Merge pull request #128 from mlemanczyk/codex/modify-evenperfectbitscanner-execution-path

Remove CPU fallbacks when GPU pow2 fails

EvenPerfectBitScanner.Benchmarks/EvenPerfectBitScanner.Benchmarks.csproj

@@ -15,7 +15,4 @@
     <PackageReference Include="BenchmarkDotNet" Version="0.13.12" />
   </ItemGroup>
 
-  <ItemGroup>
-    <Compile Remove=".temp.cs" />
-  </ItemGroup>
 </Project>

EvenPerfectBitScanner.Benchmarks/GpuPow2ModBenchmarks.cs

@@ -50,7 +50,7 @@ public class GpuPow2ModBenchmarks
         return Pow2ModSingleBit(Input.Exponent, Input.Modulus);
     }
 
-    private static GpuUInt128 Pow2ModSingleBit(ulong exponent, GpuUInt128 modulus)
+    private static GpuUInt128 Pow2ModSingleBit(ulong exponent, in GpuUInt128 modulus)
     {
         if (modulus.IsZero || modulus == GpuUInt128.One)
         {

EvenPerfectBitScanner.Benchmarks/MulMod64Benchmarks.cs

@@ -158,16 +158,36 @@ public class MulMod64Benchmarks
     }
 
     /// <summary>
-    /// Uses the production <see cref="ULongExtensions.MulMod64(ulong, ulong, ulong)"/> helper shared by CPU and GPU callers.
+    /// Uses the GPU-friendly <see cref="ULongExtensions.MulMod64GpuCompatible(ulong, ulong, ulong)"/> helper; dense inputs now
+    /// cost 24.4–61.9 ns, while NearFullRange and SparseOperands finish near 7.27 ns and ZeroOperands stay baseline-fast at
+    /// 3.59 ns.
     /// </summary>
     /// <remarks>
-    /// Observed means prior to the helper consolidation: CrossWordBlend 24.422 ns, MixedBitPattern 61.850 ns, NearFullRange 7.267 ns,
+    /// Observed means: CrossWordBlend 24.422 ns (6.86×), MixedBitPattern 61.850 ns, NearFullRange 7.267 ns,
     /// PrimeSizedModulus 55.114 ns, SparseOperands 7.261 ns, ZeroOperands 3.587 ns.
     /// </remarks>
     [Benchmark]
-    public ulong UnifiedMulMod()
+    public ulong GpuCompatibleBaseline()
     {
-        return Input.Left.MulMod64(Input.Right, Input.Modulus);
+        // TODO: Callers in production should migrate to ULongExtensions.MulMod64 where GPU parity
+        // is not required; that baseline stayed 6–17× faster for dense 64-bit operands.
+        return Input.Left.MulMod64GpuCompatible(Input.Right, Input.Modulus);
+    }
+
+    /// <summary>
+    /// Uses the deferred native-modulo helper <see cref="ULongExtensions.MulMod64GpuCompatibleDeferred(ulong, ulong, ulong)"/>;
+    /// excels when operands are tiny (2.01 ns on ZeroOperands) yet remains the slowest choice on dense data at 146–295 ns.
+    /// </summary>
+    /// <remarks>
+    /// Observed means: CrossWordBlend 146.559 ns (40.63×), MixedBitPattern 255.555 ns, NearFullRange 21.390 ns,
+    /// PrimeSizedModulus 294.486 ns, SparseOperands 18.182 ns, ZeroOperands 2.009 ns.
+    /// </remarks>
+    [Benchmark]
+    public ulong GpuCompatibleDeferred()
+    {
+        // TODO: Retire the deferred GPU shim from runtime paths and keep it only for benchmarks;
+        // ULongExtensions.MulMod64 avoids the 6×–82× slowdown on real-world operand mixes.
+        return Input.Left.MulMod64GpuCompatibleDeferred(Input.Right, Input.Modulus);
     }
 
     [MethodImpl(MethodImplOptions.AggressiveInlining)]

EvenPerfectBitScanner.Benchmarks/MutableUInt128AddBenchmarks.cs

@@ -1,75 +0,0 @@
-using System.Runtime.CompilerServices;
-using BenchmarkDotNet.Attributes;
-using BenchmarkDotNet.Jobs;
-using PerfectNumbers.Core;
-
-namespace EvenPerfectBitScanner.Benchmarks;
-
-[ShortRunJob(RuntimeMoniker.Net80)]
-[MemoryDiagnoser]
-public class MutableUInt128AddBenchmarks
-{
-	private static readonly BenchmarkInput[] Inputs =
-	[
-		new(Create(0UL, 0x0000_0000_0000_0007UL), Create(0, 0x3UL), "TinyOperands"),
-		new(Create(0x1234_5678_9ABC_DEF0UL, 0x0FED_CBA9_8765_4321UL), Create(0xFFFF_FFFF, 0x0000_0001UL), "MixedOperands"),
-		new(Create(0xFFFF_FFFF_FFFF_FFFFUL, 0xFFFF_FFFF_FFFF_FFFFUL), Create(0xFFFF_FFFF,  0xFFFF_FFFBUL), "NearMaxOperands"),
-    ];
-
-    [ParamsSource(nameof(GetInputs))]
-    public BenchmarkInput Input { get; set; }
-
-    public static IEnumerable<BenchmarkInput> GetInputs() => Inputs;
-
-	private static MutableUInt128 Create(ulong high, ulong low) => new(high, low);
-
-	[MethodImpl(MethodImplOptions.AggressiveInlining)]
-	private static void StandardInternal(ref MutableUInt128 source, MutableUInt128 value)
-	{
-        ulong temp = source.Low + value.Low;
-        source.Low = temp;
-        temp = temp < source.Low ? 1UL : 0UL;
-		temp += source.High;
-		temp += value.High;
-        source.High = temp;
-	}
-
-	/// <summary>
-	/// Baseline MultiplyAdd implementation that chains Multiply and Add; measured 3.370 ns with mixed operands,
-	/// 3.361 ns when both limbs and inputs are near their maxima, and 3.368 ns with tiny operands.
-	/// </summary>
-	/// <remarks>
-	/// Observed means: MixedOperands 3.370 ns, NearMaxOperands 3.361 ns, TinyOperands 3.368 ns.
-	/// </remarks>
-	[Benchmark(Baseline = true)]
-	public MutableUInt128 Standard()
-	{
-		MutableUInt128 value = Input.InitialValue;
-		StandardInternal(ref value, Input.Addend);
-		return value;
-	}
-
-    /// <summary>
-    /// Fully inlined MultiplyAdd implementation that kept all arithmetic inside the method; landed at 3.175 ns on mixed operands,
-    /// 3.177 ns on near-max inputs, and 3.178 ns on tiny operands.
-    /// </summary>
-    /// <remarks>
-    /// Observed means: MixedOperands 3.175 ns, NearMaxOperands 3.177 ns, TinyOperands 3.178 ns.
-	/// Status: Implemented
-    /// </remarks>
-    [Benchmark]
-    public MutableUInt128 Optimized()
-    {
-        MutableUInt128 value = Input.InitialValue;
-        value.Add(Input.Addend);
-        return value;
-    }
-
-    public readonly record struct BenchmarkInput(MutableUInt128 InitialValue, MutableUInt128 Addend, string Name)
-    {
-        public override string ToString()
-        {
-            return Name;
-        }
-    }
-}

EvenPerfectBitScanner.Benchmarks/MutableUInt128Mul64Benchmarks.cs

@@ -1,100 +0,0 @@
-using System.Runtime.CompilerServices;
-using BenchmarkDotNet.Attributes;
-using BenchmarkDotNet.Jobs;
-using PerfectNumbers.Core;
-
-namespace EvenPerfectBitScanner.Benchmarks;
-
-[ShortRunJob(RuntimeMoniker.Net80)]
-[MemoryDiagnoser]
-public class MutableUInt128Mul64Benchmarks
-{
-	private static readonly BenchmarkInput[] Inputs =
-	[
-		new(Create(0UL, 0x0000_0000_0000_0007UL), 0x3UL, "TinyOperands"),
-		new(Create(0x1234_5678_9ABC_DEF0UL, 0x0FED_CBA9_8765_4321UL), 0xFFFF_FFFF_0000_0001UL, "MixedOperands"),
-		new(Create(0xFFFF_FFFF_FFFF_FFFFUL, 0xFFFF_FFFF_FFFF_FFFFUL), 0xFFFF_FFFF_FFFF_FFFBUL, "NearMaxOperands"),
-    ];
-
-    [ParamsSource(nameof(GetInputs))]
-    public BenchmarkInput Input { get; set; }
-
-    public static IEnumerable<BenchmarkInput> GetInputs() => Inputs;
-
-	private static MutableUInt128 Create(ulong high, ulong low) => new(high, low);
-
-    [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    private static ulong Mul64(ulong left, ulong right, out ulong low)
-    {
-        const ulong Mask32 = 0xFFFF_FFFFUL;
-
-        ulong leftLow = left & Mask32;
-        ulong leftHigh = left >> 32;
-        ulong rightLow = right & Mask32;
-        ulong rightHigh = right >> 32;
-
-        ulong lowLow = leftLow * rightLow;
-        ulong lowHigh = leftLow * rightHigh;
-        ulong highLow = leftHigh * rightLow;
-        ulong highHigh = leftHigh * rightHigh;
-
-        ulong carry = (lowLow >> 32) + (lowHigh & Mask32) + (highLow & Mask32);
-        ulong high = highHigh + (lowHigh >> 32) + (highLow >> 32) + (carry >> 32);
-
-        low = ((carry & Mask32) << 32) | (lowLow & Mask32);
-        return high;
-    }
-
-	[MethodImpl(MethodImplOptions.AggressiveInlining)]
-	private static void StandardInternal(ref MutableUInt128 source, ulong value)
-	{
-        ulong currentHigh = source.High;
-        ulong currentLow = source.Low;
-
-        ulong carry = Mul64(currentLow, value, out ulong newLow);
-        Mul64(currentHigh, value, out ulong shiftedHigh);
-        ulong newHigh = shiftedHigh + carry;
-
-        source.Low = newLow;
-        source.High = newHigh;
-	}
-
-	/// <summary>
-	/// Baseline MultiplyAdd implementation that chains Multiply and Add; measured 3.113 ns with mixed operands,
-	/// 3.128 ns when both limbs and inputs are near their maxima, and 3.131 ns with tiny operands.
-	/// </summary>
-	/// <remarks>
-	/// Observed means: MixedOperands 3.113 ns, NearMaxOperands 3.128 ns, TinyOperands 3.131 ns.
-	/// </remarks>
-	[Benchmark(Baseline = true)]
-	public MutableUInt128 Standard()
-	{
-		MutableUInt128 value = Input.InitialValue;
-		StandardInternal(ref value, Input.Addend);
-		return value;
-	}
-
-    /// <summary>
-    /// Fully inlined MultiplyAdd implementation that kept all arithmetic inside the method; landed at 2.969 ns on mixed operands,
-    /// 2.912 ns on near-max inputs, and 2.935 ns on tiny operands.
-    /// </summary>
-    /// <remarks>
-    /// Observed means: MixedOperands 2.969 ns, NearMaxOperands 2.912 ns, TinyOperands 2.935 ns.
-	/// Status: Implemented
-    /// </remarks>
-    [Benchmark]
-    public MutableUInt128 Optimized()
-    {
-        MutableUInt128 value = Input.InitialValue;
-        value.Mul(Input.Addend);
-        return value;
-    }
-
-    public readonly record struct BenchmarkInput(MutableUInt128 InitialValue, ulong Addend, string Name)
-    {
-        public override string ToString()
-        {
-            return Name;
-        }
-    }
-}

EvenPerfectBitScanner.Benchmarks/MutableUInt128MultiplyAddBenchmarks.cs

@@ -1,76 +0,0 @@
-using System.Collections.Generic;
-using System.Numerics;
-using System.Runtime.CompilerServices;
-using BenchmarkDotNet.Attributes;
-using BenchmarkDotNet.Jobs;
-using PerfectNumbers.Core;
-
-namespace EvenPerfectBitScanner.Benchmarks;
-
-[ShortRunJob(RuntimeMoniker.Net80)]
-[MemoryDiagnoser]
-public class MutableUInt128MultiplyAddBenchmarks
-{
-    private static readonly MultiplyAddInput[] Inputs =
-    [
-        new(Create(0UL, 0x0000_0000_0000_0007UL), 0x3UL, 0x5UL, "TinyOperands"),
-        new(Create(0x1234_5678_9ABC_DEF0UL, 0x0FED_CBA9_8765_4321UL), 0xFFFF_FFFF_0000_0001UL, 0x1357_9BDF_2468_ACEEUL, "MixedOperands"),
-        new(Create(0xFFFF_FFFF_FFFF_FFFFUL, 0xFFFF_FFFF_FFFF_FFFFUL), 0xFFFF_FFFF_FFFF_FFFBUL, 0xFFFF_FFFF_FFFF_FFF7UL, "NearMaxOperands"),
-    ];
-
-    [ParamsSource(nameof(GetInputs))]
-    public MultiplyAddInput Input { get; set; }
-
-    public static IEnumerable<MultiplyAddInput> GetInputs() => Inputs;
-
-    private static UInt128 Create(ulong high, ulong low)
-    {
-        return ((UInt128)high << 64) | low;
-    }
-
-    [MethodImpl(MethodImplOptions.AggressiveInlining)]
-	private static void MultiplyAdd(ref MutableUInt128 source, ulong multiplier, ulong addend)
-	{
-		source.Mul(multiplier);
-		source.Add(addend);
-	}
-
-	/// <summary>
-	/// Baseline MultiplyAdd implementation that chains Multiply and Add; measured 3.370 ns with mixed operands,
-	/// 3.361 ns when both limbs and inputs are near their maxima, and 3.368 ns with tiny operands.
-	/// </summary>
-	/// <remarks>
-	/// Observed means: MixedOperands 3.370 ns, NearMaxOperands 3.361 ns, TinyOperands 3.368 ns.
-	/// </remarks>
-	[Benchmark(Baseline = true)]
-	public MutableUInt128 MultiplyAddChained()
-	{
-		MutableUInt128 value = new(Input.InitialValue);
-		MultiplyAdd(ref value, Input.Multiplier, Input.Addend);
-		return value;
-	}
-
-    /// <summary>
-    /// Fully inlined MultiplyAdd implementation that kept all arithmetic inside the method; landed at 3.175 ns on mixed operands,
-    /// 3.177 ns on near-max inputs, and 3.178 ns on tiny operands.
-    /// </summary>
-    /// <remarks>
-    /// Observed means: MixedOperands 3.175 ns, NearMaxOperands 3.177 ns, TinyOperands 3.178 ns.
-	/// Status: Implemented
-    /// </remarks>
-    [Benchmark]
-    public MutableUInt128 MultiplyAddFullyInlined()
-    {
-        MutableUInt128 value = new(Input.InitialValue);
-        value.MulAdd(Input.Multiplier, Input.Addend);
-        return value;
-    }
-
-    public readonly record struct MultiplyAddInput(UInt128 InitialValue, ulong Multiplier, ulong Addend, string Name)
-    {
-        public override string ToString()
-        {
-            return Name;
-        }
-    }
-}

EvenPerfectBitScanner.Benchmarks/MutableUInt128SubMutableBenchmarks.cs

@@ -1,75 +0,0 @@
-using System.Runtime.CompilerServices;
-using BenchmarkDotNet.Attributes;
-using BenchmarkDotNet.Jobs;
-using PerfectNumbers.Core;
-
-namespace EvenPerfectBitScanner.Benchmarks;
-
-[ShortRunJob(RuntimeMoniker.Net80)]
-[MemoryDiagnoser]
-public class MutableUInt128SubMutableBenchmarks
-{
-	private static readonly BenchmarkInput[] Inputs =
-	[
-		new(Create(0UL, 0x0000_0000_0000_0007UL), Create(0, 0x3UL), "TinyOperands"),
-		new(Create(0x1234_5678_9ABC_DEF0UL, 0x0FED_CBA9_8765_4321UL), Create(0xFFFF_FFFF, 0x0000_0001UL), "MixedOperands"),
-		new(Create(0xFFFF_FFFF_FFFF_FFFFUL, 0xFFFF_FFFF_FFFF_FFFFUL), Create(0xFFFF_FFFF,  0xFFFF_FFFBUL), "NearMaxOperands"),
-    ];
-
-    [ParamsSource(nameof(GetInputs))]
-    public BenchmarkInput Input { get; set; }
-
-    public static IEnumerable<BenchmarkInput> GetInputs() => Inputs;
-
-	private static MutableUInt128 Create(ulong high, ulong low) => new(high, low);
-
-	[MethodImpl(MethodImplOptions.AggressiveInlining)]
-	private static void StandardInternal(ref MutableUInt128 source, MutableUInt128 value)
-	{
-        ulong low = source.Low;
-        ulong result = low - value.Low;
-        ulong borrow = low < value.Low ? 1UL : 0UL;
-
-        source.Low = result;
-        source.High = source.High - value.High - borrow;
-	}
-
-	/// <summary>
-	/// Baseline MultiplyAdd implementation that chains Multiply and Add; measured 3.370 ns with mixed operands,
-	/// 3.361 ns when both limbs and inputs are near their maxima, and 3.368 ns with tiny operands.
-	/// </summary>
-	/// <remarks>
-	/// Observed means: MixedOperands 3.370 ns, NearMaxOperands 3.361 ns, TinyOperands 3.368 ns.
-	/// </remarks>
-	[Benchmark(Baseline = true)]
-	public MutableUInt128 Standard()
-	{
-		MutableUInt128 value = Input.InitialValue;
-		StandardInternal(ref value, Input.Addend);
-		return value;
-	}
-
-    /// <summary>
-    /// Fully inlined MultiplyAdd implementation that kept all arithmetic inside the method; landed at 3.175 ns on mixed operands,
-    /// 3.177 ns on near-max inputs, and 3.178 ns on tiny operands.
-    /// </summary>
-    /// <remarks>
-    /// Observed means: MixedOperands 3.175 ns, NearMaxOperands 3.177 ns, TinyOperands 3.178 ns.
-	/// Status: Implemented
-    /// </remarks>
-    [Benchmark]
-    public MutableUInt128 Optimized()
-    {
-        MutableUInt128 value = Input.InitialValue;
-        value.Sub(Input.Addend);
-        return value;
-    }
-
-    public readonly record struct BenchmarkInput(MutableUInt128 InitialValue, MutableUInt128 Addend, string Name)
-    {
-        public override string ToString()
-        {
-            return Name;
-        }
-    }
-}

EvenPerfectBitScanner.Benchmarks/Pow2MontgomeryModBenchmarks.cs

@@ -52,31 +52,14 @@ public class Pow2MontgomeryModBenchmarks
     /// Baseline right-to-left Montgomery ladder; measured 36.36 μs on the small sample set and 117.32 μs on the large one.
     /// </summary>
     [Benchmark(Baseline = true)]
-    public ulong BaselineCpu()
+    public ulong Baseline()
     {
         GetData(out var exponents, out var divisors, out _);
         ulong checksum = 0UL;
 
         for (int i = 0; i < SampleCount; i++)
         {
-            checksum ^= exponents[i].Pow2MontgomeryModWindowedCpu(divisors[i], keepMontgomery: false);
-        }
-
-        return checksum;
-    }
-
-    /// <summary>
-    /// Baseline right-to-left Montgomery ladder; measured ??.?? μs on the small sample set and ??.?? μs on the large one.
-    /// </summary>
-    [Benchmark]
-    public ulong BaselineGpu()
-    {
-        GetData(out var exponents, out var divisors, out _);
-        ulong checksum = 0UL;
-
-        for (int i = 0; i < SampleCount; i++)
-        {
-            checksum ^= exponents[i].Pow2MontgomeryModWindowedGpu(divisors[i], keepMontgomery: false);
+            checksum ^= exponents[i].Pow2MontgomeryModWindowed(divisors[i], keepMontgomery: false);
         }
 
         return checksum;
@@ -86,31 +69,14 @@ public class Pow2MontgomeryModBenchmarks
     /// Uses a known divisor cycle to fold the exponent first, dropping runtimes to 26.04 μs on small inputs (0.72× baseline) and 9.53 μs on large inputs (0.08× baseline).
     /// </summary>
     [Benchmark]
-    public ulong BaselineWithKnownCycleCpu()
+    public ulong BaselineWithKnownCycle()
     {
         GetData(out var exponents, out var divisors, out var cycles);
         ulong checksum = 0UL;
 
         for (int i = 0; i < SampleCount; i++)
         {
-            checksum ^= exponents[i].Pow2MontgomeryModWithCycleCpu(cycles[i], divisors[i]);
-        }
-
-        return checksum;
-    }
-
-    /// <summary>
-    /// Uses a known divisor cycle to fold the exponent first, dropping runtimes to 26.04 μs on small inputs (0.72× baseline) and 9.53 μs on large inputs (0.08× baseline).
-    /// </summary>
-    [Benchmark]
-    public ulong BaselineWithKnownCycleGpu()
-    {
-        GetData(out var exponents, out var divisors, out var cycles);
-        ulong checksum = 0UL;
-
-        for (int i = 0; i < SampleCount; i++)
-        {
-            checksum ^= exponents[i].Pow2MontgomeryModWithCycleGpu(cycles[i], divisors[i]);
+            checksum ^= exponents[i].Pow2MontgomeryModWithCycle(cycles[i], divisors[i]);
         }
 
         return checksum;

