BitContradiction filtering changes to cut invalid divisors during checks.

EvenPerfectBitScanner/CliArguments.cs

@@ -5,18 +5,6 @@ using PerfectNumbers.Core.Gpu;
 
 namespace EvenPerfectBitScanner;
 
-internal enum ByDivisorKIncrementMode
-{
-    Sequential,
-    Pow2Groups,
-    Predictive,
-    Percentile,
-    Additive,
-    TopDown,
-    BitContradiction,
-    BitTree,
-}
-
 internal readonly struct CliArguments
 {
     private const string DefaultCyclesPath = "divisor_cycles.bin";

EvenPerfectBitScanner/Program.cs

@@ -42,7 +42,7 @@ namespace EvenPerfectBitScanner;
 internal static class Program
 {
 	// private static ThreadLocal<PrimeTester> PrimeTesters = null!;
-private static ThreadLocal<object>? MersenneTesters;
+	private static ThreadLocal<object>? MersenneTesters;
 	private const ulong InitialP = PerfectNumberConstants.BiggestKnownEvenPerfectP;
 	private static MersenneNumberDivisorGpuTester? _divisorTester;
 	private static UInt128 _divisor;
@@ -111,37 +111,31 @@ private static ThreadLocal<object>? MersenneTesters;
 				return;
 			}
 
-		Console.WriteLine("Initializing");
-		int gpuPrimeThreads = Math.Max(1, _cliArguments.GpuPrimeThreads * PerfectNumberConstants.ThreadsByAccelerator);
-		int threadCount = Math.Max(1, _cliArguments.ThreadCount);
+			Console.WriteLine("Initializing");
 
-		EnvironmentConfiguration.MinK = _cliArguments.MinK;
-		EnvironmentConfiguration.GpuBatchSize = Math.Max(1, _cliArguments.ScanBatchSize);
-		EnvironmentConfiguration.GpuRatio = _cliArguments.GpuRatio;
-		EnvironmentConfiguration.MersenneDevice = _cliArguments.MersenneDevice;
-		ComputationDevice orderDevice = _cliArguments.OrderDevice;
-		if (_cliArguments.MersenneDevice == ComputationDevice.Gpu &&
-			_cliArguments.ByDivisorKIncrementMode == ByDivisorKIncrementMode.BitContradiction)
-		{
-			orderDevice = ComputationDevice.Gpu;
-		}
+			EnvironmentConfiguration.ByDivisorKIncrementMode = _cliArguments.ByDivisorKIncrementMode;
+			EnvironmentConfiguration.ByDivisorSpecialRange = _cliArguments.ByDivisorSpecialRange;
+			EnvironmentConfiguration.CalculationMethod =
+				_cliArguments.UseByDivisor ? CalculationMethod.ByDivisor
+				: _cliArguments.UseDivisor ? CalculationMethod.Divisor
+				: _cliArguments.UseResidue ? CalculationMethod.Residue
+				: _cliArguments.UseLucas ? CalculationMethod.LucasLehmer
+				: _cliArguments.KernelType == GpuKernelType.Pow2Mod ? CalculationMethod.Pow2Mod
+				: CalculationMethod.Incremental;
+			EnvironmentConfiguration.GpuBatchSize = Math.Max(1, _cliArguments.ScanBatchSize);
+			EnvironmentConfiguration.GpuPrimeThreads = Math.Max(1, _cliArguments.GpuPrimeThreads * PerfectNumberConstants.ThreadsByAccelerator);
+			EnvironmentConfiguration.GpuRatio = _cliArguments.GpuRatio;
+			EnvironmentConfiguration.MersenneDevice = _cliArguments.MersenneDevice;
+			EnvironmentConfiguration.MinK = _cliArguments.MinK;
+			EnvironmentConfiguration.OrderDevice = _cliArguments.OrderDevice;
+			EnvironmentConfiguration.ThreadCount = Math.Max(1, _cliArguments.ThreadCount);
 
-		EnvironmentConfiguration.OrderDevice = orderDevice;
-		EnvironmentConfiguration.UsePow2GroupDivisors = _cliArguments.ByDivisorKIncrementMode == ByDivisorKIncrementMode.Pow2Groups;
-		EnvironmentConfiguration.ByDivisorSpecialRange = _cliArguments.ByDivisorSpecialRange;
+			EnvironmentConfiguration.Initialize();
 
-		bool useGpu = orderDevice == ComputationDevice.Hybrid || orderDevice == ComputationDevice.Gpu ||
-				_cliArguments.MersenneDevice == ComputationDevice.Hybrid || _cliArguments.MersenneDevice == ComputationDevice.Gpu;
-
-		EnvironmentConfiguration.RollingAccelerators = useGpu 
-			? Math.Min(Math.Min(PerfectNumberConstants.RollingAccelerators, gpuPrimeThreads), threadCount)
-			: 0;
-
-		EnvironmentConfiguration.Initialize();
-
-		PrimeOrderCalculatorAccelerator.SkipOrderTablesAndKernels = _cliArguments.MersenneDevice == ComputationDevice.Gpu &&
-			_cliArguments.UseByDivisor &&
-			_cliArguments.ByDivisorKIncrementMode == ByDivisorKIncrementMode.BitContradiction;
+			PrimeOrderCalculatorAccelerator.SkipOrderTablesAndKernels =
+				EnvironmentConfiguration.MersenneDevice == ComputationDevice.Gpu &&
+				EnvironmentConfiguration.CalculationMethod == CalculationMethod.ByDivisor &&
+				EnvironmentConfiguration.ByDivisorKIncrementMode == ByDivisorKIncrementMode.BitContradiction;
 
 			// NttGpuMath.GpuTransformBackend = _cliArguments.NttBackend;
 			// NttGpuMath.ReductionMode = _cliArguments.ModReductionMode;
@@ -151,6 +145,9 @@ private static ThreadLocal<object>? MersenneTesters;
 			ulong remainder = currentP % 6UL;
 			bool startPrimeProvided = _cliArguments.StartPrimeProvided;
 			int gpuPrimeBatch = Math.Max(1, _cliArguments.GpuPrimeBatch);
+			bool useGpu = EnvironmentConfiguration.UseGpu;
+			int threadCount = EnvironmentConfiguration.ThreadCount;
+			int gpuPrimeThreads = EnvironmentConfiguration.GpuPrimeThreads;
 			// int gpuPrimeThreads = SharedGpuContext.Device.MaxNumThreads;
 			Console.WriteLine("Setting up parameters");
 			UnboundedTaskScheduler.ConfigureThreadCount(threadCount);
@@ -606,16 +603,16 @@ private static ThreadLocal<object>? MersenneTesters;
 				EnvironmentConfiguration.ByDivisorPow2Minus1Repository ??= Pow2Minus1StateRepository.Open(Path.Combine(checksDir, "pow2minus1.faster"), pow2MinusLmdb, checksDir);
 				EnvironmentConfiguration.BitContradictionStateRepository ??= BitContradictionStateRepository.Open(checksDir);
 
-			if (kIncrementMode == ByDivisorKIncrementMode.Pow2Groups)
-			{
-				EnvironmentConfiguration.ByDivisorSpecialStateRepository = KStateRepository.Open(checksDir, "pow2groups.special.kstate.faster", Path.Combine(checksDir, "pow2groups.special.kstate.lmdb"), checksDir);
-				EnvironmentConfiguration.ByDivisorGroupsStateRepository = KStateRepository.Open(checksDir, "pow2groups.groups.kstate.faster", Path.Combine(checksDir, "pow2groups.groups.kstate.lmdb"), checksDir);
-			}
-			else
-			{
-				EnvironmentConfiguration.ByDivisorSpecialStateRepository = null;
-				EnvironmentConfiguration.ByDivisorGroupsStateRepository = null;
-			}
+				if (kIncrementMode == ByDivisorKIncrementMode.Pow2Groups)
+				{
+					EnvironmentConfiguration.ByDivisorSpecialStateRepository = KStateRepository.Open(checksDir, "pow2groups.special.kstate.faster", Path.Combine(checksDir, "pow2groups.special.kstate.lmdb"), checksDir);
+					EnvironmentConfiguration.ByDivisorGroupsStateRepository = KStateRepository.Open(checksDir, "pow2groups.groups.kstate.faster", Path.Combine(checksDir, "pow2groups.groups.kstate.lmdb"), checksDir);
+				}
+				else
+				{
+					EnvironmentConfiguration.ByDivisorSpecialStateRepository = null;
+					EnvironmentConfiguration.ByDivisorGroupsStateRepository = null;
+				}
 			}
 
