
Vector128<float> v0 = MakeVectorE0E1(z, z);
Vector128<float> v1 = MakeVectorE0E1(q == 0u ? a : 0ul, q == 1u ? a : 0ul);
SingleOpcode(opcode, v0: v0, v1: v1);
CompareAgainstUnicorn();
}
[Test, Pairwise, Description("SQABS <V><d>, <V><n>")] [Test, Pairwise, Description("SQABS <V><d>, <V><n>")]
public void Sqabs_S_B_H_S_D([Values(0u)] uint rd, public void Sqabs_S_B_H_S_D([Values(0u)] uint rd,
[Values(1u, 0u)] uint rn, [Values(1u, 0u)] uint rn,

168
Ryujinx.Tests/Cpu/CpuTestSimdReg.cs
View file

@ -206,6 +206,7 @@ namespace Ryujinx.Tests.Cpu
{ {
return new uint[] return new uint[]
{ {
0x7EA2D420u, // FABD S0, S1, S2
0x1E222820u, // FADD S0, S1, S2 0x1E222820u, // FADD S0, S1, S2
0x1E221820u, // FDIV S0, S1, S2 0x1E221820u, // FDIV S0, S1, S2
0x1E220820u, // FMUL S0, S1, S2 0x1E220820u, // FMUL S0, S1, S2
@ -218,6 +219,7 @@ namespace Ryujinx.Tests.Cpu
{ {
return new uint[] return new uint[]
{ {
0x7EE2D420u, // FABD D0, D1, D2
0x1E622820u, // FADD D0, D1, D2 0x1E622820u, // FADD D0, D1, D2
0x1E621820u, // FDIV D0, D1, D2 0x1E621820u, // FDIV D0, D1, D2
0x1E620820u, // FMUL D0, D1, D2 0x1E620820u, // FMUL D0, D1, D2
@ -226,11 +228,13 @@ namespace Ryujinx.Tests.Cpu
}; };
} }
private static uint[] _F_Add_Div_Mul_Mulx_Sub_V_2S_4S_() private static uint[] _F_Add_Div_Mul_Mulx_Sub_P_V_2S_4S_()
{ {
return new uint[] return new uint[]
{ {
0x2EA0D400u, // FABD V0.2S, V0.2S, V0.2S
0x0E20D400u, // FADD V0.2S, V0.2S, V0.2S 0x0E20D400u, // FADD V0.2S, V0.2S, V0.2S
0x2E20D400u, // FADDP V0.2S, V0.2S, V0.2S
0x2E20FC00u, // FDIV V0.2S, V0.2S, V0.2S 0x2E20FC00u, // FDIV V0.2S, V0.2S, V0.2S
0x2E20DC00u, // FMUL V0.2S, V0.2S, V0.2S 0x2E20DC00u, // FMUL V0.2S, V0.2S, V0.2S
0x0E20DC00u, // FMULX V0.2S, V0.2S, V0.2S 0x0E20DC00u, // FMULX V0.2S, V0.2S, V0.2S
@ -238,11 +242,13 @@ namespace Ryujinx.Tests.Cpu
}; };
} }
private static uint[] _F_Add_Div_Mul_Mulx_Sub_V_2D_() private static uint[] _F_Add_Div_Mul_Mulx_Sub_P_V_2D_()
{ {
return new uint[] return new uint[]
{ {
0x6EE0D400u, // FABD V0.2D, V0.2D, V0.2D
0x4E60D400u, // FADD V0.2D, V0.2D, V0.2D 0x4E60D400u, // FADD V0.2D, V0.2D, V0.2D
0x6E60D400u, // FADDP V0.2D, V0.2D, V0.2D
0x6E60FC00u, // FDIV V0.2D, V0.2D, V0.2D 0x6E60FC00u, // FDIV V0.2D, V0.2D, V0.2D
0x6E60DC00u, // FMUL V0.2D, V0.2D, V0.2D 0x6E60DC00u, // FMUL V0.2D, V0.2D, V0.2D
0x4E60DC00u, // FMULX V0.2D, V0.2D, V0.2D 0x4E60DC00u, // FMULX V0.2D, V0.2D, V0.2D
@ -250,6 +256,46 @@ namespace Ryujinx.Tests.Cpu
}; };
} }
private static uint[] _F_Cm_EqGeGt_S_S_()
{
return new uint[]
{
0x5E22E420u, // FCMEQ S0, S1, S2
0x7E22E420u, // FCMGE S0, S1, S2
0x7EA2E420u // FCMGT S0, S1, S2
};
}
private static uint[] _F_Cm_EqGeGt_S_D_()
{
return new uint[]
{
0x5E62E420u, // FCMEQ D0, D1, D2
0x7E62E420u, // FCMGE D0, D1, D2
0x7EE2E420u // FCMGT D0, D1, D2
};
