Ryujinx/ARMeilleure/Decoders/Decoder.cs
riperiperi d904706fc0
Use a Jump Table for direct and indirect calls/jumps, removing transitions to managed (#975)
* Implement Jump Table for Native Calls

NOTE: this slows down rejit considerably! Not recommended to be used
without codegen optimisation or AOT.

- Does not work on Linux
- A32 needs an additional commit.

* A32 Support

(WIP)

* Actually write Direct Call pointers to the table

That would help.

* Direct Calls: Rather than returning to the translator, attempt to keep within the native stack frame.

A return to the translator can still happen, but only by exceptionally
bubbling up to it.

Also:
- Always translate lowCq as a function. Faster interop with the direct
jumps, and this will be useful in future if we want to do speculative
translation.
- Tail Call Detection: after the decoding stage, detect if we do a tail
call, and avoid translating into it. Detected if a jump is made to an
address outwith the contiguous sequence of blocks surrounding the entry
point. The goal is to reduce code touched by jit and rejit.

* A32 Support

* Use smaller max function size for lowCq, fix exceptional returns

When a return has an unexpected value and there is no code block
following this one, we now return the value rather than continuing.

* CompareAndSwap (buggy)

* Ensure CompareAndSwap does not get optimized away.

* Use CompareAndSwap to make the dynamic table thread safe.

* Tail call for linux, throw on too many arguments.

* Combine CompareAndSwap 128 and 32/64.

They emit different IR instructions since their PreAllocator behaviour
is different, but now they just have one function on EmitterContext.

* Fix issues separating from optimisations.

* Use a stub to find and execute missing functions.

This allows us to skip doing many runtime comparisons and branches, and reduces the amount of code we need to emit significantly.

For the indirect call table, this stub also does the work of moving in the highCq address to the table when one is found.

* Make Jump Tables and Jit Cache dynmically resize

Reserve virtual memory, commit as needed.

* Move TailCallRemover to its own class.

* Multithreaded Translation (based on heuristic)

A poor one, at that. Need to get core count for a better one, which
means a lot of OS specific garbage.

* Better priority management for background threads.

* Bound core limit a bit more

Past a certain point the load is not paralellizable and starts stealing from the main thread. Likely due to GC, memory, heap allocation thread contention. Reduce by one core til optimisations come to improve the situation.

* Fix memory management on linux.

* Temporary solution to some sync problems.

This will make sure threads exit correctly, most of the time. There is a potential race where setting the sync counter to 0 does nothing (counter stays at what it was before, thread could take too long to exit), but we need to find a better way to do this anyways. Synchronization frequency has been tightened as we never enter blockwise segments of code. Essentially this means, check every x functions or loop iterations, before lowcq blocks existed and were worth just as much. Ideally it should be done in a better way, since functions can be anywhere from 1 to 5000 instructions. (maybe based on host timer, or an interrupt flag from a scheduler thread)

* Address feedback minus CompareAndSwap change.

* Use default ReservedRegion granularity.

* Merge CompareAndSwap with its V128 variant.

* We already got the source, no need to do it again.

* Make sure all background translation threads exit.

* Fix CompareAndSwap128

Detection criteria was a bit scuffed.

