From e21ebbf666f10d39d44a0856e5a44143d3d69d0d Mon Sep 17 00:00:00 2001
From: gdkchan <gab.dark.100@gmail.com>
Date: Wed, 27 Feb 2019 23:03:31 -0300
Subject: 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
---
 ChocolArm64/Translation/Translator.cs | 56 ++++++++++++++++-------------------
 1 file changed, 26 insertions(+), 30 deletions(-)

(limited to 'ChocolArm64/Translation/Translator.cs')

diff --git a/ChocolArm64/Translation/Translator.cs b/ChocolArm64/Translation/Translator.cs
index dd1215f5..bda0bca0 100644
--- a/ChocolArm64/Translation/Translator.cs
+++ b/ChocolArm64/Translation/Translator.cs
@@ -63,46 +63,34 @@ namespace ChocolArm64.Translation
                     CpuTrace?.Invoke(this, new CpuTraceEventArgs(position));
                 }
 
-                TranslatedSub subroutine = GetOrTranslateSubroutine(state, position);
+                if (!_cache.TryGetSubroutine(position, out TranslatedSub sub))
+                {
+                    sub = TranslateLowCq(position, state.GetExecutionMode());
+                }
 
-                position = subroutine.Execute(state, _memory);
+                position = sub.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)
+        internal ArmSubroutine GetOrTranslateSubroutine(CpuThreadState state, long position, CallType cs)
         {
             if (!_cache.TryGetSubroutine(position, out TranslatedSub sub))
             {
                 sub = TranslateLowCq(position, state.GetExecutionMode());
             }
 
-            if (sub.Tier == TranslationTier.Tier0)
+            if (sub.IsWorthOptimizing())
             {
-                _queue.Enqueue(new TranslatorQueueItem(position, state.GetExecutionMode(), TranslationTier.Tier1));
-            }
-
-            return sub.Delegate;
-        }
+                bool isComplete = cs == CallType.Call ||
+                                  cs == CallType.VirtualCall;
 
-        internal TranslatedSub GetOrTranslateSubroutine(CpuThreadState state, long position)
-        {
-            if (!_cache.TryGetSubroutine(position, out TranslatedSub subroutine))
-            {
-                subroutine = TranslateLowCq(position, state.GetExecutionMode());
+                _queue.Enqueue(position, state.GetExecutionMode(), TranslationTier.Tier1, isComplete);
             }
 
-            return subroutine;
+            return sub.Delegate;
         }
 
         private void TranslateQueuedSubs()
@@ -124,7 +112,7 @@ namespace ChocolArm64.Translation
                     }
                     else
                     {
-                        TranslateHighCq(item.Position, item.Mode);
+                        TranslateHighCq(item.Position, item.Mode, item.IsComplete);
                     }
                 }
                 else
@@ -142,14 +130,16 @@ namespace ChocolArm64.Translation
 
             string subName = GetSubroutineName(position);
 
-            ILMethodBuilder ilMthdBuilder = new ILMethodBuilder(context.GetILBlocks(), subName);
+            bool isAarch64 = mode == ExecutionMode.Aarch64;
+
+            ILMethodBuilder ilMthdBuilder = new ILMethodBuilder(context.GetILBlocks(), subName, isAarch64);
 
-            TranslatedSub subroutine = ilMthdBuilder.GetSubroutine(TranslationTier.Tier0);
+            TranslatedSub subroutine = ilMthdBuilder.GetSubroutine(TranslationTier.Tier0, isWorthOptimizing: true);
 
             return _cache.GetOrAdd(position, subroutine, block.OpCodes.Count);
         }
 
-        private void TranslateHighCq(long position, ExecutionMode mode)
+        private TranslatedSub TranslateHighCq(long position, ExecutionMode mode, bool isComplete)
         {
             Block graph = Decoder.DecodeSubroutine(_memory, position, mode);
 
@@ -159,9 +149,13 @@ namespace ChocolArm64.Translation
 
             string subName = GetSubroutineName(position);
 
-            ILMethodBuilder ilMthdBuilder = new ILMethodBuilder(ilBlocks, subName);
+            bool isAarch64 = mode == ExecutionMode.Aarch64;
+
+            isComplete &= !context.HasIndirectJump;
 
-            TranslatedSub subroutine = ilMthdBuilder.GetSubroutine(TranslationTier.Tier1);
+            ILMethodBuilder ilMthdBuilder = new ILMethodBuilder(ilBlocks, subName, isAarch64, isComplete);
+
+            TranslatedSub subroutine = ilMthdBuilder.GetSubroutine(TranslationTier.Tier1, context.HasSlowCall);
 
             int ilOpCount = 0;
 
@@ -170,9 +164,11 @@ namespace ChocolArm64.Translation
                 ilOpCount += ilBlock.Count;
             }
 
+            ForceAheadOfTimeCompilation(subroutine);
+
             _cache.AddOrUpdate(position, subroutine, ilOpCount);
 
-            ForceAheadOfTimeCompilation(subroutine);
+            return subroutine;
         }
 
         private string GetSubroutineName(long position)
-- 
cgit v1.2.3