 			bool useFilter = !testMode && !string.IsNullOrEmpty(filterFile);
@@ -1409,48 +1406,48 @@ private static ThreadLocal<object>? MersenneTesters;
 			// Windowed pow2mod kernel handles by-divisor scans.
 			ByDivisorKIncrementMode mode = _cliArguments.ByDivisorKIncrementMode;
 			if (EnvironmentConfiguration.UseCpuOrder)
-				{
-					return mode switch
-					{
-						ByDivisorKIncrementMode.Pow2Groups => _byDivisorCpuOrderTesterPow2.IsPrime(p, out detailedCheck, out _),
-						ByDivisorKIncrementMode.Predictive => _byDivisorCpuOrderTesterPredictive.IsPrime(p, out detailedCheck, out _),
-						ByDivisorKIncrementMode.Percentile => _byDivisorCpuOrderTesterPercentile.IsPrime(p, out detailedCheck, out _),
-						ByDivisorKIncrementMode.Additive => _byDivisorCpuOrderTesterAdditive.IsPrime(p, out detailedCheck, out _),
-						ByDivisorKIncrementMode.TopDown => _byDivisorCpuOrderTesterTopDown.IsPrime(p, out detailedCheck, out _),
-						ByDivisorKIncrementMode.BitContradiction => _byDivisorCpuOrderTesterBitContradiction.IsPrime(p, out detailedCheck, out _),
-						ByDivisorKIncrementMode.BitTree => _byDivisorCpuOrderTesterBitTree.IsPrime(p, out detailedCheck, out _),
-						_ => _byDivisorCpuOrderTester.IsPrime(p, out detailedCheck, out _),
-					};
-				}
-
-				if (EnvironmentConfiguration.UseHybridOrder)
-				{
-					return mode switch
-					{
-						ByDivisorKIncrementMode.Pow2Groups => _byDivisorHybridOrderTesterPow2.IsPrime(p, out detailedCheck, out _),
-						ByDivisorKIncrementMode.Predictive => _byDivisorHybridOrderTesterPredictive.IsPrime(p, out detailedCheck, out _),
-						ByDivisorKIncrementMode.Percentile => _byDivisorHybridOrderTesterPercentile.IsPrime(p, out detailedCheck, out _),
-						ByDivisorKIncrementMode.Additive => _byDivisorHybridOrderTesterAdditive.IsPrime(p, out detailedCheck, out _),
-						ByDivisorKIncrementMode.TopDown => _byDivisorHybridOrderTesterTopDown.IsPrime(p, out detailedCheck, out _),
-						ByDivisorKIncrementMode.BitContradiction => _byDivisorHybridOrderTesterBitContradiction.IsPrime(p, out detailedCheck, out _),
-						ByDivisorKIncrementMode.BitTree => _byDivisorHybridOrderTesterBitTree.IsPrime(p, out detailedCheck, out _),
-						_ => _byDivisorHybridOrderTester.IsPrime(p, out detailedCheck, out _),
-					};
-				}
-
+			{
 				return mode switch
 				{
-					ByDivisorKIncrementMode.Pow2Groups => _byDivisorGpuOrderTesterPow2.IsPrime(p, out detailedCheck, out _),
-					ByDivisorKIncrementMode.Predictive => _byDivisorGpuOrderTesterPredictive.IsPrime(p, out detailedCheck, out _),
-					ByDivisorKIncrementMode.Percentile => _byDivisorGpuOrderTesterPercentile.IsPrime(p, out detailedCheck, out _),
-					ByDivisorKIncrementMode.Additive => _byDivisorGpuOrderTesterAdditive.IsPrime(p, out detailedCheck, out _),
-					ByDivisorKIncrementMode.TopDown => _byDivisorGpuOrderTesterTopDown.IsPrime(p, out detailedCheck, out _),
-					ByDivisorKIncrementMode.BitContradiction => _byDivisorGpuOrderTesterBitContradiction.IsPrime(p, out detailedCheck, out _),
-					ByDivisorKIncrementMode.BitTree => _byDivisorGpuOrderTesterBitTree.IsPrime(p, out detailedCheck, out _),
-					_ => _byDivisorGpuOrderTester.IsPrime(p, out detailedCheck, out _),
+					ByDivisorKIncrementMode.Pow2Groups => _byDivisorCpuOrderTesterPow2.IsPrime(p, out detailedCheck, out _),
+					ByDivisorKIncrementMode.Predictive => _byDivisorCpuOrderTesterPredictive.IsPrime(p, out detailedCheck, out _),
+					ByDivisorKIncrementMode.Percentile => _byDivisorCpuOrderTesterPercentile.IsPrime(p, out detailedCheck, out _),
+					ByDivisorKIncrementMode.Additive => _byDivisorCpuOrderTesterAdditive.IsPrime(p, out detailedCheck, out _),
+					ByDivisorKIncrementMode.TopDown => _byDivisorCpuOrderTesterTopDown.IsPrime(p, out detailedCheck, out _),
+					ByDivisorKIncrementMode.BitContradiction => _byDivisorCpuOrderTesterBitContradiction.IsPrime(p, out detailedCheck, out _),
+					ByDivisorKIncrementMode.BitTree => _byDivisorCpuOrderTesterBitTree.IsPrime(p, out detailedCheck, out _),
+					_ => _byDivisorCpuOrderTester.IsPrime(p, out detailedCheck, out _),
 				};
 			}
 
+			if (EnvironmentConfiguration.UseHybridOrder)
+			{
+				return mode switch
+				{
+					ByDivisorKIncrementMode.Pow2Groups => _byDivisorHybridOrderTesterPow2.IsPrime(p, out detailedCheck, out _),
+					ByDivisorKIncrementMode.Predictive => _byDivisorHybridOrderTesterPredictive.IsPrime(p, out detailedCheck, out _),
+					ByDivisorKIncrementMode.Percentile => _byDivisorHybridOrderTesterPercentile.IsPrime(p, out detailedCheck, out _),
+					ByDivisorKIncrementMode.Additive => _byDivisorHybridOrderTesterAdditive.IsPrime(p, out detailedCheck, out _),
+					ByDivisorKIncrementMode.TopDown => _byDivisorHybridOrderTesterTopDown.IsPrime(p, out detailedCheck, out _),
+					ByDivisorKIncrementMode.BitContradiction => _byDivisorHybridOrderTesterBitContradiction.IsPrime(p, out detailedCheck, out _),
+					ByDivisorKIncrementMode.BitTree => _byDivisorHybridOrderTesterBitTree.IsPrime(p, out detailedCheck, out _),
+					_ => _byDivisorHybridOrderTester.IsPrime(p, out detailedCheck, out _),
+				};
+			}
+
+			return mode switch
+			{
+				ByDivisorKIncrementMode.Pow2Groups => _byDivisorGpuOrderTesterPow2.IsPrime(p, out detailedCheck, out _),
+				ByDivisorKIncrementMode.Predictive => _byDivisorGpuOrderTesterPredictive.IsPrime(p, out detailedCheck, out _),
+				ByDivisorKIncrementMode.Percentile => _byDivisorGpuOrderTesterPercentile.IsPrime(p, out detailedCheck, out _),
+				ByDivisorKIncrementMode.Additive => _byDivisorGpuOrderTesterAdditive.IsPrime(p, out detailedCheck, out _),
+				ByDivisorKIncrementMode.TopDown => _byDivisorGpuOrderTesterTopDown.IsPrime(p, out detailedCheck, out _),
+				ByDivisorKIncrementMode.BitContradiction => _byDivisorGpuOrderTesterBitContradiction.IsPrime(p, out detailedCheck, out _),
+				ByDivisorKIncrementMode.BitTree => _byDivisorGpuOrderTesterBitTree.IsPrime(p, out detailedCheck, out _),
+				_ => _byDivisorGpuOrderTester.IsPrime(p, out detailedCheck, out _),
+			};
+		}
+
 		detailedCheck = IsMersennePrime(gpu, p);
 		return detailedCheck;
 	}

PerfectNumbers.Core/ByDivisor/BitContradictionSolverWithQMultiplier.cs

@@ -9,13 +9,19 @@ public readonly struct ColumnBounds(int forcedOnes, int possibleOnes)
 	public readonly int PossibleOnes = possibleOnes;
 }
 
-public readonly struct CarryRange(long min, long max)
+public readonly struct CarryRange(long min, long max, byte mod4Mask = byte.MinValue)
 {
 	public readonly long Min = min;
 	public readonly long Max = max;
+	public readonly byte Mod4Mask = mod4Mask;
 
-	public static CarryRange Single(long value) => new(value, value);
-	public static readonly CarryRange Zero = Single(0);
+	public static CarryRange Single(long value) => new(value, value, byte.MinValue);
+	public static CarryRange Single(long value, byte mod4Mask) => new(value, value, mod4Mask);
+
+	public static readonly CarryRange Zero = Single(0, 0);
+
+	[MethodImpl(MethodImplOptions.AggressiveInlining)]
+    public bool IsEmpty() => Min > Max || Mod4Mask == 0;
 }
 
 public enum ContradictionReason

PerfectNumbers.Core/ByDivisorKIncrementMode.cs

@@ -0,0 +1,16 @@
+namespace PerfectNumbers.Core;
+#pragma warning disable CA2211 // The fields are settable on purpose for the best performance
+
+public enum ByDivisorKIncrementMode
+{
+    Sequential,
+    Pow2Groups,
+    Predictive,
+    Percentile,
+    Additive,
+    TopDown,
+    BitContradiction,
+    BitTree,
+}
+
+#pragma warning restore CA2211 // Non-constant fields should not be visible

PerfectNumbers.Core/CacheStatus.cs

@@ -0,0 +1,7 @@
+namespace PerfectNumbers.Core;
+
+public enum CacheStatus
+{
+    Disabled,
+    Enabled,
+}

PerfectNumbers.Core/CalculationMethod.cs

@@ -0,0 +1,11 @@
+namespace PerfectNumbers.Core;
+
+public enum CalculationMethod
+{
+    ByDivisor,
+    Divisor,
+    Incremental,
+    LucasLehmer,
+    Pow2Mod,
+    Residue,
+}

PerfectNumbers.Core/Cpu/MersenneNumberDivisorByDivisorCpuTesterWithTemplate.cs

@@ -63,7 +63,7 @@ public struct MersenneNumberDivisorByDivisorCpuTesterWithTemplate() : IMersenneN
 	private static readonly ConcurrentDictionary<ulong, byte> _checkedPow2MinusOne = new();
 	private ulong _currentPrime;
 
-#if DEVICE_GPU
+#if DEVICE_GPU && DivisorSet_BitContradiction
 	[MethodImpl(MethodImplOptions.AggressiveOptimization)]
 	private bool CheckDivisorsBitContradictionGpu(
 		ulong prime,
@@ -78,6 +78,17 @@ public struct MersenneNumberDivisorByDivisorCpuTesterWithTemplate() : IMersenneN
 		foundDivisor = BigInteger.Zero;
 
 		BigInteger currentK = minK;
+
+#if OVERRIDE_MIN_K
+		// Regression test convenience: ensure we actually scan the range that contains the known divisor for p=138000001.
+		if (currentK > 1)
+		{
+			Console.WriteLine($"[GPU OVERRIDE] p={prime} forcing minK from {currentK} to 1 for known-divisor regression test.");
+			currentK = BigInteger.One;
+		}
+#endif
+		const int TargetBatchIndex = 273;
+
 		bool hasLimit = !allowedMax.IsZero;
 		BigInteger maxK = hasLimit ? (allowedMax - BigInteger.One) / step : BigInteger.Zero;
 
