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This repository has been archived on 2024-10-12.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

684 lines
22 KiB
C#
Raw Normal View History

New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-17 23:57:08 +00:00
using Ryujinx.Graphics.Shader.Decoders;
using Ryujinx.Graphics.Shader.IntermediateRepresentation;
using Ryujinx.Graphics.Shader.Translation;
using System;
using static Ryujinx.Graphics.Shader.Instructions.InstEmitHelper;
using static Ryujinx.Graphics.Shader.Instructions.InstEmitAluHelper;
using static Ryujinx.Graphics.Shader.IntermediateRepresentation.OperandHelper;
namespace Ryujinx.Graphics.Shader.Instructions
{
static partial class InstEmit
{
public static void Bfe(EmitterContext context)
{
OpCodeAlu op = (OpCodeAlu)context.CurrOp;
bool isReverse = op.RawOpCode.Extract(40);
bool isSigned = op.RawOpCode.Extract(48);
Operand srcA = GetSrcA(context);
Operand srcB = GetSrcB(context);
if (isReverse)
{
srcA = context.BitfieldReverse(srcA);
}
Operand position = context.BitwiseAnd(srcB, Const(0xff));
Operand size = context.BitfieldExtractU32(srcB, Const(8), Const(8));
Operand res = isSigned
? context.BitfieldExtractS32(srcA, position, size)
: context.BitfieldExtractU32(srcA, position, size);
context.Copy(GetDest(context), res);
// TODO: CC, X, corner cases
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-17 23:57:08 +00:00
}
public static void Iadd(EmitterContext context)
{
OpCodeAlu op = (OpCodeAlu)context.CurrOp;
bool negateA = false, negateB = false;
if (!(op is OpCodeAluImm32))
{
negateB = op.RawOpCode.Extract(48);
negateA = op.RawOpCode.Extract(49);
}
Operand srcA = context.INegate(GetSrcA(context), negateA);
Operand srcB = context.INegate(GetSrcB(context), negateB);
Operand res = context.IAdd(srcA, srcB);
bool isSubtraction = negateA || negateB;
if (op.Extended)
{
// Add carry, or subtract borrow.
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-17 23:57:08 +00:00
res = context.IAdd(res, isSubtraction
? context.BitwiseNot(GetCF(context))
: context.BitwiseAnd(GetCF(context), Const(1)));
}
SetIaddFlags(context, res, srcA, srcB, op.SetCondCode, op.Extended, isSubtraction);
context.Copy(GetDest(context), res);
}
public static void Iadd3(EmitterContext context)
{
OpCodeAlu op = (OpCodeAlu)context.CurrOp;
IntegerHalfPart partC = (IntegerHalfPart)op.RawOpCode.Extract(31, 2);
IntegerHalfPart partB = (IntegerHalfPart)op.RawOpCode.Extract(33, 2);
IntegerHalfPart partA = (IntegerHalfPart)op.RawOpCode.Extract(35, 2);
IntegerShift mode = (IntegerShift)op.RawOpCode.Extract(37, 2);
bool negateC = op.RawOpCode.Extract(49);
bool negateB = op.RawOpCode.Extract(50);
bool negateA = op.RawOpCode.Extract(51);
Operand Extend(Operand src, IntegerHalfPart part)
{
if (!(op is OpCodeAluReg) || part == IntegerHalfPart.B32)
{
return src;
}
if (part == IntegerHalfPart.H0)
{
return context.BitwiseAnd(src, Const(0xffff));
}
else if (part == IntegerHalfPart.H1)
{
return context.ShiftRightU32(src, Const(16));
}
else
{
// TODO: Warning.
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-17 23:57:08 +00:00
}
return src;
}
Operand srcA = context.INegate(Extend(GetSrcA(context), partA), negateA);
Operand srcB = context.INegate(Extend(GetSrcB(context), partB), negateB);
Operand srcC = context.INegate(Extend(GetSrcC(context), partC), negateC);
Operand res = context.IAdd(srcA, srcB);
if (op is OpCodeAluReg && mode != IntegerShift.NoShift)
{
if (mode == IntegerShift.ShiftLeft)
{
res = context.ShiftLeft(res, Const(16));
}
else if (mode == IntegerShift.ShiftRight)
{
res = context.ShiftRightU32(res, Const(16));
}
else
{
// TODO: Warning.
