Batch size made dynamic - dependent on the value

PerfectNumbers.Core/Gpu/Kernels/PrimeTesterKernels.cs

@@ -71,7 +71,16 @@ internal static class PrimeTesterKernels
     public static void HeuristicTrialDivisionKernel(Index1D index, ArrayView<ulong> divisors, ulong n, ArrayView<int> resultFlag)
     {
 		ulong divisor = divisors[index];
-        Atomic.Or(ref resultFlag[0], divisor <= 1UL ? 0 : (n % divisor == 0UL ? 1 : 0));
+		if (divisor <= 1UL)
+		{
+			return;
+		}
+
+		ulong residue = n % divisor;
+		if (residue == 0UL)
+		{
+			resultFlag[0] = 1;
+		}
     }
 
     public static void SharesFactorKernel(Index1D index, ArrayView<ulong> numbers, ArrayView<byte> results)

PerfectNumbers.Core/HeuristicCombinedPrimeTester.cs

@@ -1,11 +1,5 @@
-using System;
 using Open.Numeric.Primes;
-using System.Collections.Generic;
-using System.Numerics;
 using System.Runtime.CompilerServices;
-using System.Threading;
-using ILGPU;
-using ILGPU.Algorithms;
 using ILGPU.Runtime;
 using PerfectNumbers.Core.Gpu;
 
@@ -48,8 +42,6 @@ public sealed class HeuristicCombinedPrimeTester
 
 	private const ulong Wheel210 = 210UL;
 
-	private static readonly bool UseHeuristicGroupBTrialDivision = false; // Temporary fallback gate for Group B.
-
 	private static readonly uint[] CombinedDivisorsEnding1TwoAOneB;
 	private static readonly uint[] CombinedDivisorsEnding3TwoAOneB;
 	private static readonly uint[] CombinedDivisorsEnding7TwoAOneB;
@@ -141,8 +133,6 @@ public sealed class HeuristicCombinedPrimeTester
 	}
 
 
-	internal static int HeuristicGpuDivisorBatchSize = 4_096;
-	internal static int HeuristicDivisorInterleaveBatchSize = 64;
 
 	[ThreadStatic]
 	private static HeuristicCombinedPrimeTester? _tester;
@@ -205,12 +195,20 @@ public sealed class HeuristicCombinedPrimeTester
 		return EvaluateWithOpenNumericFallback(n);
 	}
 
+	[MethodImpl(MethodImplOptions.AggressiveInlining)]
+	internal static int GetBatchSize(ulong n) => n switch
+	{
+		< 1_000 => 8,
+		< 10_000 => 16,
+		< 100_000 => 32,
+		< 1_000_000 => 128,
+		_ => 256,
+	};
+
 	private static bool HeuristicTrialDivisionGpuDetectsDivisor(ulong n, ulong maxDivisorSquare, byte nMod10)
 	{
-	        int batchCapacity = HeuristicGpuDivisorBatchSize;
+		int batchCapacity = GetBatchSize(n);
-	        var divisorPool = ThreadStaticPools.UlongPool;
 
-	        ulong[] divisorArray = divisorPool.Rent(batchCapacity);
 
 		var limiter = GpuPrimeWorkLimiter.Acquire();
 		var gpu = PrimeTester.PrimeTesterGpuContextPool.Rent(batchCapacity);
@@ -222,6 +220,7 @@ public sealed class HeuristicCombinedPrimeTester
 
 		var inputView = gpu.Input.View;
 		var flagView = gpu.HeuristicFlag.View;
+		ulong[] divisorArray = gpu.DivisorArray;
 
 		bool ProcessBatch(int length)
 		{
@@ -271,7 +270,6 @@ public sealed class HeuristicCombinedPrimeTester
 
 
 	Cleanup:
-	        divisorPool.Return(divisorArray);
 		gpu.Dispose();
 		limiter.Dispose();
 		return compositeDetected;
@@ -451,7 +449,7 @@ Cleanup:
 			}
 		}
 
