0
0
Fork 0
mirror of https://github.com/GreemDev/Ryujinx.git synced 2024-12-23 10:25:47 +00:00
Ryujinx/ChocolArm64/Translation/Translator.cs
gdkchan e21ebbf666 Misc. CPU optimizations (#575)
* Add optimizations related to caller/callee saved registers, thread synchronization and disable tier 0

* Refactoring

* Add a config entry to enable or disable the reg load/store opt.

* Remove unnecessary register state stores for calls when the callee is know

* Rename IoType to VarType

* Enable tier 0 while fixing some perf issues related to tier 0

* Small tweak -- Compile before adding to the cache, to avoid lags

* Add required config entry
2019-02-28 13:03:31 +11:00

184 lines
No EOL
5.5 KiB
C#

using ChocolArm64.Decoders;
using ChocolArm64.Events;
using ChocolArm64.Memory;
using ChocolArm64.State;
using System;
using System.Threading;
namespace ChocolArm64.Translation
{
public class Translator
{
private MemoryManager _memory;
private CpuThreadState _dummyThreadState;
private TranslatorCache _cache;
private TranslatorQueue _queue;
private Thread _backgroundTranslator;
public event EventHandler<CpuTraceEventArgs> CpuTrace;
public bool EnableCpuTrace { get; set; }
private volatile int _threadCount;
public Translator(MemoryManager memory)
{
_memory = memory;
_dummyThreadState = new CpuThreadState();
_dummyThreadState.Running = false;
_cache = new TranslatorCache();
_queue = new TranslatorQueue();
}
internal void ExecuteSubroutine(CpuThread thread, long position)
{
if (Interlocked.Increment(ref _threadCount) == 1)
{
_backgroundTranslator = new Thread(TranslateQueuedSubs);
_backgroundTranslator.Start();
}
ExecuteSubroutine(thread.ThreadState, position);
if (Interlocked.Decrement(ref _threadCount) == 0)
{
_queue.ForceSignal();
}
}
private void ExecuteSubroutine(CpuThreadState state, long position)
{
state.CurrentTranslator = this;
do
{
if (EnableCpuTrace)
{
CpuTrace?.Invoke(this, new CpuTraceEventArgs(position));
}
if (!_cache.TryGetSubroutine(position, out TranslatedSub sub))
{
sub = TranslateLowCq(position, state.GetExecutionMode());
}
position = sub.Execute(state, _memory);
}
while (position != 0 && state.Running);
state.CurrentTranslator = null;
}
internal ArmSubroutine GetOrTranslateSubroutine(CpuThreadState state, long position, CallType cs)
{
if (!_cache.TryGetSubroutine(position, out TranslatedSub sub))
{
sub = TranslateLowCq(position, state.GetExecutionMode());
}
if (sub.IsWorthOptimizing())
{
bool isComplete = cs == CallType.Call ||
cs == CallType.VirtualCall;
_queue.Enqueue(position, state.GetExecutionMode(), TranslationTier.Tier1, isComplete);
}
return sub.Delegate;
}
private void TranslateQueuedSubs()
{
while (_threadCount != 0)
{
if (_queue.TryDequeue(out TranslatorQueueItem item))
{
bool isCached = _cache.TryGetSubroutine(item.Position, out TranslatedSub sub);
if (isCached && item.Tier <= sub.Tier)
{
continue;
}
if (item.Tier == TranslationTier.Tier0)
{
TranslateLowCq(item.Position, item.Mode);
}
else
{
TranslateHighCq(item.Position, item.Mode, item.IsComplete);
}
}
else
{
_queue.WaitForItems();
}
}
}
private TranslatedSub TranslateLowCq(long position, ExecutionMode mode)
{
Block block = Decoder.DecodeBasicBlock(_memory, position, mode);
ILEmitterCtx context = new ILEmitterCtx(_memory, _cache, _queue, TranslationTier.Tier0, block);
string subName = GetSubroutineName(position);
bool isAarch64 = mode == ExecutionMode.Aarch64;
ILMethodBuilder ilMthdBuilder = new ILMethodBuilder(context.GetILBlocks(), subName, isAarch64);
TranslatedSub subroutine = ilMthdBuilder.GetSubroutine(TranslationTier.Tier0, isWorthOptimizing: true);
return _cache.GetOrAdd(position, subroutine, block.OpCodes.Count);
}
private TranslatedSub TranslateHighCq(long position, ExecutionMode mode, bool isComplete)
{
Block graph = Decoder.DecodeSubroutine(_memory, position, mode);
ILEmitterCtx context = new ILEmitterCtx(_memory, _cache, _queue, TranslationTier.Tier1, graph);
ILBlock[] ilBlocks = context.GetILBlocks();
string subName = GetSubroutineName(position);
bool isAarch64 = mode == ExecutionMode.Aarch64;
isComplete &= !context.HasIndirectJump;
ILMethodBuilder ilMthdBuilder = new ILMethodBuilder(ilBlocks, subName, isAarch64, isComplete);
TranslatedSub subroutine = ilMthdBuilder.GetSubroutine(TranslationTier.Tier1, context.HasSlowCall);
int ilOpCount = 0;
foreach (ILBlock ilBlock in ilBlocks)
{
ilOpCount += ilBlock.Count;
}
ForceAheadOfTimeCompilation(subroutine);
_cache.AddOrUpdate(position, subroutine, ilOpCount);
return subroutine;
}
private string GetSubroutineName(long position)
{
return $"Sub{position:x16}";
}
private void ForceAheadOfTimeCompilation(TranslatedSub subroutine)
{
subroutine.Execute(_dummyThreadState, null);
}
}
}