Ryujinx/Ryujinx.Graphics.Shader/Instructions/InstEmitMove.cs
gdkchan e44a43c7e1
Implement VMAD shader instruction and improve InvocationInfo and ISBERD handling (#3251)
* Implement VMAD shader instruction and improve InvocationInfo and ISBERD handling

* Shader cache version bump

* Fix typo
2022-04-08 12:42:39 +02:00

237 lines
8.3 KiB
C#

using Ryujinx.Graphics.Shader.Decoders;
using Ryujinx.Graphics.Shader.IntermediateRepresentation;
using Ryujinx.Graphics.Shader.Translation;
using static Ryujinx.Graphics.Shader.Instructions.InstEmitHelper;
using static Ryujinx.Graphics.Shader.IntermediateRepresentation.OperandHelper;
namespace Ryujinx.Graphics.Shader.Instructions
{
static partial class InstEmit
{
public static void MovR(EmitterContext context)
{
InstMovR op = context.GetOp<InstMovR>();
context.Copy(GetDest(op.Dest), GetSrcReg(context, op.SrcA));
}
public static void MovI(EmitterContext context)
{
InstMovI op = context.GetOp<InstMovI>();
context.Copy(GetDest(op.Dest), GetSrcImm(context, op.Imm20));
}
public static void MovC(EmitterContext context)
{
InstMovC op = context.GetOp<InstMovC>();
context.Copy(GetDest(op.Dest), GetSrcCbuf(context, op.CbufSlot, op.CbufOffset));
}
public static void Mov32i(EmitterContext context)
{
InstMov32i op = context.GetOp<InstMov32i>();
context.Copy(GetDest(op.Dest), GetSrcImm(context, op.Imm32));
}
public static void R2pR(EmitterContext context)
{
InstR2pR op = context.GetOp<InstR2pR>();
Operand value = GetSrcReg(context, op.SrcA);
Operand mask = GetSrcReg(context, op.SrcB);
EmitR2p(context, value, mask, op.ByteSel, op.Ccpr);
}
public static void R2pI(EmitterContext context)
{
InstR2pI op = context.GetOp<InstR2pI>();
Operand value = GetSrcReg(context, op.SrcA);
Operand mask = GetSrcImm(context, Imm20ToSInt(op.Imm20));
EmitR2p(context, value, mask, op.ByteSel, op.Ccpr);
}
public static void R2pC(EmitterContext context)
{
InstR2pC op = context.GetOp<InstR2pC>();
Operand value = GetSrcReg(context, op.SrcA);
Operand mask = GetSrcCbuf(context, op.CbufSlot, op.CbufOffset);
EmitR2p(context, value, mask, op.ByteSel, op.Ccpr);
}
public static void S2r(EmitterContext context)
{
InstS2r op = context.GetOp<InstS2r>();
Operand src;
switch (op.SReg)
{
case SReg.LaneId:
src = Attribute(AttributeConsts.LaneId);
break;
case SReg.InvocationId:
src = Attribute(AttributeConsts.InvocationId);
break;
case SReg.YDirection:
src = ConstF(1); // TODO: Use value from Y direction GPU register.
break;
case SReg.ThreadKill:
src = context.Config.Stage == ShaderStage.Fragment ? Attribute(AttributeConsts.ThreadKill) : Const(0);
break;
case SReg.InvocationInfo:
if (context.Config.Stage != ShaderStage.Compute && context.Config.Stage != ShaderStage.Fragment)
{
// Note: Lowest 8-bits seems to contain some primitive index,
// but it seems to be NVIDIA implementation specific as it's only used
// to calculate ISBE offsets, so we can just keep it as zero.
if (context.Config.Stage == ShaderStage.TessellationControl ||
context.Config.Stage == ShaderStage.TessellationEvaluation)
{
src = context.ShiftLeft(Attribute(AttributeConsts.PatchVerticesIn), Const(16));