EvenPerfectBitScanner.Benchmarks/Pow2MontgomeryModCycleComputationBenchmarks.cs

@@ -22,391 +22,320 @@ namespace EvenPerfectBitScanner.Benchmarks;
 [SimpleJob(RuntimeMoniker.Net80, launchCount: 1, warmupCount: 1, iterationCount: 5)]
 public class Pow2MontgomeryModCycleComputationBenchmarks
 {
-	private const int SampleCount = 256;
-	private const ulong OneShiftedLeft63 = 1UL << 63;
-	private const long OneShiftedLeft60 = 1L << 60;
-	private readonly MontgomeryDivisorData[] _smallDivisors = new MontgomeryDivisorData[SampleCount];
-	private readonly MontgomeryDivisorData[] _largeDivisors = new MontgomeryDivisorData[SampleCount];
-	private readonly MontgomeryDivisorData[] _veryLargeDivisors = new MontgomeryDivisorData[SampleCount];
-	private readonly UInt128[] _wideModuli = new UInt128[SampleCount];
-	private readonly ulong[] _smallExponents = new ulong[SampleCount];
-	private readonly ulong[] _largeExponents = new ulong[SampleCount];
-	private readonly ulong[] _veryLargeExponents = new ulong[SampleCount];
-	private readonly ulong[] _smallCycles = new ulong[SampleCount];
-	private readonly ulong[] _largeCycles = new ulong[SampleCount];
-	private readonly ulong[] _veryLargeCycles = new ulong[SampleCount];
-	private readonly UInt128[] _wideCycles = new UInt128[SampleCount];
+    private const int SampleCount = 256;
+    private const ulong OneShiftedLeft63 = 1UL << 63;
+    private const long OneShiftedLeft60 = 1L << 60;
+    private readonly MontgomeryDivisorData[] _smallDivisors = new MontgomeryDivisorData[SampleCount];
+    private readonly MontgomeryDivisorData[] _largeDivisors = new MontgomeryDivisorData[SampleCount];
+    private readonly MontgomeryDivisorData[] _veryLargeDivisors = new MontgomeryDivisorData[SampleCount];
+    private readonly UInt128[] _wideModuli = new UInt128[SampleCount];
+    private readonly ulong[] _smallExponents = new ulong[SampleCount];
+    private readonly ulong[] _largeExponents = new ulong[SampleCount];
+    private readonly ulong[] _veryLargeExponents = new ulong[SampleCount];
+    private readonly ulong[] _smallCycles = new ulong[SampleCount];
+    private readonly ulong[] _largeCycles = new ulong[SampleCount];
+    private readonly ulong[] _veryLargeCycles = new ulong[SampleCount];
+    private readonly UInt128[] _wideCycles = new UInt128[SampleCount];
 
-	private readonly Random _random = new(113);
-	private ulong? _previousPrimeOrder;
-	private UInt128? _previousWidePrimeOrder;
+    private readonly Random _random = new(113);
+    private ulong? _previousPrimeOrder;
+    private UInt128? _previousWidePrimeOrder;
 
-	public enum InputScale
-	{
-		Small,
-		Large,
-		VeryLarge
-	}
+    public enum InputScale
+    {
+        Small,
+        Large,
+        VeryLarge
+    }
 
-	[ParamsAllValues]
-	public InputScale Scale { get; set; }
+    [ParamsAllValues]
+    public InputScale Scale { get; set; }
 
-	[GlobalSetup]
-	public void Setup()
-	{
-		for (int i = 0; i < SampleCount; i++)
-		{
+    [GlobalSetup]
+    public void Setup()
+    {
+        for (int i = 0; i < SampleCount; i++)
+        {
 #if DEBUG
-			Console.WriteLine($"Generating sample {i}");
+            Console.WriteLine($"Generating sample {i}");
 #endif
-			ulong smallModulus = NextSmallOddModulus();
-			_smallDivisors[i] = CreateMontgomeryDivisorData(smallModulus);
-			_smallExponents[i] = NextSmallExponent();
-			_smallCycles[i] = CalculateCycleLengthWithHeuristics(smallModulus);
+            ulong smallModulus = NextSmallOddModulus();
+            _smallDivisors[i] = CreateMontgomeryDivisorData(smallModulus);
+            _smallExponents[i] = NextSmallExponent();
+            _smallCycles[i] = CalculateCycleLengthWithHeuristics(smallModulus);
 
-			(ulong largeModulus, ulong largeCycle) = NextLargeModulusAndCycle();
-			_largeDivisors[i] = CreateMontgomeryDivisorData(largeModulus);
-			_largeExponents[i] = NextLargeExponent();
-			_largeCycles[i] = CalculateCycleLengthWithHeuristics(largeModulus, largeCycle);
+            (ulong largeModulus, ulong largeCycle) = NextLargeModulusAndCycle();
+            _largeDivisors[i] = CreateMontgomeryDivisorData(largeModulus);
+            _largeExponents[i] = NextLargeExponent();
+            _largeCycles[i] = CalculateCycleLengthWithHeuristics(largeModulus, largeCycle);
 
-			ulong veryLargeModulus = NextVeryLargeOddModulus();
-			_veryLargeDivisors[i] = CreateMontgomeryDivisorData(veryLargeModulus);
-			_veryLargeExponents[i] = NextVeryLargeExponent();
+            ulong veryLargeModulus = NextVeryLargeOddModulus();
+            _veryLargeDivisors[i] = CreateMontgomeryDivisorData(veryLargeModulus);
+            _veryLargeExponents[i] = NextVeryLargeExponent();
 #if DEBUG
-			Console.WriteLine($"Calculating cycle length {veryLargeModulus}");
+            Console.WriteLine($"Calculating cycle length {veryLargeModulus}");
 #endif
-			_veryLargeCycles[i] = CalculateCycleLengthWithHeuristics(veryLargeModulus);
+            _veryLargeCycles[i] = CalculateCycleLengthWithHeuristics(veryLargeModulus);
 
-			Console.WriteLine($"Generating wide sample {i}");
-			UInt128 wideModulus = NextWideOddModulus();
-			_wideModuli[i] = wideModulus;
-			_wideCycles[i] = CalculateCycleLengthWithHeuristics(wideModulus);
-		}
-	}
+            Console.WriteLine($"Generating wide sample {i}");
+            UInt128 wideModulus = NextWideOddModulus();
+            _wideModuli[i] = wideModulus;
+            _wideCycles[i] = CalculateCycleLengthWithHeuristics(wideModulus);
+        }
+    }
 
-	/// <summary>
-	/// Baseline Montgomery reduction without precomputed cycle data; measured 35.41 μs on the small sample and 117.33 μs on the
-	/// large set.
-	/// </summary>
-	// [Benchmark(Baseline = true)]
-	public ulong MontgomeryWithoutCycleCpu()
-	{
-		GetData(out ulong[] exponents, out MontgomeryDivisorData[] divisors, out _);
-		ulong checksum = 0UL;
+    /// <summary>
+    /// Baseline Montgomery reduction without precomputed cycle data; measured 35.41 μs on the small sample and 117.33 μs on the
+    /// large set.
+    /// </summary>
+    // [Benchmark(Baseline = true)]
+    public ulong MontgomeryWithoutCycle()
+    {
+        GetData(out ulong[] exponents, out MontgomeryDivisorData[] divisors, out _);
+        ulong checksum = 0UL;
 
-		for (int i = 0; i < SampleCount; i++)
-		{
-			checksum ^= exponents[i].Pow2MontgomeryModWindowedCpu(divisors[i], keepMontgomery: false);
-		}
+        for (int i = 0; i < SampleCount; i++)
+        {
+            checksum ^= exponents[i].Pow2MontgomeryModWindowed(divisors[i], keepMontgomery: false);
+        }
 
-		return checksum;
-	}
+        return checksum;
+    }
 
-	/// <summary>
-	/// Baseline Montgomery reduction without precomputed cycle data; measured 35.41 μs on the small sample and 117.33 μs on the
-	/// large set.
-	/// </summary>
-	// [Benchmark(Baseline = true)]
-	public ulong MontgomeryWithoutCycleGpu()
-	{
-		GetData(out ulong[] exponents, out MontgomeryDivisorData[] divisors, out _);
-		ulong checksum = 0UL;
+    /// <summary>
+    /// Montgomery reduction with a known cycle length (cycle lookup only); 25.81 μs on the small set (0.73× baseline) and 9.12 μs
+    /// on the large one (0.08×).
+    /// </summary>
+    [Benchmark]
+    public ulong MontgomeryWithPrecomputedCycle()
+    {
+        GetData(out ulong[] exponents, out MontgomeryDivisorData[] divisors, out ulong[] cycles);
+        ulong checksum = 0UL;
 
-		for (int i = 0; i < SampleCount; i++)
-		{
-			checksum ^= exponents[i].Pow2MontgomeryModWindowedGpu(divisors[i], keepMontgomery: false);
-		}
+        for (int i = 0; i < SampleCount; i++)
+        {
+            checksum ^= exponents[i].Pow2MontgomeryModWithCycle(cycles[i], divisors[i]);
+        }
 
-		return checksum;
-	}
+        return checksum;
+    }
 
-	/// <summary>
-	/// Montgomery reduction with a known cycle length (cycle lookup only); 25.81 μs on the small set (0.73× baseline) and 9.12 μs
-	/// on the large one (0.08×).
-	/// </summary>
-	[Benchmark]
-	public ulong MontgomeryWithPrecomputedCycleCpu()
-	{
-		GetData(out ulong[] exponents, out MontgomeryDivisorData[] divisors, out ulong[] cycles);
-		ulong checksum = 0UL;
+    /// <summary>
+    /// Montgomery reduction with the cycle computed on the fly using the GPU helper; costs 54.09 ms on the small benchmark due to
+    /// cycle discovery, but drops to 12.91 μs on the large set (0.11× baseline).
+    /// </summary>
+    // [Benchmark]
+    public ulong MontgomeryWithGpuCycleComputation()
+    {
+        GetData(out ulong[] exponents, out MontgomeryDivisorData[] divisors, out _);
+        ulong checksum = 0UL;
 
-		for (int i = 0; i < SampleCount; i++)
-		{
-			checksum ^= exponents[i].Pow2MontgomeryModWithCycleCpu(cycles[i], divisors[i]);
-		}
+        for (int i = 0; i < SampleCount; i++)
+        {
+            ulong cycle = MersenneDivisorCycles.CalculateCycleLengthGpu(divisors[i].Modulus);
+            checksum ^= exponents[i].Pow2MontgomeryModWithCycle(cycle, divisors[i]);
+        }
 
-		return checksum;
-	}
-
-	/// <summary>
-	/// Montgomery reduction with a known cycle length (cycle lookup only); 25.81 μs on the small set (0.73× baseline) and 9.12 μs
-	/// on the large one (0.08×).
-	/// </summary>
-	[Benchmark]
-	public ulong MontgomeryWithPrecomputedCycleGpu()
-	{
-		GetData(out ulong[] exponents, out MontgomeryDivisorData[] divisors, out ulong[] cycles);
-		ulong checksum = 0UL;
-
-		for (int i = 0; i < SampleCount; i++)
-		{
-			checksum ^= exponents[i].Pow2MontgomeryModWithCycleGpu(cycles[i], divisors[i]);
-		}
-
-		return checksum;
-	}
-
-	/// <summary>
-	/// Montgomery reduction with the cycle computed on the fly using the GPU helper; costs 54.09 ms on the small benchmark due to
-	/// cycle discovery, but drops to 12.91 μs on the large set (0.11× baseline).
-	/// </summary>
-	// [Benchmark]
-	public ulong MontgomeryWithGpuCycleComputationCpu()
-	{
-		GetData(out ulong[] exponents, out MontgomeryDivisorData[] divisors, out _);
-		ulong checksum = 0UL;
-
-		for (int i = 0; i < SampleCount; i++)
-		{
-			ulong cycle = MersenneDivisorCycles.CalculateCycleLengthGpu(divisors[i].Modulus);
-			checksum ^= exponents[i].Pow2MontgomeryModWithCycleCpu(cycle, divisors[i]);
-		}
-
-		return checksum;
-	}
-
-	/// <summary>
-	/// Montgomery reduction with the cycle computed on the fly using the GPU helper; costs 54.09 ms on the small benchmark due to
-	/// cycle discovery, but drops to 12.91 μs on the large set (0.11× baseline).
-	/// </summary>
-	// [Benchmark]
-	public ulong MontgomeryWithGpuCycleComputationGpu()
-	{
-		GetData(out ulong[] exponents, out MontgomeryDivisorData[] divisors, out _);
-		ulong checksum = 0UL;
-
-		for (int i = 0; i < SampleCount; i++)
-		{
-			ulong cycle = MersenneDivisorCycles.CalculateCycleLengthGpu(divisors[i].Modulus);
-			checksum ^= exponents[i].Pow2MontgomeryModWithCycleGpu(cycle, divisors[i]);
-		}
-
-		return checksum;
-	}
-
-	[Benchmark]
-	public ulong MontgomeryWithHeuristicCycleComputationCpu()
-	{
-		GetData(out ulong[] exponents, out MontgomeryDivisorData[] divisors, out _);
-		ulong checksum = 0UL;
-
-		for (int i = 0; i < SampleCount; i++)
-		{
-			ulong cycle = CalculateCycleLengthWithHeuristics(divisors[i].Modulus);
-			checksum ^= exponents[i].Pow2MontgomeryModWithCycleCpu(cycle, divisors[i]);
-		}
-
-		return checksum;
-	}
-
-	[Benchmark]
-	public ulong MontgomeryWithHeuristicCycleComputationGpu()
-	{
-		GetData(out ulong[] exponents, out MontgomeryDivisorData[] divisors, out _);
-		ulong checksum = 0UL;
-
-		for (int i = 0; i < SampleCount; i++)
-		{
-			ulong cycle = CalculateCycleLengthWithHeuristics(divisors[i].Modulus);
-			checksum ^= exponents[i].Pow2MontgomeryModWithCycleGpu(cycle, divisors[i]);
-		}
-
-		return checksum;
-	}
-
-	[Benchmark]
-	public UInt128 HeuristicWideCycleComputation()
-	{
-		UInt128 checksum = UInt128.Zero;
-
-		for (int i = 0; i < SampleCount; i++)
-		{
-			UInt128 cycle = CalculateCycleLengthWithHeuristics(_wideModuli[i]);
-			_wideCycles[i] = cycle;
-			checksum ^= cycle;
-		}
-
-		return checksum;
-	}
-
-	private void GetData(out ulong[] exponents, out MontgomeryDivisorData[] divisors, out ulong[] cycles)
-	{
-		switch (Scale)
-		{
-			case InputScale.Small:
-				exponents = _smallExponents;
-				divisors = _smallDivisors;
-				cycles = _smallCycles;
-				break;
-			case InputScale.Large:
-				exponents = _largeExponents;
-				divisors = _largeDivisors;
-				cycles = _largeCycles;
-				break;
-			case InputScale.VeryLarge:
-				exponents = _veryLargeExponents;
-				divisors = _veryLargeDivisors;
-				cycles = _veryLargeCycles;
-				break;
-			default:
-				throw new ArgumentOutOfRangeException(nameof(Scale), Scale, null);
-		}
-	}
-
-	private ulong NextSmallExponent() => (ulong)_random.NextInt64(1L, 1_000_000L);
-
-	private ulong NextLargeExponent() => (ulong)_random.NextInt64(OneShiftedLeft60, long.MaxValue);
-
-	private ulong NextVeryLargeExponent()
-	{
-		Span<byte> buffer = stackalloc byte[8];
-		_random.NextBytes(buffer);
-		ulong value = checked(BinaryPrimitives.ReadUInt64LittleEndian(buffer) | OneShiftedLeft63);
-		return value;
-	}
-
-	private ulong NextSmallOddModulus()
-	{
-		ulong value = (ulong)_random.NextInt64(3L, 1_000_000L);
-		return value | 1UL;
-	}
-
-	private (ulong modulus, ulong cycleLength) NextLargeModulusAndCycle()
-	{
-		int bitLength = _random.Next(48, 64);
-		ulong modulus = (1UL << bitLength) - 1UL;
-		return (modulus, (ulong)bitLength);
-	}
-
-	private ulong NextVeryLargeOddModulus()
-	{
-		Span<byte> buffer = stackalloc byte[8];
-		_random.NextBytes(buffer);
-		ulong value = BinaryPrimitives.ReadUInt64LittleEndian(buffer) | OneShiftedLeft63 | 1UL;
-		return value;
-	}
-
-	private UInt128 NextWideOddModulus()
-	{
-		Span<byte> buffer = stackalloc byte[16];
-		_random.NextBytes(buffer);
-		ulong low = BinaryPrimitives.ReadUInt64LittleEndian(buffer);
-		Span<byte> highSlice = buffer.Slice(8, 8);
-		ulong high = BinaryPrimitives.ReadUInt64LittleEndian(highSlice) | (1UL << 63);
-		UInt128 value = ((UInt128)high << 64) | low;
-		return value | UInt128.One;
-	}
-
-	private static MontgomeryDivisorData CreateMontgomeryDivisorData(ulong modulus)
-	{
-		if (modulus <= 1UL || (modulus & 1UL) == 0UL)
-		{
-			return new MontgomeryDivisorData(modulus, 0UL, 0UL, 0UL, 0UL);
-		}
-
-		ulong nPrime = ComputeMontgomeryNPrime(modulus);
-		ulong montgomeryOne = ComputeMontgomeryResidue(1UL, modulus);
-		ulong montgomeryTwo = ComputeMontgomeryResidue(2UL, modulus);
-		ulong montgomeryTwoSquared = ULongExtensions.MontgomeryMultiply(montgomeryTwo, montgomeryTwo, modulus, nPrime);
-
-		return new MontgomeryDivisorData(
-			modulus,
-			nPrime,
-			montgomeryOne,
-			montgomeryTwo,
-			montgomeryTwoSquared);
-	}
+        return checksum;
+    }
 
 
-	private ulong CalculateCycleLengthWithHeuristics(ulong modulus, ulong fallbackOrder = 0UL)
-	{
-		MontgomeryDivisorData divisorData = MontgomeryDivisorData.FromModulus(modulus);
-		if (modulus <= 1UL || (modulus & 1UL) == 0UL)
-		{
-			return fallbackOrder != 0UL ? fallbackOrder : MersenneDivisorCycles.CalculateCycleLength(modulus, divisorData);
-		}
+    [Benchmark]
+    public ulong MontgomeryWithHeuristicCycleComputation()
+    {
+        GetData(out ulong[] exponents, out MontgomeryDivisorData[] divisors, out _);
+        ulong checksum = 0UL;
 
