Ryujinx/ChocolArm64/Decoders/Decoder.cs
gdkchan a694420d11
Implement speculative translation on the CPU (#515)
* Implement speculative translation on the cpu, and change the way how branches to unknown or untranslated addresses works

* Port t0opt changes and other cleanups

* Change namespace from translation related classes to ChocolArm64.Translation, other minor tweaks

* Fix typo

* Translate higher quality code for indirect jumps aswell, and on some cases that were missed when lower quality (tier 0) code was available

* Remove debug print

* Remove direct argument passing optimization, and enable tail calls for BR instructions

* Call delegates directly with Callvirt rather than calling Execute, do not emit calls for tier 0 code

* Remove unused property

* Rename argument on ArmSubroutine delegate
2019-02-04 18:26:05 -03:00

333 lines
11 KiB
C#

using ChocolArm64.Instructions;
using ChocolArm64.Memory;
using ChocolArm64.State;
using System;
using System.Collections.Concurrent;
using System.Collections.Generic;
using System.Reflection.Emit;
namespace ChocolArm64.Decoders
{
static class Decoder
{
private delegate object OpActivator(Inst inst, long position, int opCode);
private static ConcurrentDictionary<Type, OpActivator> _opActivators;
static Decoder()
{
_opActivators = new ConcurrentDictionary<Type, OpActivator>();
}
public static Block DecodeBasicBlock(MemoryManager memory, long start, ExecutionMode mode)
{
Block block = new Block(start);
FillBlock(memory, mode, block);
OpCode64 lastOp = block.GetLastOp();
if (IsBranch(lastOp) && !IsCall(lastOp) && lastOp is IOpCodeBImm op)
{
//It's possible that the branch on this block lands on the middle of the block.
//This is more common on tight loops. In this case, we can improve the codegen
//a bit by changing the CFG and either making the branch point to the same block
//(which indicates that the block is a loop that jumps back to the start), and the
//other possible case is a jump somewhere on the middle of the block, which is
//also a loop, but in this case we need to split the block in half.
if (op.Imm == start)
{
block.Branch = block;
}
else if ((ulong)op.Imm > (ulong)start &&
(ulong)op.Imm < (ulong)block.EndPosition)
{
Block botBlock = new Block(op.Imm);
int botBlockIndex = 0;
long currPosition = start;
while ((ulong)currPosition < (ulong)op.Imm)
{
currPosition += block.OpCodes[botBlockIndex++].OpCodeSizeInBytes;
}
botBlock.OpCodes.AddRange(block.OpCodes);
botBlock.OpCodes.RemoveRange(0, botBlockIndex);
block.OpCodes.RemoveRange(botBlockIndex, block.OpCodes.Count - botBlockIndex);
botBlock.EndPosition = block.EndPosition;
block.EndPosition = op.Imm;
botBlock.Branch = botBlock;
block.Next = botBlock;
}
}
return block;
}
public static Block DecodeSubroutine(MemoryManager memory, long start, ExecutionMode mode)
{
Dictionary<long, Block> visited = new Dictionary<long, Block>();
Dictionary<long, Block> visitedEnd = new Dictionary<long, Block>();
Queue<Block> blocks = new Queue<Block>();
Block Enqueue(long position)
{
if (!visited.TryGetValue(position, out Block output))
{
output = new Block(position);
blocks.Enqueue(output);
visited.Add(position, output);
}
return output;
}
Block entry = Enqueue(start);
while (blocks.Count > 0)
{
Block current = blocks.Dequeue();
FillBlock(memory, mode, current);
//Set child blocks. "Branch" is the block the branch instruction
//points to (when taken), "Next" is the block at the next address,
//executed when the branch is not taken. For Unconditional Branches
//(except BL/BLR that are sub calls) or end of executable, Next is null.
if (current.OpCodes.Count > 0)
{
OpCode64 lastOp = current.GetLastOp();
bool isCall = IsCall(lastOp);
if (lastOp is IOpCodeBImm op && !isCall)
{
current.Branch = Enqueue(op.Imm);
}
if (!IsUnconditionalBranch(lastOp) || isCall)
{
current.Next = Enqueue(current.EndPosition);
}
}
//If we have on the graph two blocks with the same end position,
//then we need to split the bigger block and have two small blocks,
//the end position of the bigger "Current" block should then be == to
//the position of the "Smaller" block.
while (visitedEnd.TryGetValue(current.EndPosition, out Block smaller))
{
if (current.Position > smaller.Position)
{
Block temp = smaller;
smaller = current;