-		uint[] result = combined.ToArray();
+		uint[] result = [.. combined];
 		ulong[] resultSquares = new ulong[result.Length];
 		for (int i = 0; i < result.Length; i++)
 		{
@@ -488,10 +486,10 @@ Cleanup:
 	private static ReadOnlySpan<uint> GetGroupBDivisors(byte ending) => ending switch
 	{
 		1 => DivisorGenerator.SmallPrimesLastOneWithoutLastThree,
-		3 => ReadOnlySpan<uint>.Empty,
+		3 => [],
 		7 => DivisorGenerator.SmallPrimesLastSevenWithoutLastThree,
 		9 => DivisorGenerator.SmallPrimesLastNineWithoutLastThree,
-		_ => ReadOnlySpan<uint>.Empty,
+		_ => [],
 	};
 
 	[MethodImpl(MethodImplOptions.AggressiveInlining)]
@@ -690,11 +688,11 @@ Cleanup:
 			residueStepper.Advance();
 		}
 
-		public ulong ProcessedCount => processedCount;
+		public readonly ulong ProcessedCount => processedCount;
 
-		public ulong LastDivisor => lastDivisor;
+		public readonly ulong LastDivisor => lastDivisor;
 
-		public bool Exhausted => !active;
+		public readonly bool Exhausted => !active;
 
 		private static HeuristicDivisorCandidate CreateCandidate(ulong value)
 		{

PerfectNumbers.Core/HeuristicPrimeTester.cs

@@ -1,13 +1,5 @@
-using System;
-using System.Buffers;
-using System.Collections.Concurrent;
 using Open.Numeric.Primes;
-using System.Collections.Generic;
-using System.Numerics;
 using System.Runtime.CompilerServices;
-using System.Threading;
-using ILGPU;
-using ILGPU.Algorithms;
 using ILGPU.Runtime;
 using PerfectNumbers.Core.Gpu;
 
@@ -68,9 +60,6 @@ public sealed class HeuristicPrimeTester
 	}
 
 
-    public static int HeuristicGpuDivisorBatchSize { get; set; } = 4_096;
-
-    public static int HeuristicDivisorInterleaveBatchSize { get; set; } = 64;
 
 	[ThreadStatic]
 	private static HeuristicPrimeTester? _tester;
@@ -120,12 +109,11 @@ public sealed class HeuristicPrimeTester
 		return !compositeDetected;
 	}
 
-
 	private static bool HeuristicTrialDivisionCpu(ulong n, ulong maxDivisorSquare, byte nMod10, bool includeGroupB)
 	{
 		ReadOnlySpan<int> groupADivisors = HeuristicPrimeSieves.GroupADivisors;
 		ReadOnlySpan<ulong> groupADivisorSquares = HeuristicPrimeSieves.GroupADivisorSquares;
-        int interleaveBatchSize = Math.Max(1, HeuristicDivisorInterleaveBatchSize);
+		int interleaveBatchSize = HeuristicCombinedPrimeTester.GetBatchSize(n);
 		int groupAIndex = 0;
 
 		if (!includeGroupB)
@@ -249,19 +237,11 @@ public sealed class HeuristicPrimeTester
 		return true;
 	}
 
-
-
-
-
-
-
 	private bool HeuristicTrialDivisionGpuDetectsDivisor(ulong n, ulong maxDivisorSquare, byte nMod10)
 	{
-        int batchCapacity = HeuristicGpuDivisorBatchSize;
+		int batchCapacity = HeuristicCombinedPrimeTester.GetBatchSize(n);
-        int interleaveBatchSize = HeuristicDivisorInterleaveBatchSize;
+		int interleaveBatchSize = batchCapacity;
-        var divisorPool = ThreadStaticPools.UlongPool;
 
-        ulong[] divisorArray = divisorPool.Rent(batchCapacity);
 
 		var limiter = GpuPrimeWorkLimiter.Acquire();
 		var gpu = PrimeTester.PrimeTesterGpuContextPool.Rent(batchCapacity);
@@ -273,6 +253,7 @@ public sealed class HeuristicPrimeTester
 
 		var inputView = gpu.Input.View;
 		var flagView = gpu.HeuristicFlag.View;
+		ulong[] divisorArray = gpu.DivisorArray;
 
 		bool ProcessBatch(int length)
 		{
@@ -393,14 +374,11 @@ public sealed class HeuristicPrimeTester
 			compositeDetected = true;
 		}
 