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-17 23:57:08 +00:00
}
}
res = context.IAdd(res, srcC);
context.Copy(GetDest(context), res);
// TODO: CC, X, corner cases
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-17 23:57:08 +00:00
}
public static void Imnmx(EmitterContext context)
{
OpCodeAlu op = (OpCodeAlu)context.CurrOp;
bool isSignedInt = op.RawOpCode.Extract(48);
Operand srcA = GetSrcA(context);
Operand srcB = GetSrcB(context);
Operand resMin = isSignedInt
? context.IMinimumS32(srcA, srcB)
: context.IMinimumU32(srcA, srcB);
Operand resMax = isSignedInt
? context.IMaximumS32(srcA, srcB)
: context.IMaximumU32(srcA, srcB);
Operand pred = GetPredicate39(context);
Operand dest = GetDest(context);
context.Copy(dest, context.ConditionalSelect(pred, resMin, resMax));
SetZnFlags(context, dest, op.SetCondCode);
// TODO: X flags.
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-17 23:57:08 +00:00
}
public static void Iscadd(EmitterContext context)
{
OpCodeAlu op = (OpCodeAlu)context.CurrOp;
bool negateA = false, negateB = false;
if (!(op is OpCodeAluImm32))
{
negateB = op.RawOpCode.Extract(48);
negateA = op.RawOpCode.Extract(49);
}
int shift = op is OpCodeAluImm32
? op.RawOpCode.Extract(53, 5)
: op.RawOpCode.Extract(39, 5);
Operand srcA = GetSrcA(context);
Operand srcB = GetSrcB(context);
srcA = context.ShiftLeft(srcA, Const(shift));
srcA = context.INegate(srcA, negateA);
srcB = context.INegate(srcB, negateB);
Operand res = context.IAdd(srcA, srcB);
context.Copy(GetDest(context), res);
// TODO: CC, X
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-17 23:57:08 +00:00
}
public static void Iset(EmitterContext context)
{
OpCodeSet op = (OpCodeSet)context.CurrOp;
bool boolFloat = op.RawOpCode.Extract(44);
bool isSigned = op.RawOpCode.Extract(48);
IntegerCondition cmpOp = (IntegerCondition)op.RawOpCode.Extract(49, 3);
Operand srcA = GetSrcA(context);
Operand srcB = GetSrcB(context);
Operand res = GetIntComparison(context, cmpOp, srcA, srcB, isSigned);
Operand pred = GetPredicate39(context);
res = GetPredLogicalOp(context, op.LogicalOp, res, pred);
Operand dest = GetDest(context);
if (boolFloat)
{
context.Copy(dest, context.ConditionalSelect(res, ConstF(1), Const(0)));
}
else
{
context.Copy(dest, res);
}
// TODO: CC, X
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-17 23:57:08 +00:00
}
public static void Isetp(EmitterContext context)
{
OpCodeSet op = (OpCodeSet)context.CurrOp;
bool isSigned = op.RawOpCode.Extract(48);
IntegerCondition cmpOp = (IntegerCondition)op.RawOpCode.Extract(49, 3);
Operand srcA = GetSrcA(context);
Operand srcB = GetSrcB(context);
Operand p0Res = GetIntComparison(context, cmpOp, srcA, srcB, isSigned);
Operand p1Res = context.BitwiseNot(p0Res);
Operand pred = GetPredicate39(context);
p0Res = GetPredLogicalOp(context, op.LogicalOp, p0Res, pred);
p1Res = GetPredLogicalOp(context, op.LogicalOp, p1Res, pred);
context.Copy(Register(op.Predicate3), p0Res);
context.Copy(Register(op.Predicate0), p1Res);
}
public static void Lop(EmitterContext context)
{
IOpCodeLop op = (IOpCodeLop)context.CurrOp;
Operand srcA = context.BitwiseNot(GetSrcA(context), op.InvertA);
Operand srcB = context.BitwiseNot(GetSrcB(context), op.InvertB);
Operand res = srcB;
switch (op.LogicalOp)
{
case LogicalOperation.And: res = context.BitwiseAnd (srcA, srcB); break;