@@ -100,8 +111,7 @@ public struct MersenneNumberDivisorByDivisorCpuTesterWithTemplate() : IMersenneN
 		BigInteger batchSizeBig = batchSize;
 		int wordCount = BitContradictionKernels.MaxQWordCount;
 		int inputCapacity = maxThreadCount * wordCount;
-		gpu.EnsureCapacity(PerfectNumberConstants.DefaultFactorsBuffer, inputCapacity);
-		gpu.EnsurePartialFactorCapacity(inputCapacity);
+		var buffers = gpu.EnsureBitContradictionCapacity(inputCapacity, inputCapacity, maxThreadCount);
 
 		var stream = AcceleratorStreamPool.Rent(gpu.AcceleratorIndex);
 		var kernels = GpuKernelPool.GetOrAddKernels(gpu.AcceleratorIndex, stream, KernelType.BitContradictionScan);
@@ -109,13 +119,158 @@ public struct MersenneNumberDivisorByDivisorCpuTesterWithTemplate() : IMersenneN
 
 		int[] found = new int[maxThreadCount];
 		ulong[] batchIndexWords = new ulong[inputCapacity];
-		using MemoryBuffer1D<int, Stride1D.Dense> delta8Keys = gpu.Accelerator.Allocate1D<int>(maxThreadCount * BitContradictionKernels.DeltaCacheSlots);
-		using MemoryBuffer1D<int, Stride1D.Dense> delta8Cache = gpu.Accelerator.Allocate1D<int>(maxThreadCount * BitContradictionKernels.DeltaCacheSlots * 4194304);
-		#if DETAILED_LOG
-			int debugWordsPerSlot = BitContradictionKernels.DebugWordCountPerSlot;
-			using MemoryBuffer1D<ulong, Stride1D.Dense> debugBuffer = gpu.Accelerator.Allocate1D<ulong>(maxThreadCount * debugWordsPerSlot);
-			ArrayView1D<ulong, Stride1D.Dense> debugView = debugBuffer.View;
-		#endif
+		ArrayView1D<int, Stride1D.Dense> delta8KeysView = buffers.Delta8CacheKeysView;
+		ArrayView1D<int, Stride1D.Dense> delta8CacheView = buffers.Delta8CacheValuesView;
+
+		Span<int> temp = stackalloc int[(int)delta8KeysView.Length];
+		temp.Fill(int.MinValue);
+		delta8KeysView.CopyFromCPU(stream, temp);
+
+
+#if INVERSION_VALIDATION
+			///
+			/// INVERSION VALIDATION
+			/// 
+			{
+				// --- Inputs (known test case)
+				ulong p = 138000001UL;
+				ulong k = 4383UL;
+
+				// --- ILGPU setup (minimal)
+				var accelerator = gpu.Accelerator;
+
+				// --- Compile kernel (Extent=1)
+				// Kernel signature must be: (Index1D, ulong p, ulong k, ArrayView1D<ulong, Stride1D.Dense> outBuf)
+				var kernel2 = accelerator.LoadAutoGroupedStreamKernel<
+					Index1D,
+					ulong,
+					ulong,
+					ArrayView1D<ulong, Stride1D.Dense>
+				>(BitContradictionKernels.BitContradictionKernelExportInverse1024);
+
+				// --- Output buffer (64 ulongs is enough: 16 qPrefix + 16 inv + 16 aLow + meta)
+				using var outBufGpu = accelerator.Allocate1D<ulong>(64);
+				var outCpu = new ulong[64];
+
+				// Optional: clear buffer
+				for (int i = 0; i < outCpu.Length; i++) outCpu[i] = 0;
+				outBufGpu.View.CopyFromCPU(stream, outCpu);
+
+				// --- Run kernel
+				kernel2(1, p, k, outBufGpu.View);
+				stream.Synchronize();
+
+				// --- Copy back
+				outBufGpu.View.CopyToCPU(stream, outCpu);
+				stream.Synchronize();
+
+				// --- Parse meta
+				ulong pOut = outCpu[48];
+				ulong kOut = outCpu[49];
+				ulong qOut = outCpu[50];
+				ulong marker = outCpu[51];
+
+				Console.WriteLine($"[GPU EXPORT] pOut={pOut} kOut={kOut} qOut={qOut} marker=0x{marker:X}");
+
+				// --- Helpers: build BigInteger from 16x ulong limbs (little-endian)
+				BigInteger FromU64Le(int startIndex)
+				{
+					// +1 byte to force positive sign for BigInteger ctor
+					byte[] bytes = new byte[16 * 8 + 1];
+					for (int i = 0; i < 16; i++)
+					{
+						ulong w = outCpu[startIndex + i];
+						int o = i * 8;
+						bytes[o + 0] = (byte)(w);
+						bytes[o + 1] = (byte)(w >> 8);
+						bytes[o + 2] = (byte)(w >> 16);
+						bytes[o + 3] = (byte)(w >> 24);
+						bytes[o + 4] = (byte)(w >> 32);
+						bytes[o + 5] = (byte)(w >> 40);
+						bytes[o + 6] = (byte)(w >> 48);
+						bytes[o + 7] = (byte)(w >> 56);
+					}
+
+					return new BigInteger(bytes);
+				}
+
+				var qPrefix = FromU64Le(0);
+				var inv = FromU64Le(16);
+				var aLow = FromU64Le(32);
+
+				var mask = (BigInteger.One << 1024) - 1;
+
+				// --- Validate inverse: (qPrefix * inv) mod 2^1024 == 1
+				var prod = (qPrefix * inv) & mask;
+				bool invOk = prod == BigInteger.One;
+
+				// --- Validate aLow: aLow == (-inv) mod 2^1024
+				var expectedALow = (mask + BigInteger.One - (inv & mask)) & mask;
+				bool aLowOk = ((aLow & mask) == expectedALow);
+
+				// --- Print concise diagnostics
+				Console.WriteLine($"[CPU VALIDATION] invOk={invOk} aLowOk={aLowOk}");
+
+				// Print low 64 bits (quick sanity)
+				ulong prodLow64 = (ulong)(prod & ((BigInteger.One << 64) - 1));
+				ulong invLow64 = (ulong)(inv & ((BigInteger.One << 64) - 1));
+				ulong aLowLow64 = (ulong)(aLow & ((BigInteger.One << 64) - 1));
+
+				Console.WriteLine($"[LOW64] qPrefix=0x{(ulong)(qPrefix & ((System.Numerics.BigInteger.One<<64)-1)):X}");
+				Console.WriteLine($"[LOW64] inv=0x{invLow64:X} aLow=0x{aLowLow64:X} prod=0x{prodLow64:X}");
+
+				// Optional: hard fail for CI/regression
+				if (!invOk || !aLowOk)
+					throw new System.InvalidOperationException($"Inverse export validation failed: invOk={invOk}, aLowOk={aLowOk}");
+			}
+#endif
+
+#if SELF_TEST
+			///
+			/// SELF-TEST
+			/// 
+			{
+				ulong p = 138000001UL;
+				ulong k = 4383UL;
+
+				var accelerator = gpu.Accelerator;
+
+				Console.WriteLine($"Compiling kernel for {prime}");
+				var kernel3 = accelerator.LoadAutoGroupedStreamKernel<
+					Index1D,
+					ulong,
+					ulong,
+					int,
+					ArrayView1D<ulong, Stride1D.Dense>
+				>(BitContradictionKernels.BitContradictionKernelDebugDpSingleQ);
+
+				var launcher3 = KernelUtil.GetKernel(kernel3).CreateLauncherDelegate<
+					Action<AcceleratorStream, Index1D, ulong, ulong, int, ArrayView1D<ulong, Stride1D.Dense>>>();
+
+				Console.WriteLine($"Preparing buffers for {prime}");
+				using var outGpu = accelerator.Allocate1D<ulong>(128);
+				var outCpu = new ulong[128];
+				Array.Clear(outCpu, 0, outCpu.Length);
+				outGpu.View.CopyFromCPU(stream, outCpu);
+
+				// First run: disable batching (scalar DP)
+				Console.WriteLine($"Launching GPU kernel for {prime}");
+				launcher3(stream, 1, p, k, 1, outGpu.View);
+				outGpu.View.CopyToCPU(stream, outCpu);
+				stream.Synchronize();
+
+				Console.WriteLine($"[DP DEBUG] marker=0x{outCpu[0]:X} p={outCpu[1]} k={outCpu[2]} q={outCpu[3]} qBitLen={outCpu[4]} offsetCount={outCpu[6]} ok={outCpu[9]} carryMin={outCpu[10]} carryMax={outCpu[11]}");
+
+				// Print first 10 step records
+				Console.WriteLine($"[DP DEBUG] marker=0x{outCpu[0]:X} p={outCpu[1]} k={outCpu[2]} q={outCpu[3]} qBitLen={outCpu[4]} offsetCount={outCpu[6]} ok={outCpu[9]} carryMin={outCpu[10]} carryMax={outCpu[11]} disableBatching={outCpu[12]}");
+				Console.WriteLine($"[DP EXTRA] maxAllowedA={outCpu[14]} maxFixed={outCpu[15]} windowSize={outCpu[16]} aWordCount={outCpu[17]}");
+				Console.WriteLine($"[DP FIRST] firstForced={unchecked((long)outCpu[18])} firstRemaining={unchecked((long)outCpu[19])} firstQWordsEff={outCpu[20]} firstQLastEff=0x{outCpu[21]:X} firstCarryMin={unchecked((long)outCpu[22])} firstCarryMax={unchecked((long)outCpu[23])}");
+				Console.WriteLine($"[DP SNAP] aLow0=0x{outCpu[24]:X} knownOne0=0x{outCpu[25]:X} knownZero0=0x{outCpu[26]:X} aOneWin0=0x{outCpu[27]:X} aZeroWin0=0x{outCpu[28]:X} qMask0=0x{outCpu[29]:X} lastMask=0x{outCpu[30]:X} off0={outCpu[31]} offLast={outCpu[32]}");
+
+				if (outCpu[0] != 0xD00DF00DUL)
+					throw new Exception("DP debug kernel did not run as expected.");
+			}
+#endif
 