-		// When the below is uncommented, the benchmark never completes due to the primality test cost on very large random set.
+        for (int i = 0; i < SampleCount; i++)
+        {
+            ulong cycle = CalculateCycleLengthWithHeuristics(divisors[i].Modulus);
+            checksum ^= exponents[i].Pow2MontgomeryModWithCycle(cycle, divisors[i]);
+        }
 
-		// bool isPrime = Open.Numeric.Primes.Prime.Numbers.IsPrime(modulus);
-		// // bool isPrime = PrimeTester.IsPrimeInternal(modulus, CancellationToken.None);
-		// if (!isPrime)
-		// {
-		//     return fallbackOrder != 0UL ? fallbackOrder : MersenneDivisorCycles.CalculateCycleLength(modulus);
-		// }
+        return checksum;
+    }
+
+    [Benchmark]
+    public UInt128 HeuristicWideCycleComputation()
+    {
+        UInt128 checksum = UInt128.Zero;
+
+        for (int i = 0; i < SampleCount; i++)
+        {
+            UInt128 cycle = CalculateCycleLengthWithHeuristics(_wideModuli[i]);
+            _wideCycles[i] = cycle;
+            checksum ^= cycle;
+        }
+
+        return checksum;
+    }
+
+    private void GetData(out ulong[] exponents, out MontgomeryDivisorData[] divisors, out ulong[] cycles)
+    {
+        switch (Scale)
+        {
+            case InputScale.Small:
+                exponents = _smallExponents;
+                divisors = _smallDivisors;
+                cycles = _smallCycles;
+                break;
+            case InputScale.Large:
+                exponents = _largeExponents;
+                divisors = _largeDivisors;
+                cycles = _largeCycles;
+                break;
+            case InputScale.VeryLarge:
+                exponents = _veryLargeExponents;
+                divisors = _veryLargeDivisors;
+                cycles = _veryLargeCycles;
+                break;
+            default:
+                throw new ArgumentOutOfRangeException(nameof(Scale), Scale, null);
+        }
+    }
+
+    private ulong NextSmallExponent() => (ulong)_random.NextInt64(1L, 1_000_000L);
+
+    private ulong NextLargeExponent() => (ulong)_random.NextInt64(OneShiftedLeft60, long.MaxValue);
+
+    private ulong NextVeryLargeExponent()
+    {
+        Span<byte> buffer = stackalloc byte[8];
+        _random.NextBytes(buffer);
+        ulong value = checked(BinaryPrimitives.ReadUInt64LittleEndian(buffer) | OneShiftedLeft63);
+        return value;
+    }
+
+    private ulong NextSmallOddModulus()
+    {
+        ulong value = (ulong)_random.NextInt64(3L, 1_000_000L);
+        return value | 1UL;
+    }
+
+    private (ulong modulus, ulong cycleLength) NextLargeModulusAndCycle()
+    {
+        int bitLength = _random.Next(48, 64);
+        ulong modulus = (1UL << bitLength) - 1UL;
+        return (modulus, (ulong)bitLength);
+    }
+
+    private ulong NextVeryLargeOddModulus()
+    {
+        Span<byte> buffer = stackalloc byte[8];
+        _random.NextBytes(buffer);
+        ulong value = BinaryPrimitives.ReadUInt64LittleEndian(buffer) | OneShiftedLeft63 | 1UL;
+        return value;
+    }
+
+    private UInt128 NextWideOddModulus()
+    {
+        Span<byte> buffer = stackalloc byte[16];
+        _random.NextBytes(buffer);
+        ulong low = BinaryPrimitives.ReadUInt64LittleEndian(buffer);
+        Span<byte> highSlice = buffer.Slice(8, 8);
+        ulong high = BinaryPrimitives.ReadUInt64LittleEndian(highSlice) | (1UL << 63);
+        UInt128 value = ((UInt128)high << 64) | low;
+        return value | UInt128.One;
+    }
+
+    private static MontgomeryDivisorData CreateMontgomeryDivisorData(ulong modulus)
+    {
+        if (modulus <= 1UL || (modulus & 1UL) == 0UL)
+        {
+            return new MontgomeryDivisorData(modulus, 0UL, 0UL, 0UL, 0UL);
+        }
+
+        ulong nPrime = ComputeMontgomeryNPrime(modulus);
+        ulong montgomeryOne = ComputeMontgomeryResidue(1UL, modulus);
+        ulong montgomeryTwo = ComputeMontgomeryResidue(2UL, modulus);
+        ulong montgomeryTwoSquared = ULongExtensions.MontgomeryMultiply(montgomeryTwo, montgomeryTwo, modulus, nPrime);
+
+        return new MontgomeryDivisorData(
+            modulus,
+            nPrime,
+            montgomeryOne,
+            montgomeryTwo,
+            montgomeryTwoSquared);
+    }
+
+
+    private ulong CalculateCycleLengthWithHeuristics(ulong modulus, ulong fallbackOrder = 0UL)
+    {
+        MontgomeryDivisorData divisorData = MontgomeryDivisorData.FromModulus(modulus);
+        if (modulus <= 1UL || (modulus & 1UL) == 0UL)
+        {
+            return fallbackOrder != 0UL ? fallbackOrder : MersenneDivisorCycles.CalculateCycleLength(modulus, divisorData);
+        }
+
+        // When the below is uncommented, the benchmark never completes due to the primality test cost on very large random set.
+
+        // bool isPrime = Open.Numeric.Primes.Prime.Numbers.IsPrime(modulus);
+        // // bool isPrime = PrimeTester.IsPrimeInternal(modulus, CancellationToken.None);
+        // if (!isPrime)
+        // {
+        //     return fallbackOrder != 0UL ? fallbackOrder : MersenneDivisorCycles.CalculateCycleLength(modulus);
+        // }
 
 #if DEBUG
-		Console.WriteLine("Trying heuristic. Prime order calculation");
+        Console.WriteLine("Trying heuristic. Prime order calculation");
 #endif
-		ulong order = PrimeOrderCalculator.Calculate(
-			modulus,
-			_previousPrimeOrder,
-			divisorData,
-			PrimeOrderCalculator.PrimeOrderSearchConfig.HeuristicDefault,
-			PrimeOrderCalculator.PrimeOrderHeuristicDevice.Gpu);
+        PrimeOrderCalculator.PrimeOrderResult orderResult = PrimeOrderCalculator.Calculate(
+            modulus,
+            _previousPrimeOrder,
+            divisorData,
+            PrimeOrderCalculator.PrimeOrderSearchConfig.HeuristicDefault);
 
-		if (order != 0UL)
-		{
-			_previousPrimeOrder = order;
-			return order;
-		}
+        if (orderResult.Order != 0UL)
+        {
+            _previousPrimeOrder = orderResult.Order;
+            return orderResult.Order;
+        }
 
-		_previousPrimeOrder = null;
+        _previousPrimeOrder = null;
 
-		Console.WriteLine($"Heuristic failed for {modulus}, falling back to full cycle calculation");
-		return fallbackOrder != 0UL ? fallbackOrder : MersenneDivisorCycles.CalculateCycleLength(modulus, divisorData);
-	}
+        Console.WriteLine($"Heuristic failed for {modulus}, falling back to full cycle calculation");
+        return fallbackOrder != 0UL ? fallbackOrder : MersenneDivisorCycles.CalculateCycleLength(modulus, divisorData);
+    }
 
-	private UInt128 CalculateCycleLengthWithHeuristics(UInt128 modulus)
-	{
-		if (modulus <= UInt128.One || (modulus & UInt128.One) == UInt128.Zero)
-		{
-			return MersenneDivisorCycles.GetCycle(modulus);
-		}
+    private UInt128 CalculateCycleLengthWithHeuristics(UInt128 modulus)
+    {
+        if (modulus <= UInt128.One || (modulus & UInt128.One) == UInt128.Zero)
+        {
+            return MersenneDivisorCycles.GetCycle(modulus);
+        }
 
 #if DEBUG
-		Console.WriteLine("Trying heuristic. Wide prime order calculation");
+        Console.WriteLine("Trying heuristic. Wide prime order calculation");
 #endif
-        UInt128 order = PrimeOrderCalculator.Calculate(
-                modulus,
-                _previousWidePrimeOrder,
-                PrimeOrderCalculator.PrimeOrderSearchConfig.HeuristicDefault,
-                PrimeOrderCalculator.PrimeOrderHeuristicDevice.Gpu);
+        PrimeOrderCalculator.PrimeOrderResultWide orderResult = PrimeOrderCalculator.Calculate(
+            modulus,
+            _previousWidePrimeOrder,
+            PrimeOrderCalculator.PrimeOrderSearchConfig.HeuristicDefault);
 
-		if (order != UInt128.Zero)
-		{
-			_previousWidePrimeOrder = order;
-			return order;
-		}
+        if (orderResult.Order != UInt128.Zero)
+        {
+            _previousWidePrimeOrder = orderResult.Order;
+            return orderResult.Order;
+        }
 
-		_previousWidePrimeOrder = null;
+        _previousWidePrimeOrder = null;
 
-		Console.WriteLine($"Wide heuristic failed for {modulus}, falling back to full cycle calculation");
-		return MersenneDivisorCycles.GetCycle(modulus);
-	}
+        Console.WriteLine($"Wide heuristic failed for {modulus}, falling back to full cycle calculation");
+        return MersenneDivisorCycles.GetCycle(modulus);
+    }
 
-	private static ulong ComputeMontgomeryResidue(ulong value, ulong modulus) => (ulong)((UInt128)value * (UInt128.One << 64) % modulus);
+    private static ulong ComputeMontgomeryResidue(ulong value, ulong modulus) => (ulong)((UInt128)value * (UInt128.One << 64) % modulus);
 
-	private static ulong ComputeMontgomeryNPrime(ulong modulus)
-	{
-		ulong inv = modulus;
-		inv *= unchecked(2UL - modulus * inv);
-		inv *= unchecked(2UL - modulus * inv);
-		inv *= unchecked(2UL - modulus * inv);
-		inv *= unchecked(2UL - modulus * inv);
-		inv *= unchecked(2UL - modulus * inv);
-		inv *= unchecked(2UL - modulus * inv);
-		return unchecked(0UL - inv);
-	}
+    private static ulong ComputeMontgomeryNPrime(ulong modulus)
+    {
+        ulong inv = modulus;
+        inv *= unchecked(2UL - modulus * inv);
+        inv *= unchecked(2UL - modulus * inv);
+        inv *= unchecked(2UL - modulus * inv);
+        inv *= unchecked(2UL - modulus * inv);
+        inv *= unchecked(2UL - modulus * inv);
+        inv *= unchecked(2UL - modulus * inv);
+        return unchecked(0UL - inv);
+    }
 }
 

EvenPerfectBitScanner.Benchmarks/Pow2MontgomeryModUInt128Benchmarks.cs

@@ -176,7 +176,7 @@ public class Pow2MontgomeryModUInt128Benchmarks
         return (UInt128)result;
     }
 
-    private static bool ShouldUseSingleBit(GpuUInt128 exponent)
+    private static bool ShouldUseSingleBit(in GpuUInt128 exponent)
     {
         if (exponent.High == 0UL && exponent.Low <= Pow2WindowFallbackThreshold)
         {
@@ -257,7 +257,7 @@ public class Pow2MontgomeryModUInt128Benchmarks
         return result;
     }
 
-    private static bool IsBitSet(GpuUInt128 value, int bitIndex)
+    private static bool IsBitSet(in GpuUInt128 value, int bitIndex)
     {
         if (bitIndex >= 64)
         {
@@ -267,7 +267,7 @@ public class Pow2MontgomeryModUInt128Benchmarks
         return ((value.Low >> bitIndex) & 1UL) != 0UL;
     }
 
-    private static ulong ExtractWindowValue(GpuUInt128 exponent, int windowStart, int windowBitCount)
+    private static ulong ExtractWindowValue(in GpuUInt128 exponent, int windowStart, int windowBitCount)
     {
         if (windowStart != 0)
         {

EvenPerfectBitScanner.Benchmarks/Program.cs

@@ -8,22 +8,16 @@ public static class Program
 {
     public static void Main(string[] args)
     {
-		if (args.Length > 0)
-		{
-			IConfig config = DefaultConfig.Instance.WithOptions(ConfigOptions.JoinSummary | ConfigOptions.DisableOptimizationsValidator);
-			BenchmarkSwitcher.FromAssembly(typeof(Program).Assembly).Run(args, config);
-			return;
-		}
+        // var config = DefaultConfig.Instance.WithOptions(ConfigOptions.JoinSummary | ConfigOptions.DisableOptimizationsValidator);
+        // BenchmarkSwitcher.FromAssembly(typeof(Program).Assembly).Run(args, config);
 
-		BenchmarkSwitcher.FromAssembly(typeof(Program).Assembly).Run();
+        Pow2MontgomeryModCycleComputationBenchmarks benchmarks = new()
+        {
+            Scale = Pow2MontgomeryModCycleComputationBenchmarks.InputScale.Large
+        };
 
-        // Pow2MontgomeryModCycleComputationBenchmarks benchmarks = new()
-        // {
-        //     Scale = Pow2MontgomeryModCycleComputationBenchmarks.InputScale.Large
-        // };
-
-        // GpuPrimeWorkLimiter.SetLimit(1);
-        // benchmarks.Setup();
+        GpuPrimeWorkLimiter.SetLimit(1);
+        benchmarks.Setup();
 
         // Console.WriteLine("Press ENTER to run");
         // Console.ReadLine();

EvenPerfectBitScanner.ResultsParser/EvenPerfectBitScanner.ResultsParser.csproj

@@ -11,7 +11,4 @@
     <PackageReference Include="Open.Numeric.Primes" Version="4.0.4" />
   </ItemGroup>
 
-  <ItemGroup>
-    <Compile Remove=".temp.cs" />
-  </ItemGroup>
 </Project>

EvenPerfectBitScanner.Tests/EvenPerfectBitScanner.Tests.csproj

@@ -22,7 +22,4 @@
   <ItemGroup>
     <ProjectReference Include="../EvenPerfectBitScanner/EvenPerfectBitScanner.csproj" />
   </ItemGroup>
-  <ItemGroup>
-    <Compile Remove=".temp.cs" />
-  </ItemGroup>
 </Project>

EvenPerfectBitScanner.Tests/ProgramTests.cs

@@ -540,7 +540,7 @@ public class ProgramTests
     public void Main_displays_help_when_requested()
     {
         var main = typeof(Program).GetMethod("Main", BindingFlags.NonPublic | BindingFlags.Static)!;
-        var writer = new StringWriter();
+        using var writer = new StringWriter();
         TextWriter original = Console.Out;
         Console.SetOut(writer);
 
@@ -554,7 +554,6 @@ public class ProgramTests
         }
 
         string output = writer.ToString();
-        writer.Dispose();
         output.Should().Contain("Usage:");
         output.Should().Contain("--mersenne-device=cpu|gpu");
         output.Should().Contain("--primes-device=cpu|gpu");

EvenPerfectBitScanner/EvenPerfectBitScanner.csproj

@@ -16,7 +16,4 @@
     <PackageReference Include="Open.Numeric" Version="3.0.1" />
     <PackageReference Include="Open.Numeric.Primes" Version="4.0.4" />
   </ItemGroup>
-  <ItemGroup>
-    <Compile Remove=".temp.cs" />
-  </ItemGroup>
 </Project>

EvenPerfectBitScanner/Program.cs

@@ -44,7 +44,6 @@ internal static class Program
     private static string? _resultsPrefix;
     private static readonly Optimized PrimeIterator = new();
     private static ulong _byDivisorStartPrime;
-    private static TimeSpan? _primeTestLimit;
 
     [ThreadStatic]
     private static bool _lastCompositeP;
@@ -75,8 +74,6 @@ internal static class Program
         bool useResidue = false;    // M_p test via residue divisors (replaces LL/incremental)
         bool useDivisor = false;     // M_p divisibility by specific divisor
         bool useByDivisor = false;   // Iterative divisor scan across primes		UInt128 divisor = UInt128.Zero;
-        ByDivisorDeltasDevice byDivisorDeltasDevice = ByDivisorDeltasDevice.Cpu;
-        ByDivisorMontgomeryDevice byDivisorMontgomeryDevice = ByDivisorMontgomeryDevice.Cpu;
                                      // Device routing
         bool useGpuCycles = true;
         bool mersenneOnGpu = true;   // controls Lucas/incremental/pow2mod device
@@ -184,51 +181,6 @@ internal static class Program
             {
                 mersenneOnGpu = !arg.AsSpan(arg.IndexOf('=') + 1).Equals("cpu", StringComparison.OrdinalIgnoreCase);
             }
-            else if (arg.StartsWith("--bydivisor-deltas-device=", StringComparison.OrdinalIgnoreCase))
-            {
-                ReadOnlySpan<char> deviceValue = arg.AsSpan(arg.IndexOf('=') + 1);
-                if (deviceValue.Equals("gpu", StringComparison.OrdinalIgnoreCase))
-                {
-                    byDivisorDeltasDevice = ByDivisorDeltasDevice.Gpu;
-                }
-                else if (deviceValue.Equals("cpu", StringComparison.OrdinalIgnoreCase))
-                {
-                    byDivisorDeltasDevice = ByDivisorDeltasDevice.Cpu;
-                }
-                else
-                {
-                    Console.WriteLine("Invalid --bydivisor-deltas-device value. Expected cpu or gpu.");
-                    return;
-                }
-            }
-            else if (arg.StartsWith("--bydivisor-montgomery-device=", StringComparison.OrdinalIgnoreCase))
-            {
-                ReadOnlySpan<char> deviceValue = arg.AsSpan(arg.IndexOf('=') + 1);
-                if (deviceValue.Equals("gpu", StringComparison.OrdinalIgnoreCase))
-                {
-                    byDivisorMontgomeryDevice = ByDivisorMontgomeryDevice.Gpu;
-                }
-                else if (deviceValue.Equals("cpu", StringComparison.OrdinalIgnoreCase))
-                {
-                    byDivisorMontgomeryDevice = ByDivisorMontgomeryDevice.Cpu;
-                }
-                else
-                {
-                    Console.WriteLine("Invalid --bydivisor-montgomery-device value. Expected cpu or gpu.");
-                    return;
-                }
-            }
-            else if (arg.StartsWith("--prime-test-limit=", StringComparison.OrdinalIgnoreCase))
-            {
-                ReadOnlySpan<char> value = arg.AsSpan(arg.IndexOf('=') + 1);
-                if (!TryParsePrimeTestLimit(value, out TimeSpan parsedLimitValue))
-                {
-                    Console.WriteLine("Failed to parse --prime-test-limit value.");
-                    return;
-                }
-
-                _primeTestLimit = parsedLimitValue;
-            }
             else if (arg.Equals("--workaround-mod", StringComparison.OrdinalIgnoreCase))
             {
                 useModuloWorkaround = true;
@@ -425,8 +377,6 @@ internal static class Program
             }
         }
 
-        GpuKernelPool.CpuCycleUsesGpu = orderOnGpu;
-
         if (useByDivisor)
         {
             _byDivisorStartPrime = startPrimeProvided ? currentP : 0UL;
@@ -565,13 +515,10 @@ internal static class Program
         {
             _byDivisorTester = mersenneOnGpu
                     ? new MersenneNumberDivisorByDivisorGpuTester()
-                    : new MersenneNumberDivisorByDivisorCpuTester();
+                    : new MersenneNumberDivisorByDivisorCpuTester(
+                        useGpuPow2Mod: !GpuContextPool.ForceCpu,
+                        useGpuCycleComputation: orderOnGpu);
             _byDivisorTester.BatchSize = scanBatchSize;
-            if (_byDivisorTester is MersenneNumberDivisorByDivisorCpuTester cpuByDivisorTester)
-            {
-                cpuByDivisorTester.DeltasDevice = byDivisorDeltasDevice;
-                cpuByDivisorTester.MontgomeryDevice = byDivisorMontgomeryDevice;
-            }
         }
 
