Use thread-static pools in CPU divisor testing

PerfectNumbers.Core/Cpu/MersenneNumberDivisorByDivisorCpuTester.cs

@@ -1,4 +1,4 @@
-using System.Collections.Concurrent;
+using System;
 using System.Runtime.CompilerServices;
 
 namespace PerfectNumbers.Core.Cpu;
@@ -6,7 +6,9 @@ namespace PerfectNumbers.Core.Cpu;
 public sealed class MersenneNumberDivisorByDivisorCpuTester : IMersenneNumberDivisorByDivisorTester
 {
     private readonly object _sync = new();
-    private readonly ConcurrentBag<DivisorScanSession> _sessionPool = [];
+    [ThreadStatic]
+    private static DivisorScanSession? _threadLocalSession; // The nested session type stays private, so pool it here.
+
     private ulong _divisorLimit;
     private ulong _lastStatusDivisor;
     private int _batchSize = 1_024;
@@ -85,8 +87,10 @@ public sealed class MersenneNumberDivisorByDivisorCpuTester : IMersenneNumberDiv
 
     public IMersenneNumberDivisorByDivisorTester.IDivisorScanSession CreateDivisorSession()
     {
-		if (_sessionPool.TryTake(out DivisorScanSession? session))
+        DivisorScanSession? session = _threadLocalSession;
+        if (session is not null)
         {
+            _threadLocalSession = null;
             session.Reset();
             return session;
         }
@@ -96,7 +100,36 @@ public sealed class MersenneNumberDivisorByDivisorCpuTester : IMersenneNumberDiv
 
     private void ReturnSession(DivisorScanSession session)
     {
-        _sessionPool.Add(session);
+        _threadLocalSession = session;
+    }
+
+    private struct FactorCacheLease
+    {
+        private Dictionary<ulong, MersenneDivisorCycles.FactorCacheEntry>? _cache;
+
+        public void EnsureInitialized(ref Dictionary<ulong, MersenneDivisorCycles.FactorCacheEntry>? cache)
+        {
+            Dictionary<ulong, MersenneDivisorCycles.FactorCacheEntry>? current = cache;
+            if (current is null)
+            {
+                current = ThreadStaticPools.TakeMersenneFactorCacheDictionary();
+                cache = current;
+            }
+
+            _cache = current;
+        }
+
+        public void Dispose()
+        {
+            Dictionary<ulong, MersenneDivisorCycles.FactorCacheEntry>? cache = _cache;
+            if (cache is null)
+            {
+                return;
+            }
+
+            ThreadStaticPools.ReturnMersenneFactorCacheDictionary(cache);
+            _cache = null;
+        }
     }
 
     private bool CheckDivisors(
@@ -108,10 +141,12 @@ public sealed class MersenneNumberDivisorByDivisorCpuTester : IMersenneNumberDiv
         processedCount = 0UL;
         processedAll = false;
 
-        if (allowedMax < 3UL)
+        // The EvenPerfectBitScanner feeds primes >= 138,000,000 here, so allowedMax >= 3 in production runs.
-        {
+        // Keeping the guard commented out documents the reasoning for benchmarks and tests.
-            return false;
+        // if (allowedMax < 3UL)
-        }
+        // {
+        //     return false;
+        // }
 
         UInt128 step = (UInt128)prime << 1;
         if (step == UInt128.Zero)
@@ -128,8 +163,9 @@ public sealed class MersenneNumberDivisorByDivisorCpuTester : IMersenneNumberDiv
             return false;
         }
 
+        FactorCacheLease factorCacheLease = new FactorCacheLease();
         Dictionary<ulong, MersenneDivisorCycles.FactorCacheEntry>? factorCache = null;
-        DivisorCycleCache cycleCache = DivisorCycleCache.Shared;
+        bool foundDivisor = false;
 
         byte step10 = (byte)(step % 10UL);
         byte step8 = (byte)(step % 8UL);
@@ -161,28 +197,22 @@ public sealed class MersenneNumberDivisorByDivisorCpuTester : IMersenneNumberDiv
             if (admissible && (remainder8 == 1 || remainder8 == 7) && remainder3 != 0 && remainder5 != 0 && remainder7 != 0 && remainder11 != 0)
             {
                 MontgomeryDivisorData divisorData = MontgomeryDivisorDataCache.Get(candidate);
-                ulong divisorCycle;
+                Dictionary<ulong, MersenneDivisorCycles.FactorCacheEntry>? cacheForCycle = null;
+                if (candidate > PerfectNumberConstants.MaxQForDivisorCycles)
+                {
+                    factorCacheLease.EnsureInitialized(ref factorCache);
+                    cacheForCycle = factorCache;
+                }
 
