A64: Add fast path for Fcvtas_Gp/S/V, Fcvtau_Gp/S/V and Frinta_S/V in… (#3712)

* A64: Add fast path for Fcvtas_Gp/S/V, Fcvtau_Gp/S/V and Frinta_S/V instructions;

they use "Round to Nearest with Ties to Away" rounding mode not supported in x86.

All instructions involved have been tested locally in both release and debug modes, in both lowcq and highcq.

The titles Mario Strikers and Super Smash Bros. U. use these instructions intensively.

* Update Ptc.cs

* A32: Add fast path for Vcvta_RM, Vrinta_RM and Vrinta_V instructions aswell.
This commit is contained in:
LDj3SNuD 2022-10-19 02:21:33 +02:00 committed by GitHub
parent 77c4291c34
commit 5af8ce7c38
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
6 changed files with 243 additions and 46 deletions

View file

@ -164,32 +164,74 @@ namespace ARMeilleure.Instructions
public static void Fcvtas_Gp(ArmEmitterContext context)
{
EmitFcvt_s_Gp(context, (op1) => EmitRoundMathCall(context, MidpointRounding.AwayFromZero, op1));
if (Optimizations.UseSse41)
{
EmitSse41Fcvts_Gp(context, FPRoundingMode.ToNearestAway, isFixed: false);
}
else
{
EmitFcvt_s_Gp(context, (op1) => EmitRoundMathCall(context, MidpointRounding.AwayFromZero, op1));
}
}
public static void Fcvtas_S(ArmEmitterContext context)
{
EmitFcvt(context, (op1) => EmitRoundMathCall(context, MidpointRounding.AwayFromZero, op1), signed: true, scalar: true);
if (Optimizations.UseSse41)
{
EmitSse41FcvtsOpF(context, FPRoundingMode.ToNearestAway, scalar: true);
}
else
{
EmitFcvt(context, (op1) => EmitRoundMathCall(context, MidpointRounding.AwayFromZero, op1), signed: true, scalar: true);
}
}
public static void Fcvtas_V(ArmEmitterContext context)
{
EmitFcvt(context, (op1) => EmitRoundMathCall(context, MidpointRounding.AwayFromZero, op1), signed: true, scalar: false);
if (Optimizations.UseSse41)
{
EmitSse41FcvtsOpF(context, FPRoundingMode.ToNearestAway, scalar: false);
}
else
{
EmitFcvt(context, (op1) => EmitRoundMathCall(context, MidpointRounding.AwayFromZero, op1), signed: true, scalar: false);
}
}
public static void Fcvtau_Gp(ArmEmitterContext context)
{
EmitFcvt_u_Gp(context, (op1) => EmitRoundMathCall(context, MidpointRounding.AwayFromZero, op1));
if (Optimizations.UseSse41)
{
EmitSse41Fcvtu_Gp(context, FPRoundingMode.ToNearestAway, isFixed: false);
}
else
{
EmitFcvt_u_Gp(context, (op1) => EmitRoundMathCall(context, MidpointRounding.AwayFromZero, op1));
}
}
public static void Fcvtau_S(ArmEmitterContext context)
{
EmitFcvt(context, (op1) => EmitRoundMathCall(context, MidpointRounding.AwayFromZero, op1), signed: false, scalar: true);
if (Optimizations.UseSse41)
{
EmitSse41FcvtuOpF(context, FPRoundingMode.ToNearestAway, scalar: true);
}
else
{
EmitFcvt(context, (op1) => EmitRoundMathCall(context, MidpointRounding.AwayFromZero, op1), signed: false, scalar: true);
}
}
public static void Fcvtau_V(ArmEmitterContext context)
{
EmitFcvt(context, (op1) => EmitRoundMathCall(context, MidpointRounding.AwayFromZero, op1), signed: false, scalar: false);
if (Optimizations.UseSse41)
{
EmitSse41FcvtuOpF(context, FPRoundingMode.ToNearestAway, scalar: false);
}
else
{
EmitFcvt(context, (op1) => EmitRoundMathCall(context, MidpointRounding.AwayFromZero, op1), signed: false, scalar: false);
}
}
public static void Fcvtl_V(ArmEmitterContext context)
@ -1223,7 +1265,14 @@ namespace ARMeilleure.Instructions
nRes = context.AddIntrinsic(Intrinsic.X86Mulps, nRes, fpScaledMask);
}
nRes = context.AddIntrinsic(Intrinsic.X86Roundps, nRes, Const(X86GetRoundControl(roundMode)));
if (roundMode != FPRoundingMode.ToNearestAway)
{
nRes = context.AddIntrinsic(Intrinsic.X86Roundps, nRes, Const(X86GetRoundControl(roundMode)));
}
else
{
nRes = EmitSse41RoundToNearestWithTiesToAwayOpF(context, nRes, scalar);