         // Load RLE blacklist (optional)
@@ -607,18 +554,18 @@ internal static class Program
                         useDivisor,
                         useByDivisor,
                         mersenneOnGpu,
-                        useOrder,
-                        useModuloWorkaround,
-                        _useGcdFilter,
-                        NttGpuMath.GpuTransformBackend,
-                        gpuPrimeThreads,
-                        sliceSize,
-                        scanBatchSize,
-                        _orderWarmupLimitOverride ?? 5_000_000UL,
-                        NttGpuMath.ReductionMode,
-                        mersenneOnGpu ? "gpu" : "cpu",
-                        (GpuContextPool.ForceCpu ? "cpu" : "gpu"),
-                        orderOnGpu ? "gpu" : "cpu");
+        useOrder,
+        useModuloWorkaround,
+        _useGcdFilter,
+        NttGpuMath.GpuTransformBackend,
+        gpuPrimeThreads,
+        sliceSize,
+        scanBatchSize,
+        _orderWarmupLimitOverride ?? 5_000_000UL,
+        NttGpuMath.ReductionMode,
+        mersenneOnGpu ? "gpu" : "cpu",
+        (GpuContextPool.ForceCpu ? "cpu" : "gpu"),
+        orderOnGpu ? "gpu" : "cpu");
 
         if (!string.IsNullOrEmpty(_resultsPrefix))
         {
@@ -720,21 +667,7 @@ internal static class Program
 
         if (_useByDivisorMode)
         {
-            if (_byDivisorTester is null)
-            {
-                throw new InvalidOperationException("By-divisor tester has not been initialized.");
-            }
-
-            MersenneNumberDivisorByDivisorTester.Run(
-                    byDivisorCandidates,
-                    _byDivisorTester,
-                    _byDivisorPreviousResults,
-                    _byDivisorStartPrime,
-                    static () => _lastCompositeP = true,
-                    static () => _lastCompositeP = false,
-                    PrintResult,
-                    threadCount,
-                    _primeTestLimit);
+            RunByDivisorMode(byDivisorCandidates, threadCount);
             FlushBuffer();
             StringBuilderPool.Return(_outputBuilder!);
             return;
@@ -875,6 +808,26 @@ internal static class Program
         return candidates;
     }
 
+    private static void RunByDivisorMode(List<ulong> candidates, int threadCount)
+    {
+        if (_byDivisorTester is null)
+        {
+            throw new InvalidOperationException("By-divisor tester has not been initialized.");
+        }
+
+        // TODO: Inline this helper by invoking MersenneNumberDivisorByDivisorTester.Run directly at call sites so
+        // the CLI avoids this extra frame during the hot composite-reporting path.
+        MersenneNumberDivisorByDivisorTester.Run(
+                candidates,
+                _byDivisorTester,
+                _byDivisorPreviousResults,
+                _byDivisorStartPrime,
+                static () => _lastCompositeP = true,
+                static () => _lastCompositeP = false,
+                PrintResult,
+                threadCount);
+    }
+
     private static unsafe void LoadResultsFile(string resultsFileName, Action<ulong, bool, bool> lineProcessorAction)
     {
         using FileStream readStream = new(resultsFileName, FileMode.Open, FileAccess.Read, FileShare.ReadWrite);
@@ -939,51 +892,6 @@ internal static class Program
         }
     }
 
-    private static bool TryParsePrimeTestLimit(ReadOnlySpan<char> value, out TimeSpan result)
-    {
-        result = TimeSpan.Zero;
-        if (value.IsEmpty)
-        {
-            return false;
-        }
-
-        ReadOnlySpan<char> magnitudeSpan = value;
-        double magnitude;
-
-        if (value.Length >= 2 && value.EndsWith("ms", StringComparison.OrdinalIgnoreCase))
-        {
-            magnitudeSpan = value[..^2];
-            if (!double.TryParse(magnitudeSpan, NumberStyles.Float, CultureInfo.InvariantCulture, out magnitude))
-            {
-                return false;
-            }
-
-            result = TimeSpan.FromMilliseconds(magnitude);
-            return true;
-        }
-
-        char suffix = char.ToLowerInvariant(value[^1]);
-        if (suffix == 's' || suffix == 'm')
-        {
-            magnitudeSpan = value[..^1];
-            if (!double.TryParse(magnitudeSpan, NumberStyles.Float, CultureInfo.InvariantCulture, out magnitude))
-            {
-                return false;
-            }
-
-            result = suffix == 's' ? TimeSpan.FromSeconds(magnitude) : TimeSpan.FromMinutes(magnitude);
-            return true;
-        }
-
-        if (!double.TryParse(value, NumberStyles.Float, CultureInfo.InvariantCulture, out magnitude))
-        {
-            return false;
-        }
-
-        result = TimeSpan.FromMilliseconds(magnitude);
-        return true;
-    }
-
     private static void PrintHelp()
     {
         Console.WriteLine("Usage: EvenPerfectBitScanner [options]");
@@ -1017,13 +925,10 @@ internal static class Program
         Console.WriteLine("  --divisor-cycles-batch=<value> batch size for cycles generation (default 512)");
         Console.WriteLine("  --divisor-cycles-continue  continue divisor cycles generation");
         Console.WriteLine("  --divisor-cycles-limit=<value> cycle search iterations when --mersenne=divisor");
-        Console.WriteLine("  --prime-test-limit=<value>   Time limit for --mersenne=bydivisor prime checks (e.g. 5s, 500ms)");
-        Console.WriteLine("  --bydivisor-deltas-device=cpu|gpu  Device for --mersenne=bydivisor residue batches (default: cpu)");
-        Console.WriteLine("  --bydivisor-montgomery-device=cpu|gpu  Device for Montgomery data when using --mersenne=bydivisor (default: cpu)");
         Console.WriteLine("  --use-order            test primality via q order");
         Console.WriteLine("  --workaround-mod       avoid '%' operator on the GPU");
         // mod-automaton removed
-        Console.WriteLine("  --filter-p=<path>      process only p from previous run results (required for --mersenne=bydivisor; primes must be strictly increasing)");
+        Console.WriteLine("  --filter-p=<path>      process only p from previous run results (required for --mersenne=bydivisor)");
         Console.WriteLine("  --write-batch-size=<value> overwrite frequency of disk writes (default 100 lines)");
         Console.WriteLine("  --gcd-filter           enable early sieve based on GCD");
         Console.WriteLine("  --help, -help, --?, -?, /?   show this help message");

PerfectNumbers.Core.Tests/Gpu/GpuContextPoolTests.cs

@@ -15,15 +15,9 @@ public class GpuContextPoolTests
     {
         GpuContextPool.DisposeAll();
 
-        GpuContextPool.GpuContextLease lease = GpuContextPool.RentPreferred(preferCpu: true);
-        try
-        {
-            lease.Accelerator.AcceleratorType.Should().Be(AcceleratorType.CPU);
-        }
-        finally
-        {
-            lease.Dispose();
-        }
+        using var lease = GpuContextPool.RentPreferred(preferCpu: true);
+
+        lease.Accelerator.AcceleratorType.Should().Be(AcceleratorType.CPU);
 
         GpuContextPool.DisposeAll();
     }
@@ -36,27 +30,17 @@ public class GpuContextPoolTests
 
         Context firstContext;
         Accelerator firstAccelerator;
-        GpuContextPool.GpuContextLease lease = GpuContextPool.RentPreferred(preferCpu: true);
-        try
+        using (var lease = GpuContextPool.RentPreferred(preferCpu: true))
         {
             firstContext = lease.Context;
             firstAccelerator = lease.Accelerator;
         }
-        finally
-        {
-            lease.Dispose();
-        }
 
-        lease = GpuContextPool.RentPreferred(preferCpu: true);
-        try
+        using (var lease = GpuContextPool.RentPreferred(preferCpu: true))
         {
             lease.Context.Should().BeSameAs(firstContext);
             lease.Accelerator.Should().BeSameAs(firstAccelerator);
         }
-        finally
-        {
-            lease.Dispose();
-        }
 
         GpuContextPool.DisposeAll();
     }
@@ -68,27 +52,17 @@ public class GpuContextPoolTests
         GpuContextPool.DisposeAll();
 
         Context firstContext;
-        GpuContextPool.GpuContextLease lease = GpuContextPool.RentPreferred(preferCpu: true);
-        try
+        using (var lease = GpuContextPool.RentPreferred(preferCpu: true))
         {
             firstContext = lease.Context;
         }
-        finally
-        {
-            lease.Dispose();
-        }
 
         GpuContextPool.DisposeAll();
 
-        lease = GpuContextPool.RentPreferred(preferCpu: true);
-        try
+        using (var lease = GpuContextPool.RentPreferred(preferCpu: true))
         {
             lease.Context.Should().NotBeSameAs(firstContext);
         }
-        finally
-        {
-            lease.Dispose();
-        }
 
         GpuContextPool.DisposeAll();
     }

PerfectNumbers.Core.Tests/Gpu/GpuKernelCompilationTests.cs

@@ -199,17 +199,10 @@ public class GpuKernelCompilationTests
     [MemberData(nameof(KernelLoaders))]
     public void Kernel_compiles_without_internal_compiler_errors(string name, Action<Accelerator> compile)
     {
-        GpuContextPool.GpuContextLease lease = GpuContextPool.RentPreferred(preferCpu: true);
-        try
-        {
-            var accelerator = lease.Accelerator;
-            Action action = () => compile(accelerator);
-            action.Should().NotThrow($"kernel {name} should compile");
-        }
-        finally
-        {
-            lease.Dispose();
-        }
+        using var lease = GpuContextPool.RentPreferred(preferCpu: true);
+        var accelerator = lease.Accelerator;
+        Action action = () => compile(accelerator);
+        action.Should().NotThrow($"kernel {name} should compile");
     }
 
     private static void CompileDivisorCycleCacheKernel(Accelerator accelerator)

PerfectNumbers.Core.Tests/Gpu/GpuPrimeLimiterTests.cs

@@ -18,14 +18,13 @@ public class GpuPrimeLimiterTests
         var startedSecond = new TaskCompletionSource<bool>();
         var sw = new Stopwatch();
 
-        var first = GpuPrimeWorkLimiter.Acquire();
+        using var first = GpuPrimeWorkLimiter.Acquire();
         sw.Start();
 
         var secondTask = Task.Run(() =>
         {
-            var second = GpuPrimeWorkLimiter.Acquire();
+            using var second = GpuPrimeWorkLimiter.Acquire();
             startedSecond.TrySetResult(true);
-            second.Dispose();
         });
 
         // Give secondTask a moment to attempt acquire (it should block).

PerfectNumbers.Core.Tests/Gpu/NttGpuMathTests.cs

@@ -150,36 +150,29 @@ public class NttGpuMathTests
         var modulus = new GpuUInt128(0UL, 17UL);
         var primitiveRoot = new GpuUInt128(0UL, 3UL);
 
-        GpuContextPool.GpuContextLease gpu = GpuContextPool.Rent();
-        try
+        using var gpu = GpuContextPool.Rent();
+        var accelerator = gpu.Accelerator;
+
+        var method = typeof(NttGpuMath).GetMethod("GetSquareCache", BindingFlags.NonPublic | BindingFlags.Static);
+        method.Should().NotBeNull();
+        var cache = method!.Invoke(null, new object[] { accelerator, 8, modulus, primitiveRoot })!;
+        var cacheType = cache.GetType();
+
+        var twiddlesMont = (MemoryBuffer1D<GpuUInt128, Stride1D.Dense>)cacheType.GetProperty("TwiddlesMont")!.GetValue(cache)!;
+        var twiddlesInvMont = (MemoryBuffer1D<GpuUInt128, Stride1D.Dense>)cacheType.GetProperty("TwiddlesInvMont")!.GetValue(cache)!;
+        var r2 = (ulong)cacheType.GetProperty("MontR2Mod64")!.GetValue(cache)!;
+        var mod = (ulong)cacheType.GetProperty("ModulusLow")!.GetValue(cache)!;
+
+        int count = 7; // length 8 -> 7 twiddles
+        var forward = new GpuUInt128[count];
+        var inverse = new GpuUInt128[count];
+        twiddlesMont.CopyToCPU(forward);
+        twiddlesInvMont.CopyToCPU(inverse);
+
+        for (int i = 0; i < count; i++)
         {
-            var accelerator = gpu.Accelerator;
-
-            var method = typeof(NttGpuMath).GetMethod("GetSquareCache", BindingFlags.NonPublic | BindingFlags.Static);
-            method.Should().NotBeNull();
-            var cache = method!.Invoke(null, new object[] { accelerator, 8, modulus, primitiveRoot })!;
-            var cacheType = cache.GetType();
-
-            var twiddlesMont = (MemoryBuffer1D<GpuUInt128, Stride1D.Dense>)cacheType.GetProperty("TwiddlesMont")!.GetValue(cache)!;
-            var twiddlesInvMont = (MemoryBuffer1D<GpuUInt128, Stride1D.Dense>)cacheType.GetProperty("TwiddlesInvMont")!.GetValue(cache)!;
-            var r2 = (ulong)cacheType.GetProperty("MontR2Mod64")!.GetValue(cache)!;
-            var mod = (ulong)cacheType.GetProperty("ModulusLow")!.GetValue(cache)!;
-
-            int count = 7; // length 8 -> 7 twiddles
-            var forward = new GpuUInt128[count];
-            var inverse = new GpuUInt128[count];
-            twiddlesMont.CopyToCPU(forward);
-            twiddlesInvMont.CopyToCPU(inverse);
-
-            for (int i = 0; i < count; i++)
-            {
-                ulong prod = (ulong)((((BigInteger)forward[i].Low) * inverse[i].Low) % mod);
-                prod.Should().Be(r2);
-            }
-        }
-        finally
-        {
-            gpu.Dispose();
+            ulong prod = (ulong)((((BigInteger)forward[i].Low) * inverse[i].Low) % mod);
+            prod.Should().Be(r2);
         }
     }
 
@@ -345,23 +338,12 @@ public class NttGpuMathTests
         values[1] = new GpuUInt128(0UL, 2UL);
         var modulus = new GpuUInt128(0UL, 17UL);
         var primitiveRoot = new GpuUInt128(0UL, 3UL);
-        GpuContextPool.GpuContextLease gpu = GpuContextPool.Rent();
-        try
+        using (var gpu = GpuContextPool.Rent())
         {
             var buffer = gpu.Accelerator.Allocate1D<GpuUInt128>(values.Length);
-            try
-            {
-                buffer.View.CopyFromCPU(ref values[0], values.Length);
-                NttGpuMath.SquareDevice(gpu.Accelerator, buffer.View, modulus, primitiveRoot);
-            }
-            finally
-            {
-                buffer.Dispose();
-            }
-        }
-        finally
-        {
-            gpu.Dispose();
+            buffer.View.CopyFromCPU(ref values[0], values.Length);
+            NttGpuMath.SquareDevice(gpu.Accelerator, buffer.View, modulus, primitiveRoot);
+            buffer.Dispose();
         }
 
         // Verify that the square cache now holds entries for the accelerator.

PerfectNumbers.Core.Tests/Gpu/NttMontgomeryTests.cs

@@ -33,9 +33,9 @@ public class NttMontgomeryTests
         NttGpuMath.Square(cpu, modulus, primitiveRoot);
 
         // Device execution
-        GpuContextPool.GpuContextLease context = GpuContextPool.Rent();
+        using var context = GpuContextPool.Rent();
         var accelerator = context.Accelerator;
-        var buffer = accelerator.Allocate1D<GpuUInt128>(gpu.Length);
+        using var buffer = accelerator.Allocate1D<GpuUInt128>(gpu.Length);
         buffer.View.CopyFromCPU(ref gpu[0], gpu.Length);
         NttGpuMath.SquareDevice(accelerator, buffer.View, modulus, primitiveRoot);
         buffer.View.CopyToCPU(ref gpu[0], gpu.Length);
@@ -44,9 +44,6 @@ public class NttMontgomeryTests
         {
             gpu[i].Should().Be(cpu[i]);
         }
-
-        buffer.Dispose();
-        context.Dispose();
     }
 
     [Fact]

PerfectNumbers.Core.Tests/Gpu/PrimeOrderGpuHeuristicsTests.cs

@@ -359,9 +359,10 @@ public class PrimeOrderGpuHeuristicsTests
             PrimeOrderGpuHeuristics.OverflowRegistry[prime] = 0;
 
             MontgomeryDivisorData divisorData = MontgomeryDivisorData.FromModulus(prime);
-            ulong result = PrimeOrderCalculator.Calculate(prime, null, divisorData, PrimeOrderCalculator.PrimeOrderSearchConfig.HeuristicDefault, PrimeOrderCalculator.PrimeOrderHeuristicDevice.Gpu);
+            PrimeOrderCalculator.PrimeOrderResult result = PrimeOrderCalculator.Calculate(prime, null, divisorData, PrimeOrderCalculator.PrimeOrderSearchConfig.HeuristicDefault);
 
-            result.Should().Be(3UL);
+            result.Status.Should().Be(PrimeOrderCalculator.PrimeOrderStatus.Found);
+            result.Order.Should().Be(3UL);
         }
         finally
         {

PerfectNumbers.Core.Tests/GpuKernelPoolTests.cs

@@ -11,21 +11,14 @@ public class GpuKernelPoolTests
     [Trait("Category", "Fast")]
     public void EnsureSmallCyclesOnDevice_returns_same_view_on_subsequent_calls()
     {
-        GpuContextPool.GpuContextLease lease = GpuContextPool.RentPreferred(preferCpu: true);
-        try
-        {
-            var accelerator = lease.Accelerator;
+        using var lease = GpuContextPool.RentPreferred(preferCpu: true);
+        var accelerator = lease.Accelerator;
 
-            var first = GpuKernelPool.EnsureSmallCyclesOnDevice(accelerator);
-            var second = GpuKernelPool.EnsureSmallCyclesOnDevice(accelerator);
+        var first = GpuKernelPool.EnsureSmallCyclesOnDevice(accelerator);
+        var second = GpuKernelPool.EnsureSmallCyclesOnDevice(accelerator);
 
-            second.Equals(first).Should().BeTrue();
-            second.Length.Should().Be(MersenneDivisorCycles.SmallDivisorsMax + 1);
-        }
-        finally
-        {
-            lease.Dispose();
-        }
+        second.Equals(first).Should().BeTrue();
+        second.Length.Should().Be(MersenneDivisorCycles.SmallDivisorsMax + 1);
     }
 
     [Fact]

PerfectNumbers.Core.Tests/GpuWorkLimiterTests.cs

@@ -17,14 +17,13 @@ public class GpuWorkLimiterTests
         var startedSecond = new TaskCompletionSource<bool>();
         var sw = new Stopwatch();
 
-        var first = GpuWorkLimiter.Acquire();
+        using var first = GpuWorkLimiter.Acquire();
         sw.Start();
 
         var secondTask = Task.Run(() =>
         {
-            var second = GpuWorkLimiter.Acquire();
+            using var second = GpuWorkLimiter.Acquire();
             startedSecond.TrySetResult(true);
-            second.Dispose();
         });
 
         await Task.Delay(50);

