Ryujinx/ARMeilleure/CodeGen/X86/CodeGenerator.cs

1703 lines
62 KiB
C#
Raw Normal View History

Add a new JIT compiler for CPU code (#693) * Start of the ARMeilleure project * Refactoring around the old IRAdapter, now renamed to PreAllocator * Optimize the LowestBitSet method * Add CLZ support and fix CLS implementation * Add missing Equals and GetHashCode overrides on some structs, misc small tweaks * Implement the ByteSwap IR instruction, and some refactoring on the assembler * Implement the DivideUI IR instruction and fix 64-bits IDIV * Correct constant operand type on CSINC * Move division instructions implementation to InstEmitDiv * Fix destination type for the ConditionalSelect IR instruction * Implement UMULH and SMULH, with new IR instructions * Fix some issues with shift instructions * Fix constant types for BFM instructions * Fix up new tests using the new V128 struct * Update tests * Move DIV tests to a separate file * Add support for calls, and some instructions that depends on them * Start adding support for SIMD & FP types, along with some of the related ARM instructions * Fix some typos and the divide instruction with FP operands * Fix wrong method call on Clz_V * Implement ARM FP & SIMD move instructions, Saddlv_V, and misc. fixes * Implement SIMD logical instructions and more misc. fixes * Fix PSRAD x86 instruction encoding, TRN, UABD and UABDL implementations * Implement float conversion instruction, merge in LDj3SNuD fixes, and some other misc. fixes * Implement SIMD shift instruction and fix Dup_V * Add SCVTF and UCVTF (vector, fixed-point) variants to the opcode table * Fix check with tolerance on tester * Implement FP & SIMD comparison instructions, and some fixes * Update FCVT (Scalar) encoding on the table to support the Half-float variants * Support passing V128 structs, some cleanup on the register allocator, merge LDj3SNuD fixes * Use old memory access methods, made a start on SIMD memory insts support, some fixes * Fix float constant passed to functions, save and restore non-volatile XMM registers, other fixes * Fix arguments count with struct return values, other fixes * More instructions * Misc. fixes and integrate LDj3SNuD fixes * Update tests * Add a faster linear scan allocator, unwinding support on windows, and other changes * Update Ryujinx.HLE * Update Ryujinx.Graphics * Fix V128 return pointer passing, RCX is clobbered * Update Ryujinx.Tests * Update ITimeZoneService * Stop using GetFunctionPointer as that can't be called from native code, misc. fixes and tweaks * Use generic GetFunctionPointerForDelegate method and other tweaks * Some refactoring on the code generator, assert on invalid operations and use a separate enum for intrinsics * Remove some unused code on the assembler * Fix REX.W prefix regression on float conversion instructions, add some sort of profiler * Add hardware capability detection * Fix regression on Sha1h and revert Fcm** changes * Add SSE2-only paths on vector extract and insert, some refactoring on the pre-allocator * Fix silly mistake introduced on last commit on CpuId * Generate inline stack probes when the stack allocation is too large * Initial support for the System-V ABI * Support multiple destination operands * Fix SSE2 VectorInsert8 path, and other fixes * Change placement of XMM callee save and restore code to match other compilers * Rename Dest to Destination and Inst to Instruction * Fix a regression related to calls and the V128 type * Add an extra space on comments to match code style * Some refactoring * Fix vector insert FP32 SSE2 path * Port over the ARM32 instructions * Avoid memory protection races on JIT Cache * Another fix on VectorInsert FP32 (thanks to LDj3SNuD * Float operands don't need to use the same register when VEX is supported * Add a new register allocator, higher quality code for hot code (tier up), and other tweaks * Some nits, small improvements on the pre allocator * CpuThreadState is gone * Allow changing CPU emulators with a config entry * Add runtime identifiers on the ARMeilleure project * Allow switching between CPUs through a config entry (pt. 2) * Change win10-x64 to win-x64 on projects * Update the Ryujinx project to use ARMeilleure * Ensure that the selected register is valid on the hybrid allocator * Allow exiting on returns to 0 (should fix test regression) * Remove register assignments for most used variables on the hybrid allocator * Do not use fixed registers as spill temp * Add missing namespace and remove unneeded using * Address PR feedback * Fix types, etc * Enable AssumeStrictAbiCompliance by default * Ensure that Spill and Fill don't load or store any more than necessary
2019-08-08 20:56:22 +02:00
using ARMeilleure.CodeGen.Optimizations;
using ARMeilleure.CodeGen.RegisterAllocators;
using ARMeilleure.CodeGen.Unwinding;
using ARMeilleure.Common;
using ARMeilleure.Diagnostics;
using ARMeilleure.IntermediateRepresentation;
using ARMeilleure.Translation;
using System;
using System.Collections.Generic;
using System.Diagnostics;
using System.IO;
namespace ARMeilleure.CodeGen.X86
{
static class CodeGenerator
{
private const int PageSize = 0x1000;
private const int StackGuardSize = 0x2000;
private static Action<CodeGenContext, Operation>[] _instTable;
static CodeGenerator()
{
_instTable = new Action<CodeGenContext, Operation>[EnumUtils.GetCount(typeof(Instruction))];
Add(Instruction.Add, GenerateAdd);
Add(Instruction.BitwiseAnd, GenerateBitwiseAnd);
Add(Instruction.BitwiseExclusiveOr, GenerateBitwiseExclusiveOr);
Add(Instruction.BitwiseNot, GenerateBitwiseNot);
Add(Instruction.BitwiseOr, GenerateBitwiseOr);
Add(Instruction.Branch, GenerateBranch);
Add(Instruction.BranchIfFalse, GenerateBranchIfFalse);
Add(Instruction.BranchIfTrue, GenerateBranchIfTrue);
Add(Instruction.ByteSwap, GenerateByteSwap);
Add(Instruction.Call, GenerateCall);
Add(Instruction.Clobber, GenerateClobber);
Add(Instruction.CompareAndSwap128, GenerateCompareAndSwap128);
Add(Instruction.CompareEqual, GenerateCompareEqual);
Add(Instruction.CompareGreater, GenerateCompareGreater);
Add(Instruction.CompareGreaterOrEqual, GenerateCompareGreaterOrEqual);
Add(Instruction.CompareGreaterOrEqualUI, GenerateCompareGreaterOrEqualUI);
Add(Instruction.CompareGreaterUI, GenerateCompareGreaterUI);
Add(Instruction.CompareLess, GenerateCompareLess);
Add(Instruction.CompareLessOrEqual, GenerateCompareLessOrEqual);
Add(Instruction.CompareLessOrEqualUI, GenerateCompareLessOrEqualUI);
Add(Instruction.CompareLessUI, GenerateCompareLessUI);
Add(Instruction.CompareNotEqual, GenerateCompareNotEqual);
Add(Instruction.ConditionalSelect, GenerateConditionalSelect);
Add(Instruction.ConvertI64ToI32, GenerateConvertI64ToI32);
Add(Instruction.ConvertToFP, GenerateConvertToFP);
Add(Instruction.Copy, GenerateCopy);
Add(Instruction.CountLeadingZeros, GenerateCountLeadingZeros);
Add(Instruction.CpuId, GenerateCpuId);
Add(Instruction.Divide, GenerateDivide);
Add(Instruction.DivideUI, GenerateDivideUI);
Add(Instruction.Fill, GenerateFill);
Add(Instruction.Load, GenerateLoad);
Add(Instruction.Load16, GenerateLoad16);
Add(Instruction.Load8, GenerateLoad8);
Add(Instruction.Multiply, GenerateMultiply);
Add(Instruction.Multiply64HighSI, GenerateMultiply64HighSI);
Add(Instruction.Multiply64HighUI, GenerateMultiply64HighUI);
Add(Instruction.Negate, GenerateNegate);
Add(Instruction.Return, GenerateReturn);
Add(Instruction.RotateRight, GenerateRotateRight);
Add(Instruction.ShiftLeft, GenerateShiftLeft);
Add(Instruction.ShiftRightSI, GenerateShiftRightSI);
Add(Instruction.ShiftRightUI, GenerateShiftRightUI);
Add(Instruction.SignExtend16, GenerateSignExtend16);
Add(Instruction.SignExtend32, GenerateSignExtend32);
Add(Instruction.SignExtend8, GenerateSignExtend8);
Add(Instruction.Spill, GenerateSpill);
Add(Instruction.SpillArg, GenerateSpillArg);
Add(Instruction.StackAlloc, GenerateStackAlloc);
Add(Instruction.Store, GenerateStore);
Add(Instruction.Store16, GenerateStore16);
Add(Instruction.Store8, GenerateStore8);
Add(Instruction.Subtract, GenerateSubtract);
Add(Instruction.VectorCreateScalar, GenerateVectorCreateScalar);
Add(Instruction.VectorExtract, GenerateVectorExtract);
Add(Instruction.VectorExtract16, GenerateVectorExtract16);
Add(Instruction.VectorExtract8, GenerateVectorExtract8);
Add(Instruction.VectorInsert, GenerateVectorInsert);
Add(Instruction.VectorInsert16, GenerateVectorInsert16);
Add(Instruction.VectorInsert8, GenerateVectorInsert8);
Add(Instruction.VectorOne, GenerateVectorOne);
Add(Instruction.VectorZero, GenerateVectorZero);
Add(Instruction.VectorZeroUpper64, GenerateVectorZeroUpper64);
Add(Instruction.VectorZeroUpper96, GenerateVectorZeroUpper96);
Add(Instruction.ZeroExtend16, GenerateZeroExtend16);
Add(Instruction.ZeroExtend32, GenerateZeroExtend32);
Add(Instruction.ZeroExtend8, GenerateZeroExtend8);
}
private static void Add(Instruction inst, Action<CodeGenContext, Operation> func)
{
_instTable[(int)inst] = func;
}
public static CompiledFunction Generate(CompilerContext cctx)
{
ControlFlowGraph cfg = cctx.Cfg;
Logger.StartPass(PassName.Optimization);
if ((cctx.Options & CompilerOptions.SsaForm) != 0 &&
(cctx.Options & CompilerOptions.Optimize) != 0)
{
Optimizer.RunPass(cfg);
}
Logger.EndPass(PassName.Optimization, cfg);
Logger.StartPass(PassName.PreAllocation);
StackAllocator stackAlloc = new StackAllocator();
PreAllocator.RunPass(cctx, stackAlloc, out int maxCallArgs);
Logger.EndPass(PassName.PreAllocation, cfg);
Logger.StartPass(PassName.RegisterAllocation);
if ((cctx.Options & CompilerOptions.SsaForm) != 0)
{
Ssa.Deconstruct(cfg);
}
IRegisterAllocator regAlloc;
if ((cctx.Options & CompilerOptions.Lsra) != 0)
{
regAlloc = new LinearScanAllocator();
}
else
{
regAlloc = new HybridAllocator();
}
RegisterMasks regMasks = new RegisterMasks(
CallingConvention.GetIntAvailableRegisters(),
CallingConvention.GetVecAvailableRegisters(),
CallingConvention.GetIntCallerSavedRegisters(),
CallingConvention.GetVecCallerSavedRegisters(),
CallingConvention.GetIntCalleeSavedRegisters(),
CallingConvention.GetVecCalleeSavedRegisters());
AllocationResult allocResult = regAlloc.RunPass(cfg, stackAlloc, regMasks);
Logger.EndPass(PassName.RegisterAllocation, cfg);
Logger.StartPass(PassName.CodeGeneration);
using (MemoryStream stream = new MemoryStream())
{
CodeGenContext context = new CodeGenContext(stream, allocResult, maxCallArgs, cfg.Blocks.Count);
UnwindInfo unwindInfo = WritePrologue(context);
for (BasicBlock block = cfg.Blocks.First; block != null; block = block.ListNext)