case LogicalOperation.Or: res = context.BitwiseOr (srcA, srcB); break;
case LogicalOperation.ExclusiveOr: res = context.BitwiseExclusiveOr(srcA, srcB); break;
}
EmitLopPredWrite(context, op, res);
Operand dest = GetDest(context);
context.Copy(dest, res);
SetZnFlags(context, dest, op.SetCondCode, op.Extended);
}
public static void Lop3(EmitterContext context)
{
IOpCodeLop op = (IOpCodeLop)context.CurrOp;
Operand srcA = GetSrcA(context);
Operand srcB = GetSrcB(context);
Operand srcC = GetSrcC(context);
bool regVariant = op is OpCodeLopReg;
int truthTable = regVariant
? op.RawOpCode.Extract(28, 8)
: op.RawOpCode.Extract(48, 8);
Operand res = Lop3Expression.GetFromTruthTable(context, srcA, srcB, srcC, truthTable);
if (regVariant)
{
EmitLopPredWrite(context, op, res);
}
Operand dest = GetDest(context);
context.Copy(dest, res);
SetZnFlags(context, dest, op.SetCondCode, op.Extended);
}
public static void Psetp(EmitterContext context)
{
OpCodePsetp op = (OpCodePsetp)context.CurrOp;
bool invertA = op.RawOpCode.Extract(15);
bool invertB = op.RawOpCode.Extract(32);
Operand srcA = context.BitwiseNot(Register(op.Predicate12), invertA);
Operand srcB = context.BitwiseNot(Register(op.Predicate29), invertB);
Operand p0Res = GetPredLogicalOp(context, op.LogicalOpAB, srcA, srcB);
Operand p1Res = context.BitwiseNot(p0Res);
Operand pred = GetPredicate39(context);
p0Res = GetPredLogicalOp(context, op.LogicalOp, p0Res, pred);
p1Res = GetPredLogicalOp(context, op.LogicalOp, p1Res, pred);
context.Copy(Register(op.Predicate3), p0Res);
context.Copy(Register(op.Predicate0), p1Res);
}
public static void Rro(EmitterContext context)
{
// This is the range reduction operator,
// we translate it as a simple move, as it
// should be always followed by a matching
// MUFU instruction.
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-17 23:57:08 +00:00
OpCodeAlu op = (OpCodeAlu)context.CurrOp;
bool negateB = op.RawOpCode.Extract(45);
bool absoluteB = op.RawOpCode.Extract(49);
Operand srcB = GetSrcB(context);
srcB = context.FPAbsNeg(srcB, absoluteB, negateB);
context.Copy(GetDest(context), srcB);
}
public static void Shl(EmitterContext context)
{
OpCodeAlu op = (OpCodeAlu)context.CurrOp;
bool isMasked = op.RawOpCode.Extract(39);
Operand srcB = GetSrcB(context);
if (isMasked)
{
srcB = context.BitwiseAnd(srcB, Const(0x1f));
}
Operand res = context.ShiftLeft(GetSrcA(context), srcB);
if (!isMasked)
{
// Clamped shift value.
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-17 23:57:08 +00:00
Operand isLessThan32 = context.ICompareLessUnsigned(srcB, Const(32));
res = context.ConditionalSelect(isLessThan32, res, Const(0));
}
// TODO: X, CC
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-17 23:57:08 +00:00
context.Copy(GetDest(context), res);
}
public static void Shr(EmitterContext context)
{
OpCodeAlu op = (OpCodeAlu)context.CurrOp;
bool isMasked = op.RawOpCode.Extract(39);
bool isReverse = op.RawOpCode.Extract(40);
bool isSigned = op.RawOpCode.Extract(48);
Operand srcA = GetSrcA(context);
Operand srcB = GetSrcB(context);
if (isReverse)
{
srcA = context.BitfieldReverse(srcA);
}
if (isMasked)
{
srcB = context.BitwiseAnd(srcB, Const(0x1f));
}
Operand res = isSigned
? context.ShiftRightS32(srcA, srcB)
: context.ShiftRightU32(srcA, srcB);
if (!isMasked)
{
// Clamped shift value.