-                if (candidate <= PerfectNumberConstants.MaxQForDivisorCycles)
+                if (!MersenneDivisorCycles.TryCalculateCycleLengthForExponent(candidate, prime, divisorData, cacheForCycle, out ulong divisorCycle))
                 {
-                    divisorCycle = cycleCache.GetCycleLength(candidate);
+                    divisorCycle = 0UL;
-                }
-                else
-                {
-                    factorCache ??= new Dictionary<ulong, MersenneDivisorCycles.FactorCacheEntry>(8);
-                    if (!MersenneDivisorCycles.TryCalculateCycleLengthForExponent(candidate, prime, divisorData, factorCache, out divisorCycle) || divisorCycle == 0UL)
-                    {
-                        divisorCycle = cycleCache.GetCycleLength(candidate);
-                    }
                 }
 
                 if (divisorCycle == prime)
                 {
-                    // A cycle equal to the tested exponent (which is prime in this path)
+                    foundDivisor = true;
-                    // guarantees that the candidate divides the corresponding Mersenne
+                    break;
-                    // number because the order of 2 modulo the divisor is exactly p.
-                    processedAll = true;
-                    return true;
                 }
 
                 if (divisorCycle == 0UL)
@@ -200,6 +230,15 @@ public sealed class MersenneNumberDivisorByDivisorCpuTester : IMersenneNumberDiv
             remainder11 = AddMod(remainder11, step11, (byte)11);
         }
 
+        factorCacheLease.Dispose();
+        factorCache = null; // The pooled dictionary was returned to the thread-static slot.
+
+        if (foundDivisor)
+        {
+            processedAll = true;
+            return true;
+        }
+
         processedAll = divisor > limit;
         return false;
     }

PerfectNumbers.Core/MersenneDivisorCycles.cs

@@ -17,13 +17,32 @@ public class MersenneDivisorCycles
     private List<(ulong divisor, ulong cycleLength)> _table = [];
     // Lightweight read-mostly cache for small divisors (<= 4,000,000). 0 => unknown
     private ulong[]? _smallCycles;
-
     public static MersenneDivisorCycles Shared { get; } = new MersenneDivisorCycles();
 
     public MersenneDivisorCycles()
     {
     }
 
+    private static Dictionary<ulong, int> RentFactorCountDictionary()
+    {
+        return ThreadStaticPools.TakeFactorCountDictionary();
+    }
+
+    private static void ReturnFactorCountDictionary(Dictionary<ulong, int> dictionary)
+    {
+        ThreadStaticPools.ReturnFactorCountDictionary(dictionary);
+    }
+
+    private static Dictionary<ulong, int> RentFactorScratchDictionary()
+    {
+        return ThreadStaticPools.TakeFactorScratchDictionary();
+    }
+
+    private static void ReturnFactorScratchDictionary(Dictionary<ulong, int> dictionary)
+    {
+        ThreadStaticPools.ReturnFactorScratchDictionary(dictionary);
+    }
+
     public void LoadFrom(string path)
     {
         EnsureSmallBuffer();
@@ -316,7 +335,10 @@ public class MersenneDivisorCycles
             return false;
         }
 
-        Dictionary<ulong, int> factorCounts = new(8);
+        Dictionary<ulong, int> factorCounts = RentFactorCountDictionary();
+
+        try
+        {
             if (twoCount > 0)
             {
                 factorCounts[2UL] = twoCount;
@@ -330,6 +352,11 @@ public class MersenneDivisorCycles
             cycleLength = ReduceOrder(divisorData, phi, factorCounts);
             return true;
         }
+        finally
+        {
+            ReturnFactorCountDictionary(factorCounts);
+        }
+    }
 
     private static bool AccumulateFactors(
         ulong value,
@@ -370,7 +397,10 @@ public class MersenneDivisorCycles
             return true;
         }
 
