Ryujinx/Ryujinx.Graphics.Shader/Instructions/InstEmitFArith.cs

470 lines
16 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.InstEmitAluHelper;
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 Fadd(EmitterContext context)
{
IOpCodeFArith op = (IOpCodeFArith)context.CurrOp;
bool absoluteA = op.AbsoluteA, absoluteB, negateA, negateB;
if (op is OpCodeFArithImm32)
{
negateB = op.RawOpCode.Extract(53);
negateA = op.RawOpCode.Extract(56);
absoluteB = op.RawOpCode.Extract(57);
}
else
{
negateB = op.RawOpCode.Extract(45);
negateA = op.RawOpCode.Extract(48);
absoluteB = op.RawOpCode.Extract(49);
}
Operand srcA = context.FPAbsNeg(GetSrcA(context), absoluteA, negateA);
Operand srcB = context.FPAbsNeg(GetSrcB(context), absoluteB, negateB);
Operand dest = GetDest(context);
context.Copy(dest, context.FPSaturate(context.FPAdd(srcA, srcB), op.Saturate));
SetFPZnFlags(context, dest, op.SetCondCode);
}
public static void Ffma(EmitterContext context)
{
IOpCodeFArith op = (IOpCodeFArith)context.CurrOp;
bool negateB = op.RawOpCode.Extract(48);
bool negateC = op.RawOpCode.Extract(49);
Operand srcA = GetSrcA(context);
Operand srcB = context.FPNegate(GetSrcB(context), negateB);
Operand srcC = context.FPNegate(GetSrcC(context), negateC);
Operand dest = GetDest(context);
context.Copy(dest, context.FPSaturate(context.FPFusedMultiplyAdd(srcA, srcB, srcC), op.Saturate));
SetFPZnFlags(context, dest, op.SetCondCode);
}
public static void Fmnmx(EmitterContext context)
{
IOpCodeFArith op = (IOpCodeFArith)context.CurrOp;
bool absoluteA = op.AbsoluteA;
bool negateB = op.RawOpCode.Extract(45);
bool negateA = op.RawOpCode.Extract(48);
bool absoluteB = op.RawOpCode.Extract(49);
Operand srcA = context.FPAbsNeg(GetSrcA(context), absoluteA, negateA);
Operand srcB = context.FPAbsNeg(GetSrcB(context), absoluteB, negateB);
Operand resMin = context.FPMinimum(srcA, srcB);
Operand resMax = context.FPMaximum(srcA, srcB);
Operand pred = GetPredicate39(context);
Operand dest = GetDest(context);
context.Copy(dest, context.ConditionalSelect(pred, resMin, resMax));
SetFPZnFlags(context, dest, op.SetCondCode);
}
public static void Fmul(EmitterContext context)
{
IOpCodeFArith op = (IOpCodeFArith)context.CurrOp;
bool isImm32 = op is OpCodeFArithImm32;
bool negateB = !isImm32 && op.RawOpCode.Extract(48);
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 srcA = GetSrcA(context);
Operand srcB = context.FPNegate(GetSrcB(context), negateB);
switch (op.Scale)
{
case FPMultiplyScale.None: break;
case FPMultiplyScale.Divide2: srcA = context.FPDivide (srcA, ConstF(2)); break;
case FPMultiplyScale.Divide4: srcA = context.FPDivide (srcA, ConstF(4)); break;
case FPMultiplyScale.Divide8: srcA = context.FPDivide (srcA, ConstF(8)); break;
case FPMultiplyScale.Multiply2: srcA = context.FPMultiply(srcA, ConstF(2)); break;
case FPMultiplyScale.Multiply4: srcA = context.FPMultiply(srcA, ConstF(4)); break;
case FPMultiplyScale.Multiply8: srcA = context.FPMultiply(srcA, ConstF(8)); break;
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
default: break; //TODO: Warning.