* Address Comments.
2020-03-12 14:20:55 +11:00

374 lines
12 KiB
C#

using ARMeilleure.Decoders.Optimizations;
using ARMeilleure.Instructions;
using ARMeilleure.Memory;
using ARMeilleure.State;
using System;
using System.Collections.Concurrent;
using System.Collections.Generic;
using System.Reflection.Emit;
namespace ARMeilleure.Decoders
{
static class Decoder
{
// We define a limit on the number of instructions that a function may have,
// this prevents functions being potentially too large, which would
// take too long to compile and use too much memory.
private const int MaxInstsPerFunction = 5000;
// For lower code quality translation, we set a lower limit since we're blocking execution.
private const int MaxInstsPerFunctionLowCq = 500;
private delegate object MakeOp(InstDescriptor inst, ulong address, int opCode);
private static ConcurrentDictionary<Type, MakeOp> _opActivators;
static Decoder()
{
_opActivators = new ConcurrentDictionary<Type, MakeOp>();
}
public static Block[] DecodeBasicBlock(MemoryManager memory, ulong address, ExecutionMode mode)
{
Block block = new Block(address);
FillBlock(memory, mode, block, ulong.MaxValue);
return new Block[] { block };
}
public static Block[] DecodeFunction(MemoryManager memory, ulong address, ExecutionMode mode, bool highCq)
{
List<Block> blocks = new List<Block>();
Queue<Block> workQueue = new Queue<Block>();
Dictionary<ulong, Block> visited = new Dictionary<ulong, Block>();
int opsCount = 0;
int instructionLimit = highCq ? MaxInstsPerFunction : MaxInstsPerFunctionLowCq;
Block GetBlock(ulong blkAddress)
{
if (!visited.TryGetValue(blkAddress, out Block block))
{
if (opsCount > instructionLimit || !memory.IsMapped((long)blkAddress))
{
return null;
}
block = new Block(blkAddress);
workQueue.Enqueue(block);
visited.Add(blkAddress, block);
}
return block;
}
GetBlock(address);
while (workQueue.TryDequeue(out Block currBlock))
{
// Check if the current block is inside another block.
if (BinarySearch(blocks, currBlock.Address, out int nBlkIndex))
{
Block nBlock = blocks[nBlkIndex];
if (nBlock.Address == currBlock.Address)
{
throw new InvalidOperationException("Found duplicate block address on the list.");
}
nBlock.Split(currBlock);
blocks.Insert(nBlkIndex + 1, currBlock);
continue;
}
// If we have a block after the current one, set the limit address.
ulong limitAddress = ulong.MaxValue;
if (nBlkIndex != blocks.Count)
{
Block nBlock = blocks[nBlkIndex];
int nextIndex = nBlkIndex + 1;
if (nBlock.Address < currBlock.Address && nextIndex < blocks.Count)
{
limitAddress = blocks[nextIndex].Address;
}
else if (nBlock.Address > currBlock.Address)
{
limitAddress = blocks[nBlkIndex].Address;
}
}
FillBlock(memory, mode, currBlock, limitAddress);
opsCount += currBlock.OpCodes.Count;
if (currBlock.OpCodes.Count != 0)
{
// 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.
OpCode lastOp = currBlock.GetLastOp();
bool isCall = IsCall(lastOp);
if (lastOp is IOpCodeBImm op && !isCall)
{
currBlock.Branch = GetBlock((ulong)op.Immediate);
}
if (!IsUnconditionalBranch(lastOp) || isCall)
{
currBlock.Next = GetBlock(currBlock.EndAddress);
}
}
// Insert the new block on the list (sorted by address).
if (blocks.Count != 0)
{
Block nBlock = blocks[nBlkIndex];
blocks.Insert(nBlkIndex + (nBlock.Address < currBlock.Address ? 1 : 0), currBlock);
}
else
{
blocks.Add(currBlock);
}
}
TailCallRemover.RunPass(address, blocks);
return blocks.ToArray();
}
public static bool BinarySearch(List<Block> blocks, ulong address, out int index)
{
index = 0;
int left = 0;
int right = blocks.Count - 1;
while (left <= right)
{
int size = right - left;
int middle = left + (size >> 1);
Block block = blocks[middle];
index = middle;
if (address >= block.Address && address < block.EndAddress)
{
return true;
}
if (address < block.Address)
{
right = middle - 1;
}
else
{
left = middle + 1;
}
}
return false;
}
private static void FillBlock(
MemoryManager memory,
ExecutionMode mode,
Block block,
ulong limitAddress)
{
ulong address = block.Address;
OpCode opCode;
do
{
if (address >= limitAddress)
{
break;
}
opCode = DecodeOpCode(memory, address, mode);
block.OpCodes.Add(opCode);
address += (ulong)opCode.OpCodeSizeInBytes;
}
while (!(IsBranch(opCode) || IsException(opCode)));
block.EndAddress = address;
}
private static bool IsBranch(OpCode opCode)
{
return opCode is OpCodeBImm ||
opCode is OpCodeBReg || IsAarch32Branch(opCode);
}
private static bool IsUnconditionalBranch(OpCode opCode)
{
return opCode is OpCodeBImmAl ||
opCode is OpCodeBReg || IsAarch32UnconditionalBranch(opCode);
}
private static bool IsAarch32UnconditionalBranch(OpCode 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(OpCode 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.Instruction.Name == InstName.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($"The type \"{opCode.GetType().Name}\" is not implemented on the decoder.");
}
if ((rt == RegisterAlias.Aarch32Pc && isLoad) ||
(rn == RegisterAlias.Aarch32Pc && wBack))
{
return true;
}
}
// Explicit branch instructions.
return opCode is IOpCode32BImm ||
opCode is IOpCode32BReg;
}
private static bool IsCall(OpCode opCode)
{
return opCode.Instruction.Name == InstName.Bl ||
opCode.Instruction.Name == InstName.Blr ||
opCode.Instruction.Name == InstName.Blx;
}
private static bool IsException(OpCode opCode)
{
return opCode.Instruction.Name == InstName.Brk ||
opCode.Instruction.Name == InstName.Svc ||
opCode.Instruction.Name == InstName.Trap ||
opCode.Instruction.Name == InstName.Und;
}
public static OpCode DecodeOpCode(MemoryManager memory, ulong address, ExecutionMode mode)
{
int opCode = memory.ReadInt32((long)address);
InstDescriptor inst;
Type type;
if (mode == ExecutionMode.Aarch64)
{
(inst, type) = OpCodeTable.GetInstA64(opCode);
}
else
{
if (mode == ExecutionMode.Aarch32Arm)
{
(inst, type) = OpCodeTable.GetInstA32(opCode);
}
else /* if (mode == ExecutionMode.Aarch32Thumb) */
{
(inst, type) = OpCodeTable.GetInstT32(opCode);
}
}
if (type != null)
{
return MakeOpCode(inst, type, address, opCode);
}
else
{
return new OpCode(inst, address, opCode);
}
}
private static OpCode MakeOpCode(InstDescriptor inst, Type type, ulong address, int opCode)
{
MakeOp createInstance = _opActivators.GetOrAdd(type, CacheOpActivator);
return (OpCode)createInstance(inst, address, opCode);
}
private static MakeOp CacheOpActivator(Type type)
{
Type[] argTypes = new Type[] { typeof(InstDescriptor), typeof(ulong), typeof(int) };
DynamicMethod mthd = new DynamicMethod($"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 (MakeOp)mthd.CreateDelegate(typeof(MakeOp));
}
}
}