Ryujinx/Ryujinx.Graphics.Shader/Decoders/OpCodeTable.cs

304 lines
17 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 16:57:08 -07:00
using Ryujinx.Graphics.Shader.Instructions;
using System;
using System.Reflection.Emit;
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 16:57:08 -07:00
namespace Ryujinx.Graphics.Shader.Decoders
{
static class OpCodeTable
{
public delegate object OpActivator(InstEmitter emitter, ulong address, long opCode);
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 16:57:08 -07:00
private const int EncodingBits = 14;
private class TableEntry
{
public InstEmitter Emitter { get; }
public Type OpCodeType { get; }
public int XBits { get; }
public OpActivator OpActivator { get; }
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 16:57:08 -07:00
public TableEntry(InstEmitter emitter, Type opCodeType, int xBits)
{
Emitter = emitter;
OpCodeType = opCodeType;
XBits = xBits;
OpActivator = CacheOpActivator(opCodeType);
}
private static OpActivator CacheOpActivator(Type type)
{
Type[] argTypes = new Type[] { typeof(InstEmitter), typeof(ulong), typeof(long) };
DynamicMethod mthd = new DynamicMethod($"Make{type.Name}", type, argTypes);
ILGenerator generator = mthd.GetILGenerator();
generator.Emit(OpCodes.Ldarg_0);
generator.Emit(OpCodes.Ldarg_1);
generator.Emit(OpCodes.Ldarg_2);
generator.Emit(OpCodes.Newobj, type.GetConstructor(argTypes));
generator.Emit(OpCodes.Ret);
return (OpActivator)mthd.CreateDelegate(typeof(OpActivator));
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 16:57:08 -07:00
}
}
private static TableEntry[] _opCodes;
static OpCodeTable()
{
_opCodes = new TableEntry[1 << EncodingBits];
#region Instructions
Set("1110111111011x", InstEmit.Ald, typeof(OpCodeAttribute));
Set("1110111111110x", InstEmit.Ast, typeof(OpCodeAttribute));
Set("11101100xxxxxx", InstEmit.Atoms, typeof(OpCodeAtom));
Set("1111000010101x", InstEmit.Bar, typeof(OpCodeBarrier));
Set("0100110000000x", InstEmit.Bfe, typeof(OpCodeAluCbuf));
Set("0011100x00000x", InstEmit.Bfe, typeof(OpCodeAluImm));
Set("0101110000000x", InstEmit.Bfe, typeof(OpCodeAluReg));
Set("0100101111110x", InstEmit.Bfi, typeof(OpCodeAluCbuf));
Set("0011011x11110x", InstEmit.Bfi, typeof(OpCodeAluImm));
Set("0101001111110x", InstEmit.Bfi, typeof(OpCodeAluRegCbuf));
Set("0101101111110x", InstEmit.Bfi, typeof(OpCodeAluReg));
Set("111000100100xx", InstEmit.Bra, typeof(OpCodeBranch));
Set("111000110100xx", InstEmit.Brk, typeof(OpCodeBranchPop));
Set("111000100101xx", InstEmit.Brx, typeof(OpCodeBranchIndir));
2019-11-24 14:49:56 -08:00
Set("0101000010100x", InstEmit.Csetp, typeof(OpCodePset));
Set("0100110001110x", InstEmit.Dadd, typeof(OpCodeFArithCbuf));
Set("0011100x01110x", InstEmit.Dadd, typeof(OpCodeDArithImm));
Set("0101110001110x", InstEmit.Dadd, typeof(OpCodeFArithReg));
Set("1111000011110x", InstEmit.Depbar, typeof(OpCode));