Add a new JIT compiler for CPU code (#693) * Start of the ARMeilleure project * Refactoring around the old IRAdapter, now renamed to PreAllocator * Optimize the LowestBitSet method * Add CLZ support and fix CLS implementation * Add missing Equals and GetHashCode overrides on some structs, misc small tweaks * Implement the ByteSwap IR instruction, and some refactoring on the assembler * Implement the DivideUI IR instruction and fix 64-bits IDIV * Correct constant operand type on CSINC * Move division instructions implementation to InstEmitDiv * Fix destination type for the ConditionalSelect IR instruction * Implement UMULH and SMULH, with new IR instructions * Fix some issues with shift instructions * Fix constant types for BFM instructions * Fix up new tests using the new V128 struct * Update tests * Move DIV tests to a separate file * Add support for calls, and some instructions that depends on them * Start adding support for SIMD & FP types, along with some of the related ARM instructions * Fix some typos and the divide instruction with FP operands * Fix wrong method call on Clz_V * Implement ARM FP & SIMD move instructions, Saddlv_V, and misc. fixes * Implement SIMD logical instructions and more misc. fixes * Fix PSRAD x86 instruction encoding, TRN, UABD and UABDL implementations * Implement float conversion instruction, merge in LDj3SNuD fixes, and some other misc. fixes * Implement SIMD shift instruction and fix Dup_V * Add SCVTF and UCVTF (vector, fixed-point) variants to the opcode table * Fix check with tolerance on tester * Implement FP & SIMD comparison instructions, and some fixes * Update FCVT (Scalar) encoding on the table to support the Half-float variants * Support passing V128 structs, some cleanup on the register allocator, merge LDj3SNuD fixes * Use old memory access methods, made a start on SIMD memory insts support, some fixes * Fix float constant passed to functions, save and restore non-volatile XMM registers, other fixes * Fix arguments count with struct return values, other fixes * More instructions * Misc. fixes and integrate LDj3SNuD fixes * Update tests * Add a faster linear scan allocator, unwinding support on windows, and other changes * Update Ryujinx.HLE * Update Ryujinx.Graphics * Fix V128 return pointer passing, RCX is clobbered * Update Ryujinx.Tests * Update ITimeZoneService * Stop using GetFunctionPointer as that can't be called from native code, misc. fixes and tweaks * Use generic GetFunctionPointerForDelegate method and other tweaks * Some refactoring on the code generator, assert on invalid operations and use a separate enum for intrinsics * Remove some unused code on the assembler * Fix REX.W prefix regression on float conversion instructions, add some sort of profiler * Add hardware capability detection * Fix regression on Sha1h and revert Fcm** changes * Add SSE2-only paths on vector extract and insert, some refactoring on the pre-allocator * Fix silly mistake introduced on last commit on CpuId * Generate inline stack probes when the stack allocation is too large * Initial support for the System-V ABI * Support multiple destination operands * Fix SSE2 VectorInsert8 path, and other fixes * Change placement of XMM callee save and restore code to match other compilers * Rename Dest to Destination and Inst to Instruction * Fix a regression related to calls and the V128 type * Add an extra space on comments to match code style * Some refactoring * Fix vector insert FP32 SSE2 path * Port over the ARM32 instructions * Avoid memory protection races on JIT Cache * Another fix on VectorInsert FP32 (thanks to LDj3SNuD * Float operands don't need to use the same register when VEX is supported * Add a new register allocator, higher quality code for hot code (tier up), and other tweaks * Some nits, small improvements on the pre allocator * CpuThreadState is gone * Allow changing CPU emulators with a config entry * Add runtime identifiers on the ARMeilleure project * Allow switching between CPUs through a config entry (pt. 2) * Change win10-x64 to win-x64 on projects * Update the Ryujinx project to use ARMeilleure * Ensure that the selected register is valid on the hybrid allocator * Allow exiting on returns to 0 (should fix test regression) * Remove register assignments for most used variables on the hybrid allocator * Do not use fixed registers as spill temp * Add missing namespace and remove unneeded using * Address PR feedback * Fix types, etc * Enable AssumeStrictAbiCompliance by default * Ensure that Spill and Fill don't load or store any more than necessary
2019-08-08 20:56:22 +02:00
{
context.EnterBlock(block);
for (Node node = block.Operations.First; node != null; node = node.ListNext)
Add a new JIT compiler for CPU code (#693) * Start of the ARMeilleure project * Refactoring around the old IRAdapter, now renamed to PreAllocator * Optimize the LowestBitSet method * Add CLZ support and fix CLS implementation * Add missing Equals and GetHashCode overrides on some structs, misc small tweaks * Implement the ByteSwap IR instruction, and some refactoring on the assembler * Implement the DivideUI IR instruction and fix 64-bits IDIV * Correct constant operand type on CSINC * Move division instructions implementation to InstEmitDiv * Fix destination type for the ConditionalSelect IR instruction * Implement UMULH and SMULH, with new IR instructions * Fix some issues with shift instructions * Fix constant types for BFM instructions * Fix up new tests using the new V128 struct * Update tests * Move DIV tests to a separate file * Add support for calls, and some instructions that depends on them * Start adding support for SIMD & FP types, along with some of the related ARM instructions * Fix some typos and the divide instruction with FP operands * Fix wrong method call on Clz_V * Implement ARM FP & SIMD move instructions, Saddlv_V, and misc. fixes * Implement SIMD logical instructions and more misc. fixes * Fix PSRAD x86 instruction encoding, TRN, UABD and UABDL implementations * Implement float conversion instruction, merge in LDj3SNuD fixes, and some other misc. fixes * Implement SIMD shift instruction and fix Dup_V * Add SCVTF and UCVTF (vector, fixed-point) variants to the opcode table * Fix check with tolerance on tester * Implement FP & SIMD comparison instructions, and some fixes * Update FCVT (Scalar) encoding on the table to support the Half-float variants * Support passing V128 structs, some cleanup on the register allocator, merge LDj3SNuD fixes * Use old memory access methods, made a start on SIMD memory insts support, some fixes * Fix float constant passed to functions, save and restore non-volatile XMM registers, other fixes * Fix arguments count with struct return values, other fixes * More instructions * Misc. fixes and integrate LDj3SNuD fixes * Update tests * Add a faster linear scan allocator, unwinding support on windows, and other changes * Update Ryujinx.HLE * Update Ryujinx.Graphics * Fix V128 return pointer passing, RCX is clobbered * Update Ryujinx.Tests * Update ITimeZoneService * Stop using GetFunctionPointer as that can't be called from native code, misc. fixes and tweaks * Use generic GetFunctionPointerForDelegate method and other tweaks * Some refactoring on the code generator, assert on invalid operations and use a separate enum for intrinsics * Remove some unused code on the assembler * Fix REX.W prefix regression on float conversion instructions, add some sort of profiler * Add hardware capability detection * Fix regression on Sha1h and revert Fcm** changes * Add SSE2-only paths on vector extract and insert, some refactoring on the pre-allocator * Fix silly mistake introduced on last commit on CpuId * Generate inline stack probes when the stack allocation is too large * Initial support for the System-V ABI * Support multiple destination operands * Fix SSE2 VectorInsert8 path, and other fixes * Change placement of XMM callee save and restore code to match other compilers * Rename Dest to Destination and Inst to Instruction * Fix a regression related to calls and the V128 type * Add an extra space on comments to match code style * Some refactoring * Fix vector insert FP32 SSE2 path * Port over the ARM32 instructions * Avoid memory protection races on JIT Cache * Another fix on VectorInsert FP32 (thanks to LDj3SNuD * Float operands don't need to use the same register when VEX is supported * Add a new register allocator, higher quality code for hot code (tier up), and other tweaks * Some nits, small improvements on the pre allocator * CpuThreadState is gone * Allow changing CPU emulators with a config entry * Add runtime identifiers on the ARMeilleure project * Allow switching between CPUs through a config entry (pt. 2) * Change win10-x64 to win-x64 on projects * Update the Ryujinx project to use ARMeilleure * Ensure that the selected register is valid on the hybrid allocator * Allow exiting on returns to 0 (should fix test regression) * Remove register assignments for most used variables on the hybrid allocator * Do not use fixed registers as spill temp * Add missing namespace and remove unneeded using * Address PR feedback * Fix types, etc * Enable AssumeStrictAbiCompliance by default * Ensure that Spill and Fill don't load or store any more than necessary
2019-08-08 20:56:22 +02:00
{
if (node is Operation operation)
{
GenerateOperation(context, operation);
}
}
}
Logger.EndPass(PassName.CodeGeneration);
return new CompiledFunction(context.GetCode(), unwindInfo);
}
}
private static void GenerateOperation(CodeGenContext context, Operation operation)
{
if (operation.Instruction == Instruction.Extended)
{
IntrinsicOperation intrinOp = (IntrinsicOperation)operation;
IntrinsicInfo info = IntrinsicTable.GetInfo(intrinOp.Intrinsic);
switch (info.Type)
{
case IntrinsicType.Comis_:
{
Operand dest = operation.Destination;
Operand src1 = operation.GetSource(0);
Operand src2 = operation.GetSource(1);
switch (intrinOp.Intrinsic)
{
case Intrinsic.X86Comisdeq:
context.Assembler.Comisd(src1, src2);
context.Assembler.Setcc(dest, X86Condition.Equal);
break;
case Intrinsic.X86Comisdge:
context.Assembler.Comisd(src1, src2);
context.Assembler.Setcc(dest, X86Condition.AboveOrEqual);
break;
case Intrinsic.X86Comisdlt:
context.Assembler.Comisd(src1, src2);
context.Assembler.Setcc(dest, X86Condition.Below);
break;
case Intrinsic.X86Comisseq:
context.Assembler.Comiss(src1, src2);
context.Assembler.Setcc(dest, X86Condition.Equal);
break;
case Intrinsic.X86Comissge:
context.Assembler.Comiss(src1, src2);