}
Operand nInt = context.AddIntrinsic(Intrinsic.X86Cvtps2dq, nRes);
@ -1265,7 +1314,14 @@ namespace ARMeilleure.Instructions
nRes = context.AddIntrinsic(Intrinsic.X86Mulpd, nRes, fpScaledMask);
}
nRes = context.AddIntrinsic(Intrinsic.X86Roundpd, nRes, Const(X86GetRoundControl(roundMode)));
if (roundMode != FPRoundingMode.ToNearestAway)
{
nRes = context.AddIntrinsic(Intrinsic.X86Roundpd, nRes, Const(X86GetRoundControl(roundMode)));
}
else
{
nRes = EmitSse41RoundToNearestWithTiesToAwayOpF(context, nRes, scalar);
}
Operand nLong = EmitSse2CvtDoubleToInt64OpF(context, nRes, scalar);
@ -1314,7 +1370,14 @@ namespace ARMeilleure.Instructions
nRes = context.AddIntrinsic(Intrinsic.X86Mulps, nRes, fpScaledMask);
}
nRes = context.AddIntrinsic(Intrinsic.X86Roundps, nRes, Const(X86GetRoundControl(roundMode)));
if (roundMode != FPRoundingMode.ToNearestAway)
{
nRes = context.AddIntrinsic(Intrinsic.X86Roundps, nRes, Const(X86GetRoundControl(roundMode)));
}
else
{
nRes = EmitSse41RoundToNearestWithTiesToAwayOpF(context, nRes, scalar);
}
Operand zero = context.VectorZero();
@ -1369,7 +1432,14 @@ namespace ARMeilleure.Instructions
nRes = context.AddIntrinsic(Intrinsic.X86Mulpd, nRes, fpScaledMask);
}
nRes = context.AddIntrinsic(Intrinsic.X86Roundpd, nRes, Const(X86GetRoundControl(roundMode)));
if (roundMode != FPRoundingMode.ToNearestAway)
{
nRes = context.AddIntrinsic(Intrinsic.X86Roundpd, nRes, Const(X86GetRoundControl(roundMode)));
}
else
{
nRes = EmitSse41RoundToNearestWithTiesToAwayOpF(context, nRes, scalar);
}
Operand zero = context.VectorZero();
@ -1424,7 +1494,14 @@ namespace ARMeilleure.Instructions
nRes = context.AddIntrinsic(Intrinsic.X86Mulss, nRes, fpScaledMask);
}
nRes = context.AddIntrinsic(Intrinsic.X86Roundss, nRes, Const(X86GetRoundControl(roundMode)));
if (roundMode != FPRoundingMode.ToNearestAway)
{
nRes = context.AddIntrinsic(Intrinsic.X86Roundss, nRes, Const(X86GetRoundControl(roundMode)));
}
else
{
nRes = EmitSse41RoundToNearestWithTiesToAwayOpF(context, nRes, scalar: true);
}
Operand nIntOrLong = op.RegisterSize == RegisterSize.Int32
? context.AddIntrinsicInt (Intrinsic.X86Cvtss2si, nRes)
@ -1464,7 +1541,14 @@ namespace ARMeilleure.Instructions
nRes = context.AddIntrinsic(Intrinsic.X86Mulsd, nRes, fpScaledMask);
}
nRes = context.AddIntrinsic(Intrinsic.X86Roundsd, nRes, Const(X86GetRoundControl(roundMode)));
if (roundMode != FPRoundingMode.ToNearestAway)
{
nRes = context.AddIntrinsic(Intrinsic.X86Roundsd, nRes, Const(X86GetRoundControl(roundMode)));
}
else
{
nRes = EmitSse41RoundToNearestWithTiesToAwayOpF(context, nRes, scalar: true);
}
Operand nIntOrLong = op.RegisterSize == RegisterSize.Int32
? context.AddIntrinsicInt (Intrinsic.X86Cvtsd2si, nRes)
@ -1512,7 +1596,14 @@ namespace ARMeilleure.Instructions
nRes = context.AddIntrinsic(Intrinsic.X86Mulss, nRes, fpScaledMask);
}
nRes = context.AddIntrinsic(Intrinsic.X86Roundss, nRes, Const(X86GetRoundControl(roundMode)));
if (roundMode != FPRoundingMode.ToNearestAway)
{
nRes = context.AddIntrinsic(Intrinsic.X86Roundss, nRes, Const(X86GetRoundControl(roundMode)));
}
else
{
nRes = EmitSse41RoundToNearestWithTiesToAwayOpF(context, nRes, scalar: true);
}
Operand zero = context.VectorZero();
@ -1567,7 +1658,14 @@ namespace ARMeilleure.Instructions
nRes = context.AddIntrinsic(Intrinsic.X86Mulsd, nRes, fpScaledMask);
}
nRes = context.AddIntrinsic(Intrinsic.X86Roundsd, nRes, Const(X86GetRoundControl(roundMode)));
if (roundMode != FPRoundingMode.ToNearestAway)
{
nRes = context.AddIntrinsic(Intrinsic.X86Roundsd, nRes, Const(X86GetRoundControl(roundMode)));
}
else
{
nRes = EmitSse41RoundToNearestWithTiesToAwayOpF(context, nRes, scalar: true);
}
Operand zero = context.VectorZero();