 		BigInteger currentBatch = (currentK - BigInteger.One) / batchSizeBig;
 		currentK = (currentBatch * batchSizeBig) + BigInteger.One;
@@ -123,7 +278,6 @@ public struct MersenneNumberDivisorByDivisorCpuTesterWithTemplate() : IMersenneN
 
 		while (true)
 		{
-
 			if (hasLimit && currentBatch > maxBatch)
 			{
 				processedAll = true;
@@ -137,6 +291,7 @@ public struct MersenneNumberDivisorByDivisorCpuTesterWithTemplate() : IMersenneN
 				break;
 			}
 
+			Console.WriteLine($"Preparing data for{prime}, batch {currentBatch}");
 			int threadCount = maxThreadCount;
 			if (remainingBatches < threadCount)
 			{
@@ -162,15 +317,28 @@ public struct MersenneNumberDivisorByDivisorCpuTesterWithTemplate() : IMersenneN
 				WriteBigIntegerToWords(batchIndex, batchIndexWords.AsSpan(t * wordCount, wordCount));
 			}
 
-			gpu.InputView.CopyFromCPU(stream, batchIndexWords.AsSpan(0, threadCount * wordCount));
-			kernel(stream, threadCount, prime, gpu.InputView, countQ, gpu.OutputIntView,
-				   #if DETAILED_LOG
-					   debugView,
-				   #endif
-				   delta8Keys.View, delta8Cache.View);
+			Console.WriteLine($"Copying data to GPU for {prime}, batch {currentBatch}");
+			buffers.BatchIndexWordsView.CopyFromCPU(stream, batchIndexWords.AsSpan(0, threadCount * wordCount));
+// #if DETAILED_LOG
+				Console.WriteLine($"Launching GPU kernel for {prime}, batch {currentBatch}");
+// #endif
+			kernel(stream, 1, prime, buffers.BatchIndexWordsView, countQ, buffers.FoundQWordsView,
+			// kernel(stream, threadCount, prime, buffers.BatchIndexWords, countQ, buffers.FoundQWords,
+					delta8KeysView, delta8CacheView);
+			buffers.FoundQWordsView.CopyToCPU(stream, found.AsSpan(0, threadCount));
 			stream.Synchronize();
-			gpu.OutputIntView.CopyToCPU(stream, found.AsSpan(0, threadCount));
 
+			_consolePrintOut++;
+			if  (_consolePrintOut >= PerfectNumberConstants.ConsoleInterval)
+			{
+				Console.WriteLine($"GPU kernel for {prime} complete, batch {currentBatch}");
+				Console.WriteLine($"[GPU FOUND RAW] batch={currentBatch} found[0]={found[0]}");
+				_consolePrintOut = 0;
+			}
+#if DETAILED_LOG || SELF_TEST
+#endif
+
+			// TODO: This loop is incorrect. We have a loop over the threads outside from here
 			for (int i = 0; i < threadCount; i++)
 			{
 				int foundIndex = found[i];
@@ -217,49 +385,92 @@ public struct MersenneNumberDivisorByDivisorCpuTesterWithTemplate() : IMersenneN
 					// if (!dividesExact)
 					// {
 					// 	throw new InvalidOperationException($"BitContradiction CPU/GPU false positive for p={prime}, q={candidateQ}, k={candidateK}, batch={batchIndex}");
-					#if DETAILED_LOG
+#if DETAILED_LOG
 						if (prime == 138000001UL)
+						// --- Fast self-check for known divisor: run scan kernel with ONE thread and SMALL countQ.
+						// This avoids waiting for other threads that would scan huge ranges and never hit the target.
 						{
-							ulong[] debugWords = new ulong[debugWordsPerSlot];
-							debugView.SubView(i * debugWordsPerSlot, debugWordsPerSlot).CopyToCPU(stream, debugWords);
-							BigInteger gpuQ = ReadBigIntegerFromWords(debugWords.AsSpan(0, 16));
-							BigInteger gpuInv = ReadBigIntegerFromWords(debugWords.AsSpan(16, 16));
-							BigInteger gpuALow = ReadBigIntegerFromWords(debugWords.AsSpan(32, 16));
-							ulong gpuQBits = debugWords[48];
-							ulong gpuQWordCount = debugWords[49];
-							ulong gpuQMask0 = debugWords[50];
-							ulong gpuQMask1 = debugWords[51];
-							ulong gpuKnownOne0 = debugWords[52];
-							ulong gpuKnownZero0 = debugWords[53];
-							ulong gpuAOneWin0 = debugWords[54];
-							ulong gpuAZeroWin0 = debugWords[55];
-							long gpuCarryMin = unchecked((long)debugWords[56]);
-							long gpuCarryMax = unchecked((long)debugWords[57]);
-							long gpuMaxAllowedA = unchecked((long)debugWords[58]);
-							long gpuMaxFixed = unchecked((long)debugWords[59]);
-							long gpuWindowSize = unchecked((long)debugWords[60]);
-							long gpuAWordCount = unchecked((long)debugWords[61]);
-							long gpuMaxKnownA = unchecked((long)debugWords[62]);
-							long gpuMaxFixedPrefilter = unchecked((long)debugWords[63]);
-							ulong preKnownOne0 = debugWords[64];
-							ulong preKnownZero0 = debugWords[65];
-							ulong preAOneWin0 = debugWords[66];
-							ulong preAZeroWin0 = debugWords[67];
-							long preCarryMin = unchecked((long)debugWords[68]);
-							long preCarryMax = unchecked((long)debugWords[69]);
-							long preSegmentBase = unchecked((long)debugWords[70]);
-							long preWindowColumn = unchecked((long)debugWords[71]);
-							long gpuOffsetCount = unchecked((long)debugWords[72]);
-							long gpuMaxOffset = unchecked((long)debugWords[73]);
-							long gpuFirstForced = unchecked((long)debugWords[74]);
-							long gpuFirstRemaining = unchecked((long)debugWords[75]);
-							long gpuFirstQWordsEff = unchecked((long)debugWords[76]);
-							ulong gpuFirstQLastEff = debugWords[77];
-							long gpuFirstCarryMin = unchecked((long)debugWords[78]);
-							long gpuFirstCarryMax = unchecked((long)debugWords[79]);
-							throw new InvalidOperationException($"BitContradiction GPU debug for p={prime}, q={candidateQ}, k={candidateK}, qBits={gpuQBits}, qWordCount={gpuQWordCount}, qMask0=0x{gpuQMask0:X16}, qMask1=0x{gpuQMask1:X16}, knownOne0=0x{gpuKnownOne0:X16}, knownZero0=0x{gpuKnownZero0:X16}, aOneWin0=0x{gpuAOneWin0:X16}, aZeroWin0=0x{gpuAZeroWin0:X16}, carryMin={gpuCarryMin}, carryMax={gpuCarryMax}, maxAllowedA={gpuMaxAllowedA}, maxFixed={gpuMaxFixed}, windowSize={gpuWindowSize}, aWordCount={gpuAWordCount}, maxKnownA={gpuMaxKnownA}, maxFixedPrefilter={gpuMaxFixedPrefilter}, preKnownOne0=0x{preKnownOne0:X16}, preKnownZero0=0x{preKnownZero0:X16}, preAOneWin0=0x{preAOneWin0:X16}, preAZeroWin0=0x{preAZeroWin0:X16}, preCarryMin={preCarryMin}, preCarryMax={preCarryMax}, preSegmentBase={preSegmentBase}, preWindowColumn={preWindowColumn}, offsetCount={gpuOffsetCount}, maxOffset={gpuMaxOffset}, firstForced={gpuFirstForced}, firstRemaining={gpuFirstRemaining}, firstQWordsEff={gpuFirstQWordsEff}, firstQLastEff=0x{gpuFirstQLastEff:X16}, firstCarryMin={gpuFirstCarryMin}, firstCarryMax={gpuFirstCarryMax}, gpuQ={gpuQ}, gpuInv={gpuInv}, gpuALow={gpuALow}");
+							const int expectedFoundIndex = 4382; // k=4383 => index=k-1 in batch 0
+							countQ = 6000; // must be > 4382, keep small to finish quickly
+
+							Array.Clear(found, 0, found.Length);
+							Array.Clear(batchIndexWords, 0, batchIndexWords.Length);
+
+							// batchIndex = 0
+							WriteBigIntegerToWords(BigInteger.Zero, batchIndexWords.AsSpan(0, wordCount));
+							buffers.BatchIndexWords.CopyFromCPU(stream, batchIndexWords.AsSpan(0, threadCount * wordCount));
+
+							Console.WriteLine($"[GPU SELFTEST] launching scan kernel (threads={threadCount}, countQ={countQ})");
+							kernel(
+								stream,
+								threadCount,
+								prime,
+								buffers.BatchIndexWords,
+								countQ,
+								buffers.FoundQWords,
+						#if DETAILED_LOG
+								debugView,
+						#endif
+								delta8Keys.View,
+								delta8Cache.View);
+
+							stream.Synchronize();
+							buffers.FoundQWords.CopyToCPU(stream, found.AsSpan(0, threadCount));
+							stream.Synchronize();
+
+							Console.WriteLine($"[GPU SELFTEST] found[0]={found[0]} (expected {expectedFoundIndex})");
+							if (found[0] != expectedFoundIndex)
+							{
+								throw new InvalidOperationException($"[GPU SELFTEST FAILED] got={found[0]} expected={expectedFoundIndex}");
+							}
+
+							// Self-test succeeded => stop here (do not start full scan which may hit TDR during debugging).
+							processedAll = false;
+							foundDivisor = BigInteger.Zero;
+							AcceleratorStreamPool.Return(gpu.AcceleratorIndex, stream);
+							return false;
 						}
-					#endif
+						// {
+						// 	ulong[] debugWords = new ulong[debugWordsPerSlot];
+						// 	debugView.SubView(i * debugWordsPerSlot, debugWordsPerSlot).CopyToCPU(stream, debugWords);
+						// 	BigInteger gpuQ = ReadBigIntegerFromWords(debugWords.AsSpan(0, 16));
+						// 	BigInteger gpuInv = ReadBigIntegerFromWords(debugWords.AsSpan(16, 16));
+						// 	BigInteger gpuALow = ReadBigIntegerFromWords(debugWords.AsSpan(32, 16));
+						// 	ulong gpuQBits = debugWords[48];
+						// 	ulong gpuQWordCount = debugWords[49];
+						// 	ulong gpuQMask0 = debugWords[50];
+						// 	ulong gpuQMask1 = debugWords[51];
+						// 	ulong gpuKnownOne0 = debugWords[52];
+						// 	ulong gpuKnownZero0 = debugWords[53];
+						// 	ulong gpuAOneWin0 = debugWords[54];
+						// 	ulong gpuAZeroWin0 = debugWords[55];
+						// 	long gpuCarryMin = unchecked((long)debugWords[56]);
+						// 	long gpuCarryMax = unchecked((long)debugWords[57]);
+						// 	long gpuMaxAllowedA = unchecked((long)debugWords[58]);
+						// 	long gpuMaxFixed = unchecked((long)debugWords[59]);
+						// 	long gpuWindowSize = unchecked((long)debugWords[60]);
+						// 	long gpuAWordCount = unchecked((long)debugWords[61]);
+						// 	long gpuMaxKnownA = unchecked((long)debugWords[62]);
+						// 	long gpuMaxFixedPrefilter = unchecked((long)debugWords[63]);
+						// 	ulong preKnownOne0 = debugWords[64];
+						// 	ulong preKnownZero0 = debugWords[65];
+						// 	ulong preAOneWin0 = debugWords[66];
+						// 	ulong preAZeroWin0 = debugWords[67];
+						// 	long preCarryMin = unchecked((long)debugWords[68]);
+						// 	long preCarryMax = unchecked((long)debugWords[69]);
+						// 	long preSegmentBase = unchecked((long)debugWords[70]);
+						// 	long preWindowColumn = unchecked((long)debugWords[71]);
+						// 	long gpuOffsetCount = unchecked((long)debugWords[72]);
+						// 	long gpuMaxOffset = unchecked((long)debugWords[73]);
+						// 	long gpuFirstForced = unchecked((long)debugWords[74]);
+						// 	long gpuFirstRemaining = unchecked((long)debugWords[75]);
+						// 	long gpuFirstQWordsEff = unchecked((long)debugWords[76]);
+						// 	ulong gpuFirstQLastEff = debugWords[77];
+						// 	long gpuFirstCarryMin = unchecked((long)debugWords[78]);
+						// 	long gpuFirstCarryMax = unchecked((long)debugWords[79]);
+						// 	throw new InvalidOperationException($"BitContradiction GPU debug for p={prime}, q={candidateQ}, k={candidateK}, qBits={gpuQBits}, qWordCount={gpuQWordCount}, qMask0=0x{gpuQMask0:X16}, qMask1=0x{gpuQMask1:X16}, knownOne0=0x{gpuKnownOne0:X16}, knownZero0=0x{gpuKnownZero0:X16}, aOneWin0=0x{gpuAOneWin0:X16}, aZeroWin0=0x{gpuAZeroWin0:X16}, carryMin={gpuCarryMin}, carryMax={gpuCarryMax}, maxAllowedA={gpuMaxAllowedA}, maxFixed={gpuMaxFixed}, windowSize={gpuWindowSize}, aWordCount={gpuAWordCount}, maxKnownA={gpuMaxKnownA}, maxFixedPrefilter={gpuMaxFixedPrefilter}, preKnownOne0=0x{preKnownOne0:X16}, preKnownZero0=0x{preKnownZero0:X16}, preAOneWin0=0x{preAOneWin0:X16}, preAZeroWin0=0x{preAZeroWin0:X16}, preCarryMin={preCarryMin}, preCarryMax={preCarryMax}, preSegmentBase={preSegmentBase}, preWindowColumn={preWindowColumn}, offsetCount={gpuOffsetCount}, maxOffset={gpuMaxOffset}, firstForced={gpuFirstForced}, firstRemaining={gpuFirstRemaining}, firstQWordsEff={gpuFirstQWordsEff}, firstQLastEff=0x{gpuFirstQLastEff:X16}, firstCarryMin={gpuFirstCarryMin}, firstCarryMax={gpuFirstCarryMax}, gpuQ={gpuQ}, gpuInv={gpuInv}, gpuALow={gpuALow}");
+						// }
+#endif
 