}
private static uint[] _F_Cm_EqGeGt_V_2S_4S_()
{
return new uint[]
{
0x0E20E400u, // FCMEQ V0.2S, V0.2S, V0.2S
0x2E20E400u, // FCMGE V0.2S, V0.2S, V0.2S
0x2EA0E400u // FCMGT V0.2S, V0.2S, V0.2S
};
}
private static uint[] _F_Cm_EqGeGt_V_2D_()
{
return new uint[]
{
0x4E60E400u, // FCMEQ V0.2D, V0.2D, V0.2D
0x6E60E400u, // FCMGE V0.2D, V0.2D, V0.2D
0x6EE0E400u // FCMGT V0.2D, V0.2D, V0.2D
};
}
private static uint[] _F_Cmp_Cmpe_S_S_() private static uint[] _F_Cmp_Cmpe_S_S_()
{ {
return new uint[] return new uint[]
@ -1285,14 +1331,14 @@ namespace Ryujinx.Tests.Cpu
} }
[Test, Pairwise] [Explicit] [Test, Pairwise] [Explicit]
public void F_Add_Div_Mul_Mulx_Sub_V_2S_4S([ValueSource("_F_Add_Div_Mul_Mulx_Sub_V_2S_4S_")] uint opcodes, public void F_Add_Div_Mul_Mulx_Sub_P_V_2S_4S([ValueSource("_F_Add_Div_Mul_Mulx_Sub_P_V_2S_4S_")] uint opcodes,
[Values(0u)] uint rd, [Values(0u)] uint rd,
[Values(1u, 0u)] uint rn, [Values(1u, 0u)] uint rn,
[Values(2u, 0u)] uint rm, [Values(2u, 0u)] uint rm,
[ValueSource("_2S_F_")] ulong z, [ValueSource("_2S_F_")] ulong z,
[ValueSource("_2S_F_")] ulong a, [ValueSource("_2S_F_")] ulong a,
[ValueSource("_2S_F_")] ulong b, [ValueSource("_2S_F_")] ulong b,
[Values(0b0u, 0b1u)] uint q) // <2S, 4S> [Values(0b0u, 0b1u)] uint q) // <2S, 4S>
{ {
opcodes |= ((rm & 31) << 16) | ((rn & 31) << 5) | ((rd & 31) << 0); opcodes |= ((rm & 31) << 16) | ((rn & 31) << 5) | ((rd & 31) << 0);
opcodes |= ((q & 1) << 30); opcodes |= ((q & 1) << 30);
@ -1312,13 +1358,13 @@ namespace Ryujinx.Tests.Cpu
} }
[Test, Pairwise] [Explicit] [Test, Pairwise] [Explicit]
public void F_Add_Div_Mul_Mulx_Sub_V_2D([ValueSource("_F_Add_Div_Mul_Mulx_Sub_V_2D_")] uint opcodes, public void F_Add_Div_Mul_Mulx_Sub_P_V_2D([ValueSource("_F_Add_Div_Mul_Mulx_Sub_P_V_2D_")] uint opcodes,
[Values(0u)] uint rd, [Values(0u)] uint rd,
[Values(1u, 0u)] uint rn, [Values(1u, 0u)] uint rn,
[Values(2u, 0u)] uint rm, [Values(2u, 0u)] uint rm,
[ValueSource("_1D_F_")] ulong z, [ValueSource("_1D_F_")] ulong z,
[ValueSource("_1D_F_")] ulong a, [ValueSource("_1D_F_")] ulong a,
[ValueSource("_1D_F_")] ulong b) [ValueSource("_1D_F_")] ulong b)
{ {
opcodes |= ((rm & 31) << 16) | ((rn & 31) << 5) | ((rd & 31) << 0); opcodes |= ((rm & 31) << 16) | ((rn & 31) << 5) | ((rd & 31) << 0);
@ -1336,6 +1382,94 @@ namespace Ryujinx.Tests.Cpu
CompareAgainstUnicorn(fpsrMask: Fpsr.Ioc | Fpsr.Dzc | Fpsr.Idc); CompareAgainstUnicorn(fpsrMask: Fpsr.Ioc | Fpsr.Dzc | Fpsr.Idc);
} }
[Test, Pairwise] [Explicit]
public void F_Cm_EqGeGt_S_S([ValueSource("_F_Cm_EqGeGt_S_S_")] uint opcodes,
[ValueSource("_1S_F_")] ulong a,
[ValueSource("_1S_F_")] ulong b)
{
ulong z = TestContext.CurrentContext.Random.NextULong();
Vector128<float> v0 = MakeVectorE0E1(z, z);
Vector128<float> v1 = MakeVectorE0(a);