PerfectNumbers.Core.Tests/MersenneDivisorCyclesGenerateGpuTests.cs

@@ -14,8 +14,8 @@ public class MersenneDivisorCyclesGenerateGpuTests
         {
             MersenneDivisorCycles.GenerateGpu(path, maxDivisor: 50UL, batchSize: 16, skipCount: 0, nextPosition: 0);
             var pairs = new List<(ulong divisor, ulong cycle)>();
-            var stream = new FileStream(path, FileMode.Open, FileAccess.Read);
-            var reader = new BinaryReader(stream);
+            using var stream = new FileStream(path, FileMode.Open, FileAccess.Read);
+            using var reader = new BinaryReader(stream);
             while (stream.Position < stream.Length)
             {
                 ulong d = reader.ReadUInt64();
@@ -41,9 +41,6 @@ public class MersenneDivisorCyclesGenerateGpuTests
                 MontgomeryDivisorData divisorData = MontgomeryDivisorData.FromModulus(pair.divisor);
                 pair.cycle.Should().Be(MersenneDivisorCycles.CalculateCycleLength(pair.divisor, divisorData));
             }
-
-            reader.Dispose();
-            stream.Dispose();
         }
         finally
         {
@@ -64,8 +61,8 @@ public class MersenneDivisorCyclesGenerateGpuTests
             MersenneDivisorCycles.Generate(path, maxDivisor: 50UL, threads: 4);
 
             var pairs = new List<(ulong divisor, ulong cycle)>();
-            var stream = new FileStream(path, FileMode.Open, FileAccess.Read);
-            var reader = new BinaryReader(stream);
+            using var stream = new FileStream(path, FileMode.Open, FileAccess.Read);
+            using var reader = new BinaryReader(stream);
             while (stream.Position < stream.Length)
             {
                 ulong divisor = reader.ReadUInt64();
@@ -98,9 +95,6 @@ public class MersenneDivisorCyclesGenerateGpuTests
 
             var (_, completeCount) = MersenneDivisorCycles.FindLast(path);
             completeCount.Should().Be(expected.Count);
-
-            reader.Dispose();
-            stream.Dispose();
         }
         finally
         {

PerfectNumbers.Core.Tests/MersenneNumberDivisorByDivisorTesterTests.cs

@@ -26,8 +26,7 @@ public class MersenneNumberDivisorByDivisorTesterTests
             markComposite: () => compositeMarks++,
             clearComposite: () => clearedMarks++,
             printResult: (prime, searched, detailed, passed) => results.Add((prime, searched, detailed, passed)),
-            threadCount: 1,
-            primeTestLimit: null);
+            threadCount: 1);
 
         tester.ConfiguredMaxPrime.Should().Be(13UL);
         compositeMarks.Should().Be(2);
@@ -60,7 +59,7 @@ public class MersenneNumberDivisorByDivisorTesterTests
             return DivisorLimit;
         }
 
-        public bool IsPrime(ulong prime, out bool divisorsExhausted, TimeSpan? timeLimit = null)
+        public bool IsPrime(ulong prime, out bool divisorsExhausted)
         {
             divisorsExhausted = true;
             return prime % 7UL != 0UL && prime % 11UL != 0UL;

PerfectNumbers.Core.Tests/MersenneNumberDivisorCpuTesterTests.cs

@@ -1,6 +1,5 @@
 using System;
 using FluentAssertions;
-using PerfectNumbers.Core;
 using PerfectNumbers.Core.Cpu;
 using Xunit;
 
@@ -8,125 +7,78 @@ namespace PerfectNumbers.Core.Tests;
 
 public class MersenneNumberDivisorCpuTesterTests
 {
-    private const ulong LargeDivisor1 = 4_000_169UL;
-    private const ulong LargeDivisor2 = 4_000_561UL;
-
-    [Fact]
-    [Trait("Category", "Fast")]
-    public void ByDivisor_tester_tracks_divisors_across_primes()
-    {
-        var tester = new MersenneNumberDivisorByDivisorCpuTester();
-        tester.ConfigureFromMaxPrime(83UL);
-
-        tester.IsPrime(59UL, out bool divisorsExhausted).Should().BeTrue();
-        divisorsExhausted.Should().BeTrue();
-
-        tester.IsPrime(61UL, out divisorsExhausted).Should().BeTrue();
-        divisorsExhausted.Should().BeTrue();
-
-        tester.IsPrime(67UL, out divisorsExhausted).Should().BeTrue();
-        divisorsExhausted.Should().BeTrue();
-    }
-
-    [Fact]
-    [Trait("Category", "Fast")]
-    public void ByDivisor_prime_test_limit_disables_divisor_exhaustion()
-    {
-        var tester = new MersenneNumberDivisorByDivisorCpuTester();
-        tester.ConfigureFromMaxPrime(83UL);
-
-        tester.IsPrime(59UL, out bool divisorsExhausted, TimeSpan.Zero).Should().BeTrue();
-        divisorsExhausted.Should().BeFalse();
-    }
-
-    [Fact]
-    [Trait("Category", "Fast")]
-    public void ByDivisor_session_checks_divisors_across_exponents()
-    {
-        var tester = new MersenneNumberDivisorByDivisorCpuTester();
-        tester.ConfigureFromMaxPrime(197UL);
-
-        var session = tester.CreateDivisorSession();
-        ulong[] exponents = { 59UL, 61UL, 67UL, 71UL, 73UL, 79UL, 117_652UL, 333_380UL };
-        byte[] hits = new byte[exponents.Length];
-
-        MontgomeryDivisorData divisor1 = MontgomeryDivisorData.FromModulus(LargeDivisor1);
-        ulong cycle1 = MersenneDivisorCycles.CalculateCycleLength(LargeDivisor1, divisor1);
-        session.CheckDivisor(LargeDivisor1, divisor1, cycle1, exponents, hits);
-
-        byte[] expected1 = new byte[exponents.Length];
-        for (int i = 0; i < exponents.Length; i++)
+        [Fact]
+        [Trait("Category", "Fast")]
+        public void ByDivisor_tester_tracks_divisors_across_primes()
         {
-            expected1[i] = exponents[i].Pow2ModWindowedCpu(divisor1.Modulus) == 1UL ? (byte)1 : (byte)0;
+                var tester = new MersenneNumberDivisorByDivisorCpuTester();
+                tester.ConfigureFromMaxPrime(11UL);
+
+                tester.IsPrime(5UL, out bool divisorsExhausted).Should().BeTrue();
+                divisorsExhausted.Should().BeTrue();
+
+                tester.IsPrime(7UL, out divisorsExhausted).Should().BeTrue();
+                divisorsExhausted.Should().BeTrue();
+
+                tester.IsPrime(11UL, out divisorsExhausted).Should().BeFalse();
+                divisorsExhausted.Should().BeTrue();
         }
 
-        hits.Should().Equal(expected1);
-
-        Array.Fill(hits, (byte)0);
-        MontgomeryDivisorData divisor2 = MontgomeryDivisorData.FromModulus(LargeDivisor2);
-        ulong cycle2 = MersenneDivisorCycles.CalculateCycleLength(LargeDivisor2, divisor2);
-        session.CheckDivisor(LargeDivisor2, divisor2, cycle2, exponents, hits);
-
-        byte[] expected2 = new byte[exponents.Length];
-        for (int i = 0; i < exponents.Length; i++)
+        [Fact]
+        [Trait("Category", "Fast")]
+        public void ByDivisor_session_checks_divisors_across_primes()
         {
-            expected2[i] = exponents[i].Pow2ModWindowedCpu(divisor2.Modulus) == 1UL ? (byte)1 : (byte)0;
+                var tester = new MersenneNumberDivisorByDivisorCpuTester();
+                tester.ConfigureFromMaxPrime(13UL);
+
+                using var session = tester.CreateDivisorSession();
+                ulong[] primes = { 5UL, 7UL, 11UL, 13UL };
+                byte[] hits = new byte[primes.Length];
+
+                ulong cycle23 = MersenneDivisorCycles.CalculateCycleLength(23UL, MontgomeryDivisorData.FromModulus(23UL));
+                session.CheckDivisor(23UL, MontgomeryDivisorData.FromModulus(23UL), cycle23, primes, hits);
+                hits.Should().ContainInOrder(new byte[] { 0, 0, 1, 0 });
+
+                Array.Fill(hits, (byte)0);
+                ulong cycle31 = MersenneDivisorCycles.CalculateCycleLength(31UL, MontgomeryDivisorData.FromModulus(31UL));
+                session.CheckDivisor(31UL, MontgomeryDivisorData.FromModulus(31UL), cycle31, primes, hits);
+                hits.Should().ContainInOrder(new byte[] { 1, 0, 0, 0 });
         }
 
-        hits.Should().Equal(expected2);
-
-        session.Dispose();
-    }
-
-    [Fact]
-    [Trait("Category", "Fast")]
-    public void ByDivisor_session_marks_mersenne_numbers_divisible_by_large_divisor1_as_composite()
-    {
-        var tester = new MersenneNumberDivisorByDivisorCpuTester();
-        tester.ConfigureFromMaxPrime(197UL);
-
-        var session = tester.CreateDivisorSession();
-        MontgomeryDivisorData divisorData = MontgomeryDivisorData.FromModulus(LargeDivisor1);
-        ulong cycle = MersenneDivisorCycles.CalculateCycleLength(LargeDivisor1, divisorData);
-
-        ulong[] exponents = { 59UL, 67UL, cycle, cycle * 2UL, cycle + 1UL };
-        byte[] hits = new byte[exponents.Length];
-        session.CheckDivisor(LargeDivisor1, divisorData, cycle, exponents, hits);
-
-        byte[] expected = new byte[exponents.Length];
-        for (int i = 0; i < exponents.Length; i++)
+        [Fact]
+        [Trait("Category", "Fast")]
+        public void ByDivisor_session_marks_mersenne_numbers_divisible_by_seven_as_composite()
         {
-            expected[i] = exponents[i].Pow2ModWindowedCpu(divisorData.Modulus) == 1UL ? (byte)1 : (byte)0;
+                var tester = new MersenneNumberDivisorByDivisorCpuTester();
+                tester.ConfigureFromMaxPrime(43UL);
+
+                using var session = tester.CreateDivisorSession();
+                ulong[] exponents = { 6UL, 7UL, 9UL, 10UL, 12UL };
+                byte[] hits = new byte[exponents.Length];
+
+                MontgomeryDivisorData divisorData = MontgomeryDivisorData.FromModulus(7UL);
+                ulong cycle = MersenneDivisorCycles.CalculateCycleLength(7UL, divisorData);
+                session.CheckDivisor(7UL, divisorData, cycle, exponents, hits);
+
+                hits.Should().Equal(new byte[] { 1, 0, 1, 0, 1 });
         }
 
-        hits.Should().Equal(expected);
-
-        session.Dispose();
-    }
-
-    [Fact]
-    [Trait("Category", "Fast")]
-    public void ByDivisor_session_marks_mersenne_numbers_divisible_by_large_divisor2_as_composite()
-    {
-        var tester = new MersenneNumberDivisorByDivisorCpuTester();
-        tester.ConfigureFromMaxPrime(197UL);
-
-        var session = tester.CreateDivisorSession();
-        MontgomeryDivisorData divisorData = MontgomeryDivisorData.FromModulus(LargeDivisor2);
-        ulong cycle = MersenneDivisorCycles.CalculateCycleLength(LargeDivisor2, divisorData);
-
-        ulong[] exponents = { 61UL, 73UL, cycle, cycle * 2UL, cycle - 1UL };
-        byte[] hits = new byte[exponents.Length];
-        session.CheckDivisor(LargeDivisor2, divisorData, cycle, exponents, hits);
-
-        byte[] expected = new byte[exponents.Length];
-        for (int i = 0; i < exponents.Length; i++)
+        [Fact]
+        [Trait("Category", "Fast")]
+        public void ByDivisor_session_marks_mersenne_numbers_divisible_by_eleven_as_composite()
         {
-            expected[i] = exponents[i].Pow2ModWindowedCpu(divisorData.Modulus) == 1UL ? (byte)1 : (byte)0;
+                var tester = new MersenneNumberDivisorByDivisorCpuTester();
+                tester.ConfigureFromMaxPrime(61UL);
+
+                using var session = tester.CreateDivisorSession();
+                ulong[] exponents = { 10UL, 11UL, 20UL, 21UL, 30UL };
+                byte[] hits = new byte[exponents.Length];
+
+                MontgomeryDivisorData divisorData = MontgomeryDivisorData.FromModulus(11UL);
+                ulong cycle = MersenneDivisorCycles.CalculateCycleLength(11UL, divisorData);
+                session.CheckDivisor(11UL, divisorData, cycle, exponents, hits);
+
+                hits.Should().Equal(new byte[] { 1, 0, 1, 0, 1 });
         }
-
-        hits.Should().Equal(expected);
-
-        session.Dispose();
-    }
 }
+

PerfectNumbers.Core.Tests/MersenneNumberDivisorGpuTesterTests.cs

@@ -12,12 +12,12 @@ namespace PerfectNumbers.Core.Tests;
 public class MersenneNumberDivisorGpuTesterTests
 {
     [Theory]
-    [InlineData(117_652UL, 4_000_169UL, true)]
-    [InlineData(333_380UL, 4_000_561UL, true)]
-    [InlineData(59UL, 4_000_169UL, false)]
-    [InlineData(61UL, 4_000_561UL, false)]
-    [InlineData(333_379UL, 4_000_561UL, false)]
-    [InlineData(117_653UL, 4_000_169UL, false)]
+    [InlineData(33UL, 7UL, true)]
+    [InlineData(35UL, 31UL, true)]
+    [InlineData(37UL, 223UL, true)]
+    [InlineData(33UL, 13UL, false)]
+    [InlineData(35UL, 73UL, false)]
+    [InlineData(37UL, 227UL, false)]
     [Trait("Category", "Fast")]
     public void IsDivisible_returns_expected(ulong exponent, ulong divisor, bool expected)
     {
@@ -86,17 +86,6 @@ public class MersenneNumberDivisorGpuTesterTests
         exhausted.Should().BeTrue();
     }
 
-    [Fact]
-    [Trait("Category", "Fast")]
-    public void ByDivisor_gpu_prime_test_limit_stops_scan_immediately()
-    {
-        var tester = new MersenneNumberDivisorByDivisorGpuTester();
-        tester.ConfigureFromMaxPrime(43UL);
-
-        tester.IsPrime(31UL, out bool divisorsExhausted, TimeSpan.Zero).Should().BeTrue();
-        divisorsExhausted.Should().BeFalse();
-    }
-
     [Fact]
     [Trait("Category", "Fast")]
     public void ByDivisor_tester_requires_configuration()
@@ -152,41 +141,23 @@ public class MersenneNumberDivisorGpuTesterTests
 
     [Fact]
     [Trait("Category", "Fast")]
-    public void ByDivisor_session_checks_divisors_across_exponents()
+    public void ByDivisor_session_checks_divisors_across_primes()
     {
         var tester = new MersenneNumberDivisorByDivisorGpuTester();
-        tester.ConfigureFromMaxPrime(197UL);
+        tester.ConfigureFromMaxPrime(43UL);
 
-        var session = tester.CreateDivisorSession();
-        ulong[] exponents = { 59UL, 61UL, 67UL, 71UL, 73UL, 79UL, 117_652UL, 333_380UL };
-        byte[] hits = new byte[exponents.Length];
+        using var session = tester.CreateDivisorSession();
+        ulong[] primes = { 31UL, 37UL, 41UL, 43UL };
+        byte[] hits = new byte[primes.Length];
 
-        MontgomeryDivisorData divisor1 = MontgomeryDivisorData.FromModulus(4_000_169UL);
-        ulong cycle1 = MersenneDivisorCycles.CalculateCycleLength(4_000_169UL, divisor1);
-        session.CheckDivisor(4_000_169UL, divisor1, cycle1, exponents, hits);
-
-        byte[] expected1 = new byte[exponents.Length];
-        for (int i = 0; i < exponents.Length; i++)
-        {
-            expected1[i] = exponents[i].Pow2ModWindowedCpu(divisor1.Modulus) == 1UL ? (byte)1 : (byte)0;
-        }
-
-        hits.Should().Equal(expected1);
+        ulong cycle223 = MersenneDivisorCycles.CalculateCycleLength(223UL, MontgomeryDivisorData.FromModulus(223UL));
+        session.CheckDivisor(223UL, MontgomeryDivisorData.FromModulus(223UL), cycle223, primes, hits);
+        hits.Should().ContainInOrder(new byte[] { 0, 1, 0, 0 });
 
         Array.Fill(hits, (byte)0);
-        MontgomeryDivisorData divisor2 = MontgomeryDivisorData.FromModulus(4_000_561UL);
-        ulong cycle2 = MersenneDivisorCycles.CalculateCycleLength(4_000_561UL, divisor2);
-        session.CheckDivisor(4_000_561UL, divisor2, cycle2, exponents, hits);
-
-        byte[] expected2 = new byte[exponents.Length];
-        for (int i = 0; i < exponents.Length; i++)
-        {
-            expected2[i] = exponents[i].Pow2ModWindowedCpu(divisor2.Modulus) == 1UL ? (byte)1 : (byte)0;
-        }
-
-        hits.Should().Equal(expected2);
-
-        session.Dispose();
+        ulong cycle13367 = MersenneDivisorCycles.CalculateCycleLength(13367UL, MontgomeryDivisorData.FromModulus(13367UL));
+        session.CheckDivisor(13367UL, MontgomeryDivisorData.FromModulus(13367UL), cycle13367, primes, hits);
+        hits.Should().ContainInOrder(new byte[] { 0, 0, 1, 0 });
     }
 
     [Fact]
@@ -198,134 +169,122 @@ public class MersenneNumberDivisorGpuTesterTests
             GpuBatchSize = 2,
         };
 
-        tester.ConfigureFromMaxPrime(197UL);
+        tester.ConfigureFromMaxPrime(47UL);
 
-        var session = tester.CreateDivisorSession();
-        ulong[] exponents = { 59UL, 61UL, 67UL, 71UL, 73UL, 79UL, 117_652UL };
-        byte[] hits = new byte[exponents.Length];
+        using var session = tester.CreateDivisorSession();
+        ulong[] primes = { 31UL, 37UL, 41UL, 43UL, 47UL };
+        byte[] hits = new byte[primes.Length];
 
-        MontgomeryDivisorData divisor = MontgomeryDivisorData.FromModulus(4_000_169UL);
-        ulong cycle = MersenneDivisorCycles.CalculateCycleLength(4_000_169UL, divisor);
-        session.CheckDivisor(4_000_169UL, divisor, cycle, exponents, hits);
+        ulong cycle223 = MersenneDivisorCycles.CalculateCycleLength(223UL, MontgomeryDivisorData.FromModulus(223UL));
+        session.CheckDivisor(223UL, MontgomeryDivisorData.FromModulus(223UL), cycle223, primes, hits);
 
-        byte[] expected = new byte[exponents.Length];
-        for (int i = 0; i < exponents.Length; i++)
-        {
-            expected[i] = exponents[i].Pow2ModWindowedCpu(divisor.Modulus) == 1UL ? (byte)1 : (byte)0;
-        }
-
-        hits.Should().Equal(expected);
-
-        session.Dispose();
+        hits.Should().ContainInOrder(new byte[] { 0, 1, 0, 0, 0 });
     }
 
     [Fact]
     [Trait("Category", "Fast")]
-    public void ByDivisor_session_marks_mersenne_numbers_divisible_by_large_divisor1_as_composite()
+    public void ByDivisor_session_marks_mersenne_numbers_divisible_by_seven_as_composite()
     {
         var tester = new MersenneNumberDivisorByDivisorGpuTester
         {
             GpuBatchSize = 5,
         };
-        tester.ConfigureFromMaxPrime(197UL);
+        tester.ConfigureFromMaxPrime(43UL);
 
-        var session = tester.CreateDivisorSession();
-        MontgomeryDivisorData divisorData = MontgomeryDivisorData.FromModulus(4_000_169UL);
-        ulong cycle = MersenneDivisorCycles.CalculateCycleLength(4_000_169UL, divisorData);
-
-        ulong[] exponents = { 59UL, 67UL, cycle, cycle * 2UL, cycle + 1UL };
+        using var session = tester.CreateDivisorSession();
+        ulong[] exponents = { 6UL, 7UL, 9UL, 10UL, 12UL };
         byte[] hits = new byte[exponents.Length];
 
-        session.CheckDivisor(4_000_169UL, divisorData, cycle, exponents, hits);
+        MontgomeryDivisorData divisorData = MontgomeryDivisorData.FromModulus(7UL);
+        ulong cycle = MersenneDivisorCycles.CalculateCycleLength(7UL, divisorData);
+
+        session.CheckDivisor(7UL, divisorData, cycle, exponents, hits);
 
         var hostBufferField = typeof(MersenneNumberDivisorByDivisorGpuTester.DivisorScanSession)
             .GetField("_hostBuffer", BindingFlags.NonPublic | BindingFlags.Instance)!;
         ulong[] hostBuffer = (ulong[])hostBufferField.GetValue(session)!;
         Span<ulong> residues = hostBuffer.AsSpan(0, exponents.Length);
-        byte[] expected = new byte[exponents.Length];
+        byte[] computedHits = new byte[exponents.Length];
         for (int i = 0; i < residues.Length; i++)
         {
-            expected[i] = residues[i] == 1UL ? (byte)1 : (byte)0;
+            computedHits[i] = residues[i] == divisorData.MontgomeryOne ? (byte)1 : (byte)0;
         }
 