context.Assembler.Setcc(dest, X86Condition.AboveOrEqual);
break;
case Intrinsic.X86Comisslt:
context.Assembler.Comiss(src1, src2);
context.Assembler.Setcc(dest, X86Condition.Below);
break;
}
context.Assembler.Movzx8(dest, dest, OperandType.I32);
break;
}
case IntrinsicType.PopCount:
{
Operand dest = operation.Destination;
Operand source = operation.GetSource(0);
EnsureSameType(dest, source);
Debug.Assert(dest.Type.IsInteger());
context.Assembler.Popcnt(dest, source, dest.Type);
break;
}
case IntrinsicType.Unary:
{
Operand dest = operation.Destination;
Operand source = operation.GetSource(0);
EnsureSameType(dest, source);
Debug.Assert(!dest.Type.IsInteger());
context.Assembler.WriteInstruction(info.Inst, dest, source);
break;
}
case IntrinsicType.UnaryToGpr:
{
Operand dest = operation.Destination;
Operand source = operation.GetSource(0);
Debug.Assert(dest.Type.IsInteger() && !source.Type.IsInteger());
if (intrinOp.Intrinsic == Intrinsic.X86Cvtsi2si)
{
if (dest.Type == OperandType.I32)
{
context.Assembler.Movd(dest, source); // int _mm_cvtsi128_si32(__m128i a)
}
else /* if (dest.Type == OperandType.I64) */
{
context.Assembler.Movq(dest, source); // __int64 _mm_cvtsi128_si64(__m128i a)
}
}
else
{
context.Assembler.WriteInstruction(info.Inst, dest, source, dest.Type);
}
Add a new JIT compiler for CPU code (#693) * Start of the ARMeilleure project * Refactoring around the old IRAdapter, now renamed to PreAllocator * Optimize the LowestBitSet method * Add CLZ support and fix CLS implementation * Add missing Equals and GetHashCode overrides on some structs, misc small tweaks * Implement the ByteSwap IR instruction, and some refactoring on the assembler * Implement the DivideUI IR instruction and fix 64-bits IDIV * Correct constant operand type on CSINC * Move division instructions implementation to InstEmitDiv * Fix destination type for the ConditionalSelect IR instruction * Implement UMULH and SMULH, with new IR instructions * Fix some issues with shift instructions * Fix constant types for BFM instructions * Fix up new tests using the new V128 struct * Update tests * Move DIV tests to a separate file * Add support for calls, and some instructions that depends on them * Start adding support for SIMD & FP types, along with some of the related ARM instructions * Fix some typos and the divide instruction with FP operands * Fix wrong method call on Clz_V * Implement ARM FP & SIMD move instructions, Saddlv_V, and misc. fixes * Implement SIMD logical instructions and more misc. fixes * Fix PSRAD x86 instruction encoding, TRN, UABD and UABDL implementations * Implement float conversion instruction, merge in LDj3SNuD fixes, and some other misc. fixes * Implement SIMD shift instruction and fix Dup_V * Add SCVTF and UCVTF (vector, fixed-point) variants to the opcode table * Fix check with tolerance on tester * Implement FP & SIMD comparison instructions, and some fixes * Update FCVT (Scalar) encoding on the table to support the Half-float variants * Support passing V128 structs, some cleanup on the register allocator, merge LDj3SNuD fixes * Use old memory access methods, made a start on SIMD memory insts support, some fixes * Fix float constant passed to functions, save and restore non-volatile XMM registers, other fixes * Fix arguments count with struct return values, other fixes * More instructions * Misc. fixes and integrate LDj3SNuD fixes * Update tests * Add a faster linear scan allocator, unwinding support on windows, and other changes * Update Ryujinx.HLE * Update Ryujinx.Graphics * Fix V128 return pointer passing, RCX is clobbered * Update Ryujinx.Tests * Update ITimeZoneService * Stop using GetFunctionPointer as that can't be called from native code, misc. fixes and tweaks * Use generic GetFunctionPointerForDelegate method and other tweaks * Some refactoring on the code generator, assert on invalid operations and use a separate enum for intrinsics * Remove some unused code on the assembler * Fix REX.W prefix regression on float conversion instructions, add some sort of profiler * Add hardware capability detection * Fix regression on Sha1h and revert Fcm** changes * Add SSE2-only paths on vector extract and insert, some refactoring on the pre-allocator * Fix silly mistake introduced on last commit on CpuId * Generate inline stack probes when the stack allocation is too large * Initial support for the System-V ABI * Support multiple destination operands * Fix SSE2 VectorInsert8 path, and other fixes * Change placement of XMM callee save and restore code to match other compilers * Rename Dest to Destination and Inst to Instruction * Fix a regression related to calls and the V128 type * Add an extra space on comments to match code style * Some refactoring * Fix vector insert FP32 SSE2 path * Port over the ARM32 instructions * Avoid memory protection races on JIT Cache * Another fix on VectorInsert FP32 (thanks to LDj3SNuD * Float operands don't need to use the same register when VEX is supported * Add a new register allocator, higher quality code for hot code (tier up), and other tweaks * Some nits, small improvements on the pre allocator * CpuThreadState is gone * Allow changing CPU emulators with a config entry * Add runtime identifiers on the ARMeilleure project * Allow switching between CPUs through a config entry (pt. 2) * Change win10-x64 to win-x64 on projects * Update the Ryujinx project to use ARMeilleure * Ensure that the selected register is valid on the hybrid allocator * Allow exiting on returns to 0 (should fix test regression) * Remove register assignments for most used variables on the hybrid allocator * Do not use fixed registers as spill temp * Add missing namespace and remove unneeded using * Address PR feedback * Fix types, etc * Enable AssumeStrictAbiCompliance by default * Ensure that Spill and Fill don't load or store any more than necessary
2019-08-08 20:56:22 +02:00
break;
}
case IntrinsicType.Binary:
{
Operand dest = operation.Destination;
Operand src1 = operation.GetSource(0);
Operand src2 = operation.GetSource(1);
EnsureSameType(dest, src1);
if (!HardwareCapabilities.SupportsVexEncoding)
{
EnsureSameReg(dest, src1);
}
Debug.Assert(!dest.Type.IsInteger());
Debug.Assert(!src2.Type.IsInteger() || src2.Kind == OperandKind.Constant);
context.Assembler.WriteInstruction(info.Inst, dest, src1, src2);
break;
}
case IntrinsicType.BinaryGpr:
{
Operand dest = operation.Destination;
Operand src1 = operation.GetSource(0);
Operand src2 = operation.GetSource(1);
EnsureSameType(dest, src1);
if (!HardwareCapabilities.SupportsVexEncoding)
{
EnsureSameReg(dest, src1);
}
Debug.Assert(!dest.Type.IsInteger() && src2.Type.IsInteger());
context.Assembler.WriteInstruction(info.Inst, dest, src1, src2, src2.Type);
break;
}
Add a new JIT compiler for CPU code (#693) * Start of the ARMeilleure project * Refactoring around the old IRAdapter, now renamed to PreAllocator * Optimize the LowestBitSet method * Add CLZ support and fix CLS implementation * Add missing Equals and GetHashCode overrides on some structs, misc small tweaks * Implement the ByteSwap IR instruction, and some refactoring on the assembler * Implement the DivideUI IR instruction and fix 64-bits IDIV * Correct constant operand type on CSINC * Move division instructions implementation to InstEmitDiv * Fix destination type for the ConditionalSelect IR instruction * Implement UMULH and SMULH, with new IR instructions * Fix some issues with shift instructions * Fix constant types for BFM instructions * Fix up new tests using the new V128 struct * Update tests * Move DIV tests to a separate file * Add support for calls, and some instructions that depends on them * Start adding support for SIMD & FP types, along with some of the related ARM instructions * Fix some typos and the divide instruction with FP operands * Fix wrong method call on Clz_V * Implement ARM FP & SIMD move instructions, Saddlv_V, and misc. fixes * Implement SIMD logical instructions and more misc. fixes * Fix PSRAD x86 instruction encoding, TRN, UABD and UABDL implementations * Implement float conversion instruction, merge in LDj3SNuD fixes, and some other misc. fixes * Implement SIMD shift instruction and fix Dup_V * Add SCVTF and UCVTF (vector, fixed-point) variants to the opcode table * Fix check with tolerance on tester * Implement FP & SIMD comparison instructions, and some fixes * Update FCVT (Scalar) encoding on the table to support the Half-float variants * Support passing V128 structs, some cleanup on the register allocator, merge LDj3SNuD fixes * Use old memory access methods, made a start on SIMD memory insts support, some fixes * Fix float constant passed to functions, save and restore non-volatile XMM registers, other fixes * Fix arguments count with struct return values, other fixes * More instructions * Misc. fixes and integrate LDj3SNuD fixes * Update tests * Add a faster linear scan allocator, unwinding support on windows, and other changes * Update Ryujinx.HLE * Update Ryujinx.Graphics * Fix V128 return pointer passing, RCX is clobbered * Update Ryujinx.Tests * Update ITimeZoneService * Stop using GetFunctionPointer as that can't be called from native code, misc. fixes and tweaks * Use generic GetFunctionPointerForDelegate method and other tweaks * Some refactoring on the code generator, assert on invalid operations and use a separate enum for intrinsics * Remove some unused code on the assembler * Fix REX.W prefix regression on float conversion instructions, add some sort of profiler * Add hardware capability detection * Fix regression on Sha1h and revert Fcm** changes * Add SSE2-only paths on vector extract and insert, some refactoring on the pre-allocator * Fix silly mistake introduced on last commit on CpuId * Generate inline stack probes when the stack allocation is too large * Initial support for the System-V ABI * Support multiple destination operands * Fix SSE2 VectorInsert8 path, and other fixes * Change placement of XMM callee save and restore code to match other compilers * Rename Dest to Destination and Inst to Instruction * Fix a regression related to calls and the V128 type * Add an extra space on comments to match code style * Some refactoring * Fix vector insert FP32 SSE2 path * Port over the ARM32 instructions * Avoid memory protection races on JIT Cache * Another fix on VectorInsert FP32 (thanks to LDj3SNuD * Float operands don't need to use the same register when VEX is supported * Add a new register allocator, higher quality code for hot code (tier up), and other tweaks * Some nits, small improvements on the pre allocator * CpuThreadState is gone * Allow changing CPU emulators with a config entry * Add runtime identifiers on the ARMeilleure project * Allow switching between CPUs through a config entry (pt. 2) * Change win10-x64 to win-x64 on projects * Update the Ryujinx project to use ARMeilleure * Ensure that the selected register is valid on the hybrid allocator * Allow exiting on returns to 0 (should fix test regression) * Remove register assignments for most used variables on the hybrid allocator * Do not use fixed registers as spill temp * Add missing namespace and remove unneeded using * Address PR feedback * Fix types, etc * Enable AssumeStrictAbiCompliance by default * Ensure that Spill and Fill don't load or store any more than necessary