-        Dictionary<ulong, int> scratch = new(8);
+        Dictionary<ulong, int> scratch = RentFactorScratchDictionary();
+
+        try
+        {
             if (!TryFactorIntoCountsInternal(value, scratch))
             {
                 factorization = default!;
@@ -398,9 +428,15 @@ public class MersenneDivisorCycles
 
             return true;
         }
+        finally
+        {
+            ReturnFactorScratchDictionary(scratch);
+        }
+    }
 
     private static bool TryFactorIntoCountsInternal(ulong value, Dictionary<ulong, int> counts)
     {
+        // This guard terminates the recursion once Pollard-Rho factorization has reduced the input to 1.
         if (value <= 1UL)
         {
             return true;

PerfectNumbers.Core/ThreadStaticPools.cs

@@ -1,4 +1,6 @@
+using System;
 using System.Buffers;
+using System.Collections.Generic;
 using System.Runtime.CompilerServices;
 
 namespace PerfectNumbers.Core
@@ -7,6 +9,7 @@ namespace PerfectNumbers.Core
     {
         [ThreadStatic]
         private static ArrayPool<FactorEntry>? _factorEntryPool;
+
         public static ArrayPool<FactorEntry> FactorEntryPool
         {
             [MethodImpl(MethodImplOptions.AggressiveInlining)]
@@ -18,6 +21,7 @@ namespace PerfectNumbers.Core
 
         [ThreadStatic]
         private static ArrayPool<FactorEntry128>? _factorEntry128Pool;
+
         public static ArrayPool<FactorEntry128> FactorEntry128Pool
         {
             [MethodImpl(MethodImplOptions.AggressiveInlining)]
@@ -29,6 +33,7 @@ namespace PerfectNumbers.Core
 
         [ThreadStatic]
         private static ArrayPool<ulong>? _ulongPool;
+
         public static ArrayPool<ulong> UlongPool
         {
             [MethodImpl(MethodImplOptions.AggressiveInlining)]
@@ -37,5 +42,77 @@ namespace PerfectNumbers.Core
                 return _ulongPool ??= ArrayPool<ulong>.Create();
             }
         }
+
+        [ThreadStatic]
+        private static Dictionary<ulong, MersenneDivisorCycles.FactorCacheEntry>? _mersenneFactorCacheDictionary;
+
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public static Dictionary<ulong, MersenneDivisorCycles.FactorCacheEntry> TakeMersenneFactorCacheDictionary()
+        {
+            Dictionary<ulong, MersenneDivisorCycles.FactorCacheEntry>? cache = _mersenneFactorCacheDictionary;
+            if (cache is null)
+            {
+                return new Dictionary<ulong, MersenneDivisorCycles.FactorCacheEntry>(8);
+            }
+
+            _mersenneFactorCacheDictionary = null;
+            cache.Clear();
+            return cache;
+        }
+
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public static void ReturnMersenneFactorCacheDictionary(Dictionary<ulong, MersenneDivisorCycles.FactorCacheEntry> cache)
+        {
+            cache.Clear();
+            _mersenneFactorCacheDictionary = cache;
+        }
+
+        [ThreadStatic]
+        private static Dictionary<ulong, int>? _factorCountDictionary;
+
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public static Dictionary<ulong, int> TakeFactorCountDictionary()
+        {
+            Dictionary<ulong, int>? dictionary = _factorCountDictionary;
+            if (dictionary is null)
+            {
+                return new Dictionary<ulong, int>(8);
+            }
+
+            _factorCountDictionary = null;
+            dictionary.Clear();
+            return dictionary;
+        }
+
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public static void ReturnFactorCountDictionary(Dictionary<ulong, int> dictionary)
+        {
+            dictionary.Clear();
+            _factorCountDictionary = dictionary;
+        }
+
+        [ThreadStatic]
+        private static Dictionary<ulong, int>? _factorScratchDictionary;
+
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public static Dictionary<ulong, int> TakeFactorScratchDictionary()
+        {
+            Dictionary<ulong, int>? dictionary = _factorScratchDictionary;
+            if (dictionary is null)
+            {
+                return new Dictionary<ulong, int>(8);
+            }
+
+            _factorScratchDictionary = null;
+            dictionary.Clear();
+            return dictionary;
+        }
+
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public static void ReturnFactorScratchDictionary(Dictionary<ulong, int> dictionary)
+        {
+            dictionary.Clear();
+            _factorScratchDictionary = dictionary;
+        }
     }
 }