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-17 23:57:08 +00:00
Operand resShiftBy32;
if (isSigned)
{
resShiftBy32 = context.ShiftRightS32(srcA, Const(31));
}
else
{
resShiftBy32 = Const(0);
}
Operand isLessThan32 = context.ICompareLessUnsigned(srcB, Const(32));
res = context.ConditionalSelect(isLessThan32, res, resShiftBy32);
}
// TODO: X, CC
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-17 23:57:08 +00:00
context.Copy(GetDest(context), res);
}
public static void Xmad(EmitterContext context)
{
OpCodeAlu op = (OpCodeAlu)context.CurrOp;
bool signedA = context.CurrOp.RawOpCode.Extract(48);
bool signedB = context.CurrOp.RawOpCode.Extract(49);
bool highA = context.CurrOp.RawOpCode.Extract(53);
bool highB = false;
XmadCMode mode;
if (op is OpCodeAluReg)
{
highB = context.CurrOp.RawOpCode.Extract(35);
mode = (XmadCMode)context.CurrOp.RawOpCode.Extract(50, 3);
}
else
{
mode = (XmadCMode)context.CurrOp.RawOpCode.Extract(50, 2);
if (!(op is OpCodeAluImm))
{
highB = context.CurrOp.RawOpCode.Extract(52);
}
}
Operand srcA = GetSrcA(context);
Operand srcB = GetSrcB(context);
Operand srcC = GetSrcC(context);
// XMAD immediates are 16-bits unsigned integers.
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-17 23:57:08 +00:00
if (srcB.Type == OperandType.Constant)
{
srcB = Const(srcB.Value & 0xffff);
}
Operand Extend16To32(Operand src, bool high, bool signed)
{
if (signed && high)
{
return context.ShiftRightS32(src, Const(16));
}
else if (signed)
{
return context.BitfieldExtractS32(src, Const(0), Const(16));
}
else if (high)
{
return context.ShiftRightU32(src, Const(16));
}
else
{
return context.BitwiseAnd(src, Const(0xffff));
}
}
srcA = Extend16To32(srcA, highA, signedA);
srcB = Extend16To32(srcB, highB, signedB);
bool productShiftLeft = false;
bool merge = false;
if (!(op is OpCodeAluRegCbuf))
{
productShiftLeft = context.CurrOp.RawOpCode.Extract(36);
merge = context.CurrOp.RawOpCode.Extract(37);
}
bool extended;
if ((op is OpCodeAluReg) || (op is OpCodeAluImm))
{
extended = context.CurrOp.RawOpCode.Extract(38);
}
else
{
extended = context.CurrOp.RawOpCode.Extract(54);
}
Operand res = context.IMultiply(srcA, srcB);
if (productShiftLeft)
{
res = context.ShiftLeft(res, Const(16));
}
switch (mode)
{
case XmadCMode.Cfull: break;
case XmadCMode.Clo: srcC = Extend16To32(srcC, high: false, signed: false); break;
case XmadCMode.Chi: srcC = Extend16To32(srcC, high: true, signed: false); break;
case XmadCMode.Cbcc:
{
srcC = context.IAdd(srcC, context.ShiftLeft(GetSrcB(context), Const(16)));
break;
}
case XmadCMode.Csfu:
{
Operand signAdjustA = context.ShiftLeft(context.ShiftRightU32(srcA, Const(31)), Const(16));
Operand signAdjustB = context.ShiftLeft(context.ShiftRightU32(srcB, Const(31)), Const(16));
srcC = context.ISubtract(srcC, context.IAdd(signAdjustA, signAdjustB));
break;
}
default: /* TODO: Warning */ break;
}
Operand product = res;
if (extended)
{
// Add with carry.
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-17 23:57:08 +00:00
res = context.IAdd(res, context.BitwiseAnd(GetCF(context), Const(1)));
}
else
{
// Add (no carry in).