-        byte[] direct = new byte[exponents.Length];
-        for (int i = 0; i < exponents.Length; i++)
-        {
-            direct[i] = exponents[i].Pow2ModWindowedCpu(divisorData.Modulus) == 1UL ? (byte)1 : (byte)0;
-        }
-
-        expected.Should().Equal(direct);
-        hits.Should().Equal(direct);
-
-        session.Dispose();
+        computedHits.Should().Equal(new byte[] { 1, 0, 1, 0, 1 });
     }
 
     [Fact]
     [Trait("Category", "Fast")]
-    public void ByDivisor_session_marks_mersenne_numbers_divisible_by_large_divisor2_as_composite()
+    public void ByDivisor_session_marks_mersenne_numbers_divisible_by_eleven_as_composite()
     {
         var tester = new MersenneNumberDivisorByDivisorGpuTester
         {
             GpuBatchSize = 5,
         };
-        tester.ConfigureFromMaxPrime(197UL);
+        tester.ConfigureFromMaxPrime(61UL);
 
-        var session = tester.CreateDivisorSession();
-        MontgomeryDivisorData divisorData = MontgomeryDivisorData.FromModulus(4_000_561UL);
-        ulong cycle = MersenneDivisorCycles.CalculateCycleLength(4_000_561UL, divisorData);
-
-        ulong[] exponents = { 61UL, 73UL, cycle, cycle * 2UL, cycle - 1UL };
+        using var session = tester.CreateDivisorSession();
+        ulong[] exponents = { 10UL, 11UL, 20UL, 21UL, 30UL };
         byte[] hits = new byte[exponents.Length];
 
-        session.CheckDivisor(4_000_561UL, divisorData, cycle, exponents, hits);
+        MontgomeryDivisorData divisorData = MontgomeryDivisorData.FromModulus(11UL);
+        ulong cycle = MersenneDivisorCycles.CalculateCycleLength(11UL, divisorData);
+
+        session.CheckDivisor(11UL, divisorData, cycle, exponents, hits);
 
         var hostBufferField = typeof(MersenneNumberDivisorByDivisorGpuTester.DivisorScanSession)
             .GetField("_hostBuffer", BindingFlags.NonPublic | BindingFlags.Instance)!;
         ulong[] hostBuffer = (ulong[])hostBufferField.GetValue(session)!;
         Span<ulong> residues = hostBuffer.AsSpan(0, exponents.Length);
-        byte[] expected = new byte[exponents.Length];
+        byte[] computedHits = new byte[exponents.Length];
         for (int i = 0; i < residues.Length; i++)
         {
-            expected[i] = residues[i] == 1UL ? (byte)1 : (byte)0;
+            computedHits[i] = residues[i] == divisorData.MontgomeryOne ? (byte)1 : (byte)0;
         }
 
-        byte[] direct = new byte[exponents.Length];
-        for (int i = 0; i < exponents.Length; i++)
-        {
-            direct[i] = exponents[i].Pow2ModWindowedCpu(divisorData.Modulus) == 1UL ? (byte)1 : (byte)0;
-        }
-
-        expected.Should().Equal(direct);
-        hits.Should().Equal(direct);
-
-        session.Dispose();
+        computedHits.Should().Equal(new byte[] { 1, 0, 1, 0, 1 });
     }
 
     [Fact]
     [Trait("Category", "Fast")]
-    public void ByDivisor_gpu_tester_generates_cycles_for_large_divisors()
+    public void ByDivisor_gpu_tester_skips_divisors_excluded_by_small_cycle_generation()
     {
-        var tester = new MersenneNumberDivisorByDivisorGpuTester();
-        tester.ConfigureFromMaxPrime(197UL);
+        var cycles = MersenneDivisorCycles.Shared;
+        var tableField = typeof(MersenneDivisorCycles).GetField("_table", BindingFlags.NonPublic | BindingFlags.Instance)!;
+        var smallCyclesField = typeof(MersenneDivisorCycles).GetField("_smallCycles", BindingFlags.NonPublic | BindingFlags.Instance)!;
 
-        var session = tester.CreateDivisorSession();
-        ulong[] exponents = { 59UL, 67UL, 117_652UL };
-        byte[] hits = new byte[exponents.Length];
+        var originalTable = (List<(ulong Divisor, ulong Cycle)>)tableField.GetValue(cycles)!;
+        var originalSmall = (ulong[]?)smallCyclesField.GetValue(cycles);
 
-        MontgomeryDivisorData divisorData = MontgomeryDivisorData.FromModulus(4_000_169UL);
-        ulong cycle = MersenneDivisorCycles.CalculateCycleLength(4_000_169UL, divisorData);
-
-        session.CheckDivisor(4_000_169UL, divisorData, cycle, exponents, hits);
-
-        byte[] expected = new byte[exponents.Length];
-        for (int i = 0; i < exponents.Length; i++)
+        try
         {
-            expected[i] = exponents[i].Pow2ModWindowedCpu(divisorData.Modulus) == 1UL ? (byte)1 : (byte)0;
+            var patchedTable = new List<(ulong Divisor, ulong Cycle)>();
+            ulong[] patchedSmall = new ulong[PerfectNumberConstants.MaxQForDivisorCycles + 1];
+
+            MontgomeryDivisorData divisorData = MontgomeryDivisorData.FromModulus(191UL);
+            patchedSmall[191] = MersenneDivisorCycles.CalculateCycleLength(191UL, divisorData);
+
+            tableField.SetValue(cycles, patchedTable);
+            smallCyclesField.SetValue(cycles, patchedSmall);
+            DivisorCycleCache.Shared.RefreshSnapshot();
+
+            var tester = new MersenneNumberDivisorByDivisorGpuTester();
+            tester.ConfigureFromMaxPrime(19UL);
+
+            typeof(MersenneNumberDivisorByDivisorGpuTester)
+                .GetField("_divisorLimit", BindingFlags.NonPublic | BindingFlags.Instance)!
+                .SetValue(tester, 200UL);
+
+            tester.IsPrime(19UL, out bool divisorsExhausted).Should().BeTrue();
+            divisorsExhausted.Should().BeTrue();
+        }
+        finally
+        {
+            tableField.SetValue(cycles, originalTable);
+            smallCyclesField.SetValue(cycles, originalSmall);
+            DivisorCycleCache.Shared.RefreshSnapshot();
+            GpuContextPool.DisposeAll();
         }
-
-        hits.Should().Equal(expected);
-
-        session.Dispose();
     }
 
     [Fact]
@@ -337,25 +296,16 @@ public class MersenneNumberDivisorGpuTesterTests
             BatchSize = 2,
         };
 
-        tester.ConfigureFromMaxPrime(197UL);
+        tester.ConfigureFromMaxPrime(47UL);
 
-        var session = tester.CreateDivisorSession();
-        ulong[] exponents = { 59UL, 61UL, 67UL, 71UL, 73UL, 79UL, 117_652UL };
-        byte[] hits = new byte[exponents.Length];
+        using var session = tester.CreateDivisorSession();
+        ulong[] primes = { 31UL, 37UL, 41UL, 43UL, 47UL };
+        byte[] hits = new byte[primes.Length];
 
-        MontgomeryDivisorData divisor = MontgomeryDivisorData.FromModulus(4_000_169UL);
-        ulong cycle = MersenneDivisorCycles.CalculateCycleLength(4_000_169UL, divisor);
-        session.CheckDivisor(4_000_169UL, divisor, cycle, exponents, hits);
+        ulong cycle223 = MersenneDivisorCycles.CalculateCycleLength(223UL, MontgomeryDivisorData.FromModulus(223UL));
+        session.CheckDivisor(223UL, MontgomeryDivisorData.FromModulus(223UL), cycle223, primes, hits);
 
-        byte[] expected = new byte[exponents.Length];
-        for (int i = 0; i < exponents.Length; i++)
-        {
-            expected[i] = exponents[i].Pow2ModWindowedCpu(divisor.Modulus) == 1UL ? (byte)1 : (byte)0;
-        }
-
-        hits.Should().Equal(expected);
-
-        session.Dispose();
+        hits.Should().ContainInOrder(new byte[] { 0, 1, 0, 0, 0 });
     }
 }
 

PerfectNumbers.Core.Tests/PerfectNumbers.Core.Tests.csproj

@@ -22,7 +22,4 @@
   <ItemGroup>
     <ProjectReference Include="../PerfectNumbers.Core/PerfectNumbers.Core.csproj" />
   </ItemGroup>
-  <ItemGroup>
-    <Compile Remove=".temp.cs" />
-  </ItemGroup>
 </Project>

PerfectNumbers.Core.Tests/PrimeOrderCalculatorTests.cs

@@ -7,9 +7,6 @@ namespace PerfectNumbers.Core.Tests;
 
 public class PrimeOrderCalculatorTests
 {
-    // EvenPerfectBitScanner never exercises the GPU heuristic device or the small-prime guards that the legacy tests cover,
-    // so those scenarios are commented out because they are unreachable in the production pipeline.
-    /*
     public static IEnumerable<object[]> SmallPrimes()
     {
         yield return new object[] { 3UL };
@@ -48,14 +45,14 @@ public class PrimeOrderCalculatorTests
     {
         ulong expected = ComputeOrderByDoubling(prime);
 
-        ulong result = PrimeOrderCalculator.Calculate(
+        PrimeOrderCalculator.PrimeOrderResult result = PrimeOrderCalculator.Calculate(
             prime,
             previousOrder: null,
             MontgomeryDivisorData.FromModulus(prime),
-            PrimeOrderCalculator.PrimeOrderSearchConfig.HeuristicDefault,
-			      PrimeOrderCalculator.PrimeOrderHeuristicDevice.Gpu);
+            PrimeOrderCalculator.PrimeOrderSearchConfig.HeuristicDefault);
 
-        result.Should().Be(expected);
+        result.Status.Should().Be(PrimeOrderCalculator.PrimeOrderStatus.Found);
+        result.Order.Should().Be(expected);
     }
 
     [Theory]
@@ -65,21 +62,21 @@ public class PrimeOrderCalculatorTests
     {
         ulong previousOrder = ComputeOrderByDoubling(previousPrime);
 
-        ulong heuristic = PrimeOrderCalculator.Calculate(
+        PrimeOrderCalculator.PrimeOrderResult heuristic = PrimeOrderCalculator.Calculate(
             prime,
             previousOrder,
             MontgomeryDivisorData.FromModulus(prime),
-            PrimeOrderCalculator.PrimeOrderSearchConfig.HeuristicDefault,
-			      PrimeOrderCalculator.PrimeOrderHeuristicDevice.Gpu);
+            PrimeOrderCalculator.PrimeOrderSearchConfig.HeuristicDefault);
 
-        ulong strict = PrimeOrderCalculator.Calculate(
+        PrimeOrderCalculator.PrimeOrderResult strict = PrimeOrderCalculator.Calculate(
             prime,
             previousOrder: null,
             MontgomeryDivisorData.FromModulus(prime),
-            PrimeOrderCalculator.PrimeOrderSearchConfig.StrictDefault,
-			      PrimeOrderCalculator.PrimeOrderHeuristicDevice.Gpu);
+            PrimeOrderCalculator.PrimeOrderSearchConfig.StrictDefault);
 
-        heuristic.Should().Be(strict);
+        heuristic.Status.Should().Be(PrimeOrderCalculator.PrimeOrderStatus.Found);
+        strict.Status.Should().Be(PrimeOrderCalculator.PrimeOrderStatus.Found);
+        heuristic.Order.Should().Be(strict.Order);
     }
 
     [Theory]
@@ -88,14 +85,14 @@ public class PrimeOrderCalculatorTests
     [InlineData(3UL, 2UL)]
     public void Calculate_handles_trivial_primes(ulong prime, ulong expectedOrder)
     {
-        ulong result = PrimeOrderCalculator.Calculate(
+        PrimeOrderCalculator.PrimeOrderResult result = PrimeOrderCalculator.Calculate(
             prime,
             previousOrder: null,
             MontgomeryDivisorData.FromModulus(prime),
-            PrimeOrderCalculator.PrimeOrderSearchConfig.HeuristicDefault,
-			      PrimeOrderCalculator.PrimeOrderHeuristicDevice.Gpu);
+            PrimeOrderCalculator.PrimeOrderSearchConfig.HeuristicDefault);
 
-        result.Should().Be(expectedOrder);
+        result.Status.Should().Be(PrimeOrderCalculator.PrimeOrderStatus.Found);
+        result.Order.Should().Be(expectedOrder);
     }
 
     [Fact]
@@ -108,14 +105,14 @@ public class PrimeOrderCalculatorTests
             maxPowChecks: 8,
             mode: PrimeOrderCalculator.PrimeOrderMode.Heuristic);
 
-        ulong result = PrimeOrderCalculator.Calculate(
+        PrimeOrderCalculator.PrimeOrderResult result = PrimeOrderCalculator.Calculate(
             prime: 13UL,
             previousOrder: null,
             MontgomeryDivisorData.FromModulus(13UL),
-            config,
-			      PrimeOrderCalculator.PrimeOrderHeuristicDevice.Gpu);
+            config);
 
-        result.Should().Be(ComputeOrderByDoubling(13UL));
+        result.Status.Should().Be(PrimeOrderCalculator.PrimeOrderStatus.HeuristicUnresolved);
+        result.Order.Should().Be(ComputeOrderByDoubling(13UL));
     }
 
     [Fact]
@@ -129,14 +126,14 @@ public class PrimeOrderCalculatorTests
             mode: PrimeOrderCalculator.PrimeOrderMode.Heuristic);
 
         MontgomeryDivisorData divisorData = MontgomeryDivisorData.FromModulus(239UL);
-        ulong heuristic = PrimeOrderCalculator.Calculate(
+        PrimeOrderCalculator.PrimeOrderResult heuristic = PrimeOrderCalculator.Calculate(
             prime: 239UL,
             previousOrder: null,
             divisorData,
-            heuristicConfig,
-			      PrimeOrderCalculator.PrimeOrderHeuristicDevice.Gpu);
+            heuristicConfig);
 
-        heuristic.Should().Be(ComputeOrderByDoubling(239UL));
+        heuristic.Status.Should().Be(PrimeOrderCalculator.PrimeOrderStatus.HeuristicUnresolved);
+        heuristic.Order.Should().Be(ComputeOrderByDoubling(239UL));
 
         var strictConfig = new PrimeOrderCalculator.PrimeOrderSearchConfig(
             smallFactorLimit: 2,
@@ -144,14 +141,14 @@ public class PrimeOrderCalculatorTests
             maxPowChecks: 1,
             mode: PrimeOrderCalculator.PrimeOrderMode.Strict);
 
-        ulong strict = PrimeOrderCalculator.Calculate(
+        PrimeOrderCalculator.PrimeOrderResult strict = PrimeOrderCalculator.Calculate(
             prime: 239UL,
             previousOrder: null,
             divisorData,
-            strictConfig,
-			      PrimeOrderCalculator.PrimeOrderHeuristicDevice.Gpu);
+            strictConfig);
 
-        strict.Should().Be(heuristic);
+        strict.Status.Should().Be(PrimeOrderCalculator.PrimeOrderStatus.Found);
+        strict.Order.Should().Be(heuristic.Order);
     }
 
     [Fact]
@@ -160,13 +157,13 @@ public class PrimeOrderCalculatorTests
     {
         UInt128 prime = UInt128.Parse("18446744073709641691");
 
-        UInt128 result = PrimeOrderCalculator.Calculate(
-                prime,
-                previousOrder: null,
-                PrimeOrderCalculator.PrimeOrderSearchConfig.HeuristicDefault,
-                PrimeOrderCalculator.PrimeOrderHeuristicDevice.Gpu);
+        PrimeOrderCalculator.PrimeOrderResultWide result = PrimeOrderCalculator.Calculate(
+            prime,
+            previousOrder: null,
+            PrimeOrderCalculator.PrimeOrderSearchConfig.HeuristicDefault);
 
-        result.Should().Be((UInt128)1229782938247309446UL);
+        result.Status.Should().Be(PrimeOrderCalculator.PrimeOrderStatus.Found);
+        result.Order.Should().Be((UInt128)1229782938247309446UL);
     }
 
     [Fact]
@@ -200,5 +197,4 @@ public class PrimeOrderCalculatorTests
 
         return order;
     }
-*/
 }

PerfectNumbers.Core/ByDivisorDeltasDevice.cs

@@ -1,7 +0,0 @@
-namespace PerfectNumbers.Core;
-
-public enum ByDivisorDeltasDevice
-{
-    Cpu,
-    Gpu
-}

PerfectNumbers.Core/ByDivisorMontgomeryDevice.cs

@@ -1,7 +0,0 @@
-namespace PerfectNumbers.Core;
-
-public enum ByDivisorMontgomeryDevice
-{
-    Cpu = 0,
-    Gpu = 1,
-}

PerfectNumbers.Core/DeterministicRandom.cs

@@ -1,49 +0,0 @@
-using System;
-using System.Threading;
-
-namespace PerfectNumbers.Core;
-
-internal static class DeterministicRandom
-{
-    [ThreadStatic]
-    private static ulong s_threadState;
-
-    private static long s_seedCounter;
-
-    public static ulong NextUInt64()
-    {
-        ulong state = s_threadState;
-        if (state == 0UL)
-        {
-            state = InitializeThreadState();
-        }
-
-        state ^= state >> 12;
-        state ^= state << 25;
-        state ^= state >> 27;
-        s_threadState = state;
-        return state * 2685821657736338717UL;
-    }
-
-    public static UInt128 NextUInt128()
-    {
-        UInt128 high = (UInt128)NextUInt64();
-        UInt128 low = (UInt128)NextUInt64();
-        return (high << 64) | low;
-    }
-
-    private static ulong InitializeThreadState()
-    {
-        ulong seed = (ulong)Interlocked.Increment(ref s_seedCounter);
-        seed += 0x9E3779B97F4A7C15UL;
-        seed = (seed ^ (seed >> 30)) * 0xBF58476D1CE4E5B9UL;
-        seed = (seed ^ (seed >> 27)) * 0x94D049BB133111EBUL;
-        seed ^= seed >> 31;
-        if (seed == 0UL)
-        {
-            seed = 0x9E3779B97F4A7C15UL;
-        }
-
-        return seed;
-    }
-}

PerfectNumbers.Core/DivisorCycleCache.cs

@@ -63,9 +63,7 @@ public sealed class DivisorCycleCache
         int[]? rentedMissing = null;
         Span<int> missingBuffer = length <= StackBufferThreshold
             ? stackalloc int[length]
-            : length < PerfectNumberConstants.PooledArrayThreshold
-                ? new Span<int>(new int[length])
-                : new Span<int>(rentedMissing = ThreadLocalArrayPool<int>.Shared.Rent(length), 0, length);
+            : new Span<int>(rentedMissing = ArrayPool<int>.Shared.Rent(length), 0, length);
 
         int missingCount = 0;
 
@@ -86,10 +84,11 @@ public sealed class DivisorCycleCache
                 }
 
                 cycles[i] = cached;
-                continue;
             }
-
-            missingBuffer[missingCount++] = i;
+            else
+            {
+                missingBuffer[missingCount++] = i;
+            }
         }
 
         if (missingCount != 0)
@@ -103,12 +102,11 @@ public sealed class DivisorCycleCache
             {
                 ComputeCyclesCpu(divisors, cycles, missingIndices);
             }
-
         }
 
         if (rentedMissing is not null)
         {
-            ThreadLocalArrayPool<int>.Shared.Return(rentedMissing, clearArray: false);
+            ArrayPool<int>.Shared.Return(rentedMissing, clearArray: false);
         }
     }
 
