L4 is now on new algo, using tricky way to init entryies

StgSharp.Common/HighPerformance/Memory/L4.CacheLine.cs

@@ -25,6 +25,7 @@
 //     
 // -----------------------------------------------------------------------
 // -----------------------------------------------------------------------
+using StgSharp.HighPerformance.ProcessorAbstraction;
 using System.Runtime.InteropServices;
 
 using Hlsf = StgSharp.HighPerformance.Memory.HybridLayerSegregatedFitAllocator;
@@ -55,12 +56,13 @@ namespace StgSharp.HighPerformance.Memory
         {
 
             [FieldOffset(0)] internal Hlsf.Entry AllocatorEntry;
+            [FieldOffset(0)] internal readonly M512 Buffer;
 
             #region reference counting and state information
 
             [FieldOffset(52)] internal int RefCount;
             [FieldOffset(60)] private readonly int Reverse;
-            [FieldOffset(56)] internal int Id;
+            [FieldOffset(56)] internal readonly int Id;
 
             #endregion
 

StgSharp.Common/HighPerformance/Memory/L4.Prefetch.cs

@@ -137,12 +137,14 @@ namespace StgSharp.HighPerformance.Memory
                         }
                     }
                     head = _head + index;
-                    head->Id = index;
                     head->Origin = address;
                     head->Profile = pHandle.MapPolicy;
                     head->RefCount = 0;
                     head->Predictor = (short)predictIndex;
 
+                    (*_predictCount)[index] = 1;
+                    (*_mapCount)[index] = 1;
+
                     // pre fetch them here
                     predict.Prefetch(head->Origin, head->Profile, new Span<byte>((byte*)head->Address, (int)head->Size));
                     _ = _map.TryAddOrSet(address, (nuint)head, out _);

StgSharp.Common/HighPerformance/Memory/L4.cs

@@ -103,6 +103,11 @@ namespace StgSharp.HighPerformance.Memory
             ptr += predictionCountSize;
             _mapCount = (MapArray*)ptr;
             ptr += predictionCountSize;
+
+            M512* mPtr = (M512*)_head;
+            for (uint i = 0; i < CacheLineCount; i++) {
+                mPtr->BroadCastFrom<ulong>(i);
+            }
         }
 
         public nuint ByteCapacity { get; init; }

StgSharp.Common/HighPerformance/ProcessorAbstraction/M512.cs

@@ -42,7 +42,13 @@ namespace StgSharp.HighPerformance.ProcessorAbstraction
     {
 
         [FieldOffset(0)] private fixed byte _buffer[64];
-        [FieldOffset(0)] private Vector512<float> Vec;
+        [FieldOffset(0)] private Vector512<byte> Vec;
+
+        public M512()
+        {
+            Unsafe.SkipInit(out this);
+            Unsafe.SkipInit(out Vec);
+        }
 
         public ref T AsRef<T>() where T : unmanaged, INumber<T>
         {
@@ -53,7 +59,7 @@ namespace StgSharp.HighPerformance.ProcessorAbstraction
                     T value
         ) where T : unmanaged, INumber<T>
         {
-            Vec = Vector512.Create(value).AsSingle();
+            Vec = Vector512.Create(value).AsByte();
         }
 
         public ref T Member<T>(