Set("010010110111xx", InstEmit.Dfma, typeof(OpCodeFArithCbuf));
Set("0011011x0111xx", InstEmit.Dfma, typeof(OpCodeDArithImm));
Set("010100110111xx", InstEmit.Dfma, typeof(OpCodeFArithRegCbuf));
Set("010110110111xx", InstEmit.Dfma, typeof(OpCodeFArithReg));
Set("0100110010000x", InstEmit.Dmul, typeof(OpCodeFArithCbuf));
Set("0011100x10000x", InstEmit.Dmul, typeof(OpCodeDArithImm));
Set("0101110010000x", InstEmit.Dmul, typeof(OpCodeFArithReg));
Set("111000110000xx", InstEmit.Exit, typeof(OpCodeExit));
Set("0100110010101x", InstEmit.F2F, typeof(OpCodeFArithCbuf));
Set("0011100x10101x", InstEmit.F2F, typeof(OpCodeFArithImm));
Set("0101110010101x", InstEmit.F2F, typeof(OpCodeFArithReg));
Set("0100110010110x", InstEmit.F2I, typeof(OpCodeFArithCbuf));
Set("0011100x10110x", InstEmit.F2I, typeof(OpCodeFArithImm));
Set("0101110010110x", InstEmit.F2I, typeof(OpCodeFArithReg));
Set("0100110001011x", InstEmit.Fadd, typeof(OpCodeFArithCbuf));
Set("0011100x01011x", InstEmit.Fadd, typeof(OpCodeFArithImm));
Set("000010xxxxxxxx", InstEmit.Fadd, typeof(OpCodeFArithImm32));
Set("0101110001011x", InstEmit.Fadd, typeof(OpCodeFArithReg));
Set("010010011xxxxx", InstEmit.Ffma, typeof(OpCodeFArithCbuf));
Set("0011001x1xxxxx", InstEmit.Ffma, typeof(OpCodeFArithImm));
Set("000011xxxxxxxx", InstEmit.Ffma32i, typeof(OpCodeFArithImm32));
Set("010100011xxxxx", InstEmit.Ffma, typeof(OpCodeFArithRegCbuf));
Set("010110011xxxxx", InstEmit.Ffma, typeof(OpCodeFArithReg));
Set("0100110000110x", InstEmit.Flo, typeof(OpCodeAluCbuf));
Set("0011100x00110x", InstEmit.Flo, typeof(OpCodeAluImm));
Set("0101110000110x", InstEmit.Flo, typeof(OpCodeAluReg));
Set("0100110001100x", InstEmit.Fmnmx, typeof(OpCodeFArithCbuf));
Set("0011100x01100x", InstEmit.Fmnmx, typeof(OpCodeFArithImm));
Set("0101110001100x", InstEmit.Fmnmx, typeof(OpCodeFArithReg));
Set("0100110001101x", InstEmit.Fmul, typeof(OpCodeFArithCbuf));
Set("0011100x01101x", InstEmit.Fmul, typeof(OpCodeFArithImm));
Set("00011110xxxxxx", InstEmit.Fmul, typeof(OpCodeFArithImm32));
Set("0101110001101x", InstEmit.Fmul, typeof(OpCodeFArithReg));
Set("0100100xxxxxxx", InstEmit.Fset, typeof(OpCodeSetCbuf));
Set("0011000xxxxxxx", InstEmit.Fset, typeof(OpCodeFsetImm));
Set("01011000xxxxxx", InstEmit.Fset, typeof(OpCodeSetReg));
Set("010010111011xx", InstEmit.Fsetp, typeof(OpCodeSetCbuf));
Set("0011011x1011xx", InstEmit.Fsetp, typeof(OpCodeFsetImm));
Set("010110111011xx", InstEmit.Fsetp, typeof(OpCodeSetReg));
Set("0101000011111x", InstEmit.Fswzadd, typeof(OpCodeAluReg));
Set("0111101x1xxxxx", InstEmit.Hadd2, typeof(OpCodeAluCbuf));
Set("0111101x0xxxxx", InstEmit.Hadd2, typeof(OpCodeAluImm2x10));
Set("0010110xxxxxxx", InstEmit.Hadd2, typeof(OpCodeAluImm32));
Set("0101110100010x", InstEmit.Hadd2, typeof(OpCodeAluReg));
Set("01110xxx1xxxxx", InstEmit.Hfma2, typeof(OpCodeHfmaCbuf));
Set("01110xxx0xxxxx", InstEmit.Hfma2, typeof(OpCodeHfmaImm2x10));