Vector128<float> v2 = MakeVectorE0(b);
int rnd = (int)TestContext.CurrentContext.Random.NextUInt();
int fpcr = rnd & (1 << (int)Fpcr.Fz);
SingleOpcode(opcodes, v0: v0, v1: v1, v2: v2, fpcr: fpcr);
CompareAgainstUnicorn(fpsrMask: Fpsr.Ioc | Fpsr.Idc);
}
[Test, Pairwise] [Explicit]
public void F_Cm_EqGeGt_S_D([ValueSource("_F_Cm_EqGeGt_S_D_")] uint opcodes,
[ValueSource("_1D_F_")] ulong a,
[ValueSource("_1D_F_")] ulong b)
{
ulong z = TestContext.CurrentContext.Random.NextULong();
Vector128<float> v0 = MakeVectorE1(z);
Vector128<float> v1 = MakeVectorE0(a);
Vector128<float> v2 = MakeVectorE0(b);
int rnd = (int)TestContext.CurrentContext.Random.NextUInt();
int fpcr = rnd & (1 << (int)Fpcr.Fz);
SingleOpcode(opcodes, v0: v0, v1: v1, v2: v2, fpcr: fpcr);
CompareAgainstUnicorn(fpsrMask: Fpsr.Ioc | Fpsr.Idc);
}
[Test, Pairwise] [Explicit]
public void F_Cm_EqGeGt_V_2S_4S([ValueSource("_F_Cm_EqGeGt_V_2S_4S_")] uint opcodes,
[Values(0u)] uint rd,
[Values(1u, 0u)] uint rn,
[Values(2u, 0u)] uint rm,
[ValueSource("_2S_F_")] ulong z,
[ValueSource("_2S_F_")] ulong a,
[ValueSource("_2S_F_")] ulong b,
[Values(0b0u, 0b1u)] uint q) // <2S, 4S>
{
opcodes |= ((rm & 31) << 16) | ((rn & 31) << 5) | ((rd & 31) << 0);
opcodes |= ((q & 1) << 30);
Vector128<float> v0 = MakeVectorE0E1(z, z);
Vector128<float> v1 = MakeVectorE0E1(a, a * q);
Vector128<float> v2 = MakeVectorE0E1(b, b * q);
int rnd = (int)TestContext.CurrentContext.Random.NextUInt();
int fpcr = rnd & (1 << (int)Fpcr.Fz);
SingleOpcode(opcodes, v0: v0, v1: v1, v2: v2, fpcr: fpcr);
CompareAgainstUnicorn(fpsrMask: Fpsr.Ioc | Fpsr.Idc);
}
[Test, Pairwise] [Explicit]
public void F_Cm_EqGeGt_V_2D([ValueSource("_F_Cm_EqGeGt_V_2D_")] uint opcodes,
[Values(0u)] uint rd,
[Values(1u, 0u)] uint rn,
[Values(2u, 0u)] uint rm,
[ValueSource("_1D_F_")] ulong z,
[ValueSource("_1D_F_")] ulong a,
[ValueSource("_1D_F_")] ulong b)
{
opcodes |= ((rm & 31) << 16) | ((rn & 31) << 5) | ((rd & 31) << 0);
Vector128<float> v0 = MakeVectorE0E1(z, z);
Vector128<float> v1 = MakeVectorE0E1(a, a);
Vector128<float> v2 = MakeVectorE0E1(b, b);
int rnd = (int)TestContext.CurrentContext.Random.NextUInt();
int fpcr = rnd & (1 << (int)Fpcr.Fz);
SingleOpcode(opcodes, v0: v0, v1: v1, v2: v2, fpcr: fpcr);
CompareAgainstUnicorn(fpsrMask: Fpsr.Ioc | Fpsr.Idc);
}
[Test, Pairwise] [Explicit] [Test, Pairwise] [Explicit]
public void F_Cmp_Cmpe_S_S([ValueSource("_F_Cmp_Cmpe_S_S_")] uint opcodes, public void F_Cmp_Cmpe_S_S([ValueSource("_F_Cmp_Cmpe_S_S_")] uint opcodes,
[ValueSource("_1S_F_")] ulong a, [ValueSource("_1S_F_")] ulong a,

96
Ryujinx.Tests/Cpu/CpuTestSimdShImm.cs
View file

@ -50,6 +50,33 @@ namespace Ryujinx.Tests.Cpu
#endregion #endregion
#region "ValueSource (Opcodes)" #region "ValueSource (Opcodes)"
private static uint[] _SU_Shll_V_8B8H_16B8H_()