-        divisorPool.Return(divisorArray);
 		gpu.Dispose();
 		limiter.Dispose();
 		return compositeDetected;
 	}
 
-
-
 	private static bool EvaluateWithOpenNumericFallback(ulong n)
 	{
 		return Prime.Numbers.IsPrime(n);
@@ -419,8 +397,6 @@ public sealed class HeuristicPrimeTester
 		_ => ReadOnlySpan<byte>.Empty,
 	};
 
-
-
 	[MethodImpl(MethodImplOptions.AggressiveInlining)]
     private static bool TrySelectNextGroupBDivisor(ReadOnlySpan<byte> endingOrder, Span<int> indices, ulong maxDivisorSquare, out ulong divisor)
 	{
@@ -444,7 +420,7 @@ public sealed class HeuristicPrimeTester
 				continue;
 			}
 
-            ulong candidate = (ulong)divisors[index];
+			ulong candidate = divisors[index];
 			if (bestEndingIndex == -1 || candidate < bestCandidate)
 			{
 				bestCandidate = candidate;
@@ -459,14 +435,11 @@ public sealed class HeuristicPrimeTester
 		}
 
 		ReadOnlySpan<uint> selectedDivisors = GetGroupBDivisors(endingOrder[bestEndingIndex]);
-        divisor = (ulong)selectedDivisors[indices[bestEndingIndex]];
+		divisor = selectedDivisors[indices[bestEndingIndex]];
 		indices[bestEndingIndex]++;
 		return true;
 	}
 
-
-
-
 	[MethodImpl(MethodImplOptions.AggressiveInlining)]
 	private static bool HasGroupBCandidates(ReadOnlySpan<byte> endingOrder, Span<int> indices, ulong maxDivisorSquare)
 	{

PerfectNumbers.Core/PrimeTester.cs

@@ -273,6 +273,7 @@ public sealed class PrimeTester
 			public MemoryBuffer1D<byte, Stride1D.Dense> Output;
 			public MemoryBuffer1D<int, Stride1D.Dense> HeuristicFlag;
 			public int BufferCapacity;
+	        public ulong[] DivisorArray;
 
 			internal PrimeTesterGpuContextLease(int minBufferCapacity)
 			{
@@ -321,16 +322,18 @@ public sealed class PrimeTester
 				DevicePrimesPow2LastNine = Accelerator.Allocate1D<ulong>(primesPow2LastNine.Length);
 				DevicePrimesPow2LastNine.View.CopyFromCPU(primesPow2LastNine);
 
-				BufferCapacity = minBufferCapacity < 1 ? 1 : minBufferCapacity;
+				BufferCapacity = minBufferCapacity;
-				Input = Accelerator.Allocate1D<ulong>(BufferCapacity);
+				Input = Accelerator.Allocate1D<ulong>(minBufferCapacity);
-				Output = Accelerator.Allocate1D<byte>(BufferCapacity);
+				Output = Accelerator.Allocate1D<byte>(minBufferCapacity);
-				HeuristicFlag = Accelerator.Allocate1D<int>(BufferCapacity);
+				HeuristicFlag = Accelerator.Allocate1D<int>(minBufferCapacity);
+				DivisorArray = ThreadStaticPools.UlongPool.Rent(minBufferCapacity);
+
 				_returned = false;
 			}
 
 			public void EnsureCapacity(int minCapacity)
 			{
-				int required = minCapacity < 1 ? 1 : minCapacity;
+				int required = minCapacity;
 				if (required <= BufferCapacity)
 				{
 					return;
@@ -339,9 +342,11 @@ public sealed class PrimeTester
 				Input.Dispose();
 				Output.Dispose();
 				HeuristicFlag.Dispose();
+				ThreadStaticPools.UlongPool.Return(DivisorArray);
 				Input = Accelerator.Allocate1D<ulong>(required);
 				Output = Accelerator.Allocate1D<byte>(required);
 				HeuristicFlag = Accelerator.Allocate1D<int>(required);
+				DivisorArray = ThreadStaticPools.UlongPool.Rent(required);
 				BufferCapacity = required;
 			}