Set("0010100xxxxxxx", InstEmit.Hfma2, typeof(OpCodeHfmaImm32));
Set("0101110100000x", InstEmit.Hfma2, typeof(OpCodeHfmaReg));
Set("01100xxx1xxxxx", InstEmit.Hfma2, typeof(OpCodeHfmaRegCbuf));
Set("0111100x1xxxxx", InstEmit.Hmul2, typeof(OpCodeAluCbuf));
Set("0111100x0xxxxx", InstEmit.Hmul2, typeof(OpCodeAluImm2x10));
Set("0010101xxxxxxx", InstEmit.Hmul2, typeof(OpCodeAluImm32));
Set("0101110100001x", InstEmit.Hmul2, typeof(OpCodeAluReg));
Set("0111110x1xxxxx", InstEmit.Hset2, typeof(OpCodeSetCbuf));
Set("0111110x0xxxxx", InstEmit.Hset2, typeof(OpCodeHsetImm2x10));
Set("0101110100011x", InstEmit.Hset2, typeof(OpCodeSetReg));
Set("0111111x1xxxxx", InstEmit.Hsetp2, typeof(OpCodeSetCbuf));
Set("0111111x0xxxxx", InstEmit.Hsetp2, typeof(OpCodeHsetImm2x10));
Set("0101110100100x", InstEmit.Hsetp2, typeof(OpCodeSetReg));
Set("0100110010111x", InstEmit.I2F, typeof(OpCodeAluCbuf));
Set("0011100x10111x", InstEmit.I2F, typeof(OpCodeAluImm));
Set("0101110010111x", InstEmit.I2F, typeof(OpCodeAluReg));
Set("0100110011100x", InstEmit.I2I, typeof(OpCodeAluCbuf));
Set("0011100x11100x", InstEmit.I2I, typeof(OpCodeAluImm));
Set("0101110011100x", InstEmit.I2I, typeof(OpCodeAluReg));
Set("0100110000010x", InstEmit.Iadd, typeof(OpCodeAluCbuf));
Set("0011100x00010x", InstEmit.Iadd, typeof(OpCodeAluImm));
Set("0001110x0xxxxx", InstEmit.Iadd, typeof(OpCodeAluImm32));
Set("0101110000010x", InstEmit.Iadd, typeof(OpCodeAluReg));
Set("010011001100xx", InstEmit.Iadd3, typeof(OpCodeAluCbuf));
Set("0011100x1100xx", InstEmit.Iadd3, typeof(OpCodeAluImm));
Set("010111001100xx", InstEmit.Iadd3, typeof(OpCodeAluReg));
Set("010010110100xx", InstEmit.Icmp, typeof(OpCodeAluCbuf));
Set("0011011x0100xx", InstEmit.Icmp, typeof(OpCodeAluImm));
Set("010110110100xx", InstEmit.Icmp, typeof(OpCodeAluReg));
Set("010100110100xx", InstEmit.Icmp, typeof(OpCodeAluRegCbuf));
Set("010010100xxxxx", InstEmit.Imad, typeof(OpCodeAluCbuf));
Set("0011010x0xxxxx", InstEmit.Imad, typeof(OpCodeAluImm));
Set("010110100xxxxx", InstEmit.Imad, typeof(OpCodeAluReg));
Set("010100100xxxxx", InstEmit.Imad, typeof(OpCodeAluRegCbuf));
Set("0100110000100x", InstEmit.Imnmx, typeof(OpCodeAluCbuf));
Set("0011100x00100x", InstEmit.Imnmx, typeof(OpCodeAluImm));
Set("0101110000100x", InstEmit.Imnmx, typeof(OpCodeAluReg));
Set("11100000xxxxxx", InstEmit.Ipa, typeof(OpCodeIpa));
Set("1110111111010x", InstEmit.Isberd, typeof(OpCodeAlu));
Set("0100110000011x", InstEmit.Iscadd, typeof(OpCodeAluCbuf));
Set("0011100x00011x", InstEmit.Iscadd, typeof(OpCodeAluImm));
Set("000101xxxxxxxx", InstEmit.Iscadd, typeof(OpCodeAluImm32));
Set("0101110000011x", InstEmit.Iscadd, typeof(OpCodeAluReg));
Set("010010110101xx", InstEmit.Iset, typeof(OpCodeSetCbuf));
Set("001101100101xx", InstEmit.Iset, typeof(OpCodeSetImm));
Set("010110110101xx", InstEmit.Iset, typeof(OpCodeSetReg));
Set("010010110110xx", InstEmit.Isetp, typeof(OpCodeSetCbuf));