 					foundDivisor = candidateQ;
 					AcceleratorStreamPool.Return(gpu.AcceleratorIndex, stream);
@@ -267,6 +478,12 @@ public struct MersenneNumberDivisorByDivisorCpuTesterWithTemplate() : IMersenneN
 				}
 			}
 
+			if (currentBatch == TargetBatchIndex)
+			{
+				Console.WriteLine($"No divisor found up to batch index = {currentBatch}, with the target batch index = {TargetBatchIndex}");
+				Environment.Exit(1);
+			}
+
 		ContinueOuter:
 			RecordState(currentK);
 			if (hasLimit && currentK > maxK)
@@ -3039,7 +3256,7 @@ public struct MersenneNumberDivisorByDivisorCpuTesterWithTemplate() : IMersenneN
 				}
 
 			ContinueOuter:
-			RecordState(currentK);
+				RecordState(currentK);
 				// if (divisorCycle == prime && !IsMersenneValue(prime, candidate))
 				if (divisorCycle == prime)
 				{
@@ -3160,7 +3377,7 @@ public struct MersenneNumberDivisorByDivisorCpuTesterWithTemplate() : IMersenneN
 			{
 				// Stopwatch sw = Stopwatch.StartNew();
 			ContinueOuter:
-			RecordState(currentK);
+				RecordState(currentK);
 				ulong powMod;
 				bool isPow2 = IsPowerOfTwo(currentK);
 #if DETAILED_LOG
@@ -3262,7 +3479,7 @@ public struct MersenneNumberDivisorByDivisorCpuTesterWithTemplate() : IMersenneN
 					if (remainder8 == 1 || remainder8 == 7)
 					{
 					ContinueOuter:
-			RecordState(currentK);
+						RecordState(currentK);
 						BigInteger powResult = BigInteger.ModPow(BigIntegerNumbers.Two, primeBig, divisor);
 						if (powResult.IsOne)
 						{
@@ -3962,7 +4179,7 @@ public struct MersenneNumberDivisorByDivisorCpuTesterWithTemplate() : IMersenneN
 					if (remainder8 == 1 || remainder8 == 7)
 					{
 					ContinueOuter:
-			RecordState(currentK);
+						RecordState(currentK);
 						BigInteger powResult = BigInteger.ModPow(BigIntegerNumbers.Two, primeBig, divisor);
 						// if (powResult.IsOne && !IsMersenneValue(prime, divisor))
 						if (powResult.IsOne)
@@ -4317,7 +4534,7 @@ public struct MersenneNumberDivisorByDivisorCpuTesterWithTemplate() : IMersenneN
 				}
 
 			ContinueOuter:
-			RecordState(currentK);
+				RecordState(currentK);
 				if (divisorCycle == prime && !IsMersenneValue(prime, divisor))
 				{
 					foundDivisor = divisor;
@@ -4891,6 +5108,9 @@ public struct MersenneNumberDivisorByDivisorCpuTesterWithTemplate() : IMersenneN
 		while (true)
 		{
 			int remainder8 = (stepMod8 * kRem8 + 1) & 7;
+			if (prime == 138000001UL && divisor == 1209708008767)
+				Console.WriteLine("[HIT] enumerated target q before remainder8 check");
+
 			if (remainder8 != 1 && remainder8 != 7)
 			{
 				goto MOVE_NEXT;
@@ -4932,6 +5152,12 @@ public struct MersenneNumberDivisorByDivisorCpuTesterWithTemplate() : IMersenneN
 				}
 			}
 
+			if (prime == 138000001UL && divisor == 1209708008767)
+			{
+				Console.WriteLine("[HIT] processed target q. Stopping.");
+				Environment.Exit(1);
+			}
+
 		MOVE_NEXT:
 		ContinueOuter:
 			RecordState(currentK);
@@ -5090,6 +5316,7 @@ public struct MersenneNumberDivisorByDivisorCpuTesterWithTemplate() : IMersenneN
 				{
 					oneOffsets[i + 1] = pow2Offsets[i] + 1;
 				}
+
 				oneCount = needed;
 				usedPow2Offsets = true;
 			}
@@ -5114,6 +5341,7 @@ public struct MersenneNumberDivisorByDivisorCpuTesterWithTemplate() : IMersenneN
 				pow2OffsetsTarget = new int[oneCount];
 				_pow2Offsets = pow2OffsetsTarget;
 			}
+
 			Array.Copy(oneOffsets, pow2OffsetsTarget, oneCount);
 			_pow2OffsetCount = oneCount;
 			_pow2K = currentK;
@@ -5128,23 +5356,25 @@ public struct MersenneNumberDivisorByDivisorCpuTesterWithTemplate() : IMersenneN
 