@@ -130,16 +128,12 @@ public sealed class DivisorCycleCache
         ulong[]? rentedDivisors = null;
         Span<ulong> workDivisors = count <= StackBufferThreshold
             ? stackalloc ulong[count]
-            : count < PerfectNumberConstants.PooledArrayThreshold
-                ? new Span<ulong>(new ulong[count])
-                : new Span<ulong>(rentedDivisors = ThreadLocalArrayPool<ulong>.Shared.Rent(count), 0, count);
+            : new Span<ulong>(rentedDivisors = ArrayPool<ulong>.Shared.Rent(count), 0, count);
 
         ulong[]? rentedResults = null;
         Span<ulong> workResults = count <= StackBufferThreshold
             ? stackalloc ulong[count]
-            : count < PerfectNumberConstants.PooledArrayThreshold
-                ? new Span<ulong>(new ulong[count])
-                : new Span<ulong>(rentedResults = ThreadLocalArrayPool<ulong>.Shared.Rent(count), 0, count);
+            : new Span<ulong>(rentedResults = ArrayPool<ulong>.Shared.Rent(count), 0, count);
 
         for (int i = 0; i < count; i++)
         {
@@ -155,12 +149,8 @@ public sealed class DivisorCycleCache
 
         if (rentedDivisors is not null)
         {
-            ThreadLocalArrayPool<ulong>.Shared.Return(rentedDivisors, clearArray: false);
-        }
-
-        if (rentedResults is not null)
-        {
-            ThreadLocalArrayPool<ulong>.Shared.Return(rentedResults, clearArray: false);
+            ArrayPool<ulong>.Shared.Return(rentedDivisors, clearArray: false);
+            ArrayPool<ulong>.Shared.Return(rentedResults!, clearArray: false);
         }
     }
 
@@ -182,30 +172,22 @@ public sealed class DivisorCycleCache
         ulong[]? rentedPow = null;
         Span<ulong> powSpan = length <= StackBufferThreshold
             ? stackalloc ulong[length]
-            : length < PerfectNumberConstants.PooledArrayThreshold
-                ? new Span<ulong>(new ulong[length])
-                : new Span<ulong>(rentedPow = ThreadLocalArrayPool<ulong>.Shared.Rent(length), 0, length);
+            : new Span<ulong>(rentedPow = ArrayPool<ulong>.Shared.Rent(length), 0, length);
 
         ulong[]? rentedOrder = null;
         Span<ulong> orderSpan = length <= StackBufferThreshold
             ? stackalloc ulong[length]
-            : length < PerfectNumberConstants.PooledArrayThreshold
-                ? new Span<ulong>(new ulong[length])
-                : new Span<ulong>(rentedOrder = ThreadLocalArrayPool<ulong>.Shared.Rent(length), 0, length);
+            : new Span<ulong>(rentedOrder = ArrayPool<ulong>.Shared.Rent(length), 0, length);
 
         ulong[]? rentedResult = null;
         Span<ulong> resultSpan = length <= StackBufferThreshold
             ? stackalloc ulong[length]
-            : length < PerfectNumberConstants.PooledArrayThreshold
-                ? new Span<ulong>(new ulong[length])
-                : new Span<ulong>(rentedResult = ThreadLocalArrayPool<ulong>.Shared.Rent(length), 0, length);
+            : new Span<ulong>(rentedResult = ArrayPool<ulong>.Shared.Rent(length), 0, length);
 
         byte[]? rentedStatus = null;
         Span<byte> statusSpan = length <= StackBufferThreshold
             ? stackalloc byte[length]
-            : length < PerfectNumberConstants.PooledArrayThreshold
-                ? new Span<byte>(new byte[length])
-                : new Span<byte>(rentedStatus = ThreadLocalArrayPool<byte>.Shared.Rent(length), 0, length);
+            : new Span<byte>(rentedStatus = ArrayPool<byte>.Shared.Rent(length), 0, length);
 
         ref ulong divisorRef = ref MemoryMarshal.GetReference(divisors);
         divisorBuffer.View.CopyFromCPU(ref divisorRef, length);
@@ -252,22 +234,10 @@ public sealed class DivisorCycleCache
         resultSpan.CopyTo(destination);
         if (rentedPow is not null)
         {
-            ThreadLocalArrayPool<ulong>.Shared.Return(rentedPow, clearArray: false);
-        }
-
-        if (rentedOrder is not null)
-        {
-            ThreadLocalArrayPool<ulong>.Shared.Return(rentedOrder, clearArray: false);
-        }
-
-        if (rentedResult is not null)
-        {
-            ThreadLocalArrayPool<ulong>.Shared.Return(rentedResult, clearArray: false);
-        }
-
-        if (rentedStatus is not null)
-        {
-            ThreadLocalArrayPool<byte>.Shared.Return(rentedStatus, clearArray: false);
+            ArrayPool<ulong>.Shared.Return(rentedPow, clearArray: false);
+            ArrayPool<ulong>.Shared.Return(rentedOrder!, clearArray: false);
+            ArrayPool<ulong>.Shared.Return(rentedResult!, clearArray: false);
+            ArrayPool<byte>.Shared.Return(rentedStatus!, clearArray: false);
         }
 
         execution.Dispose();

PerfectNumbers.Core/ExponentRemainderStepper.cs

@@ -47,7 +47,7 @@ internal struct ExponentRemainderStepper
             // TODO: Route these state resets through the ProcessEightBitWindows helper once the scalar
             // Pow2MontgomeryMod implementation adopts it so delta stepping inherits the benchmarked
             // 2× gains for large exponents.
-            _currentMontgomery = exponent.Pow2MontgomeryModWindowedCpu(_divisor, keepMontgomery: true);
+            _currentMontgomery = exponent.Pow2MontgomeryModWindowed(_divisor, keepMontgomery: true);
             _previousExponent = exponent;
             _hasState = true;
             return ReduceCurrent();
@@ -59,7 +59,7 @@ internal struct ExponentRemainderStepper
         // cached Montgomery residue ladder highlighted in MersenneDivisorCycleLengthGpuBenchmarks.
         // TODO: Replace this per-delta powmod with the upcoming windowed ladder so incremental
         // updates stop paying the single-bit cost highlighted in GpuPow2ModBenchmarks.
-        ulong multiplier = delta.Pow2MontgomeryModWindowedCpu(_divisor, keepMontgomery: true);
+        ulong multiplier = delta.Pow2MontgomeryModWindowed(_divisor, keepMontgomery: true);
         _currentMontgomery = _currentMontgomery.MontgomeryMultiply(multiplier, _modulus, _nPrime);
         _previousExponent = exponent;
         return ReduceCurrent();
@@ -77,7 +77,7 @@ internal struct ExponentRemainderStepper
         {
             // TODO: Switch this reload to the shared windowed pow2 helper once available so CPU
             // residue checks align with the optimized ProcessEightBitWindows timings.
-            _currentMontgomery = exponent.Pow2MontgomeryModWindowedCpu(_divisor, keepMontgomery: true);
+            _currentMontgomery = exponent.Pow2MontgomeryModWindowed(_divisor, keepMontgomery: true);
             _previousExponent = exponent;
             _hasState = true;
             return _currentMontgomery == _montgomeryOne;
@@ -88,7 +88,7 @@ internal struct ExponentRemainderStepper
         // lookups so this branch also avoids recomputing powmods when the snapshot lacks the divisor.
         // TODO: Use the windowed delta pow2 helper here as well to avoid the single-bit ladder that
         // currently lags behind the benchmarked implementation for huge divisor cycles.
-        ulong multiplier = delta.Pow2MontgomeryModWindowedCpu(_divisor, keepMontgomery: true);
+        ulong multiplier = delta.Pow2MontgomeryModWindowed(_divisor, keepMontgomery: true);
         _currentMontgomery = _currentMontgomery.MontgomeryMultiply(multiplier, _modulus, _nPrime);
         _previousExponent = exponent;
         return _currentMontgomery == _montgomeryOne;
@@ -123,7 +123,7 @@ internal struct ExponentRemainderStepper
         {
             // TODO: Replace this fallback path with the upcoming ProcessEightBitWindows helper so
             // fresh Montgomery states also benefit from the faster pow2 ladder measured on CPUs.
-            _currentMontgomery = exponent.Pow2MontgomeryModWindowedCpu(_divisor, keepMontgomery: true);
+            _currentMontgomery = exponent.Pow2MontgomeryModWindowed(_divisor, keepMontgomery: true);
             _previousExponent = exponent;
             _hasState = true;
             isUnity = _currentMontgomery == _montgomeryOne;

PerfectNumbers.Core/FactorEntry.cs

@@ -1,19 +0,0 @@
-namespace PerfectNumbers.Core
-{
-    public readonly struct FactorEntry
-	{
-		public readonly ulong Value;
-		public readonly int Exponent = 1;
-
-		public FactorEntry(ulong value, int exponent)
-		{
-			Value = value;
-			Exponent = exponent;
-		}
-
-		public FactorEntry(ulong value)
-		{
-			Value = value;
-		}
-	}
-}

PerfectNumbers.Core/Gpu/GpuContextPool.cs

@@ -6,159 +6,156 @@ namespace PerfectNumbers.Core.Gpu;
 
 public static class GpuContextPool
 {
-    private static readonly bool PoolingEnabled = true;
-    // TODO: Introduce accelerator-specific warmup so pooled contexts precompile the ProcessEightBitWindows kernels and load
-    // divisor-cycle data before the first scan begins.
-    // Default device preference for generic GPU kernels (prime scans, NTT, etc.)
-    public static bool ForceCpu { get; set; } = false;
+        private static readonly bool PoolingEnabled = true;
+        // TODO: Introduce accelerator-specific warmup so pooled contexts precompile the ProcessEightBitWindows kernels and load
+        // divisor-cycle data before the first scan begins.
+        // Default device preference for generic GPU kernels (prime scans, NTT, etc.)
+        public static bool ForceCpu { get; set; } = false;
 
-    internal sealed class PooledContext
-    {
-        private bool disposed;
-
-        public Context Context { get; }
-        public Accelerator Accelerator { get; }
-        public bool IsCpu { get; }
-        public object ExecutionLock { get; } = new();
-
-        public PooledContext(bool preferCpu)
+        internal sealed class PooledContext
         {
-            Context = Context.CreateDefault();
-            Accelerator = Context.GetPreferredDevice(preferCpu).CreateAccelerator(Context);
-            IsCpu = Accelerator.AcceleratorType == AcceleratorType.CPU;
-            // NOTE: Avoid loading/compiling any kernel here to prevent implicit
-            // CL stream/queue creation during accelerator construction.
-            // Some OpenCL drivers are fragile when a queue is created immediately
-            // after device init. Real kernels will be JIT-loaded at first use.
+                private bool disposed;
+
+                public Context Context { get; }
+                public Accelerator Accelerator { get; }
+                public bool IsCpu { get; }
+                public object ExecutionLock { get; } = new();
+
+		public PooledContext(bool preferCpu)
+		{
+			Context = Context.CreateDefault();
+			Accelerator = Context.GetPreferredDevice(preferCpu).CreateAccelerator(Context);
+			IsCpu = Accelerator.AcceleratorType == AcceleratorType.CPU;
+			// NOTE: Avoid loading/compiling any kernel here to prevent implicit
+			// CL stream/queue creation during accelerator construction.
+			// Some OpenCL drivers are fragile when a queue is created immediately
+			// after device init. Real kernels will be JIT-loaded at first use.
+		}
+
+		public void Dispose()
+		{
+			if (disposed)
+			{
+				return;
+			}
+
+			// Ensure all cached GPU buffers for this accelerator are released.
+			NttGpuMath.ClearCaches(Accelerator);
+			// Release PrimeTester GPU state for this accelerator (kernel/device primes)
+			// Clear any per-accelerator cached resources to avoid releasing
+			// after the accelerator is destroyed.
+			PrimeTester.ClearGpuCaches(Accelerator);
+
+			Accelerator.Dispose();
+			Context.Dispose();
+			disposed = true;
+		}
+	}
+
+	private static readonly ConcurrentQueue<PooledContext> CpuPool = new();
+	private static readonly ConcurrentQueue<PooledContext> GpuPool = new();
+	private static readonly object CreationLock = new();
+
+	public static GpuContextLease Rent()
+	{
+		return RentPreferred(ForceCpu);
+	}
+
+	// Allows callers to choose CPU/GPU per use-case, decoupled from ForceCpu.
+	public static GpuContextLease RentPreferred(bool preferCpu)
+	{
+		if (PoolingEnabled)
+		{
+			if (preferCpu && CpuPool.TryDequeue(out var cpu))
+			{
+				return new GpuContextLease(cpu, CreationLock);
+			}
+
+			if (!preferCpu && GpuPool.TryDequeue(out var gpu))
+			{
+				return new GpuContextLease(gpu, CreationLock);
+			}
+		}
+
+		// Serialize accelerator creation to avoid concurrent CL device/stream setup across threads.
+		lock (CreationLock)
+		{
+			return new GpuContextLease(new PooledContext(preferCpu), CreationLock);
+		}
+	}
+
+	public static void DisposeAll()
+	{
+		if (!PoolingEnabled)
+		{
+			return;
+		}
+
+		while (CpuPool.TryDequeue(out var cpu))
+		{
+			cpu.Dispose();
+		}
+
+		while (GpuPool.TryDequeue(out var gpu))
+		{
+			gpu.Dispose();
+		}
+	}
+
+	private static void Return(PooledContext ctx)
+	{
+		if (PoolingEnabled)
+		{
+			ctx.Accelerator.Synchronize();
+			if (ctx.IsCpu)
+			{
+				CpuPool.Enqueue(ctx);
+			}
+			else
+			{
+				GpuPool.Enqueue(ctx);
+			}
+		}
+
+                else
+                {
+                        lock (CreationLock)
+                        {
+                                ctx.Dispose();
+                        }
+                }
         }
 
-        public void Dispose()
-        {
-            if (disposed)
-            {
-                return;
-            }
-
-            // Ensure all cached GPU buffers for this accelerator are released.
-            NttGpuMath.ClearCaches(Accelerator);
-            // Release PrimeTester GPU state for this accelerator (kernel/device primes)
-            // Clear any per-accelerator cached resources to avoid releasing
-            // after the accelerator is destroyed.
-            PrimeTester.ClearGpuCaches(Accelerator);
-
-            Accelerator.Dispose();
-            Context.Dispose();
-            disposed = true;
-        }
-    }
-
-    private static readonly ConcurrentQueue<PooledContext> CpuPool = new();
-    private static readonly ConcurrentQueue<PooledContext> GpuPool = new();
-    private static readonly object CreationLock = new();
-
-    public static GpuContextLease Rent()
-    {
-        return RentPreferred(ForceCpu);
-    }
-
-    // Allows callers to choose CPU/GPU per use-case, decoupled from ForceCpu.
-    public static GpuContextLease RentPreferred(bool preferCpu)
-    {
-        if (PoolingEnabled)
-        {
-            if (preferCpu && CpuPool.TryDequeue(out var cpu))
-            {
-                return new GpuContextLease(cpu, CreationLock);
-            }
-
-            if (!preferCpu && GpuPool.TryDequeue(out var gpu))
-            {
-                return new GpuContextLease(gpu, CreationLock);
-            }
-        }
-
-        // Serialize accelerator creation to avoid concurrent CL device/stream setup across threads.
-        lock (CreationLock)
-        {
-            return new GpuContextLease(new PooledContext(preferCpu), CreationLock);
-        }
-    }
-
-    public static void DisposeAll()
-    {
-        if (!PoolingEnabled)
-        {
-            return;
-        }
-
-        while (CpuPool.TryDequeue(out var cpu))
-        {
-            cpu.Dispose();
-        }
-
-        while (GpuPool.TryDequeue(out var gpu))
-        {
-            gpu.Dispose();
-        }
-    }
-
-    private static void Return(PooledContext ctx)
-    {
-        if (PoolingEnabled)
-        {
-            ctx.Accelerator.Synchronize();
-            if (ctx.IsCpu)
-            {
-                CpuPool.Enqueue(ctx);
-            }
-            else
-            {
-                GpuPool.Enqueue(ctx);
-            }
-        }
-
-        else
-        {
-            lock (CreationLock)
-            {
-                ctx.Dispose();
-            }
-        }
-    }
-
-    public struct GpuContextLease
+    public struct GpuContextLease : IDisposable
     {
         private readonly PooledContext _ctx;
-        private bool disposed;
+                private bool disposed;
 
-        internal GpuContextLease(PooledContext ctx, object kernelInitLock)
+                internal GpuContextLease(PooledContext ctx, object kernelInitLock)
         {
             _ctx = ctx;
-            KernelInitLock = kernelInitLock;
-        }
+                        KernelInitLock = kernelInitLock;
+                }
 
         public Context Context => _ctx.Context;
 
         public Accelerator Accelerator => _ctx.Accelerator;
 
-        public object KernelInitLock { get; }
+                public object KernelInitLock { get; }
 
-        public object ExecutionLock => _ctx.ExecutionLock;
+                public object ExecutionLock => _ctx.ExecutionLock;
 
-        internal void ReturnToPool()
-        {
-            if (disposed)
-            {
-                return;
-            }
 
-            Return(_ctx);
-            disposed = true;
-        }
+                public void Dispose()
+                {
+                        if (disposed)
+			{
+				return;
+			}
 
-        public void Dispose()
-        {
-            ReturnToPool();
+			Return(_ctx);
+			disposed = true;
         }
     }
 }
+

PerfectNumbers.Core/Gpu/GpuKernelPool.cs

@@ -37,7 +37,6 @@ public sealed class KernelContainer
             ResidueAutomatonArgs, ArrayView<int>, ArrayView1D<ulong, Stride1D.Dense>>? IncrementalOrder;
     public Action<AcceleratorStream, Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong,
         ResidueAutomatonArgs, ArrayView<int>, ArrayView1D<ulong, Stride1D.Dense>>? Pow2ModOrder;
-    public Action<AcceleratorStream, Index1D, ArrayView1D<MontgomeryDivisorData, Stride1D.Dense>, ArrayView1D<int, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>, int>? MontgomeryOddPowers;
 
     // Optional device buffer with small divisor cycles (<= 4M). Index = divisor, value = cycle length.
     public MemoryBuffer1D<ulong, Stride1D.Dense>? SmallCycles;
@@ -76,10 +75,10 @@ public readonly struct ResiduePrimeViews(
     public readonly ArrayView1D<ulong, Stride1D.Dense> LastSevenPow2 = lastSevenPow2;
 }
 
-public ref struct GpuKernelLease(GpuPrimeWorkLimiter.Lease limiter, GpuContextLease gpu, KernelContainer kernels)
+public ref struct GpuKernelLease(IDisposable limiter, GpuContextLease gpu, KernelContainer kernels)
 {
     private bool disposedValue;
-    private GpuPrimeWorkLimiter.Lease? _limiter = limiter;
+    private IDisposable? _limiter = limiter;
     private GpuContextLease? _gpu = gpu;
     private KernelContainer _kernels = kernels;
     private AcceleratorStream? _stream;
@@ -110,15 +109,15 @@ public ref struct GpuKernelLease(GpuPrimeWorkLimiter.Lease limiter, GpuContextLe
         get
         {
             var accel = Accelerator; // avoid capturing 'this' in lambda
-            // lock (_gpu!.Value.KernelInitLock)
-            // {
+            lock (_gpu!.Value.KernelInitLock)
+            {
                 return KernelContainer.InitOnce(ref _kernels.Order, () =>
                 {
                     var loaded = accel.LoadAutoGroupedStreamKernel<Index1D, ulong, ulong, ArrayView<GpuUInt128>, ArrayView<ulong>>(OrderKernelScan);
                     var kernel = KernelUtil.GetKernel(loaded);
                     return kernel.CreateLauncherDelegate<Action<AcceleratorStream, Index1D, ulong, ulong, ArrayView<GpuUInt128>, ArrayView<ulong>>>();
                 });
-            // }
+            }
         }
     }
 
