Merge pull request #210 from mlemanczyk/codex/optimize-partialfactor-method-for-isprime-caching

Cache composite primality results during partial factorization

EvenPerfectBitScanner/Program.cs

@@ -55,6 +55,7 @@ internal static class Program
 			ulong remainder = currentP % 6UL;
 			bool startPrimeProvided = _cliArguments.StartPrimeProvided;
 			int threadCount = Math.Max(1, _cliArguments.ThreadCount);
+			UnboundedTaskScheduler.ConfigureThreadCount(threadCount);
 			int blockSize = Math.Max(1, _cliArguments.BlockSize);
 			int gpuPrimeThreads = Math.Max(1, _cliArguments.GpuPrimeThreads);
 			int gpuPrimeBatch = Math.Max(1, _cliArguments.GpuPrimeBatch);

PerfectNumbers.Core/PrimeOrderCalculator.Cpu.cs

@@ -10,10 +10,35 @@ namespace PerfectNumbers.Core;
 internal static partial class PrimeOrderCalculator
 {
 
-	internal readonly struct PendingEntry(ulong value, bool knownComposite)
+	internal readonly struct PendingEntry
 	{
-		public readonly ulong Value = value;
+		public readonly ulong Value;
-		public readonly bool KnownComposite = knownComposite;
+		public readonly bool KnownComposite;
+		public readonly bool HasKnownPrimality;
+		public readonly bool IsPrime;
+
+		public PendingEntry(ulong value, bool knownComposite)
+		{
+			Value = value;
+			KnownComposite = knownComposite;
+			HasKnownPrimality = knownComposite;
+			IsPrime = false;
+		}
+
+		private PendingEntry(ulong value, bool knownComposite, bool hasKnownPrimality, bool isPrime)
+		{
+			Value = value;
+			KnownComposite = knownComposite;
+			HasKnownPrimality = hasKnownPrimality;
+			IsPrime = isPrime;
+		}
+
+		[MethodImpl(MethodImplOptions.AggressiveInlining)]
+		public PendingEntry WithPrimality(bool isPrime)
+		{
+			bool knownComposite = KnownComposite || !isPrime;
+			return new PendingEntry(Value, knownComposite, true, isPrime);
+		}
 	}
 
 	private static ulong CalculateInternal(ulong prime, ulong? previousOrder, in MontgomeryDivisorData divisorData, in PrimeOrderSearchConfig config)
@@ -720,6 +745,7 @@ internal static partial class PrimeOrderCalculator
 		return isPrime;
 	}
 
+
 	private static PartialFactorResult PartialFactor(ulong value, in PrimeOrderSearchConfig config)
 	{
 		if (value <= 1UL)
@@ -756,7 +782,6 @@ internal static partial class PrimeOrderCalculator
 		List<PendingEntry> pending = ThreadStaticPools.RentPrimeOrderPendingEntryList(2);
 		Stack<ulong>? compositeStack = null;
 		PartialFactorResult result;
-
 		if (!gpuFactored)
 		{
 			// We don't need to worry about leftovers, because we always use indexes within the calculated counts
@@ -829,6 +854,14 @@ internal static partial class PrimeOrderCalculator
 						continue;
 					}
 
+					long currentTimestamp = Stopwatch.GetTimestamp();
+					if (currentTimestamp > deadlineTimestamp)
+					{
+						pollardRhoDeadlineReached = true;
+						pending.Add(new PendingEntry(composite, knownComposite: true));
+						continue;
+					}
+
 					if (!TryPollardRho(composite, deadlineTimestamp, out ulong factor))
 					{
 						pending.Add(new PendingEntry(composite, knownComposite: true));
@@ -840,8 +873,11 @@ internal static partial class PrimeOrderCalculator
 					compositeStack.Push(quotient);
 				}
 
+				if (!pollardRhoDeadlineReached)
+				{
 					pollardRhoDeadlineReached = Stopwatch.GetTimestamp() > deadlineTimestamp;
 				}
+			}
 
 			if (pollardRhoDeadlineReached)
 			{
@@ -868,9 +904,14 @@ internal static partial class PrimeOrderCalculator
 				continue;
 			}
 
+			if (!entry.HasKnownPrimality)
+			{
 				bool isPrime = GetOrComputePrimality(composite);
+				entry = entry.WithPrimality(isPrime);
+				pending[index] = entry;
+			}
 
-			if (isPrime)
+			if (entry.IsPrime)
 			{
 				AddFactorToCollector(ref useDictionary, ref counts, primeSlots, exponentSlots, ref factorCount, composite, 1);
 			}
@@ -968,6 +1009,7 @@ internal static partial class PrimeOrderCalculator
 		// ThreadStaticPools.UlongPool.Return(primeSlotsArray);
 		// ThreadStaticPools.IntPool.Return(exponentSlotsArray);
 