2019-08-08 20:56:22 +02:00
case IntrinsicType.BinaryImm:
{
Operand dest = operation.Destination;
Operand src1 = operation.GetSource(0);
Operand src2 = operation.GetSource(1);
EnsureSameType(dest, src1);
if (!HardwareCapabilities.SupportsVexEncoding)
{
EnsureSameReg(dest, src1);
}
Debug.Assert(!dest.Type.IsInteger() && src2.Kind == OperandKind.Constant);
context.Assembler.WriteInstruction(info.Inst, dest, src1, src2.AsByte());
break;
}
case IntrinsicType.Ternary:
{
Operand dest = operation.Destination;
Operand src1 = operation.GetSource(0);
Operand src2 = operation.GetSource(1);
Operand src3 = operation.GetSource(2);
EnsureSameType(dest, src1, src2, src3);
Debug.Assert(!dest.Type.IsInteger());
if (info.Inst == X86Instruction.Blendvpd && HardwareCapabilities.SupportsVexEncoding)
{
context.Assembler.WriteInstruction(X86Instruction.Vblendvpd, dest, src1, src2, src3);
}
else if (info.Inst == X86Instruction.Blendvps && HardwareCapabilities.SupportsVexEncoding)
{
context.Assembler.WriteInstruction(X86Instruction.Vblendvps, dest, src1, src2, src3);
}
else if (info.Inst == X86Instruction.Pblendvb && HardwareCapabilities.SupportsVexEncoding)
Add a new JIT compiler for CPU code (#693) * Start of the ARMeilleure project * Refactoring around the old IRAdapter, now renamed to PreAllocator * Optimize the LowestBitSet method * Add CLZ support and fix CLS implementation * Add missing Equals and GetHashCode overrides on some structs, misc small tweaks * Implement the ByteSwap IR instruction, and some refactoring on the assembler * Implement the DivideUI IR instruction and fix 64-bits IDIV * Correct constant operand type on CSINC * Move division instructions implementation to InstEmitDiv * Fix destination type for the ConditionalSelect IR instruction * Implement UMULH and SMULH, with new IR instructions * Fix some issues with shift instructions * Fix constant types for BFM instructions * Fix up new tests using the new V128 struct * Update tests * Move DIV tests to a separate file * Add support for calls, and some instructions that depends on them * Start adding support for SIMD & FP types, along with some of the related ARM instructions * Fix some typos and the divide instruction with FP operands * Fix wrong method call on Clz_V * Implement ARM FP & SIMD move instructions, Saddlv_V, and misc. fixes * Implement SIMD logical instructions and more misc. fixes * Fix PSRAD x86 instruction encoding, TRN, UABD and UABDL implementations * Implement float conversion instruction, merge in LDj3SNuD fixes, and some other misc. fixes * Implement SIMD shift instruction and fix Dup_V * Add SCVTF and UCVTF (vector, fixed-point) variants to the opcode table * Fix check with tolerance on tester * Implement FP & SIMD comparison instructions, and some fixes * Update FCVT (Scalar) encoding on the table to support the Half-float variants * Support passing V128 structs, some cleanup on the register allocator, merge LDj3SNuD fixes * Use old memory access methods, made a start on SIMD memory insts support, some fixes * Fix float constant passed to functions, save and restore non-volatile XMM registers, other fixes * Fix arguments count with struct return values, other fixes * More instructions * Misc. fixes and integrate LDj3SNuD fixes * Update tests * Add a faster linear scan allocator, unwinding support on windows, and other changes * Update Ryujinx.HLE * Update Ryujinx.Graphics * Fix V128 return pointer passing, RCX is clobbered * Update Ryujinx.Tests * Update ITimeZoneService * Stop using GetFunctionPointer as that can't be called from native code, misc. fixes and tweaks * Use generic GetFunctionPointerForDelegate method and other tweaks * Some refactoring on the code generator, assert on invalid operations and use a separate enum for intrinsics * Remove some unused code on the assembler * Fix REX.W prefix regression on float conversion instructions, add some sort of profiler * Add hardware capability detection * Fix regression on Sha1h and revert Fcm** changes * Add SSE2-only paths on vector extract and insert, some refactoring on the pre-allocator * Fix silly mistake introduced on last commit on CpuId * Generate inline stack probes when the stack allocation is too large * Initial support for the System-V ABI * Support multiple destination operands * Fix SSE2 VectorInsert8 path, and other fixes * Change placement of XMM callee save and restore code to match other compilers * Rename Dest to Destination and Inst to Instruction * Fix a regression related to calls and the V128 type * Add an extra space on comments to match code style * Some refactoring * Fix vector insert FP32 SSE2 path * Port over the ARM32 instructions * Avoid memory protection races on JIT Cache * Another fix on VectorInsert FP32 (thanks to LDj3SNuD * Float operands don't need to use the same register when VEX is supported * Add a new register allocator, higher quality code for hot code (tier up), and other tweaks * Some nits, small improvements on the pre allocator * CpuThreadState is gone * Allow changing CPU emulators with a config entry * Add runtime identifiers on the ARMeilleure project * Allow switching between CPUs through a config entry (pt. 2) * Change win10-x64 to win-x64 on projects * Update the Ryujinx project to use ARMeilleure * Ensure that the selected register is valid on the hybrid allocator * Allow exiting on returns to 0 (should fix test regression) * Remove register assignments for most used variables on the hybrid allocator * Do not use fixed registers as spill temp * Add missing namespace and remove unneeded using * Address PR feedback * Fix types, etc * Enable AssumeStrictAbiCompliance by default * Ensure that Spill and Fill don't load or store any more than necessary
2019-08-08 20:56:22 +02:00
{
context.Assembler.WriteInstruction(X86Instruction.Vpblendvb, dest, src1, src2, src3);
}
else
{
EnsureSameReg(dest, src1);
Debug.Assert(src3.GetRegister().Index == 0);
context.Assembler.WriteInstruction(info.Inst, dest, src1, src2);
}
break;
}
case IntrinsicType.TernaryImm:
{
Operand dest = operation.Destination;
Operand src1 = operation.GetSource(0);
Operand src2 = operation.GetSource(1);
Operand src3 = operation.GetSource(2);
EnsureSameType(dest, src1, src2);
if (!HardwareCapabilities.SupportsVexEncoding)
{
EnsureSameReg(dest, src1);
}
Debug.Assert(!dest.Type.IsInteger() && src3.Kind == OperandKind.Constant);
context.Assembler.WriteInstruction(info.Inst, dest, src1, src2, src3.AsByte());
break;
}
}
}
else
{
Action<CodeGenContext, Operation> func = _instTable[(int)operation.Instruction];
if (func != null)
{
func(context, operation);
}
else
{
throw new ArgumentException($"Invalid instruction \"{operation.Instruction}\".");
}
}
}
private static void GenerateAdd(CodeGenContext context, Operation operation)
{
Operand dest = operation.Destination;
Operand src1 = operation.GetSource(0);
Operand src2 = operation.GetSource(1);
ValidateBinOp(dest, src1, src2);
if (dest.Type.IsInteger())
{
context.Assembler.Add(dest, src2, dest.Type);
}
else if (dest.Type == OperandType.FP32)
{
context.Assembler.Addss(dest, src1, src2);
}
else /* if (dest.Type == OperandType.FP64) */
{
context.Assembler.Addsd(dest, src1, src2);
}
}
private static void GenerateBitwiseAnd(CodeGenContext context, Operation operation)
{
Operand dest = operation.Destination;
Operand src1 = operation.GetSource(0);
Operand src2 = operation.GetSource(1);
ValidateBinOp(dest, src1, src2);
Debug.Assert(dest.Type.IsInteger());
context.Assembler.And(dest, src2, dest.Type);
}
private static void GenerateBitwiseExclusiveOr(CodeGenContext context, Operation operation)
{
Operand dest = operation.Destination;
Operand src1 = operation.GetSource(0);
Operand src2 = operation.GetSource(1);
ValidateBinOp(dest, src1, src2);
if (dest.Type.IsInteger())
{
context.Assembler.Xor(dest, src2, dest.Type);
}
else
{
context.Assembler.Xorps(dest, src1, src2);
}
}
private static void GenerateBitwiseNot(CodeGenContext context, Operation operation)
{
Operand dest = operation.Destination;
Operand source = operation.GetSource(0);
ValidateUnOp(dest, source);
Debug.Assert(dest.Type.IsInteger());
context.Assembler.Not(dest);
}
private static void GenerateBitwiseOr(CodeGenContext context, Operation operation)
{
Operand dest = operation.Destination;
Operand src1 = operation.GetSource(0);
Operand src2 = operation.GetSource(1);
ValidateBinOp(dest, src1, src2);
Debug.Assert(dest.Type.IsInteger());
context.Assembler.Or(dest, src2, dest.Type);
}
private static void GenerateBranch(CodeGenContext context, Operation operation)
{
context.JumpTo(context.CurrBlock.Branch);
}
private static void GenerateBranchIfFalse(CodeGenContext context, Operation operation)
{
Operand source = operation.GetSource(0);
context.Assembler.Test(source, source, source.Type);
context.JumpTo(X86Condition.Equal, context.CurrBlock.Branch);
}
private static void GenerateBranchIfTrue(CodeGenContext context, Operation operation)
{
Operand source = operation.GetSource(0);
context.Assembler.Test(source, source, source.Type);
context.JumpTo(X86Condition.NotEqual, context.CurrBlock.Branch);
}
private static void GenerateByteSwap(CodeGenContext context, Operation operation)
{
Operand dest = operation.Destination;
Operand source = operation.GetSource(0);
ValidateUnOp(dest, source);
Debug.Assert(dest.Type.IsInteger());
context.Assembler.Bswap(dest);
}
private static void GenerateCall(CodeGenContext context, Operation operation)
{
context.Assembler.Call(operation.GetSource(0));
}
private static void GenerateClobber(CodeGenContext context, Operation operation)
{
// This is only used to indicate that a register is clobbered to the
// register allocator, we don't need to produce any code.