current = temp;
}
current.EndPosition = smaller.Position;
current.Next = smaller;
current.Branch = null;
current.OpCodes.RemoveRange(
current.OpCodes.Count - smaller.OpCodes.Count,
smaller.OpCodes.Count);
visitedEnd[smaller.EndPosition] = smaller;
}
visitedEnd.Add(current.EndPosition, current);
}
return entry;
}
private static void FillBlock(MemoryManager memory, ExecutionMode mode, Block block)
{
long position = block.Position;
OpCode64 opCode;
do
{
opCode = DecodeOpCode(memory, position, mode);
block.OpCodes.Add(opCode);
position += opCode.OpCodeSizeInBytes;
}
while (!(IsBranch(opCode) || IsException(opCode)));
block.EndPosition = position;
}
private static bool IsBranch(OpCode64 opCode)
{
return opCode is OpCodeBImm64 ||
opCode is OpCodeBReg64 || IsAarch32Branch(opCode);
}
private static bool IsUnconditionalBranch(OpCode64 opCode)
{
return opCode is OpCodeBImmAl64 ||
opCode is OpCodeBReg64 || IsAarch32UnconditionalBranch(opCode);
}
private static bool IsAarch32UnconditionalBranch(OpCode64 opCode)
{
if (!(opCode is OpCode32 op))
{
return false;
}
//Note: On ARM32, most instructions have conditional execution,
//so there's no "Always" (unconditional) branch like on ARM64.
//We need to check if the condition is "Always" instead.
return IsAarch32Branch(op) && op.Cond >= Condition.Al;
}
private static bool IsAarch32Branch(OpCode64 opCode)
{
//Note: On ARM32, most ALU operations can write to R15 (PC),
//so we must consider such operations as a branch in potential aswell.
if (opCode is IOpCode32Alu opAlu && opAlu.Rd == RegisterAlias.Aarch32Pc)
{
return true;
}
//Same thing for memory operations. We have the cases where PC is a target
//register (Rt == 15 or (mask & (1 << 15)) != 0), and cases where there is
//a write back to PC (wback == true && Rn == 15), however the later may
//be "undefined" depending on the CPU, so compilers should not produce that.
if (opCode is IOpCode32Mem || opCode is IOpCode32MemMult)
{
int rt, rn;
bool wBack, isLoad;
if (opCode is IOpCode32Mem opMem)
{
rt = opMem.Rt;
rn = opMem.Rn;
wBack = opMem.WBack;
isLoad = opMem.IsLoad;
//For the dual load, we also need to take into account the
//case were Rt2 == 15 (PC).
if (rt == 14 && opMem.Emitter == InstEmit32.Ldrd)
{
rt = RegisterAlias.Aarch32Pc;
}
}
else if (opCode is IOpCode32MemMult opMemMult)
{
const int pcMask = 1 << RegisterAlias.Aarch32Pc;
rt = (opMemMult.RegisterMask & pcMask) != 0 ? RegisterAlias.Aarch32Pc : 0;
rn = opMemMult.Rn;
wBack = opMemMult.PostOffset != 0;
isLoad = opMemMult.IsLoad;
}
else
{
throw new NotImplementedException($"The type \"{opCode.GetType().Name}\" is not implemented on the decoder.");
}
if ((rt == RegisterAlias.Aarch32Pc && isLoad) ||
(rn == RegisterAlias.Aarch32Pc && wBack))
{
return true;
}
}
//Explicit branch instructions.
return opCode is IOpCode32BImm ||
opCode is IOpCode32BReg;
}
private static bool IsCall(OpCode64 opCode)
{
//TODO (CQ): ARM32 support.
return opCode.Emitter == InstEmit.Bl ||
opCode.Emitter == InstEmit.Blr;
}
private static bool IsException(OpCode64 opCode)
{
return opCode.Emitter == InstEmit.Brk ||
opCode.Emitter == InstEmit.Svc ||
opCode.Emitter == InstEmit.Und;
}
public static OpCode64 DecodeOpCode(MemoryManager memory, long position, ExecutionMode mode)
{
int opCode = memory.ReadInt32(position);
Inst inst;
if (mode == ExecutionMode.Aarch64)
{
inst = OpCodeTable.GetInstA64(opCode);
}
else
{
if (mode == ExecutionMode.Aarch32Arm)
{
inst = OpCodeTable.GetInstA32(opCode);
}
else /* if (mode == ExecutionMode.Aarch32Thumb) */
{
inst = OpCodeTable.GetInstT32(opCode);
}
}
OpCode64 decodedOpCode = new OpCode64(Inst.Undefined, position, opCode);
if (inst.Type != null)
{
decodedOpCode = MakeOpCode(inst.Type, inst, position, opCode);
}
return decodedOpCode;
}
private static OpCode64 MakeOpCode(Type type, Inst inst, long position, int opCode)
{
if (type == null)
{
throw new ArgumentNullException(nameof(type));
}
OpActivator createInstance = _opActivators.GetOrAdd(type, CacheOpActivator);
return (OpCode64)createInstance(inst, position, opCode);
}
private static OpActivator CacheOpActivator(Type type)
{
Type[] argTypes = new Type[] { typeof(Inst), typeof(long), typeof(int) };
DynamicMethod mthd = new DynamicMethod($"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));
}
}
}