+
 		if (counts is not null)
 		{
 			ThreadStaticPools.ReturnUlongIntDictionary(counts);

PerfectNumbers.Core/PrimeOrderCalculator.cs

@@ -1,3 +1,5 @@
+using System.Runtime.CompilerServices;
+
 namespace PerfectNumbers.Core;
 
 internal static partial class PrimeOrderCalculator
@@ -10,9 +12,27 @@ internal static partial class PrimeOrderCalculator
 
 	internal readonly struct PrimeOrderSearchConfig(uint smallFactorLimit, int pollardRhoMilliseconds, int maxPowChecks, PrimeOrderMode mode)
 	{
-		public static readonly PrimeOrderSearchConfig HeuristicDefault = new(smallFactorLimit: 100_000, pollardRhoMilliseconds: 24, maxPowChecks: 24, PrimeOrderMode.Heuristic);
+		private static PrimeOrderSearchConfig _heuristicDefault = CreateHeuristicDefault(UnboundedTaskScheduler.ConfiguredThreadCount);
+
+		public static PrimeOrderSearchConfig HeuristicDefault
+		{
+				[MethodImpl(MethodImplOptions.AggressiveInlining)]
+				get => _heuristicDefault;
+		}
+
 		public static readonly PrimeOrderSearchConfig StrictDefault = new(smallFactorLimit: 1_000_000, pollardRhoMilliseconds: 0, maxPowChecks: 0, PrimeOrderMode.Strict);
 
+		public static void ConfigureHeuristicDefault(int threadCount)
+		{
+				_heuristicDefault = CreateHeuristicDefault(threadCount);
+		}
+
+		public static PrimeOrderSearchConfig CreateHeuristicDefault(int threadCount)
+		{
+				int normalizedThreadCount = threadCount < 1 ? 1 : threadCount;
+				return new(smallFactorLimit: 100_000, pollardRhoMilliseconds: normalizedThreadCount * 24, maxPowChecks: 24, PrimeOrderMode.Heuristic);
+		}
+
 		public readonly uint SmallFactorLimit = smallFactorLimit;
 		public readonly int PollardRhoMilliseconds = pollardRhoMilliseconds;
 		public readonly int MaxPowChecks = maxPowChecks;

PerfectNumbers.Core/UnboundedTaskScheduler.cs

@@ -6,13 +6,31 @@ namespace PerfectNumbers.Core;
 
 public sealed class UnboundedTaskScheduler : TaskScheduler
 {
-    public static UnboundedTaskScheduler Instance { get; } = new();
+    private static UnboundedTaskScheduler? _instance;
+    private static int _configuredThreadCount = Environment.ProcessorCount;
 
-    private readonly TaskThreadPool _threadPool;
+    public static UnboundedTaskScheduler Instance => _instance ??= new UnboundedTaskScheduler(_configuredThreadCount);
 
-    private UnboundedTaskScheduler()
+    public static int ConfiguredThreadCount => _configuredThreadCount;
+
+    private TaskThreadPool _threadPool;
+
+    private UnboundedTaskScheduler(int threadCount)
     {
-        _threadPool = new TaskThreadPool(10_340, ExecuteTask);
+        _threadPool = new TaskThreadPool(threadCount, ExecuteTask);
+    }
+
+    public static void ConfigureThreadCount(int threadCount)
+    {
+        _configuredThreadCount = threadCount;
+        PrimeOrderCalculator.PrimeOrderSearchConfig.ConfigureHeuristicDefault(threadCount);
+
+        // On EvenPerfectBitScanner's execution path the scheduler instance is created after configuration,
+        // so updating an existing pool would never run. Leave the update path disabled for now.
+        // if (_instance is not null)
+        // {
+        //     _instance.UpdateThreadPool(threadCount);
+        // }
     }
 
     public override int MaximumConcurrencyLevel => int.MaxValue;
@@ -22,6 +40,19 @@ public sealed class UnboundedTaskScheduler : TaskScheduler
         _threadPool.Queue(task);
     }
 
+    private void UpdateThreadPool(int threadCount)
+    {
+        if (_threadPool.ThreadCount == threadCount)
+        {
+            return;
+        }
+
+        TaskThreadPool newPool = new(threadCount, ExecuteTask);
+        TaskThreadPool oldPool = _threadPool;
+        _threadPool = newPool;
+        oldPool.Dispose();
+    }
+
     private void ExecuteTask(Task task)
     {
         TryExecuteTask(task);
@@ -36,4 +67,5 @@ public sealed class UnboundedTaskScheduler : TaskScheduler
     {
         return _threadPool.GetScheduledTasks();
     }
+
 }