}
private static void GenerateCompareAndSwap128(CodeGenContext context, Operation operation)
{
Operand source = operation.GetSource(0);
MemoryOperand memOp = new MemoryOperand(OperandType.I64, source);
context.Assembler.Cmpxchg16b(memOp);
}
private static void GenerateCompareEqual(CodeGenContext context, Operation operation)
{
GenerateCompare(context, operation, X86Condition.Equal);
}
private static void GenerateCompareGreater(CodeGenContext context, Operation operation)
{
GenerateCompare(context, operation, X86Condition.Greater);
}
private static void GenerateCompareGreaterOrEqual(CodeGenContext context, Operation operation)
{
GenerateCompare(context, operation, X86Condition.GreaterOrEqual);
}
private static void GenerateCompareGreaterOrEqualUI(CodeGenContext context, Operation operation)
{
GenerateCompare(context, operation, X86Condition.AboveOrEqual);
}
private static void GenerateCompareGreaterUI(CodeGenContext context, Operation operation)
{
GenerateCompare(context, operation, X86Condition.Above);
}
private static void GenerateCompareLess(CodeGenContext context, Operation operation)
{
GenerateCompare(context, operation, X86Condition.Less);
}
private static void GenerateCompareLessOrEqual(CodeGenContext context, Operation operation)
{
GenerateCompare(context, operation, X86Condition.LessOrEqual);
}
private static void GenerateCompareLessOrEqualUI(CodeGenContext context, Operation operation)
{
GenerateCompare(context, operation, X86Condition.BelowOrEqual);
}
private static void GenerateCompareLessUI(CodeGenContext context, Operation operation)
{
GenerateCompare(context, operation, X86Condition.Below);
}
private static void GenerateCompareNotEqual(CodeGenContext context, Operation operation)
{
GenerateCompare(context, operation, X86Condition.NotEqual);
}
private static void GenerateCompare(CodeGenContext context, Operation operation, X86Condition condition)
{
Operand dest = operation.Destination;
Operand src1 = operation.GetSource(0);
Operand src2 = operation.GetSource(1);
EnsureSameType(src1, src2);
Debug.Assert(dest.Type == OperandType.I32);
context.Assembler.Cmp(src1, src2, src1.Type);
context.Assembler.Setcc(dest, condition);
context.Assembler.Movzx8(dest, dest, OperandType.I32);
}
private static void GenerateConditionalSelect(CodeGenContext context, Operation operation)
{
Operand dest = operation.Destination;
Operand src1 = operation.GetSource(0);
Operand src2 = operation.GetSource(1);
Operand src3 = operation.GetSource(2);
EnsureSameReg (dest, src3);
EnsureSameType(dest, src2, src3);
Debug.Assert(dest.Type.IsInteger());
Debug.Assert(src1.Type == OperandType.I32);
context.Assembler.Test (src1, src1, src1.Type);
context.Assembler.Cmovcc(dest, src2, dest.Type, X86Condition.NotEqual);
}
private static void GenerateConvertI64ToI32(CodeGenContext context, Operation operation)
{
Operand dest = operation.Destination;
Operand source = operation.GetSource(0);
Debug.Assert(dest.Type == OperandType.I32 && source.Type == OperandType.I64);
context.Assembler.Mov(dest, source, OperandType.I32);
}
private static void GenerateConvertToFP(CodeGenContext context, Operation operation)
{
Operand dest = operation.Destination;
Operand source = operation.GetSource(0);
Debug.Assert(dest.Type == OperandType.FP32 || dest.Type == OperandType.FP64);
if (dest.Type == OperandType.FP32)
{
Debug.Assert(source.Type.IsInteger() || source.Type == OperandType.FP64);
if (source.Type.IsInteger())
{
context.Assembler.Xorps (dest, dest, dest);
context.Assembler.Cvtsi2ss(dest, dest, source, source.Type);
}
else /* if (source.Type == OperandType.FP64) */
{
context.Assembler.Cvtsd2ss(dest, dest, source);
GenerateZeroUpper96(context, dest, dest);
}
}
else /* if (dest.Type == OperandType.FP64) */
{
Debug.Assert(source.Type.IsInteger() || source.Type == OperandType.FP32);
if (source.Type.IsInteger())
{
context.Assembler.Xorps (dest, dest, dest);
context.Assembler.Cvtsi2sd(dest, dest, source, source.Type);
}
else /* if (source.Type == OperandType.FP32) */
{
context.Assembler.Cvtss2sd(dest, dest, source);
GenerateZeroUpper64(context, dest, dest);
}
}
}
private static void GenerateCopy(CodeGenContext context, Operation operation)
{
Operand dest = operation.Destination;
Operand source = operation.GetSource(0);
EnsureSameType(dest, source);
Debug.Assert(dest.Type.IsInteger() || source.Kind != OperandKind.Constant);
// Moves to the same register are useless.
if (dest.Kind == source.Kind && dest.Value == source.Value)
{
return;
}
if (dest.Kind == OperandKind.Register &&
source.Kind == OperandKind.Constant && source.Value == 0)
{
// Assemble "mov reg, 0" as "xor reg, reg" as the later is more efficient.
context.Assembler.Xor(dest, dest, OperandType.I32);
}
else if (dest.Type.IsInteger())
{
context.Assembler.Mov(dest, source, dest.Type);
}
else
{
context.Assembler.Movdqu(dest, source);
}
}
private static void GenerateCountLeadingZeros(CodeGenContext context, Operation operation)
{
Operand dest = operation.Destination;
Operand source = operation.GetSource(0);
EnsureSameType(dest, source);
Debug.Assert(dest.Type.IsInteger());
context.Assembler.Bsr(dest, source, dest.Type);
int operandSize = dest.Type == OperandType.I32 ? 32 : 64;
int operandMask = operandSize - 1;
// When the input operand is 0, the result is undefined, however the
// ZF flag is set. We are supposed to return the operand size on that
// case. So, add an additional jump to handle that case, by moving the
// operand size constant to the destination register.
context.JumpToNear(X86Condition.NotEqual);
context.Assembler.Mov(dest, new Operand(operandSize | operandMask), OperandType.I32);
context.JumpHere();
// BSR returns the zero based index of the last bit set on the operand,
// starting from the least significant bit. However we are supposed to
// return the number of 0 bits on the high end. So, we invert the result
// of the BSR using XOR to get the correct value.
context.Assembler.Xor(dest, new Operand(operandMask), OperandType.I32);
}
private static void GenerateCpuId(CodeGenContext context, Operation operation)
{
context.Assembler.Cpuid();
}
private static void GenerateDivide(CodeGenContext context, Operation operation)
{
Operand dest = operation.Destination;
Operand dividend = operation.GetSource(0);
Operand divisor = operation.GetSource(1);
if (!dest.Type.IsInteger())
{
ValidateBinOp(dest, dividend, divisor);
}
if (dest.Type.IsInteger())
{
divisor = operation.GetSource(2);
EnsureSameType(dest, divisor);
if (divisor.Type == OperandType.I32)
{
context.Assembler.Cdq();
}
else
{
context.Assembler.Cqo();
}
context.Assembler.Idiv(divisor);
}
else if (dest.Type == OperandType.FP32)
{
context.Assembler.Divss(dest, dividend, divisor);
}
else /* if (dest.Type == OperandType.FP64) */
{
context.Assembler.Divsd(dest, dividend, divisor);
}
}
private static void GenerateDivideUI(CodeGenContext context, Operation operation)
{
Operand divisor = operation.GetSource(2);
Operand rdx = Register(X86Register.Rdx);
Debug.Assert(divisor.Type.IsInteger());
context.Assembler.Xor(rdx, rdx, OperandType.I32);
context.Assembler.Div(divisor);
}
private static void GenerateFill(CodeGenContext context, Operation operation)
{
Operand dest = operation.Destination;
Operand offset = operation.GetSource(0);
Debug.Assert(offset.Kind == OperandKind.Constant);
int offs = offset.AsInt32() + context.CallArgsRegionSize;
Operand rsp = Register(X86Register.Rsp);
MemoryOperand memOp = new MemoryOperand(dest.Type, rsp, null, Multiplier.x1, offs);
GenerateLoad(context, memOp, dest);
}
private static void GenerateLoad(CodeGenContext context, Operation operation)
{
Operand value = operation.Destination;
Operand address = Memory(operation.GetSource(0), value.Type);
GenerateLoad(context, address, value);
}
private static void GenerateLoad16(CodeGenContext context, Operation operation)
{
Operand value = operation.Destination;
Operand address = Memory(operation.GetSource(0), value.Type);
Debug.Assert(value.Type.IsInteger());
context.Assembler.Movzx16(value, address, value.Type);
}
private static void GenerateLoad8(CodeGenContext context, Operation operation)
{
Operand value = operation.Destination;
Operand address = Memory(operation.GetSource(0), value.Type);
Debug.Assert(value.Type.IsInteger());
context.Assembler.Movzx8(value, address, value.Type);
}
private static void GenerateMultiply(CodeGenContext context, Operation operation)
{
Operand dest = operation.Destination;
Operand src1 = operation.GetSource(0);
Operand src2 = operation.GetSource(1);
if (src2.Kind != OperandKind.Constant)
{
EnsureSameReg(dest, src1);
}
EnsureSameType(dest, src1, src2);
if (dest.Type.IsInteger())
{
if (src2.Kind == OperandKind.Constant)
{
context.Assembler.Imul(dest, src1, src2, dest.Type);
}
else
{
context.Assembler.Imul(dest, src2, dest.Type);
}
}
else if (dest.Type == OperandType.FP32)
{
context.Assembler.Mulss(dest, src1, src2);
}
else /* if (dest.Type == OperandType.FP64) */
{
context.Assembler.Mulsd(dest, src1, src2);
}
}
private static void GenerateMultiply64HighSI(CodeGenContext context, Operation operation)
{
Operand source = operation.GetSource(1);
Debug.Assert(source.Type == OperandType.I64);
context.Assembler.Imul(source);
}
private static void GenerateMultiply64HighUI(CodeGenContext context, Operation operation)
{
Operand source = operation.GetSource(1);
Debug.Assert(source.Type == OperandType.I64);
context.Assembler.Mul(source);
}
private static void GenerateNegate(CodeGenContext context, Operation operation)
{
Operand dest = operation.Destination;
Operand source = operation.GetSource(0);
ValidateUnOp(dest, source);
Debug.Assert(dest.Type.IsInteger());
context.Assembler.Neg(dest);
}
private static void GenerateReturn(CodeGenContext context, Operation operation)
{
WriteEpilogue(context);
context.Assembler.Return();
}
private static void GenerateRotateRight(CodeGenContext context, Operation operation)
{
Operand dest = operation.Destination;
Operand src1 = operation.GetSource(0);
Operand src2 = operation.GetSource(1);
ValidateShift(dest, src1, src2);
context.Assembler.Ror(dest, src2, dest.Type);
}
private static void GenerateShiftLeft(CodeGenContext context, Operation operation)
{
Operand dest = operation.Destination;
Operand src1 = operation.GetSource(0);
Operand src2 = operation.GetSource(1);
ValidateShift(dest, src1, src2);
context.Assembler.Shl(dest, src2, dest.Type);
}
private static void GenerateShiftRightSI(CodeGenContext context, Operation operation)
{
Operand dest = operation.Destination;
Operand src1 = operation.GetSource(0);
Operand src2 = operation.GetSource(1);
ValidateShift(dest, src1, src2);
context.Assembler.Sar(dest, src2, dest.Type);
}
private static void GenerateShiftRightUI(CodeGenContext context, Operation operation)
{
Operand dest = operation.Destination;
Operand src1 = operation.GetSource(0);
Operand src2 = operation.GetSource(1);
ValidateShift(dest, src1, src2);
context.Assembler.Shr(dest, src2, dest.Type);
}
private static void GenerateSignExtend16(CodeGenContext context, Operation operation)
{
Operand dest = operation.Destination;
Operand source = operation.GetSource(0);
Debug.Assert(dest.Type.IsInteger() && source.Type.IsInteger());
context.Assembler.Movsx16(dest, source, dest.Type);
}
private static void GenerateSignExtend32(CodeGenContext context, Operation operation)
{
Operand dest = operation.Destination;
Operand source = operation.GetSource(0);
Debug.Assert(dest.Type.IsInteger() && source.Type.IsInteger());
context.Assembler.Movsx32(dest, source, dest.Type);
}
private static void GenerateSignExtend8(CodeGenContext context, Operation operation)
{
Operand dest = operation.Destination;
Operand source = operation.GetSource(0);
Debug.Assert(dest.Type.IsInteger() && source.Type.IsInteger());
context.Assembler.Movsx8(dest, source, dest.Type);
}
private static void GenerateSpill(CodeGenContext context, Operation operation)
{
GenerateSpill(context, operation, context.CallArgsRegionSize);
}
private static void GenerateSpillArg(CodeGenContext context, Operation operation)
{
GenerateSpill(context, operation, 0);
}
private static void GenerateSpill(CodeGenContext context, Operation operation, int baseOffset)
{
Operand offset = operation.GetSource(0);
Operand source = operation.GetSource(1);
Debug.Assert(offset.Kind == OperandKind.Constant);
int offs = offset.AsInt32() + baseOffset;
Operand rsp = Register(X86Register.Rsp);
MemoryOperand memOp = new MemoryOperand(source.Type, rsp, null, Multiplier.x1, offs);
GenerateStore(context, memOp, source);
}
private static void GenerateStackAlloc(CodeGenContext context, Operation operation)
{
Operand dest = operation.Destination;
Operand offset = operation.GetSource(0);