 		ReadOnlySpan<int> oneOffsetsSlice = oneOffsets[..oneCount];
 		// BitContradictionSolverWithQMultiplier.SetDebugEnabled(true);
-		#if DETAILED_LOG
+#if DETAILED_LOG
 			bool decided = BitContradictionSolverWithAMultiplier.TryCheckDivisibilityFromOneOffsets(oneOffsetsSlice, prime, currentK, isPowerOfTwo, out divides, out var reason);
-		#else
+#else
 			bool decided = BitContradictionSolverWithAMultiplier.TryCheckDivisibilityFromOneOffsets(oneOffsetsSlice, prime, currentK, isPowerOfTwo, out divides);
-		#endif
+#endif
 		// BitContradictionSolverWithQMultiplier.SetDebugEnabled(false);
 		if (decided && divides)
-		{
+		{			
 			// Verify to avoid false positives on structurally admissible but non-dividing q.
 			bool structuralOk = ((divisor - BigInteger.One) % (((BigInteger)prime) << 1)) == BigInteger.Zero;
 			bool dividesExact = structuralOk && BigInteger.ModPow(2, prime, divisor) == BigInteger.One;
 			if (!dividesExact)
 			{
+				throw new InvalidOperationException($"We reported divisor q={divisor} as dividing {prime} but it doesn't divide it");
+
 				divides = false;
-				#if DETAILED_LOG
+#if DETAILED_LOG
 					reason = ContradictionReason.ParityUnreachable;
-				#endif
+#endif
 			}
 		}
 		if (decided && !divides)

PerfectNumbers.Core/EnvironmentConfiguration.cs

@@ -4,18 +4,21 @@ using PerfectNumbers.Core.Gpu;
 
 namespace PerfectNumbers.Core;
 
-#pragma warning disable CA2211 // The fields are settable on purpose for the best performance
+#pragma warning disable CA2211 // Non-constant fields should not be visible
 
 public static class EnvironmentConfiguration
 {
 	public static ulong DivisorLimit = ulong.MaxValue;
 	public static int GpuBatchSize = GpuConstants.ScanBatchSize;
+	public static int GpuPrimeThreads = 1;
 	public static int GpuRatio = PerfectNumberConstants.GpuRatio;
 	public static ComputationDevice MersenneDevice = ComputationDevice.Cpu;
 	public static BigInteger MinK = BigInteger.One;
 	public static ComputationDevice OrderDevice = ComputationDevice.Cpu;
 	public static int RollingAccelerators = PerfectNumberConstants.RollingAccelerators;
 	public static string? StateFilePath;
+	public static int ThreadCount;
+	public static bool UseGpu;
 	public static bool UseGpuOrder;
 	public static bool UseHybridOrder;
 	public static bool UseCpuOrder;
@@ -24,6 +27,7 @@ public static class EnvironmentConfiguration
 	public static ByDivisor.ByDivisorClassModel? ByDivisorClassModel;
 	public static ByDivisor.ByDivisorScanTuning ByDivisorScanTuning = ByDivisor.ByDivisorScanTuning.Default;
 	public static double ByDivisorCheapKLimit = 10_000d;
+	public static ByDivisorKIncrementMode ByDivisorKIncrementMode = ByDivisorKIncrementMode.Sequential;
 	public static Persistence.ByDivisorClassModelRepository? ByDivisorClassModelRepository;
 	public static Persistence.KStateRepository? ByDivisorKStateRepository;
 	public static Persistence.KStateRepository? ByDivisorCheapKStateRepository;
@@ -31,13 +35,31 @@ public static class EnvironmentConfiguration
 	public static Persistence.KStateRepository? ByDivisorGroupsStateRepository;
 	public static Persistence.Pow2Minus1StateRepository? ByDivisorPow2Minus1Repository;
 	public static Persistence.BitContradictionStateRepository? BitContradictionStateRepository;
+	public static CalculationMethod CalculationMethod = CalculationMethod.Residue;
 
 	[MethodImpl(MethodImplOptions.AggressiveInlining | MethodImplOptions.AggressiveOptimization)]
 	public static void Initialize()
 	{
+		if (MersenneDevice == ComputationDevice.Gpu &&
+			ByDivisorKIncrementMode == ByDivisorKIncrementMode.BitContradiction)
+		{
+			OrderDevice = ComputationDevice.Gpu;
+		}
+
 		UseGpuOrder = OrderDevice == ComputationDevice.Gpu;
 		UseHybridOrder = OrderDevice == ComputationDevice.Hybrid;
 		UseCpuOrder = OrderDevice == ComputationDevice.Cpu;
+
+		UseGpu =
+			UseGpuOrder || UseHybridOrder ||
+			MersenneDevice is ComputationDevice.Hybrid or ComputationDevice.Gpu;
+
+		RollingAccelerators = UseGpu 
+			? Math.Min(Math.Min(
+			PerfectNumberConstants.RollingAccelerators, GpuPrimeThreads), ThreadCount)
+			: 0;
+
+		UsePow2GroupDivisors = ByDivisorKIncrementMode == ByDivisorKIncrementMode.Pow2Groups;
 	}
 }
 

PerfectNumbers.Core/Gpu/Accelerators/PrimeOrderCalculatorAccelerator.cs

@@ -5,6 +5,238 @@ using PerfectNumbers.Core.Gpu.Kernels;
 
 namespace PerfectNumbers.Core.Gpu.Accelerators;
 