}
Operand dest = GetDest(context);
bool saturate = isImm32 ? op.RawOpCode.Extract(55) : op.Saturate;
context.Copy(dest, context.FPSaturate(context.FPMultiply(srcA, srcB), saturate));
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
SetFPZnFlags(context, dest, op.SetCondCode);
}
public static void Fset(EmitterContext context)
{
OpCodeSet op = (OpCodeSet)context.CurrOp;
Condition cmpOp = (Condition)op.RawOpCode.Extract(48, 4);
bool negateA = op.RawOpCode.Extract(43);
bool absoluteB = op.RawOpCode.Extract(44);
bool boolFloat = op.RawOpCode.Extract(52);
bool negateB = op.RawOpCode.Extract(53);
bool absoluteA = op.RawOpCode.Extract(54);
Operand srcA = context.FPAbsNeg(GetSrcA(context), absoluteA, negateA);
Operand srcB = context.FPAbsNeg(GetSrcB(context), absoluteB, negateB);
Operand res = GetFPComparison(context, cmpOp, srcA, srcB);
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 Fsetp(EmitterContext context)
{
OpCodeSet op = (OpCodeSet)context.CurrOp;
Condition cmpOp = (Condition)op.RawOpCode.Extract(48, 4);
2019-10-13 06:02:07 +00:00
bool negateB = op.RawOpCode.Extract(6);
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
bool absoluteA = op.RawOpCode.Extract(7);
bool negateA = op.RawOpCode.Extract(43);
bool absoluteB = op.RawOpCode.Extract(44);
Operand srcA = context.FPAbsNeg(GetSrcA(context), absoluteA, negateA);
Operand srcB = context.FPAbsNeg(GetSrcB(context), absoluteB, negateB);
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 p0Res = GetFPComparison(context, cmpOp, 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 Fswzadd(EmitterContext context)
{
OpCodeAlu op = (OpCodeAlu)context.CurrOp;
int mask = op.RawOpCode.Extract(28, 8);
Operand srcA = GetSrcA(context);
Operand srcB = GetSrcB(context);
Operand dest = GetDest(context);
context.Copy(dest, context.FPSwizzleAdd(srcA, srcB, mask));
SetFPZnFlags(context, dest, op.SetCondCode);
}
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 Hadd2(EmitterContext context)
{
Hadd2Hmul2Impl(context, isAdd: true);
}
public static void Hfma2(EmitterContext context)
{
IOpCodeHfma op = (IOpCodeHfma)context.CurrOp;
Operand[] srcA = GetHfmaSrcA(context);
Operand[] srcB = GetHfmaSrcB(context);
Operand[] srcC = GetHfmaSrcC(context);
Operand[] res = new Operand[2];
for (int index = 0; index < res.Length; index++)
{
res[index] = context.FPFusedMultiplyAdd(srcA[index], srcB[index], srcC[index]);
res[index] = context.FPSaturate(res[index], op.Saturate);
}
context.Copy(GetDest(context), GetHalfPacked(context, res));
}
public static void Hmul2(EmitterContext context)
{
Hadd2Hmul2Impl(context, isAdd: false);
}
private static void Hadd2Hmul2Impl(EmitterContext context, bool isAdd)
{
OpCode op = context.CurrOp;
bool saturate = op.RawOpCode.Extract(op is IOpCodeReg ? 32 : 52);
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[] srcA = GetHalfSrcA(context, isAdd);
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[] srcB = GetHalfSrcB(context);
Operand[] res = new Operand[2];
for (int index = 0; index < res.Length; index++)
{
if (isAdd)
{
res[index] = context.FPAdd(srcA[index], srcB[index]);
}
else
{
res[index] = context.FPMultiply(srcA[index], srcB[index]);
}
res[index] = context.FPSaturate(res[index], saturate);
}
context.Copy(GetDest(context), GetHalfPacked(context, res));
}
public static void Hset2(EmitterContext context)
2019-10-13 06:02:07 +00:00
{
OpCodeSet op = (OpCodeSet)context.CurrOp;
bool isRegVariant = op is IOpCodeReg;
2019-10-13 06:02:07 +00:00
bool boolFloat = isRegVariant
? op.RawOpCode.Extract(49)
: op.RawOpCode.Extract(53);
Condition cmpOp = isRegVariant
2019-10-13 06:02:07 +00:00
? (Condition)op.RawOpCode.Extract(35, 4)
: (Condition)op.RawOpCode.Extract(49, 4);
Operand[] srcA = GetHalfSrcA(context);
Operand[] srcB = GetHalfSrcB(context);
Operand[] res = new Operand[2];
res[0] = GetFPComparison(context, cmpOp, srcA[0], srcB[0]);
res[1] = GetFPComparison(context, cmpOp, srcA[1], srcB[1]);
2019-10-13 06:02:07 +00:00
Operand pred = GetPredicate39(context);
res[0] = GetPredLogicalOp(context, op.LogicalOp, res[0], pred);
res[1] = GetPredLogicalOp(context, op.LogicalOp, res[1], pred);
if (boolFloat)
{
res[0] = context.ConditionalSelect(res[0], ConstF(1), Const(0));
res[1] = context.ConditionalSelect(res[1], ConstF(1), Const(0));
context.Copy(GetDest(context), context.PackHalf2x16(res[0], res[1]));