Set("0011011x0110xx", InstEmit.Isetp, typeof(OpCodeSetImm));
Set("010110110110xx", InstEmit.Isetp, typeof(OpCodeSetReg));
Set("111000110011xx", InstEmit.Kil, typeof(OpCodeExit));
Set("1110111101000x", InstEmit.Ld, typeof(OpCodeMemory));
Set("1110111110010x", InstEmit.Ldc, typeof(OpCodeLdc));
Set("1110111011010x", InstEmit.Ldg, typeof(OpCodeMemory));
Set("1110111101001x", InstEmit.Lds, typeof(OpCodeMemory));
Set("0100110001000x", InstEmit.Lop, typeof(OpCodeLopCbuf));
Set("0011100001000x", InstEmit.Lop, typeof(OpCodeLopImm));
Set("000001xxxxxxxx", InstEmit.Lop, typeof(OpCodeLopImm32));
Set("0101110001000x", InstEmit.Lop, typeof(OpCodeLopReg));
Set("0010000xxxxxxx", InstEmit.Lop3, typeof(OpCodeLopCbuf));
Set("001111xxxxxxxx", InstEmit.Lop3, typeof(OpCodeLopImm));
Set("0101101111100x", InstEmit.Lop3, typeof(OpCodeLopReg));
Set("1110111110011x", InstEmit.Membar, typeof(OpCodeMemoryBarrier));
Set("0100110010011x", InstEmit.Mov, typeof(OpCodeAluCbuf));
Set("0011100x10011x", InstEmit.Mov, typeof(OpCodeAluImm));
Set("000000010000xx", InstEmit.Mov, typeof(OpCodeAluImm32));
Set("0101110010011x", InstEmit.Mov, typeof(OpCodeAluReg));
Set("0101000010000x", InstEmit.Mufu, typeof(OpCodeFArith));
Set("0101000010110x", InstEmit.Nop, typeof(OpCode));
Set("1111101111100x", InstEmit.Out, typeof(OpCode));
Set("111000101010xx", InstEmit.Pbk, typeof(OpCodePush));
Set("0100110000001x", InstEmit.Popc, typeof(OpCodeAluCbuf));
Set("0011100x00001x", InstEmit.Popc, typeof(OpCodeAluImm));
Set("0101110000001x", InstEmit.Popc, typeof(OpCodeAluReg));
2019-11-24 14:49:56 -08:00
Set("0101000010001x", InstEmit.Pset, typeof(OpCodePset));
Set("0101000010010x", InstEmit.Psetp, typeof(OpCodePset));
2019-11-19 05:45:46 -08:00
Set("0100110011110x", InstEmit.R2p, typeof(OpCodeAluCbuf));
Set("0011100x11110x", InstEmit.R2p, typeof(OpCodeAluImm));
Set("0101110011110x", InstEmit.R2p, typeof(OpCodeAluReg));
Set("1110101111111x", InstEmit.Red, typeof(OpCodeRed));
Set("0100110010010x", InstEmit.Rro, typeof(OpCodeFArithCbuf));
Set("0011100x10010x", InstEmit.Rro, typeof(OpCodeFArithImm));
Set("0101110010010x", InstEmit.Rro, typeof(OpCodeFArithReg));
Set("1111000011001x", InstEmit.S2r, typeof(OpCodeAlu));
Set("0100110010100x", InstEmit.Sel, typeof(OpCodeAluCbuf));
Set("0011100x10100x", InstEmit.Sel, typeof(OpCodeAluImm));
Set("0101110010100x", InstEmit.Sel, typeof(OpCodeAluReg));
Set("1110111100010x", InstEmit.Shfl, typeof(OpCodeShuffle));
Set("0100110001001x", InstEmit.Shl, typeof(OpCodeAluCbuf));
Set("0011100x01001x", InstEmit.Shl, typeof(OpCodeAluImm));
Set("0101110001001x", InstEmit.Shl, typeof(OpCodeAluReg));
Set("0100110000101x", InstEmit.Shr, typeof(OpCodeAluCbuf));
Set("0011100x00101x", InstEmit.Shr, typeof(OpCodeAluImm));
Set("0101110000101x", InstEmit.Shr, typeof(OpCodeAluReg));
Set("111000101001xx", InstEmit.Ssy, typeof(OpCodePush));
Set("1110111101010x", InstEmit.St, typeof(OpCodeMemory));