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-17 23:57:08 +00:00
res = context.IAdd(res, srcC);
}
SetIaddFlags(context, res, product, srcC, op.SetCondCode, extended);
if (merge)
{
res = context.BitwiseAnd(res, Const(0xffff));
res = context.BitwiseOr(res, context.ShiftLeft(GetSrcB(context), Const(16)));
}
context.Copy(GetDest(context), res);
}
private static Operand GetIntComparison(
EmitterContext context,
IntegerCondition cond,
Operand srcA,
Operand srcB,
bool isSigned)
{
Operand res;
if (cond == IntegerCondition.Always)
{
res = Const(IrConsts.True);
}
else if (cond == IntegerCondition.Never)
{
res = Const(IrConsts.False);
}
else
{
Instruction inst;
switch (cond)
{
case IntegerCondition.Less: inst = Instruction.CompareLessU32; break;
case IntegerCondition.Equal: inst = Instruction.CompareEqual; break;
case IntegerCondition.LessOrEqual: inst = Instruction.CompareLessOrEqualU32; break;
case IntegerCondition.Greater: inst = Instruction.CompareGreaterU32; break;
case IntegerCondition.NotEqual: inst = Instruction.CompareNotEqual; break;
case IntegerCondition.GreaterOrEqual: inst = Instruction.CompareGreaterOrEqualU32; break;
default: throw new InvalidOperationException($"Unexpected condition \"{cond}\".");
}
if (isSigned)
{
switch (cond)
{
case IntegerCondition.Less: inst = Instruction.CompareLess; break;
case IntegerCondition.LessOrEqual: inst = Instruction.CompareLessOrEqual; break;
case IntegerCondition.Greater: inst = Instruction.CompareGreater; break;
case IntegerCondition.GreaterOrEqual: inst = Instruction.CompareGreaterOrEqual; break;
}
}
res = context.Add(inst, Local(), srcA, srcB);
}
return res;
}
private static void EmitLopPredWrite(EmitterContext context, IOpCodeLop op, Operand result)
{
if (op is OpCodeLop opLop && !opLop.Predicate48.IsPT)
{
Operand pRes;
if (opLop.CondOp == ConditionalOperation.False)
{
pRes = Const(IrConsts.False);
}
else if (opLop.CondOp == ConditionalOperation.True)
{
pRes = Const(IrConsts.True);
}
else if (opLop.CondOp == ConditionalOperation.Zero)
{
pRes = context.ICompareEqual(result, Const(0));
}
else /* if (opLop.CondOp == ConditionalOperation.NotZero) */
{
pRes = context.ICompareNotEqual(result, Const(0));
}
context.Copy(Register(opLop.Predicate48), pRes);
}
}
private static void SetIaddFlags(
EmitterContext context,
Operand res,
Operand srcA,
Operand srcB,
bool setCC,
bool extended,
bool isSubtraction = false)
{
if (!setCC)
{
return;
}
if (!extended || isSubtraction)
{
// C = d < a
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-17 23:57:08 +00:00
context.Copy(GetCF(context), context.ICompareLessUnsigned(res, srcA));
}
else
{
// C = (d == a && CIn) || d < a
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-17 23:57:08 +00:00
Operand tempC0 = context.ICompareEqual (res, srcA);
Operand tempC1 = context.ICompareLessUnsigned(res, srcA);
tempC0 = context.BitwiseAnd(tempC0, GetCF(context));
context.Copy(GetCF(context), context.BitwiseOr(tempC0, tempC1));
}
// V = (d ^ a) & ~(a ^ b) < 0
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-17 23:57:08 +00:00
Operand tempV0 = context.BitwiseExclusiveOr(res, srcA);
Operand tempV1 = context.BitwiseExclusiveOr(srcA, srcB);
tempV1 = context.BitwiseNot(tempV1);
Operand tempV = context.BitwiseAnd(tempV0, tempV1);
context.Copy(GetVF(context), context.ICompareLess(tempV, Const(0)));
SetZnFlags(context, res, setCC: true, extended: extended);
}
}
}