Debug.Assert(offset.Kind == OperandKind.Constant);
int offs = offset.AsInt32() + context.CallArgsRegionSize;
Operand rsp = Register(X86Register.Rsp);
MemoryOperand memOp = new MemoryOperand(OperandType.I64, rsp, null, Multiplier.x1, offs);
context.Assembler.Lea(dest, memOp, OperandType.I64);
}
private static void GenerateStore(CodeGenContext context, Operation operation)
{
Operand value = operation.GetSource(1);
Operand address = Memory(operation.GetSource(0), value.Type);
GenerateStore(context, address, value);
}
private static void GenerateStore16(CodeGenContext context, Operation operation)
{
Operand value = operation.GetSource(1);
Operand address = Memory(operation.GetSource(0), value.Type);
Debug.Assert(value.Type.IsInteger());
context.Assembler.Mov16(address, value);
}
private static void GenerateStore8(CodeGenContext context, Operation operation)
{
Operand value = operation.GetSource(1);
Operand address = Memory(operation.GetSource(0), value.Type);
Debug.Assert(value.Type.IsInteger());
context.Assembler.Mov8(address, value);
}
private static void GenerateSubtract(CodeGenContext context, Operation operation)
{
Operand dest = operation.Destination;
Operand src1 = operation.GetSource(0);
Operand src2 = operation.GetSource(1);
ValidateBinOp(dest, src1, src2);
if (dest.Type.IsInteger())
{
context.Assembler.Sub(dest, src2, dest.Type);
}
else if (dest.Type == OperandType.FP32)
{
context.Assembler.Subss(dest, src1, src2);
}
else /* if (dest.Type == OperandType.FP64) */
{
context.Assembler.Subsd(dest, src1, src2);
}
}
private static void GenerateVectorCreateScalar(CodeGenContext context, Operation operation)
{
Operand dest = operation.Destination;
Operand source = operation.GetSource(0);
Debug.Assert(!dest.Type.IsInteger() && source.Type.IsInteger());
if (source.Type == OperandType.I32)
{
context.Assembler.Movd(dest, source); // (__m128i _mm_cvtsi32_si128(int a))
Add a new JIT compiler for CPU code (#693) * Start of the ARMeilleure project * Refactoring around the old IRAdapter, now renamed to PreAllocator * Optimize the LowestBitSet method * Add CLZ support and fix CLS implementation * Add missing Equals and GetHashCode overrides on some structs, misc small tweaks * Implement the ByteSwap IR instruction, and some refactoring on the assembler * Implement the DivideUI IR instruction and fix 64-bits IDIV * Correct constant operand type on CSINC * Move division instructions implementation to InstEmitDiv * Fix destination type for the ConditionalSelect IR instruction * Implement UMULH and SMULH, with new IR instructions * Fix some issues with shift instructions * Fix constant types for BFM instructions * Fix up new tests using the new V128 struct * Update tests * Move DIV tests to a separate file * Add support for calls, and some instructions that depends on them * Start adding support for SIMD & FP types, along with some of the related ARM instructions * Fix some typos and the divide instruction with FP operands * Fix wrong method call on Clz_V * Implement ARM FP & SIMD move instructions, Saddlv_V, and misc. fixes * Implement SIMD logical instructions and more misc. fixes * Fix PSRAD x86 instruction encoding, TRN, UABD and UABDL implementations * Implement float conversion instruction, merge in LDj3SNuD fixes, and some other misc. fixes * Implement SIMD shift instruction and fix Dup_V * Add SCVTF and UCVTF (vector, fixed-point) variants to the opcode table * Fix check with tolerance on tester * Implement FP & SIMD comparison instructions, and some fixes * Update FCVT (Scalar) encoding on the table to support the Half-float variants * Support passing V128 structs, some cleanup on the register allocator, merge LDj3SNuD fixes * Use old memory access methods, made a start on SIMD memory insts support, some fixes * Fix float constant passed to functions, save and restore non-volatile XMM registers, other fixes * Fix arguments count with struct return values, other fixes * More instructions * Misc. fixes and integrate LDj3SNuD fixes * Update tests * Add a faster linear scan allocator, unwinding support on windows, and other changes * Update Ryujinx.HLE * Update Ryujinx.Graphics * Fix V128 return pointer passing, RCX is clobbered * Update Ryujinx.Tests * Update ITimeZoneService * Stop using GetFunctionPointer as that can't be called from native code, misc. fixes and tweaks * Use generic GetFunctionPointerForDelegate method and other tweaks * Some refactoring on the code generator, assert on invalid operations and use a separate enum for intrinsics * Remove some unused code on the assembler * Fix REX.W prefix regression on float conversion instructions, add some sort of profiler * Add hardware capability detection * Fix regression on Sha1h and revert Fcm** changes * Add SSE2-only paths on vector extract and insert, some refactoring on the pre-allocator * Fix silly mistake introduced on last commit on CpuId * Generate inline stack probes when the stack allocation is too large * Initial support for the System-V ABI * Support multiple destination operands * Fix SSE2 VectorInsert8 path, and other fixes * Change placement of XMM callee save and restore code to match other compilers * Rename Dest to Destination and Inst to Instruction * Fix a regression related to calls and the V128 type * Add an extra space on comments to match code style * Some refactoring * Fix vector insert FP32 SSE2 path * Port over the ARM32 instructions * Avoid memory protection races on JIT Cache * Another fix on VectorInsert FP32 (thanks to LDj3SNuD * Float operands don't need to use the same register when VEX is supported * Add a new register allocator, higher quality code for hot code (tier up), and other tweaks * Some nits, small improvements on the pre allocator * CpuThreadState is gone * Allow changing CPU emulators with a config entry * Add runtime identifiers on the ARMeilleure project * Allow switching between CPUs through a config entry (pt. 2) * Change win10-x64 to win-x64 on projects * Update the Ryujinx project to use ARMeilleure * Ensure that the selected register is valid on the hybrid allocator * Allow exiting on returns to 0 (should fix test regression) * Remove register assignments for most used variables on the hybrid allocator * Do not use fixed registers as spill temp * Add missing namespace and remove unneeded using * Address PR feedback * Fix types, etc * Enable AssumeStrictAbiCompliance by default * Ensure that Spill and Fill don't load or store any more than necessary
2019-08-08 20:56:22 +02:00
}
else /* if (source.Type == OperandType.I64) */
{
context.Assembler.Movq(dest, source); // (__m128i _mm_cvtsi64_si128(__int64 a))
Add a new JIT compiler for CPU code (#693) * Start of the ARMeilleure project * Refactoring around the old IRAdapter, now renamed to PreAllocator * Optimize the LowestBitSet method * Add CLZ support and fix CLS implementation * Add missing Equals and GetHashCode overrides on some structs, misc small tweaks * Implement the ByteSwap IR instruction, and some refactoring on the assembler * Implement the DivideUI IR instruction and fix 64-bits IDIV * Correct constant operand type on CSINC * Move division instructions implementation to InstEmitDiv * Fix destination type for the ConditionalSelect IR instruction * Implement UMULH and SMULH, with new IR instructions * Fix some issues with shift instructions * Fix constant types for BFM instructions * Fix up new tests using the new V128 struct * Update tests * Move DIV tests to a separate file * Add support for calls, and some instructions that depends on them * Start adding support for SIMD & FP types, along with some of the related ARM instructions * Fix some typos and the divide instruction with FP operands * Fix wrong method call on Clz_V * Implement ARM FP & SIMD move instructions, Saddlv_V, and misc. fixes * Implement SIMD logical instructions and more misc. fixes * Fix PSRAD x86 instruction encoding, TRN, UABD and UABDL implementations * Implement float conversion instruction, merge in LDj3SNuD fixes, and some other misc. fixes * Implement SIMD shift instruction and fix Dup_V * Add SCVTF and UCVTF (vector, fixed-point) variants to the opcode table * Fix check with tolerance on tester * Implement FP & SIMD comparison instructions, and some fixes * Update FCVT (Scalar) encoding on the table to support the Half-float variants * Support passing V128 structs, some cleanup on the register allocator, merge LDj3SNuD fixes * Use old memory access methods, made a start on SIMD memory insts support, some fixes * Fix float constant passed to functions, save and restore non-volatile XMM registers, other fixes * Fix arguments count with struct return values, other fixes * More instructions * Misc. fixes and integrate LDj3SNuD fixes * Update tests * Add a faster linear scan allocator, unwinding support on windows, and other changes * Update Ryujinx.HLE * Update Ryujinx.Graphics * Fix V128 return pointer passing, RCX is clobbered * Update Ryujinx.Tests * Update ITimeZoneService * Stop using GetFunctionPointer as that can't be called from native code, misc. fixes and tweaks * Use generic GetFunctionPointerForDelegate method and other tweaks * Some refactoring on the code generator, assert on invalid operations and use a separate enum for intrinsics * Remove some unused code on the assembler * Fix REX.W prefix regression on float conversion instructions, add some sort of profiler * Add hardware capability detection * Fix regression on Sha1h and revert Fcm** changes * Add SSE2-only paths on vector extract and insert, some refactoring on the pre-allocator * Fix silly mistake introduced on last commit on CpuId * Generate inline stack probes when the stack allocation is too large * Initial support for the System-V ABI * Support multiple destination operands * Fix SSE2 VectorInsert8 path, and other fixes * Change placement of XMM callee save and restore code to match other compilers * Rename Dest to Destination and Inst to Instruction * Fix a regression related to calls and the V128 type * Add an extra space on comments to match code style * Some refactoring * Fix vector insert FP32 SSE2 path * Port over the ARM32 instructions * Avoid memory protection races on JIT Cache * Another fix on VectorInsert FP32 (thanks to LDj3SNuD * Float operands don't need to use the same register when VEX is supported * Add a new register allocator, higher quality code for hot code (tier up), and other tweaks * Some nits, small improvements on the pre allocator * CpuThreadState is gone * Allow changing CPU emulators with a config entry * Add runtime identifiers on the ARMeilleure project * Allow switching between CPUs through a config entry (pt. 2) * Change win10-x64 to win-x64 on projects * Update the Ryujinx project to use ARMeilleure * Ensure that the selected register is valid on the hybrid allocator * Allow exiting on returns to 0 (should fix test regression) * Remove register assignments for most used variables on the hybrid allocator * Do not use fixed registers as spill temp * Add missing namespace and remove unneeded using * Address PR feedback * Fix types, etc * Enable AssumeStrictAbiCompliance by default * Ensure that Spill and Fill don't load or store any more than necessary
2019-08-08 20:56:22 +02:00
}
}
private static void GenerateVectorExtract(CodeGenContext context, Operation operation)
{
Operand dest = operation.Destination; //Value
Operand src1 = operation.GetSource(0); //Vector
Operand src2 = operation.GetSource(1); //Index
Debug.Assert(src1.Type == OperandType.V128);
Debug.Assert(src2.Kind == OperandKind.Constant);
byte index = src2.AsByte();
if (dest.Type == OperandType.I32)
{
Debug.Assert(index < 4);
if (HardwareCapabilities.SupportsSse41)
{
context.Assembler.Pextrd(dest, src1, index);
}
else
{
if (index != 0)
{
int mask0 = 0b11_10_01_00;
int mask1 = 0b11_10_01_00;
mask0 = BitUtils.RotateRight(mask0, index * 2, 8);
mask1 = BitUtils.RotateRight(mask1, 8 - index * 2, 8);
context.Assembler.Pshufd(src1, src1, (byte)mask0);
context.Assembler.Movd (dest, src1);
context.Assembler.Pshufd(src1, src1, (byte)mask1);
}
else
{
context.Assembler.Movd(dest, src1);
}
}
}
else if (dest.Type == OperandType.I64)
{
Debug.Assert(index < 2);
if (HardwareCapabilities.SupportsSse41)
{
context.Assembler.Pextrq(dest, src1, index);
}
else
{
if (index != 0)
{
const byte mask = 0b01_00_11_10;
context.Assembler.Pshufd(src1, src1, mask);
context.Assembler.Movq (dest, src1);
context.Assembler.Pshufd(src1, src1, mask);
}
else
{
context.Assembler.Movq(dest, src1);
}
}
}
else
{
Debug.Assert(index < (dest.Type == OperandType.FP32 ? 4 : 2));
// Floating-point types.