+public static class PrimeOrderCalculatorAcceleratorCapacityProviders
+{
+	[MethodImpl(MethodImplOptions.AggressiveOptimization)]
+	public static void EnsureCalculateOrderWideCapacity(this PrimeOrderCalculatorAccelerator gpu, int newSize)
+	{
+		if (newSize > gpu.Input.Length)
+		{
+			gpu.CalculateOrderWideExponentBuffer.Dispose();
+			gpu.CalculateOrderWideRemainderBuffer.Dispose();
+
+			var accelerator = gpu.Accelerator;
+			// lock (accelerator)
+			// {
+			gpu.CalculateOrderWideExponentBuffer = accelerator.Allocate1D<GpuUInt128>(newSize);
+			gpu.CalculateOrderWideRemainderBuffer = accelerator.Allocate1D<GpuUInt128>(newSize);
+			gpu.CalculateOrderWideExponentBufferView = gpu.CalculateOrderWideExponentBuffer.View;
+			gpu.CalculateOrderWideRemainderBufferView = gpu.CalculateOrderWideRemainderBuffer.View;
+			// }			
+		}
+	}
+
+	[MethodImpl(MethodImplOptions.AggressiveOptimization)]
+	public static void EnsurePrimeOrderCalculatorCapacity(this PrimeOrderCalculatorAccelerator gpu, int factorsCount, int primeTesterCapacity)
+	{
+		Accelerator accelerator = gpu.Accelerator;
+
+		MemoryBuffer1D<KeyValuePair<ulong, int>, Stride1D.Dense> pow2ModEntriesToTestOnDevice = gpu.Pow2ModEntriesToTestOnDevice;
+		if (pow2ModEntriesToTestOnDevice.Length < factorsCount)
+		{
+			// lock (accelerator)
+			// {
+				pow2ModEntriesToTestOnDevice.Dispose();
+				pow2ModEntriesToTestOnDevice = accelerator.Allocate1D<KeyValuePair<ulong, int>>(factorsCount);
+				gpu.Pow2ModEntriesToTestOnDevice = pow2ModEntriesToTestOnDevice;
+				gpu.Pow2ModEntriesToTestOnDeviceView = pow2ModEntriesToTestOnDevice.View;
+			// }
+		}
+
+		MemoryBuffer1D<ulong, Stride1D.Dense> input = gpu.Input;
+		if (input.Length < primeTesterCapacity)
+		{
+			// lock (accelerator)
+			// {
+				input.Dispose();
+
+				input = accelerator.Allocate1D<ulong>(primeTesterCapacity);
+				gpu.Input = input;
+				gpu.InputView = input.View;
+			// }
+		}
+
+		MemoryBuffer1D<byte, Stride1D.Dense> outputByte = gpu.OutputByte;
+		if (outputByte.Length < primeTesterCapacity)
+		{
+			// lock (accelerator)
+			// {
+				outputByte.Dispose();
+
+				outputByte = accelerator.Allocate1D<byte>(primeTesterCapacity);
+				gpu.OutputByte = outputByte;
+				gpu.OutputByteView = outputByte.View;
+			// }
+		}
+
+		MemoryBuffer1D<int, Stride1D.Dense> outputInt = gpu.OutputInt;
+		if (outputInt.Length < primeTesterCapacity)
+		{
+			// lock (accelerator)
+			// {
+				outputInt.Dispose();
+
+				outputInt = accelerator.Allocate1D<int>(primeTesterCapacity);
+				gpu.OutputInt = outputInt;
+				gpu.OutputIntView = outputInt.View;
+			// }
+		}
+	}
+
+	public static void EnsurePartialFactorCapacity(this PrimeOrderCalculatorAccelerator gpu, int newSize)
+	{
+		var accelerator = gpu.Accelerator;
+
+		var outputUlong = gpu.OutputUlong;
+		if (newSize > outputUlong.Length)
+		{
+			outputUlong.Dispose();
+
+			outputUlong = accelerator.Allocate1D<ulong>(newSize);
+			gpu.OutputUlong = outputUlong;
+			gpu.OutputUlongView = outputUlong.View;
+		}
+
+		var outputInt = gpu.OutputInt;
+		if (newSize > outputInt.Length)
+		{
+			outputInt.Dispose();
+			
+			outputInt = accelerator.Allocate1D<int>(newSize);
+			gpu.OutputInt = outputInt;
+			gpu.OutputIntView = outputInt.View;
+		}
+	}
+
+	public static void EnsureSmallPrimeFactorSlotsCapacity(this PrimeOrderCalculatorAccelerator gpu, int newSize)
+	{
+		// Console.WriteLine($"Resizing GPU scratch buffer from pool ({buffer.SmallPrimeFactorPrimeSlots.Length} / {smallPrimeFactorSlotCount}), ({buffer.SpecialMaxFactors.Length}/{specialMaxFactorCapacity})");
+		var accelerator = gpu.Accelerator;
+		MemoryBuffer1D<ulong, Stride1D.Dense> outputUlong = gpu.OutputUlong;
+		if (outputUlong.Length < newSize)
+		{
+			// lock (accelerator)
+			// {
+				outputUlong.Dispose();
+
+				outputUlong = accelerator.Allocate1D<ulong>(newSize);
+				gpu.OutputUlong = outputUlong;
+				gpu.OutputUlongView = outputUlong.View;
+			// }
+		}
+
+		MemoryBuffer1D<int, Stride1D.Dense> outputInt = gpu.OutputInt;
+		if (outputInt.Length < newSize)
+		{
+			// lock (accelerator)
+			// {
+				outputInt.Dispose();
+
+				outputInt = accelerator.Allocate1D<int>(newSize);
+				gpu.OutputInt = outputInt;
+				gpu.OutputIntView = outputInt.View;
+			// }
+		}
+	}
+
+	[MethodImpl(MethodImplOptions.AggressiveOptimization)]
+	public static void EnsureUlongInputOutputCapacity(this PrimeOrderCalculatorAccelerator gpu, int newSize)
+	{
+		var accelerator = gpu.Accelerator;
+		MemoryBuffer1D<ulong, Stride1D.Dense> buffer = gpu.Input;
+		if (buffer.Length < newSize)
+		{
+			// lock (accelerator)
+			// {
+				buffer.Dispose();
+
+				buffer = accelerator.Allocate1D<ulong>(newSize);
+				gpu.Input = buffer;
+				gpu.InputView = buffer.View;
+			// }
+		}
+
+		buffer = gpu.OutputUlong;
+		if (buffer.Length < newSize)
+		{
+			// lock (accelerator)
+			// {
+				buffer.Dispose();
+
+				buffer = accelerator.Allocate1D<ulong>(newSize);
+				gpu.OutputUlong = buffer;
+				gpu.OutputUlongView = buffer.View;
+			// }
+		}
+	}
+
+	[MethodImpl(MethodImplOptions.AggressiveOptimization)]
+	public static BitContradictionBuffers EnsureBitContradictionCapacity(this PrimeOrderCalculatorAccelerator gpu, int maxBatchIndexWords, int maxQWords, int delta8CacheConcurrentThreads)
+	{
+		var accelerator = gpu.Accelerator;
+
+		MemoryBuffer1D<ulong, Stride1D.Dense> bufferUlong = gpu.Input;
+		if (bufferUlong.Length < maxBatchIndexWords)
+		{
+			// lock (accelerator)
+			// {
+				bufferUlong.Dispose();
+
+				bufferUlong = accelerator.Allocate1D<ulong>(maxBatchIndexWords);
+				gpu.Input = bufferUlong;
+
+			// }
+		}
+
+		MemoryBuffer1D<int, Stride1D.Dense> bufferInt = gpu.OutputInt;
+		maxQWords <<= 1;
+		if (bufferUlong.Length < maxQWords)
+		{
+			// lock (accelerator)
+			// {
+				bufferInt.Dispose();
+
+				bufferInt = accelerator.Allocate1D<int>(maxQWords);
+				gpu.OutputInt = bufferInt;
+			// }
+		}
+
+		bufferInt = gpu.StackIndexBuffer;
+		int requiredCapacity = BitContradictionKernels.GetDeltaCacheKeyLength(delta8CacheConcurrentThreads);
+		if (bufferInt.Length < requiredCapacity)
+		{
+			bufferInt.Dispose();
+			bufferInt = accelerator.Allocate1D<int>(requiredCapacity);
+			gpu.StackIndexBuffer = bufferInt;
+
+			bufferInt = gpu.StackExponentBuffer;
+			bufferInt.Dispose();
+			bufferInt = accelerator.Allocate1D<int>(BitContradictionKernels.GetDeltaCacheIntLength(delta8CacheConcurrentThreads));
+			gpu.StackExponentBuffer = bufferInt;
+		}
+
+		return new(
+			gpu.Input.View,
+			gpu.OutputInt.View,
+			gpu.StackIndexBuffer.View,
+			gpu.StackExponentBuffer.View
+		);
+	}
+}
+
+public readonly struct BitContradictionBuffers(
+	in ArrayView1D<ulong, Stride1D.Dense> batchIndexWords,
+	in ArrayView1D<int, Stride1D.Dense> foundQWords,
+	in ArrayView1D<int, Stride1D.Dense> delta8CacheKeys,
+	in ArrayView1D<int, Stride1D.Dense> delta8CacheValues
+)
+{
+	public readonly ArrayView1D<ulong, Stride1D.Dense> BatchIndexWordsView = batchIndexWords;
+	public readonly ArrayView1D<int,  Stride1D.Dense> FoundQWordsView = foundQWords;
+	public readonly ArrayView1D<int, Stride1D.Dense> Delta8CacheKeysView = delta8CacheKeys;
+	public readonly ArrayView1D<int, Stride1D.Dense> Delta8CacheValuesView = delta8CacheValues;
+}
+
 public sealed class PrimeOrderCalculatorAccelerator
 {
 	#region Pool
@@ -18,7 +250,7 @@ public sealed class PrimeOrderCalculatorAccelerator
 		var pool = _pool ??= new(PerfectNumberConstants.DefaultThreadPoolCapacity);
 		if (pool.TryPop(out var gpu))
 		{
-			gpu.EnsureCapacity(PerfectNumberConstants.DefaultFactorsBuffer, primeTesterCapacity);
+			gpu.EnsurePrimeOrderCalculatorCapacity(PerfectNumberConstants.DefaultFactorsBuffer, primeTesterCapacity);
 			return gpu;
 		}
 
@@ -137,8 +369,8 @@ public sealed class PrimeOrderCalculatorAccelerator
 
 	public readonly MemoryBuffer1D<ulong, Stride1D.Dense> WorkFactorBuffer;
 	public readonly MemoryBuffer1D<ulong, Stride1D.Dense> CandidateBuffer;
-	public readonly MemoryBuffer1D<int, Stride1D.Dense> StackIndexBuffer;
-	public readonly MemoryBuffer1D<int, Stride1D.Dense> StackExponentBuffer;
+	public MemoryBuffer1D<int, Stride1D.Dense> StackIndexBuffer;
+	public MemoryBuffer1D<int, Stride1D.Dense> StackExponentBuffer;
 	public readonly MemoryBuffer1D<ulong, Stride1D.Dense> StackProductBuffer;
 
 	public OrderKernelBuffers CalculateOrderKernelBuffers;
@@ -148,15 +380,11 @@ public sealed class PrimeOrderCalculatorAccelerator
 	public readonly ArrayView1D<int, Stride1D.Dense> StackExponentBufferView;
 	public readonly ArrayView1D<ulong, Stride1D.Dense> StackProductBufferView;
 
-
-
-
 	public MemoryBuffer1D<GpuUInt128, Stride1D.Dense> CalculateOrderWideExponentBuffer;
 	public MemoryBuffer1D<GpuUInt128, Stride1D.Dense> CalculateOrderWideRemainderBuffer;
 	public ArrayView1D<GpuUInt128, Stride1D.Dense> CalculateOrderWideExponentBufferView;
 	public ArrayView1D<GpuUInt128, Stride1D.Dense> CalculateOrderWideRemainderBufferView;
 
-
 	public Dictionary<ulong, int> Pow2ModEntriesToTestOnHost = [];
 
 	private static readonly CalculatorKernels[] _kernels = new CalculatorKernels[_accelerators.Length];
@@ -205,169 +433,6 @@ public sealed class PrimeOrderCalculatorAccelerator
 	public readonly Action<AcceleratorStream, Index1D, ArrayView<ulong>, ArrayView<uint>, ArrayView<uint>, ArrayView<uint>, ArrayView<uint>, ArrayView<ulong>, ArrayView<ulong>, ArrayView<ulong>, ArrayView<ulong>, ArrayView<byte>> SmallPrimeSieveKernelLauncher;
 	public readonly Action<AcceleratorStream, Index1D, ulong, ArrayView<ulong>, int, ulong, ArrayView<ulong>> SpecialMaxKernelLauncher;
 