Set("1110111011011x", InstEmit.Stg, typeof(OpCodeMemory));
Set("1110111101011x", InstEmit.Sts, typeof(OpCodeMemory));
Set("11101011001xxx", InstEmit.Sust, typeof(OpCodeImage));
Set("1111000011111x", InstEmit.Sync, typeof(OpCodeBranchPop));
Set("110000xxxx111x", InstEmit.Tex, typeof(OpCodeTex));
Set("1101111010111x", InstEmit.TexB, typeof(OpCodeTexB));
Set("1101x00xxxxxxx", InstEmit.Texs, typeof(OpCodeTexs));
Set("1101x01xxxxxxx", InstEmit.Texs, typeof(OpCodeTlds));
Set("11011111x0xxxx", InstEmit.Texs, typeof(OpCodeTld4s));
Set("11011100xx111x", InstEmit.Tld, typeof(OpCodeTld));
Set("11011101xx111x", InstEmit.TldB, typeof(OpCodeTld));
Set("110010xxxx111x", InstEmit.Tld4, typeof(OpCodeTld4));
Set("1101111011111x", InstEmit.Tld4, typeof(OpCodeTld4B));
Set("110111100x1110", InstEmit.Txd, typeof(OpCodeTxd));
Set("1101111101001x", InstEmit.Txq, typeof(OpCodeTex));
Set("1101111101010x", InstEmit.TxqB, typeof(OpCodeTex));
Set("01011111xxxxxx", InstEmit.Vmad, typeof(OpCodeVideo));
Set("0011101xxxxxxx", InstEmit.Vmnmx, typeof(OpCodeVideo));
Set("0101000011011x", InstEmit.Vote, typeof(OpCodeVote));
Set("0100111xxxxxxx", InstEmit.Xmad, typeof(OpCodeAluCbuf));
Set("0011011x00xxxx", InstEmit.Xmad, typeof(OpCodeAluImm));
Set("010100010xxxxx", InstEmit.Xmad, typeof(OpCodeAluRegCbuf));
Set("0101101100xxxx", InstEmit.Xmad, typeof(OpCodeAluReg));
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 16:57:08 -07:00
#endregion
}
private static void Set(string encoding, InstEmitter emitter, Type opCodeType)
{
if (encoding.Length != EncodingBits)
{
throw new ArgumentException(nameof(encoding));
}
int bit = encoding.Length - 1;
int value = 0;
int xMask = 0;
int xBits = 0;
int[] xPos = new int[encoding.Length];
for (int index = 0; index < encoding.Length; index++, bit--)
{
char chr = encoding[index];
if (chr == '1')
{
value |= 1 << bit;
}
else if (chr == 'x')
{
xMask |= 1 << bit;
xPos[xBits++] = bit;
}
}
xMask = ~xMask;
TableEntry entry = new TableEntry(emitter, opCodeType, xBits);
for (int index = 0; index < (1 << xBits); index++)
{
value &= xMask;
2019-10-12 23:02:07 -07:00
for (int x = 0; x < xBits; 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 16:57:08 -07:00
{
2019-10-12 23:02:07 -07:00
value |= ((index >> x) & 1) << xPos[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 16:57:08 -07:00
}
if (_opCodes[value] == null || _opCodes[value].XBits > xBits)
{
_opCodes[value] = entry;
}
}
}
public static (InstEmitter emitter, OpActivator opActivator) GetEmitter(long opCode)
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 16:57:08 -07:00
{
2019-10-12 23:02:07 -07:00
TableEntry entry = _opCodes[(ulong)opCode >> (64 - EncodingBits)];
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 16:57:08 -07:00
if (entry != null)
{
return (entry.Emitter, entry.OpActivator);
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 16:57:08 -07:00
}
return (null, null);
}
}
}