if ((index >= 2 && dest.Type == OperandType.FP32) ||
(index == 1 && dest.Type == OperandType.FP64))
{
context.Assembler.Movhlps(dest, dest, src1);
context.Assembler.Movq (dest, dest);
}
else
{
context.Assembler.Movq(dest, src1);
}
if (dest.Type == OperandType.FP32)
{
context.Assembler.Pshufd(dest, dest, (byte)(0xfc | (index & 1)));
}
}
}
private static void GenerateVectorExtract16(CodeGenContext context, Operation operation)
{
Operand dest = operation.Destination; //Value
Operand src1 = operation.GetSource(0); //Vector
Operand src2 = operation.GetSource(1); //Index
Debug.Assert(src1.Type == OperandType.V128);
Debug.Assert(src2.Kind == OperandKind.Constant);
byte index = src2.AsByte();
Debug.Assert(index < 8);
context.Assembler.Pextrw(dest, src1, index);
}
private static void GenerateVectorExtract8(CodeGenContext context, Operation operation)
{
Operand dest = operation.Destination; //Value
Operand src1 = operation.GetSource(0); //Vector
Operand src2 = operation.GetSource(1); //Index
Debug.Assert(src1.Type == OperandType.V128);
Debug.Assert(src2.Kind == OperandKind.Constant);
byte index = src2.AsByte();
Debug.Assert(index < 16);
if (HardwareCapabilities.SupportsSse41)
{
context.Assembler.Pextrb(dest, src1, index);
}
else
{
context.Assembler.Pextrw(dest, src1, (byte)(index >> 1));
if ((index & 1) != 0)
{
context.Assembler.Shr(dest, new Operand(8), OperandType.I32);
}
else
{
context.Assembler.Movzx8(dest, dest, OperandType.I32);
}
}
}
private static void GenerateVectorInsert(CodeGenContext context, Operation operation)
{
Operand dest = operation.Destination;
Operand src1 = operation.GetSource(0); //Vector
Operand src2 = operation.GetSource(1); //Value
Operand src3 = operation.GetSource(2); //Index
if (!HardwareCapabilities.SupportsVexEncoding)
{
EnsureSameReg(dest, src1);
}
Debug.Assert(src1.Type == OperandType.V128);
Debug.Assert(src3.Kind == OperandKind.Constant);
byte index = src3.AsByte();
void InsertIntSse2(int words)
{
if (dest.GetRegister() != src1.GetRegister())
{
context.Assembler.Movdqu(dest, src1);
}
for (int word = 0; word < words; word++)
{
// Insert lower 16-bits.
context.Assembler.Pinsrw(dest, dest, src2, (byte)(index * words + word));
// Move next word down.
context.Assembler.Ror(src2, new Operand(16), src2.Type);
}
}
if (src2.Type == OperandType.I32)
{
Debug.Assert(index < 4);
if (HardwareCapabilities.SupportsSse41)
{
context.Assembler.Pinsrd(dest, src1, src2, index);
}
else
{
InsertIntSse2(2);
}
}
else if (src2.Type == OperandType.I64)
{
Debug.Assert(index < 2);
if (HardwareCapabilities.SupportsSse41)
{
context.Assembler.Pinsrq(dest, src1, src2, index);
}
else
{
InsertIntSse2(4);
}
}
else if (src2.Type == OperandType.FP32)
{
Debug.Assert(index < 4);
if (index != 0)
{
if (HardwareCapabilities.SupportsSse41)
{
context.Assembler.Insertps(dest, src1, src2, (byte)(index << 4));
}
else
{
if (src1.GetRegister() == src2.GetRegister())
{
int mask = 0b11_10_01_00;
mask &= ~(0b11 << index * 2);
context.Assembler.Pshufd(dest, src1, (byte)mask);
}
else
{
int mask0 = 0b11_10_01_00;
int mask1 = 0b11_10_01_00;
mask0 = BitUtils.RotateRight(mask0, index * 2, 8);
mask1 = BitUtils.RotateRight(mask1, 8 - index * 2, 8);
context.Assembler.Pshufd(src1, src1, (byte)mask0); // Lane to be inserted in position 0.
context.Assembler.Movss (dest, src1, src2); // dest[127:0] = src1[127:32] | src2[31:0]
context.Assembler.Pshufd(dest, dest, (byte)mask1); // Inserted lane in original position.
if (dest.GetRegister() != src1.GetRegister())
{
context.Assembler.Pshufd(src1, src1, (byte)mask1); // Restore src1.
}
}
}
}
else
{
context.Assembler.Movss(dest, src1, src2);
}
}
else /* if (src2.Type == OperandType.FP64) */
{
Debug.Assert(index < 2);
if (index != 0)
{
context.Assembler.Movlhps(dest, src1, src2);
}
else
{
context.Assembler.Movsd(dest, src1, src2);
}
}
}
private static void GenerateVectorInsert16(CodeGenContext context, Operation operation)
{
Operand dest = operation.Destination;
Operand src1 = operation.GetSource(0); //Vector
Operand src2 = operation.GetSource(1); //Value
Operand src3 = operation.GetSource(2); //Index
if (!HardwareCapabilities.SupportsVexEncoding)
{
EnsureSameReg(dest, src1);
}
Debug.Assert(src1.Type == OperandType.V128);
Debug.Assert(src3.Kind == OperandKind.Constant);
byte index = src3.AsByte();
context.Assembler.Pinsrw(dest, src1, src2, index);
}
private static void GenerateVectorInsert8(CodeGenContext context, Operation operation)
{
Operand dest = operation.Destination;
Operand src1 = operation.GetSource(0); //Vector
Operand src2 = operation.GetSource(1); //Value
Operand src3 = operation.GetSource(2); //Index
// It's not possible to emulate this instruction without
// SSE 4.1 support without the use of a temporary register,
// so we instead handle that case on the pre-allocator when
// SSE 4.1 is not supported on the CPU.