{
return new uint[]
{
0x0F08A400u, // SSHLL V0.8H, V0.8B, #0
0x2F08A400u // USHLL V0.8H, V0.8B, #0
};
}
private static uint[] _SU_Shll_V_4H4S_8H4S_()
{
return new uint[]
{
0x0F10A400u, // SSHLL V0.4S, V0.4H, #0
0x2F10A400u // USHLL V0.4S, V0.4H, #0
};
}
private static uint[] _SU_Shll_V_2S2D_4S2D_()
{
return new uint[]
{
0x0F20A400u, // SSHLL V0.2D, V0.2S, #0
0x2F20A400u // USHLL V0.2D, V0.2S, #0
};
}
private static uint[] _ShrImm_S_D_() private static uint[] _ShrImm_S_D_()
{ {
return new uint[] return new uint[]
@ -344,6 +371,75 @@ namespace Ryujinx.Tests.Cpu
CompareAgainstUnicorn(); CompareAgainstUnicorn();
} }
[Test, Pairwise]
public void SU_Shll_V_8B8H_16B8H([ValueSource("_SU_Shll_V_8B8H_16B8H_")] uint opcodes,
[Values(0u)] uint rd,
[Values(1u, 0u)] uint rn,
[ValueSource("_8B_")] [Random(RndCnt)] ulong z,
[ValueSource("_8B_")] [Random(RndCnt)] ulong a,
[Range(0u, 7u)] uint shift,
[Values(0b0u, 0b1u)] uint q) // <8B8H, 16B8H>
{
uint immHb = (8 + shift) & 0x7F;
opcodes |= ((rn & 31) << 5) | ((rd & 31) << 0);
opcodes |= (immHb << 16);
opcodes |= ((q & 1) << 30);
Vector128<float> v0 = MakeVectorE0E1(z, z);
Vector128<float> v1 = MakeVectorE0E1(q == 0u ? a : 0ul, q == 1u ? a : 0ul);
SingleOpcode(opcodes, v0: v0, v1: v1);
CompareAgainstUnicorn();
}
[Test, Pairwise]
public void SU_Shll_V_4H4S_8H4S([ValueSource("_SU_Shll_V_4H4S_8H4S_")] uint opcodes,
[Values(0u)] uint rd,
[Values(1u, 0u)] uint rn,
[ValueSource("_4H_")] [Random(RndCnt)] ulong z,
[ValueSource("_4H_")] [Random(RndCnt)] ulong a,
[Range(0u, 15u)] uint shift,
[Values(0b0u, 0b1u)] uint q) // <4H4S, 8H4S>
{
uint immHb = (16 + shift) & 0x7F;
opcodes |= ((rn & 31) << 5) | ((rd & 31) << 0);
opcodes |= (immHb << 16);
opcodes |= ((q & 1) << 30);
Vector128<float> v0 = MakeVectorE0E1(z, z);
Vector128<float> v1 = MakeVectorE0E1(q == 0u ? a : 0ul, q == 1u ? a : 0ul);
SingleOpcode(opcodes, v0: v0, v1: v1);
CompareAgainstUnicorn();
}
[Test, Pairwise]
public void SU_Shll_V_2S2D_4S2D([ValueSource("_SU_Shll_V_2S2D_4S2D_")] uint opcodes,
[Values(0u)] uint rd,
[Values(1u, 0u)] uint rn,
[ValueSource("_2S_")] [Random(RndCnt)] ulong z,
[ValueSource("_2S_")] [Random(RndCnt)] ulong a,
[Range(0u, 31u)] uint shift,
[Values(0b0u, 0b1u)] uint q) // <2S2D, 4S2D>
{
uint immHb = (32 + shift) & 0x7F;
opcodes |= ((rn & 31) << 5) | ((rd & 31) << 0);
opcodes |= (immHb << 16);
opcodes |= ((q & 1) << 30);
Vector128<float> v0 = MakeVectorE0E1(z, z);
Vector128<float> v1 = MakeVectorE0E1(q == 0u ? a : 0ul, q == 1u ? a : 0ul);
SingleOpcode(opcodes, v0: v0, v1: v1);
CompareAgainstUnicorn();
}
[Test, Pairwise] [Test, Pairwise]
public void ShrImm_S_D([ValueSource("_ShrImm_S_D_")] uint opcodes, public void ShrImm_S_D([ValueSource("_ShrImm_S_D_")] uint opcodes,
[Values(0u)] uint rd, [Values(0u)] uint rd,