mirror of
https://github.com/Ryujinx/Ryujinx.git
synced 2024-08-20 07:38:42 +02:00
* Implement a new physical memory manager and replace DeviceMemory * Proper generic constraints * Fix debug build * Add memory tests * New CPU memory manager and general code cleanup * Remove host memory management from CPU project, use Ryujinx.Memory instead * Fix tests * Document exceptions on MemoryBlock * Fix leak on unix memory allocation * Proper disposal of some objects on tests * Fix JitCache not being set as initialized * GetRef without checks for 8-bits and 16-bits CAS * Add MemoryBlock destructor * Throw in separate method to improve codegen * Address PR feedback * QueryModified improvements * Fix memory write tracking not marking all pages as modified in some cases * Simplify MarkRegionAsModified * Remove XML doc for ghost param * Add back optimization to avoid useless buffer updates * Add Ryujinx.Cpu project, move MemoryManager there and remove MemoryBlockWrapper * Some nits * Do not perform address translation when size is 0 * Address PR feedback and format NativeInterface class * Remove ghost parameter description * Update Ryujinx.Cpu to .NET Core 3.1 * Address PR feedback * Fix build * Return a well defined value for GetPhysicalAddress with invalid VA, and do not return unmapped ranges as modified * Typo
339 lines
11 KiB
C#
339 lines
11 KiB
C#
using ARMeilleure.Decoders.Optimizations;
|
|
using ARMeilleure.Instructions;
|
|
using ARMeilleure.Memory;
|
|
using ARMeilleure.State;
|
|
using System;
|
|
using System.Collections.Generic;
|
|
|
|
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;
|
|
|
|
public static Block[] DecodeBasicBlock(IMemoryManager memory, ulong address, ExecutionMode mode)
|
|
{
|
|
Block block = new Block(address);
|
|
|
|
FillBlock(memory, mode, block, ulong.MaxValue);
|
|
|
|
return new Block[] { block };
|
|
}
|
|
|
|
public static Block[] DecodeFunction(IMemoryManager 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(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(
|
|
IMemoryManager 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(IMemoryManager memory, ulong address, ExecutionMode mode)
|
|
{
|
|
int opCode = memory.Read<int>(address);
|
|
|
|
InstDescriptor inst;
|
|
|
|
OpCodeTable.MakeOp makeOp;
|
|
|
|
if (mode == ExecutionMode.Aarch64)
|
|
{
|
|
(inst, makeOp) = OpCodeTable.GetInstA64(opCode);
|
|
}
|
|
else
|
|
{
|
|
if (mode == ExecutionMode.Aarch32Arm)
|
|
{
|
|
(inst, makeOp) = OpCodeTable.GetInstA32(opCode);
|
|
}
|
|
else /* if (mode == ExecutionMode.Aarch32Thumb) */
|
|
{
|
|
(inst, makeOp) = OpCodeTable.GetInstT32(opCode);
|
|
}
|
|
}
|
|
|
|
if (makeOp != null)
|
|
{
|
|
return (OpCode)makeOp(inst, address, opCode);
|
|
}
|
|
else
|
|
{
|
|
return new OpCode(inst, address, opCode);
|
|
}
|
|
}
|
|
}
|
|
} |