}
else
{
Operand low = context.BitwiseAnd(res[0], Const(0xffff));
Operand high = context.ShiftLeft (res[1], Const(16));
Operand packed = context.BitwiseOr(low, high);
context.Copy(GetDest(context), packed);
}
}
public static void Hsetp2(EmitterContext context)
{
OpCodeSet op = (OpCodeSet)context.CurrOp;
bool isRegVariant = op is IOpCodeReg;
bool hAnd = isRegVariant
? op.RawOpCode.Extract(49)
: op.RawOpCode.Extract(53);
Condition cmpOp = isRegVariant
? (Condition)op.RawOpCode.Extract(35, 4)
: (Condition)op.RawOpCode.Extract(49, 4);
Operand[] srcA = GetHalfSrcA(context);
Operand[] srcB = GetHalfSrcB(context);
2019-10-13 06:02:07 +00:00
Operand p0Res = GetFPComparison(context, cmpOp, srcA[0], srcB[0]);
Operand p1Res = GetFPComparison(context, cmpOp, srcA[1], srcB[1]);
if (hAnd)
{
p0Res = context.BitwiseAnd(p0Res, p1Res);
p1Res = context.BitwiseNot(p0Res);
}
Operand pred = GetPredicate39(context);
2019-10-13 06:02:07 +00:00
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);
}
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 Mufu(EmitterContext context)
{
IOpCodeFArith op = (IOpCodeFArith)context.CurrOp;
bool negateB = op.RawOpCode.Extract(48);
Operand res = context.FPAbsNeg(GetSrcA(context), op.AbsoluteA, negateB);
MufuOperation subOp = (MufuOperation)context.CurrOp.RawOpCode.Extract(20, 4);
switch (subOp)
{
case MufuOperation.Cosine:
res = context.FPCosine(res);
break;
case MufuOperation.Sine:
res = context.FPSine(res);
break;
case MufuOperation.ExponentB2:
res = context.FPExponentB2(res);
break;
case MufuOperation.LogarithmB2:
res = context.FPLogarithmB2(res);
break;
case MufuOperation.Reciprocal:
res = context.FPReciprocal(res);
break;
case MufuOperation.ReciprocalSquareRoot:
res = context.FPReciprocalSquareRoot(res);
break;
case MufuOperation.SquareRoot:
res = context.FPSquareRoot(res);
break;
default: /* TODO */ break;
}
context.Copy(GetDest(context), context.FPSaturate(res, op.Saturate));
}
private static Operand GetFPComparison(
EmitterContext context,
Condition cond,
Operand srcA,
Operand srcB)
{
Operand res;
if (cond == Condition.Always)
{
res = Const(IrConsts.True);
}
else if (cond == Condition.Never)
{
res = Const(IrConsts.False);
}
else if (cond == Condition.Nan || cond == Condition.Number)
{
res = context.BitwiseOr(context.IsNan(srcA), context.IsNan(srcB));
if (cond == Condition.Number)
{
res = context.BitwiseNot(res);
}
}
else
{
Instruction inst;
switch (cond & ~Condition.Nan)
{
case Condition.Less: inst = Instruction.CompareLess; break;
case Condition.Equal: inst = Instruction.CompareEqual; break;
case Condition.LessOrEqual: inst = Instruction.CompareLessOrEqual; break;
case Condition.Greater: inst = Instruction.CompareGreater; break;
case Condition.NotEqual: inst = Instruction.CompareNotEqual; break;
case Condition.GreaterOrEqual: inst = Instruction.CompareGreaterOrEqual; break;
default: throw new InvalidOperationException($"Unexpected condition \"{cond}\".");
}
res = context.Add(inst | Instruction.FP, Local(), srcA, srcB);
if ((cond & Condition.Nan) != 0)
{
res = context.BitwiseOr(res, context.IsNan(srcA));
res = context.BitwiseOr(res, context.IsNan(srcB));
}
}
return res;
}
private static void SetFPZnFlags(EmitterContext context, Operand dest, bool setCC)
{
if (setCC)
{
context.Copy(GetZF(context), context.FPCompareEqual(dest, ConstF(0)));
context.Copy(GetNF(context), context.FPCompareLess (dest, ConstF(0)));
}
}
private static Operand[] GetHfmaSrcA(EmitterContext context)
{
IOpCodeHfma op = (IOpCodeHfma)context.CurrOp;
2019-10-13 06:02:07 +00:00
return GetHalfUnpacked(context, GetSrcA(context), op.SwizzleA);
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
}
private static Operand[] GetHfmaSrcB(EmitterContext context)
{
IOpCodeHfma op = (IOpCodeHfma)context.CurrOp;
2019-10-13 06:02:07 +00:00
Operand[] operands = GetHalfUnpacked(context, GetSrcB(context), op.SwizzleB);
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 FPAbsNeg(context, operands, false, op.NegateB);
}
private static Operand[] GetHfmaSrcC(EmitterContext context)
{
IOpCodeHfma op = (IOpCodeHfma)context.CurrOp;
2019-10-13 06:02:07 +00:00
Operand[] operands = GetHalfUnpacked(context, GetSrcC(context), op.SwizzleC);
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 FPAbsNeg(context, operands, false, op.NegateC);
}
}
}