-	[MethodImpl(MethodImplOptions.AggressiveOptimization)]
-	public void EnsureCalculateOrderWideCapacity(int newSize)
-	{
-		if (newSize > Input.Length)
-		{
-			CalculateOrderWideExponentBuffer.Dispose();
-			CalculateOrderWideRemainderBuffer.Dispose();
-
-			var accelerator = Accelerator;
-			// lock (accelerator)
-			// {
-			CalculateOrderWideExponentBuffer = accelerator.Allocate1D<GpuUInt128>(newSize);
-			CalculateOrderWideRemainderBuffer = accelerator.Allocate1D<GpuUInt128>(newSize);
-			CalculateOrderWideExponentBufferView = CalculateOrderWideExponentBuffer.View;
-			CalculateOrderWideRemainderBufferView = CalculateOrderWideRemainderBuffer.View;
-			// }			
-		}
-	}
-
-	[MethodImpl(MethodImplOptions.AggressiveOptimization)]
-	public void EnsureCapacity(int factorsCount, int primeTesterCapacity)
-	{
-		Accelerator accelerator = Accelerator;
-
-		MemoryBuffer1D<KeyValuePair<ulong, int>, Stride1D.Dense> pow2ModEntriesToTestOnDevice = Pow2ModEntriesToTestOnDevice;
-		if (pow2ModEntriesToTestOnDevice.Length < factorsCount)
-		{
-			// lock (accelerator)
-			// {
-				pow2ModEntriesToTestOnDevice.Dispose();
-				pow2ModEntriesToTestOnDevice = accelerator.Allocate1D<KeyValuePair<ulong, int>>(factorsCount);
-				Pow2ModEntriesToTestOnDevice = pow2ModEntriesToTestOnDevice;
-				Pow2ModEntriesToTestOnDeviceView = pow2ModEntriesToTestOnDevice.View;
-			// }
-		}
-
-		MemoryBuffer1D<ulong, Stride1D.Dense> input = Input;
-		if (input.Length < primeTesterCapacity)
-		{
-			// lock (accelerator)
-			// {
-				input.Dispose();
-
-				input = accelerator.Allocate1D<ulong>(primeTesterCapacity);
-				Input = input;
-				InputView = input.View;
-			// }
-		}
-
-		MemoryBuffer1D<byte, Stride1D.Dense> outputByte = OutputByte;
-		if (outputByte.Length < primeTesterCapacity)
-		{
-			// lock (accelerator)
-			// {
-				outputByte.Dispose();
-
-				outputByte = accelerator.Allocate1D<byte>(primeTesterCapacity);
-				OutputByte = outputByte;
-				OutputByteView = outputByte.View;
-			// }
-		}
-
-		MemoryBuffer1D<int, Stride1D.Dense> outputInt = OutputInt;
-		if (outputInt.Length < primeTesterCapacity)
-		{
-			// lock (accelerator)
-			// {
-				outputInt.Dispose();
-
-				outputInt = accelerator.Allocate1D<int>(primeTesterCapacity);
-				OutputInt = outputInt;
-				OutputIntView = outputInt.View;
-			// }
-		}
-	}
-
-	public void EnsurePartialFactorCapacity(int newSize)
-	{
-		var accelerator = Accelerator;
-
-		var outputUlong = OutputUlong;
-		if (newSize > outputUlong.Length)
-		{
-			outputUlong.Dispose();
-
-			outputUlong = accelerator.Allocate1D<ulong>(newSize);
-			OutputUlong = outputUlong;
-			OutputUlongView = outputUlong.View;
-		}
-
-		var outputInt = OutputInt;
-		if (newSize > outputInt.Length)
-		{
-			outputInt.Dispose();
-			
-			outputInt = accelerator.Allocate1D<int>(newSize);
-			OutputInt = outputInt;
-			OutputIntView = outputInt.View;
-		}
-	}
-
-	public void EnsureSmallPrimeFactorSlotsCapacity(int newSize)
-	{
-		// Console.WriteLine($"Resizing GPU scratch buffer from pool ({buffer.SmallPrimeFactorPrimeSlots.Length} / {smallPrimeFactorSlotCount}), ({buffer.SpecialMaxFactors.Length}/{specialMaxFactorCapacity})");
-		var accelerator = Accelerator;
-		MemoryBuffer1D<ulong, Stride1D.Dense> outputUlong = OutputUlong;
-		if (outputUlong.Length < newSize)
-		{
-			// lock (accelerator)
-			// {
-				outputUlong.Dispose();
-
-				outputUlong = accelerator.Allocate1D<ulong>(newSize);
-				OutputUlong = outputUlong;
-				OutputUlongView = outputUlong.View;
-			// }
-		}
-
-		MemoryBuffer1D<int, Stride1D.Dense> outputInt = OutputInt;
-		if (outputInt.Length < newSize)
-		{
-			// lock (accelerator)
-			// {
-				outputInt.Dispose();
-
-				outputInt = accelerator.Allocate1D<int>(newSize);
-				OutputInt = outputInt;
-				OutputIntView = outputInt.View;
-			// }
-		}
-	}
-
-	[MethodImpl(MethodImplOptions.AggressiveOptimization)]
-	public void EnsureUlongInputOutputCapacity(int newSize)
-	{
-		var accelerator = Accelerator;
-		MemoryBuffer1D<ulong, Stride1D.Dense> buffer = Input;
-		if (buffer.Length < newSize)
-		{
-			// lock (accelerator)
-			// {
-				buffer.Dispose();
-
-				buffer = accelerator.Allocate1D<ulong>(newSize);
-				Input = buffer;
-				InputView = buffer.View;
-			// }
-		}
-
-		buffer = OutputUlong;
-		if (buffer.Length < newSize)
-		{
-			// lock (accelerator)
-			// {
-				buffer.Dispose();
-
-				buffer = accelerator.Allocate1D<ulong>(newSize);
-				OutputUlong = buffer;
-				OutputUlongView = buffer.View;
-			// }
-		}
-	}
-
 	public PrimeOrderCalculatorAccelerator(int acceleratorIndex, int factorsCount, int primeTesterCapacity, int smallPrimeFactorSlotCount, int specialMaxFactorCapacity)
 	{
 		var accelerator = _accelerators[acceleratorIndex];

PerfectNumbers.Core/Gpu/GpuKernelPool.cs

@@ -125,7 +125,7 @@ public class GpuKernelPool
 			{
 				var loaded = accelerator.LoadAutoGroupedStreamKernel<
 					Index1D, ulong, ArrayView1D<ulong, Stride1D.Dense>, int, ArrayView1D<int, Stride1D.Dense>,
-					#if DETAILED_LOG
+					#if DETAILED_LOG || SELF_TEST
 						ArrayView1D<ulong, Stride1D.Dense>,
 					#endif
 					ArrayView1D<int, Stride1D.Dense>, ArrayView1D<int, Stride1D.Dense>
@@ -135,7 +135,7 @@ public class GpuKernelPool
 
 				return kernel.CreateLauncherDelegate<Action<AcceleratorStream, Index1D, ulong, ArrayView1D<ulong, Stride1D.Dense>, int,
 					ArrayView1D<int, Stride1D.Dense>,
-					#if DETAILED_LOG
+					#if DETAILED_LOG || SELF_TEST
 						ArrayView1D<ulong, Stride1D.Dense>,
 					#endif
 					ArrayView1D<int, Stride1D.Dense>, ArrayView1D<int, Stride1D.Dense>>>();

PerfectNumbers.Core/Gpu/KernelContainer.cs

@@ -23,7 +23,7 @@ public sealed class KernelContainer
 		ResidueAutomatonArgs, ArrayView<int>, ArrayView1D<ulong, Stride1D.Dense>>? Pow2ModOrder;
 
 	public Action<AcceleratorStream, Index1D, ulong, ArrayView1D<ulong, Stride1D.Dense>, int, ArrayView1D<int, Stride1D.Dense>,
-		  #if DETAILED_LOG
+		  #if DETAILED_LOG || SELF_TEST
 			  ArrayView1D<ulong, Stride1D.Dense>,
 		  #endif
 		  ArrayView1D<int, Stride1D.Dense>, ArrayView1D<int, Stride1D.Dense>>? BitContradiction;

PerfectNumbers.Core/GpuKernelType.cs

@@ -0,0 +1,7 @@
+namespace PerfectNumbers.Core;
+
+public enum GpuKernelType
+{
+    Incremental,
+    Pow2Mod,
+}

PerfectNumbers.Core/MersenneNumberTesterTemplate.cs

@@ -11,35 +11,6 @@ using PerfectNumbers.Core.Gpu.Accelerators;
 
 namespace PerfectNumbers.Core;
 
-public enum GpuKernelType
-{
-    Incremental,
-    Pow2Mod,
-}
-
-public enum CalculationMethod
-{
-    ByDivisor,
-    Divisor,
-    Incremental,
-    LucasLehmer,
-    Pow2Mod,
-    Residue,
-}
-
-public enum CacheStatus
-{
-    Disabled,
-    Enabled,
-}
-
-public enum OrderDevice
-{
-    Cpu,
-    Hybrid,
-    Gpu,
-}
-
 [EnumDependentTemplate(typeof(CalculationMethod))]
 [EnumDependentTemplate(typeof(CacheStatus))]
 [EnumDependentTemplate(typeof(OrderDevice))]

PerfectNumbers.Core/OrderDevice.cs

@@ -0,0 +1,8 @@
+namespace PerfectNumbers.Core;
+
+public enum OrderDevice
+{
+    Cpu,
+    Hybrid,
+    Gpu,
+}

PerfectNumbers.Core/PerfectNumberConstants.cs

@@ -5,6 +5,7 @@ namespace PerfectNumbers.Core;
 public static class PerfectNumberConstants
 {
     public const ulong BiggestKnownEvenPerfectP = 136279841UL;
+	// public const int ConsoleInterval = 1;
 	public const int ConsoleInterval = 100_000;
 	public const int DefaultFactorsBuffer = 32;
 	public const int DefaultPoolCapacity = 10_240;
@@ -26,7 +27,10 @@ public static class PerfectNumberConstants
 	public const int PollardRhoMilliseconds = 100;
 	public const int ByDivisorMersenneTailCount = 8;
 	public const int ByDivisorMersenneBitLimit = 4096;
-	public const int BitContradictionGpuBatchCount = 134217728;
+	public const int BitContradictionGpuBatchCount = 16;
+	// public const int BitContradictionGpuBatchCount = 512;
+	// public const int BitContradictionGpuBatchCount = 131_072;
+	// public const int BitContradictionGpuBatchCount = 2_097_152;
 	public const int BitContradictionGpuThreads = 1;
     // TODO: Load these limits from the benchmark-driven configuration so CPU and GPU scans stay aligned with the optimal
     // divisor-cycle datasets we generate offline.

PerfectNumbers.Core/PrimeTester.cs

@@ -303,7 +303,7 @@ public sealed class PrimeTester
         int length = values.Length;
 
         int acceleratorIndex = gpu.AcceleratorIndex;
-        gpu.EnsureCapacity(0, length);
+        gpu.EnsurePrimeOrderCalculatorCapacity(0, length);
         var inputBufferView = gpu.InputView;
         var resultBufferView = gpu.OutputByteView;
         var kernelLauncher = gpu.SharesFactorKernelLauncher;

PerfectNumbers.Core/PrimeTesterSeverOrOthers.cs

@@ -223,7 +223,7 @@ public sealed class PrimeTesterSeverOrOthers
 		int length = values.Length;
 
 		int acceleratorIndex = gpu.AcceleratorIndex;
-		gpu.EnsureCapacity(0, length);
+		gpu.EnsurePrimeOrderCalculatorCapacity(0, length);
 		var inputBufferView = gpu.InputView;
 		var resultBufferView = gpu.OutputByteView;
 		var kernelLauncher = gpu.SharesFactorKernelLauncher;