drashna/ryujinx
1
0
Fork 0
Code Activity Actions

Implement speculative translation on the CPU (#515)

Browse source 
* 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
This commit is contained in:
gdkchan 2019-02-04 18:26:05 -03:00 • committed by GitHub
parent f5b4f6ccc4
commit a694420d11
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
21 changed files with 656 additions and 376 deletions
Download patch file Download diff file Expand all files Collapse all files

188
ChocolArm64/Translation/Translator.cs Normal file
View file

@ -0,0 +1,188 @@
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));
}
TranslatedSub subroutine = GetOrTranslateSubroutine(state, position);
position = subroutine.Execute(state, _memory);
}
while (position != 0 && state.Running);
state.CurrentTranslator = null;
}
internal void TranslateVirtualSubroutine(CpuThreadState state, long position)
{
if (!_cache.TryGetSubroutine(position, out TranslatedSub sub) || sub.Tier == TranslationTier.Tier0)
{
_queue.Enqueue(new TranslatorQueueItem(position, state.GetExecutionMode(), TranslationTier.Tier1));
}
}
internal ArmSubroutine GetOrTranslateVirtualSubroutine(CpuThreadState state, long position)
{
if (!_cache.TryGetSubroutine(position, out TranslatedSub sub))
{
sub = TranslateLowCq(position, state.GetExecutionMode());
}
if (sub.Tier == TranslationTier.Tier0)
{
_queue.Enqueue(new TranslatorQueueItem(position, state.GetExecutionMode(), TranslationTier.Tier1));
}
return sub.Delegate;
}
internal TranslatedSub GetOrTranslateSubroutine(CpuThreadState state, long position)
{
if (!_cache.TryGetSubroutine(position, out TranslatedSub subroutine))
{
subroutine = TranslateLowCq(position, state.GetExecutionMode());
}
return subroutine;
}
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);
}
}
else
{
_queue.WaitForItems();
}
}
}
private TranslatedSub TranslateLowCq(long position, ExecutionMode mode)
{
Block block = Decoder.DecodeBasicBlock(_memory, position, mode);
ILEmitterCtx context = new ILEmitterCtx(_cache, _queue, TranslationTier.Tier0, block);
string subName = GetSubroutineName(position);
ILMethodBuilder ilMthdBuilder = new ILMethodBuilder(context.GetILBlocks(), subName);
TranslatedSub subroutine = ilMthdBuilder.GetSubroutine(TranslationTier.Tier0);
return _cache.GetOrAdd(position, subroutine, block.OpCodes.Count);
}
private void TranslateHighCq(long position, ExecutionMode mode)
{
Block graph = Decoder.DecodeSubroutine(_memory, position, mode);
ILEmitterCtx context = new ILEmitterCtx(_cache, _queue, TranslationTier.Tier1, graph);
ILBlock[] ilBlocks = context.GetILBlocks();
string subName = GetSubroutineName(position);
ILMethodBuilder ilMthdBuilder = new ILMethodBuilder(ilBlocks, subName);
TranslatedSub subroutine = ilMthdBuilder.GetSubroutine(TranslationTier.Tier1);
int ilOpCount = 0;
foreach (ILBlock ilBlock in ilBlocks)
{
ilOpCount += ilBlock.Count;
}
_cache.AddOrUpdate(position, subroutine, ilOpCount);
ForceAheadOfTimeCompilation(subroutine);
}
private string GetSubroutineName(long position)
{
return $"Sub{position:x16}";
}
private void ForceAheadOfTimeCompilation(TranslatedSub subroutine)
{
subroutine.Execute(_dummyThreadState, null);
}
}
}