}
else
{
src = Const(context.Config.GpuAccessor.QueryPrimitiveTopology().ToInputVertices() << 16);
}
}
else
{
src = Const(0);
}
break;
case SReg.TId:
Operand tidX = Attribute(AttributeConsts.ThreadIdX);
Operand tidY = Attribute(AttributeConsts.ThreadIdY);
Operand tidZ = Attribute(AttributeConsts.ThreadIdZ);
tidY = context.ShiftLeft(tidY, Const(16));
tidZ = context.ShiftLeft(tidZ, Const(26));
src = context.BitwiseOr(tidX, context.BitwiseOr(tidY, tidZ));
break;
case SReg.TIdX:
src = Attribute(AttributeConsts.ThreadIdX);
break;
case SReg.TIdY:
src = Attribute(AttributeConsts.ThreadIdY);
break;
case SReg.TIdZ:
src = Attribute(AttributeConsts.ThreadIdZ);
break;
case SReg.CtaIdX:
src = Attribute(AttributeConsts.CtaIdX);
break;
case SReg.CtaIdY:
src = Attribute(AttributeConsts.CtaIdY);
break;
case SReg.CtaIdZ:
src = Attribute(AttributeConsts.CtaIdZ);
break;
case SReg.EqMask:
src = Attribute(AttributeConsts.EqMask);
break;
case SReg.LtMask:
src = Attribute(AttributeConsts.LtMask);
break;
case SReg.LeMask:
src = Attribute(AttributeConsts.LeMask);
break;
case SReg.GtMask:
src = Attribute(AttributeConsts.GtMask);
break;
case SReg.GeMask:
src = Attribute(AttributeConsts.GeMask);
break;
default:
src = Const(0);
break;
}
context.Copy(GetDest(op.Dest), src);
}
public static void SelR(EmitterContext context)
{
InstSelR op = context.GetOp<InstSelR>();
Operand srcA = GetSrcReg(context, op.SrcA);
Operand srcB = GetSrcReg(context, op.SrcB);
Operand srcPred = GetPredicate(context, op.SrcPred, op.SrcPredInv);
EmitSel(context, srcA, srcB, srcPred, op.Dest);
}
public static void SelI(EmitterContext context)
{
InstSelI op = context.GetOp<InstSelI>();
Operand srcA = GetSrcReg(context, op.SrcA);
Operand srcB = GetSrcImm(context, Imm20ToSInt(op.Imm20));
Operand srcPred = GetPredicate(context, op.SrcPred, op.SrcPredInv);
EmitSel(context, srcA, srcB, srcPred, op.Dest);
}
public static void SelC(EmitterContext context)
{
InstSelC op = context.GetOp<InstSelC>();
Operand srcA = GetSrcReg(context, op.SrcA);
Operand srcB = GetSrcCbuf(context, op.CbufSlot, op.CbufOffset);
Operand srcPred = GetPredicate(context, op.SrcPred, op.SrcPredInv);
EmitSel(context, srcA, srcB, srcPred, op.Dest);
}
private static void EmitR2p(EmitterContext context, Operand value, Operand mask, ByteSel byteSel, bool ccpr)
{
Operand Test(Operand value, int bit)
{
return context.ICompareNotEqual(context.BitwiseAnd(value, Const(1 << bit)), Const(0));
}
if (ccpr)
{
// TODO: Support Register to condition code flags copy.
context.Config.GpuAccessor.Log("R2P.CC not implemented.");
}
else
{
int shift = (int)byteSel * 8;
for (int bit = 0; bit < RegisterConsts.PredsCount; bit++)
{
Operand pred = Register(bit, RegisterType.Predicate);
Operand res = context.ConditionalSelect(Test(mask, bit), Test(value, bit + shift), pred);
context.Copy(pred, res);
}
}
}
private static void EmitSel(EmitterContext context, Operand srcA, Operand srcB, Operand srcPred, int rd)
{
Operand res = context.ConditionalSelect(srcPred, srcA, srcB);
context.Copy(GetDest(rd), res);
}
}
}