Debug.Assert(HardwareCapabilities.SupportsSse41);
if (!HardwareCapabilities.SupportsVexEncoding)
{
EnsureSameReg(dest, src1);
}
Debug.Assert(src1.Type == OperandType.V128);
Debug.Assert(src3.Kind == OperandKind.Constant);
byte index = src3.AsByte();
context.Assembler.Pinsrb(dest, src1, src2, index);
}
private static void GenerateVectorOne(CodeGenContext context, Operation operation)
{
Operand dest = operation.Destination;
Debug.Assert(!dest.Type.IsInteger());
context.Assembler.Pcmpeqw(dest, dest, dest);
}
private static void GenerateVectorZero(CodeGenContext context, Operation operation)
{
Operand dest = operation.Destination;
Debug.Assert(!dest.Type.IsInteger());
context.Assembler.Xorps(dest, dest, dest);
}
private static void GenerateVectorZeroUpper64(CodeGenContext context, Operation operation)
{
Operand dest = operation.Destination;
Operand source = operation.GetSource(0);
Debug.Assert(dest.Type == OperandType.V128 && source.Type == OperandType.V128);
GenerateZeroUpper64(context, dest, source);
}
private static void GenerateVectorZeroUpper96(CodeGenContext context, Operation operation)
{
Operand dest = operation.Destination;
Operand source = operation.GetSource(0);
Debug.Assert(dest.Type == OperandType.V128 && source.Type == OperandType.V128);
GenerateZeroUpper96(context, dest, source);
}
private static void GenerateZeroExtend16(CodeGenContext context, Operation operation)
{
Operand dest = operation.Destination;
Operand source = operation.GetSource(0);
Debug.Assert(dest.Type.IsInteger() && source.Type.IsInteger());
context.Assembler.Movzx16(dest, source, OperandType.I32);
}
private static void GenerateZeroExtend32(CodeGenContext context, Operation operation)
{
Operand dest = operation.Destination;
Operand source = operation.GetSource(0);
Debug.Assert(dest.Type.IsInteger() && source.Type.IsInteger());
context.Assembler.Mov(dest, source, OperandType.I32);
}
private static void GenerateZeroExtend8(CodeGenContext context, Operation operation)
{
Operand dest = operation.Destination;
Operand source = operation.GetSource(0);
Debug.Assert(dest.Type.IsInteger() && source.Type.IsInteger());
context.Assembler.Movzx8(dest, source, OperandType.I32);
}
private static void GenerateLoad(CodeGenContext context, Operand address, Operand value)
{
switch (value.Type)
{
case OperandType.I32: context.Assembler.Mov (value, address, OperandType.I32); break;
case OperandType.I64: context.Assembler.Mov (value, address, OperandType.I64); break;
case OperandType.FP32: context.Assembler.Movd (value, address); break;
case OperandType.FP64: context.Assembler.Movq (value, address); break;
case OperandType.V128: context.Assembler.Movdqu(value, address); break;
default: Debug.Assert(false); break;
}
}
private static void GenerateStore(CodeGenContext context, Operand address, Operand value)
{
switch (value.Type)
{
case OperandType.I32: context.Assembler.Mov (address, value, OperandType.I32); break;
case OperandType.I64: context.Assembler.Mov (address, value, OperandType.I64); break;
case OperandType.FP32: context.Assembler.Movd (address, value); break;
case OperandType.FP64: context.Assembler.Movq (address, value); break;
case OperandType.V128: context.Assembler.Movdqu(address, value); break;
default: Debug.Assert(false); break;
}
}
private static void GenerateZeroUpper64(CodeGenContext context, Operand dest, Operand source)
{
context.Assembler.Movq(dest, source);
}
private static void GenerateZeroUpper96(CodeGenContext context, Operand dest, Operand source)
{
context.Assembler.Movq(dest, source);
context.Assembler.Pshufd(dest, dest, 0xfc);
}
private static void ValidateUnOp(Operand dest, Operand source)
{
#if DEBUG
EnsureSameReg (dest, source);
EnsureSameType(dest, source);
#endif
}
private static void ValidateBinOp(Operand dest, Operand src1, Operand src2)
{
#if DEBUG
EnsureSameReg (dest, src1);
EnsureSameType(dest, src1, src2);
#endif
}
private static void ValidateShift(Operand dest, Operand src1, Operand src2)
{
#if DEBUG
EnsureSameReg (dest, src1);
EnsureSameType(dest, src1);
Debug.Assert(dest.Type.IsInteger() && src2.Type == OperandType.I32);
#endif
}
private static void EnsureSameReg(Operand op1, Operand op2)
{
if (!op1.Type.IsInteger() && HardwareCapabilities.SupportsVexEncoding)
{
return;
}
Debug.Assert(op1.Kind == OperandKind.Register || op1.Kind == OperandKind.Memory);
Debug.Assert(op1.Kind == op2.Kind);
Debug.Assert(op1.Value == op2.Value);
}
private static void EnsureSameType(Operand op1, Operand op2)
{
Debug.Assert(op1.Type == op2.Type);
}
private static void EnsureSameType(Operand op1, Operand op2, Operand op3)
{
Debug.Assert(op1.Type == op2.Type);
Debug.Assert(op1.Type == op3.Type);
}
private static void EnsureSameType(Operand op1, Operand op2, Operand op3, Operand op4)
{
Debug.Assert(op1.Type == op2.Type);
Debug.Assert(op1.Type == op3.Type);
Debug.Assert(op1.Type == op4.Type);
}
private static UnwindInfo WritePrologue(CodeGenContext context)
{
List<UnwindPushEntry> pushEntries = new List<UnwindPushEntry>();
Operand rsp = Register(X86Register.Rsp);
int mask = CallingConvention.GetIntCalleeSavedRegisters() & context.AllocResult.IntUsedRegisters;
while (mask != 0)
{
int bit = BitUtils.LowestBitSet(mask);
context.Assembler.Push(Register((X86Register)bit));
pushEntries.Add(new UnwindPushEntry(bit, RegisterType.Integer, context.StreamOffset));
mask &= ~(1 << bit);
}
int reservedStackSize = context.CallArgsRegionSize + context.AllocResult.SpillRegionSize;
reservedStackSize += context.XmmSaveRegionSize;
if (reservedStackSize >= StackGuardSize)
{
GenerateInlineStackProbe(context, reservedStackSize);
}
if (reservedStackSize != 0)
{
context.Assembler.Sub(rsp, new Operand(reservedStackSize), OperandType.I64);
}
int offset = reservedStackSize;
mask = CallingConvention.GetVecCalleeSavedRegisters() & context.AllocResult.VecUsedRegisters;
while (mask != 0)
{
int bit = BitUtils.LowestBitSet(mask);
offset -= 16;
MemoryOperand memOp = new MemoryOperand(OperandType.V128, rsp, null, Multiplier.x1, offset);
context.Assembler.Movdqu(memOp, Xmm((X86Register)bit));
pushEntries.Add(new UnwindPushEntry(bit, RegisterType.Vector, context.StreamOffset));
mask &= ~(1 << bit);
}
return new UnwindInfo(pushEntries.ToArray(), context.StreamOffset, reservedStackSize);
}
private static void WriteEpilogue(CodeGenContext context)
{
Operand rsp = Register(X86Register.Rsp);
int reservedStackSize = context.CallArgsRegionSize + context.AllocResult.SpillRegionSize;
reservedStackSize += context.XmmSaveRegionSize;
int offset = reservedStackSize;
int mask = CallingConvention.GetVecCalleeSavedRegisters() & context.AllocResult.VecUsedRegisters;
while (mask != 0)
{
int bit = BitUtils.LowestBitSet(mask);
offset -= 16;
MemoryOperand memOp = new MemoryOperand(OperandType.V128, rsp, null, Multiplier.x1, offset);
context.Assembler.Movdqu(Xmm((X86Register)bit), memOp);
mask &= ~(1 << bit);
}
if (reservedStackSize != 0)
{
context.Assembler.Add(rsp, new Operand(reservedStackSize), OperandType.I64);
}
mask = CallingConvention.GetIntCalleeSavedRegisters() & context.AllocResult.IntUsedRegisters;
while (mask != 0)
{
int bit = BitUtils.HighestBitSet(mask);
context.Assembler.Pop(Register((X86Register)bit));
mask &= ~(1 << bit);
}
}
private static void GenerateInlineStackProbe(CodeGenContext context, int size)
{
// Windows does lazy stack allocation, and there are just 2
// guard pages on the end of the stack. So, if the allocation
// size we make is greater than this guard size, we must ensure
// that the OS will map all pages that we'll use. We do that by
// doing a dummy read on those pages, forcing a page fault and
// the OS to map them. If they are already mapped, nothing happens.
const int pageMask = PageSize - 1;
size = (size + pageMask) & ~pageMask;
Operand rsp = Register(X86Register.Rsp);
Operand temp = Register(CallingConvention.GetIntReturnRegister());
for (int offset = PageSize; offset < size; offset += PageSize)
{
Operand memOp = new MemoryOperand(OperandType.I32, rsp, null, Multiplier.x1, -offset);
Add a new JIT compiler for CPU code (#693) * Start of the ARMeilleure project * Refactoring around the old IRAdapter, now renamed to PreAllocator * Optimize the LowestBitSet method * Add CLZ support and fix CLS implementation * Add missing Equals and GetHashCode overrides on some structs, misc small tweaks * Implement the ByteSwap IR instruction, and some refactoring on the assembler * Implement the DivideUI IR instruction and fix 64-bits IDIV * Correct constant operand type on CSINC * Move division instructions implementation to InstEmitDiv * Fix destination type for the ConditionalSelect IR instruction * Implement UMULH and SMULH, with new IR instructions * Fix some issues with shift instructions * Fix constant types for BFM instructions * Fix up new tests using the new V128 struct * Update tests * Move DIV tests to a separate file * Add support for calls, and some instructions that depends on them * Start adding support for SIMD & FP types, along with some of the related ARM instructions * Fix some typos and the divide instruction with FP operands * Fix wrong method call on Clz_V * Implement ARM FP & SIMD move instructions, Saddlv_V, and misc. fixes * Implement SIMD logical instructions and more misc. fixes * Fix PSRAD x86 instruction encoding, TRN, UABD and UABDL implementations * Implement float conversion instruction, merge in LDj3SNuD fixes, and some other misc. fixes * Implement SIMD shift instruction and fix Dup_V * Add SCVTF and UCVTF (vector, fixed-point) variants to the opcode table * Fix check with tolerance on tester * Implement FP & SIMD comparison instructions, and some fixes * Update FCVT (Scalar) encoding on the table to support the Half-float variants * Support passing V128 structs, some cleanup on the register allocator, merge LDj3SNuD fixes * Use old memory access methods, made a start on SIMD memory insts support, some fixes * Fix float constant passed to functions, save and restore non-volatile XMM registers, other fixes * Fix arguments count with struct return values, other fixes * More instructions * Misc. fixes and integrate LDj3SNuD fixes * Update tests * Add a faster linear scan allocator, unwinding support on windows, and other changes * Update Ryujinx.HLE * Update Ryujinx.Graphics * Fix V128 return pointer passing, RCX is clobbered * Update Ryujinx.Tests * Update ITimeZoneService * Stop using GetFunctionPointer as that can't be called from native code, misc. fixes and tweaks * Use generic GetFunctionPointerForDelegate method and other tweaks * Some refactoring on the code generator, assert on invalid operations and use a separate enum for intrinsics * Remove some unused code on the assembler * Fix REX.W prefix regression on float conversion instructions, add some sort of profiler * Add hardware capability detection * Fix regression on Sha1h and revert Fcm** changes * Add SSE2-only paths on vector extract and insert, some refactoring on the pre-allocator * Fix silly mistake introduced on last commit on CpuId * Generate inline stack probes when the stack allocation is too large * Initial support for the System-V ABI * Support multiple destination operands * Fix SSE2 VectorInsert8 path, and other fixes * Change placement of XMM callee save and restore code to match other compilers * Rename Dest to Destination and Inst to Instruction * Fix a regression related to calls and the V128 type * Add an extra space on comments to match code style * Some refactoring * Fix vector insert FP32 SSE2 path * Port over the ARM32 instructions * Avoid memory protection races on JIT Cache * Another fix on VectorInsert FP32 (thanks to LDj3SNuD * Float operands don't need to use the same register when VEX is supported * Add a new register allocator, higher quality code for hot code (tier up), and other tweaks * Some nits, small improvements on the pre allocator * CpuThreadState is gone * Allow changing CPU emulators with a config entry * Add runtime identifiers on the ARMeilleure project * Allow switching between CPUs through a config entry (pt. 2) * Change win10-x64 to win-x64 on projects * Update the Ryujinx project to use ARMeilleure * Ensure that the selected register is valid on the hybrid allocator * Allow exiting on returns to 0 (should fix test regression) * Remove register assignments for most used variables on the hybrid allocator * Do not use fixed registers as spill temp * Add missing namespace and remove unneeded using * Address PR feedback * Fix types, etc * Enable AssumeStrictAbiCompliance by default * Ensure that Spill and Fill don't load or store any more than necessary
2019-08-08 20:56:22 +02:00
context.Assembler.Mov(temp, memOp, OperandType.I32);
}
}
private static MemoryOperand Memory(Operand operand, OperandType type)
{
if (operand.Kind == OperandKind.Memory)
{
return operand as MemoryOperand;
}
return new MemoryOperand(type, operand);
}
private static Operand Register(X86Register register, OperandType type = OperandType.I64)
{
return new Operand((int)register, RegisterType.Integer, type);
}
private static Operand Xmm(X86Register register)
{
return new Operand((int)register, RegisterType.Vector, OperandType.V128);
}
}
}