@@ -127,15 +126,15 @@ public ref struct GpuKernelLease(GpuPrimeWorkLimiter.Lease limiter, GpuContextLe
         get
         {
             var accel = Accelerator;
-            // lock (_gpu!.Value.KernelInitLock)
-            // {
+            lock (_gpu!.Value.KernelInitLock)
+            {
                 return KernelContainer.InitOnce(ref _kernels.Pow2Mod, () =>
                 {
                     var loaded = accel.LoadAutoGroupedStreamKernel<Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ResidueAutomatonArgs, ArrayView<ulong>, ArrayView1D<ulong, Stride1D.Dense>, ArrayView1D<uint, Stride1D.Dense>, ArrayView1D<uint, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>>(Pow2ModKernelScan);
                     var kernel = KernelUtil.GetKernel(loaded);
                     return kernel.CreateLauncherDelegate<Action<AcceleratorStream, Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ResidueAutomatonArgs, ArrayView<ulong>, ArrayView1D<ulong, Stride1D.Dense>, ArrayView1D<uint, Stride1D.Dense>, ArrayView1D<uint, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>>>();
                 });
-            // }
+            }
         }
     }
 
@@ -144,15 +143,15 @@ public ref struct GpuKernelLease(GpuPrimeWorkLimiter.Lease limiter, GpuContextLe
         get
         {
             var accel = Accelerator;
-            // lock (_gpu!.Value.KernelInitLock)
-            // {
+            lock (_gpu!.Value.KernelInitLock)
+            {
                 return KernelContainer.InitOnce(ref _kernels.Incremental, () =>
                 {
                     var loaded = accel.LoadAutoGroupedStreamKernel<Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ulong, ulong, ulong, ulong, ArrayView<ulong>, ArrayView1D<ulong, Stride1D.Dense>>(IncrementalKernelScan);
                     var kernel = KernelUtil.GetKernel(loaded);
                     return kernel.CreateLauncherDelegate<Action<AcceleratorStream, Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ulong, ulong, ulong, ulong, ArrayView<ulong>, ArrayView1D<ulong, Stride1D.Dense>>>();
                 });
-            // }
+            }
         }
     }
 
@@ -161,15 +160,15 @@ public ref struct GpuKernelLease(GpuPrimeWorkLimiter.Lease limiter, GpuContextLe
         get
         {
             var accel = Accelerator;
-            // lock (_gpu!.Value.KernelInitLock)
-            // {
+            lock (_gpu!.Value.KernelInitLock)
+            {
                 return KernelContainer.InitOnce(ref _kernels.IncrementalOrder, () =>
                 {
                     var loaded = accel.LoadAutoGroupedStreamKernel<Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ResidueAutomatonArgs, ArrayView<int>, ArrayView1D<ulong, Stride1D.Dense>>(IncrementalOrderKernelScan);
                     var kernel = KernelUtil.GetKernel(loaded);
                     return kernel.CreateLauncherDelegate<Action<AcceleratorStream, Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ResidueAutomatonArgs, ArrayView<int>, ArrayView1D<ulong, Stride1D.Dense>>>();
                 });
-            // }
+            }
         }
     }
 
@@ -178,66 +177,15 @@ public ref struct GpuKernelLease(GpuPrimeWorkLimiter.Lease limiter, GpuContextLe
         get
         {
             var accel = Accelerator;
-            // lock (_gpu!.Value.KernelInitLock)
-            // {
+            lock (_gpu!.Value.KernelInitLock)
+            {
                 return KernelContainer.InitOnce(ref _kernels.Pow2ModOrder, () =>
                 {
                     var loaded = accel.LoadAutoGroupedStreamKernel<Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ResidueAutomatonArgs, ArrayView<int>, ArrayView1D<ulong, Stride1D.Dense>>(Pow2ModOrderKernelScan);
                     var kernel = KernelUtil.GetKernel(loaded);
                     return kernel.CreateLauncherDelegate<Action<AcceleratorStream, Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ResidueAutomatonArgs, ArrayView<int>, ArrayView1D<ulong, Stride1D.Dense>>>();
                 });
-            // }
-        }
-    }
-
-    public readonly Action<AcceleratorStream, Index1D, ArrayView1D<MontgomeryDivisorData, Stride1D.Dense>, ArrayView1D<int, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>, int> MontgomeryOddPowerKernel
-    {
-        get
-        {
-            var accel = Accelerator;
-            // lock (_gpu!.Value.KernelInitLock)
-            // {
-                return KernelContainer.InitOnce(ref _kernels.MontgomeryOddPowers, () =>
-                {
-                    var loaded = accel.LoadAutoGroupedStreamKernel<Index1D, ArrayView1D<MontgomeryDivisorData, Stride1D.Dense>, ArrayView1D<int, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>, int>(MontgomeryOddPowerKernelScan);
-                    var kernel = KernelUtil.GetKernel(loaded);
-                    return kernel.CreateLauncherDelegate<Action<AcceleratorStream, Index1D, ArrayView1D<MontgomeryDivisorData, Stride1D.Dense>, ArrayView1D<int, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>, int>>();
-                });
-            // }
-        }
-    }
-
-    private static void MontgomeryOddPowerKernelScan(
-        Index1D index,
-        ArrayView1D<MontgomeryDivisorData, Stride1D.Dense> divisors,
-        ArrayView1D<int, Stride1D.Dense> counts,
-        ArrayView1D<ulong, Stride1D.Dense> results,
-        int stride)
-    {
-        int tableIndex = index.X;
-        int baseOffset = tableIndex * stride;
-        MontgomeryDivisorData divisor = divisors[tableIndex];
-        int oddPowerCount = counts[tableIndex];
-        if (oddPowerCount <= 0)
-        {
-            return;
-        }
-
-        ulong montgomeryTwo = divisor.MontgomeryTwo;
-        results[baseOffset] = montgomeryTwo;
-        if (oddPowerCount == 1)
-        {
-            return;
-        }
-
-        ulong modulus = divisor.Modulus;
-        ulong nPrime = divisor.NPrime;
-        ulong square = divisor.MontgomeryTwoSquared;
-        ulong current = montgomeryTwo;
-        for (int i = 1; i < oddPowerCount; i++)
-        {
-            current = ULongExtensions.MontgomeryMultiply(current, square, modulus, nPrime);
-            results[baseOffset + i] = current;
+            }
         }
     }
 
@@ -396,10 +344,7 @@ public ref struct GpuKernelLease(GpuPrimeWorkLimiter.Lease limiter, GpuContextLe
                 _stream?.Dispose();
                 _stream = null;
 
-                if (_gpu.HasValue)
-                {
-                    _gpu.Value.ReturnToPool();
-                }
+                _gpu?.Dispose();
                 _gpu = null;
 
                 _executionLock = null;
@@ -732,25 +677,25 @@ public ref struct GpuKernelLease(GpuPrimeWorkLimiter.Lease limiter, GpuContextLe
         // GC.SuppressFinalize(this);
     }
 
-    public readonly struct ExecutionScope
+    public readonly struct ExecutionScope : IDisposable
     {
         private readonly object? _lock;
 
         public ExecutionScope(object? sync)
         {
             _lock = sync;
-            // if (_lock is not null)
-            // {
-            //     Monitor.Enter(_lock);
-            // }
+            if (_lock is not null)
+            {
+                Monitor.Enter(_lock);
+            }
         }
 
         public void Dispose()
         {
-            // if (_lock is not null)
-            // {
-            //     Monitor.Exit(_lock);
-            // }
+            if (_lock is not null)
+            {
+                Monitor.Exit(_lock);
+            }
         }
     }
 }
@@ -759,8 +704,6 @@ public class GpuKernelPool
 {
     private static readonly ConcurrentDictionary<Accelerator, KernelContainer> KernelCache = new(); // TODO: Replace this concurrent map with a simple accelerator-indexed lookup once kernel launchers are prewarmed during startup so we can drop the thread-safe wrapper entirely.
 
-    public static bool CpuCycleUsesGpu { get; set; }
-
     private static KernelContainer GetKernels(Accelerator accelerator)
     {
         return KernelCache.GetOrAdd(accelerator, _ => new KernelContainer());
@@ -777,8 +720,8 @@ public class GpuKernelPool
         }
 
         // Ensure single upload per accelerator even if multiple threads race here.
-        // lock (kernels) // TODO: Remove this lock by pre-uploading the immutable small-cycle snapshot during initialization; once no mutation happens at runtime, the pool must expose a simple reference without synchronization.
-        // {
+        lock (kernels) // TODO: Remove this lock by pre-uploading the immutable small-cycle snapshot during initialization; once no mutation happens at runtime, the pool must expose a simple reference without synchronization.
+        {
             if (kernels.SmallCycles is { } existing)
             {
                 return existing.View;
@@ -789,7 +732,7 @@ public class GpuKernelPool
             device.View.CopyFromCPU(host);
             kernels.SmallCycles = device;
             return device.View;
-        // }
+        }
     }
 
     public static ResiduePrimeViews EnsureSmallPrimesOnDevice(Accelerator accelerator)
@@ -803,8 +746,8 @@ public class GpuKernelPool
             return new ResiduePrimeViews(lastOne.View, lastSeven.View, lastOnePow2.View, lastSevenPow2.View);
         }
 
-        // lock (kernels) // TODO: Inline these small-prime uploads into startup initialization alongside the small-cycle snapshot so we can drop runtime locking and keep the GPU pool free of synchronization.
-        // {
+        lock (kernels) // TODO: Inline these small-prime uploads into startup initialization alongside the small-cycle snapshot so we can drop runtime locking and keep the GPU pool free of synchronization.
+        {
             if (kernels.SmallPrimesLastOne is { } existingLastOne &&
                 kernels.SmallPrimesLastSeven is { } existingLastSeven &&
                 kernels.SmallPrimesPow2LastOne is { } existingLastOnePow2 &&
@@ -833,7 +776,7 @@ public class GpuKernelPool
             kernels.SmallPrimesPow2LastSeven = deviceLastSevenPow2;
 
             return new ResiduePrimeViews(deviceLastOne.View, deviceLastSeven.View, deviceLastOnePow2.View, deviceLastSevenPow2.View);
-        // }
+        }
     }
 
     public static GpuKernelLease GetKernel(bool useGpuOrder)

PerfectNumbers.Core/Gpu/GpuPrimeWorkLimiter.cs

@@ -6,13 +6,13 @@ public static class GpuPrimeWorkLimiter
     private static readonly object _lock = new();
     private static int _currentLimit = 1;
 
-    public static Lease Acquire()
+    public static IDisposable Acquire()
     {
         // TODO: Inline the pooled limiter guard from the GpuLimiterThroughputBenchmarks so prime-sieve
         // GPU jobs stop allocating Releaser instances on every acquisition.
         var sem = _semaphore;
         sem.Wait();
-        return new Lease(sem);
+        return new Releaser(sem);
     }
 
     public static void SetLimit(int value)
@@ -36,12 +36,12 @@ public static class GpuPrimeWorkLimiter
         }
     }
 
-    public sealed class Lease
+    private sealed class Releaser : IDisposable
     {
         private readonly SemaphoreSlim _sem;
         private bool _disposed;
 
-        public Lease(SemaphoreSlim sem)
+        public Releaser(SemaphoreSlim sem)
         {
             _sem = sem;
         }

PerfectNumbers.Core/Gpu/GpuUInt128.cs

@@ -267,7 +267,7 @@ public struct GpuUInt128 : IComparable<GpuUInt128>, IEquatable<GpuUInt128>
     public void Add(in ReadOnlyGpuUInt128 other) => AddInternal(other.High, other.Low);
 
     [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    public void Add(GpuUInt128 other) => AddInternal(other.High, other.Low);
+    public void Add(in GpuUInt128 other) => AddInternal(other.High, other.Low);
 
     [MethodImpl(MethodImplOptions.AggressiveInlining)]
     private void AddInternal(ulong otherHigh, ulong otherLow)
@@ -294,17 +294,17 @@ public struct GpuUInt128 : IComparable<GpuUInt128>, IEquatable<GpuUInt128>
     public void AddMod(in ReadOnlyGpuUInt128 value, in ReadOnlyGpuUInt128 modulus) => AddModInternal(value.High, value.Low, in modulus);
 
     [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    public void AddMod(GpuUInt128 value, in ReadOnlyGpuUInt128 modulus) => AddModInternal(value.High, value.Low, in modulus);
+    public void AddMod(in GpuUInt128 value, in ReadOnlyGpuUInt128 modulus) => AddModInternal(value.High, value.Low, in modulus);
 
     [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    public void AddMod(in ReadOnlyGpuUInt128 value, GpuUInt128 modulus)
+    public void AddMod(in ReadOnlyGpuUInt128 value, in GpuUInt128 modulus)
     {
         ReadOnlyGpuUInt128 readOnlyMod = modulus.AsReadOnly();
         AddModInternal(value.High, value.Low, in readOnlyMod);
     }
 
     [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    public void AddMod(GpuUInt128 value, GpuUInt128 modulus)
+    public void AddMod(in GpuUInt128 value, in GpuUInt128 modulus)
     {
         ReadOnlyGpuUInt128 readOnlyMod = modulus.AsReadOnly();
         AddModInternal(value.High, value.Low, in readOnlyMod);
@@ -390,7 +390,7 @@ public struct GpuUInt128 : IComparable<GpuUInt128>, IEquatable<GpuUInt128>
     }
 
     [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    public void AddMod(GpuUInt128 value, ulong modulus) => AddMod(value.Low, modulus);
+    public void AddMod(in GpuUInt128 value, ulong modulus) => AddMod(value.Low, modulus);
 
     [MethodImpl(MethodImplOptions.AggressiveInlining)]
     public void AddMod(ulong value, ulong modulus)
@@ -414,7 +414,7 @@ public struct GpuUInt128 : IComparable<GpuUInt128>, IEquatable<GpuUInt128>
     public void Sub(in ReadOnlyGpuUInt128 other) => SubInternal(other.High, other.Low);
 
     [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    public void Sub(GpuUInt128 other) => SubInternal(other.High, other.Low);
+    public void Sub(in GpuUInt128 other) => SubInternal(other.High, other.Low);
 
     [MethodImpl(MethodImplOptions.AggressiveInlining)]
     private void SubInternal(ulong otherHigh, ulong otherLow)
@@ -461,7 +461,7 @@ public struct GpuUInt128 : IComparable<GpuUInt128>, IEquatable<GpuUInt128>
     public void Mul64(in ReadOnlyGpuUInt128 other) => Mul64Internal(other.High, other.Low);
 
     [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    public void Mul64(GpuUInt128 other) => Mul64Internal(other.High, other.Low);
+    public void Mul64(in GpuUInt128 other) => Mul64Internal(other.High, other.Low);
 
     [MethodImpl(MethodImplOptions.AggressiveInlining)]
     private void Mul64Internal(ulong otherHigh, ulong otherLow)
@@ -588,7 +588,7 @@ public struct GpuUInt128 : IComparable<GpuUInt128>, IEquatable<GpuUInt128>
     }
 
     [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    public static int TrailingZeroCount(GpuUInt128 value)
+    public static int TrailingZeroCount(in GpuUInt128 value)
     {
         if (value.Low != 0UL)
         {
@@ -646,17 +646,17 @@ public struct GpuUInt128 : IComparable<GpuUInt128>, IEquatable<GpuUInt128>
     public void SubMod(in ReadOnlyGpuUInt128 value, in ReadOnlyGpuUInt128 modulus) => SubModInternal(value.High, value.Low, in modulus);
 
     [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    public void SubMod(GpuUInt128 value, in ReadOnlyGpuUInt128 modulus) => SubModInternal(value.High, value.Low, in modulus);
+    public void SubMod(in GpuUInt128 value, in ReadOnlyGpuUInt128 modulus) => SubModInternal(value.High, value.Low, in modulus);
 
     [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    public void SubMod(in ReadOnlyGpuUInt128 value, GpuUInt128 modulus)
+    public void SubMod(in ReadOnlyGpuUInt128 value, in GpuUInt128 modulus)
     {
         ReadOnlyGpuUInt128 readOnlyMod = modulus.AsReadOnly();
         SubModInternal(value.High, value.Low, in readOnlyMod);
     }
 
     [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    public void SubMod(GpuUInt128 value, GpuUInt128 modulus)
+    public void SubMod(in GpuUInt128 value, in GpuUInt128 modulus)
     {
         ReadOnlyGpuUInt128 readOnlyMod = modulus.AsReadOnly();
         SubModInternal(value.High, value.Low, in readOnlyMod);
@@ -727,7 +727,7 @@ public struct GpuUInt128 : IComparable<GpuUInt128>, IEquatable<GpuUInt128>
     }
 
     [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    public void Mul(GpuUInt128 other)
+    public void Mul(in GpuUInt128 other)
     {
         ReadOnlyGpuUInt128 readOnlyThis = AsReadOnly();
         ReadOnlyGpuUInt128 readOnlyOther = other.AsReadOnly();

PerfectNumbers.Core/Gpu/GpuWorkLimiter.cs

@@ -6,14 +6,14 @@ public static class GpuWorkLimiter
     private static readonly object _lock = new();
     private static int _currentLimit = 1;
 
-    public static Lease Acquire()
+    public static IDisposable Acquire()
     {
         // TODO: Replace the per-call Releaser allocation with the pooled struct-based guard from the
         // GpuLimiterThroughputBenchmarks so limiter acquisition does not allocate when we enter the
         // GPU scanning hot path.
         var sem = _semaphore;
         sem.Wait();
-        return new Lease(sem);
+        return new Releaser(sem);
     }
 
     public static void SetLimit(int value)
@@ -40,12 +40,12 @@ public static class GpuWorkLimiter
         }
     }
 
-    public sealed class Lease
+    private sealed class Releaser : IDisposable
     {
         private readonly SemaphoreSlim _sem;
         private bool _disposed;
 
-        public Lease(SemaphoreSlim sem)
+        public Releaser(SemaphoreSlim sem)
         {
             _sem = sem;
         }

PerfectNumbers.Core/Gpu/MersenneNumberDivisorByDivisorGpuTester.cs

@@ -1,13 +1,13 @@
 using System;
 using System.Buffers;
 using System.Collections.Concurrent;
-using System.Collections.Generic;
 using System.Runtime.CompilerServices;
 using System.Runtime.InteropServices;
 using System.Numerics;
 using ILGPU;
 using ILGPU.Runtime;
 using MontgomeryDivisorData = PerfectNumbers.Core.MontgomeryDivisorData;
+using MontgomeryDivisorDataCache = PerfectNumbers.Core.MontgomeryDivisorDataCache;
 using CycleRemainderStepper = PerfectNumbers.Core.CycleRemainderStepper;
 
 namespace PerfectNumbers.Core.Gpu;
@@ -20,6 +20,7 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
     private const int RemainderSlotCount = 4;
 
     private int _gpuBatchSize = GpuConstants.ScanBatchSize;
+    private readonly object _sync = new();
     private ulong _divisorLimit;
     private bool _isConfigured;
 
@@ -35,7 +36,7 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
         _kernelCache.GetOrAdd(accelerator, acc => acc.LoadAutoGroupedStreamKernel<Index1D, ArrayView<MontgomeryDivisorData>, ArrayView<ulong>, ArrayView<byte>>(CheckKernel));
 
     private Action<Index1D, MontgomeryDivisorData, ArrayView<ulong>, ArrayView<ulong>> GetKernelByPrimeExponent(Accelerator accelerator) =>
-        _kernelExponentCache.GetOrAdd(accelerator, acc => acc.LoadAutoGroupedStreamKernel<Index1D, MontgomeryDivisorData, ArrayView<ulong>, ArrayView<ulong>>(ComputeExponentKernel));
+        _kernelExponentCache.GetOrAdd(accelerator, acc => acc.LoadAutoGroupedStreamKernel<Index1D, MontgomeryDivisorData, ArrayView<ulong>, ArrayView<ulong>>(ComputeMontgomeryExponentKernel));
 
     private Action<Index1D, ArrayView<ulong>, ArrayView<byte>, byte> GetRemainderDeltaKernel(Accelerator accelerator) =>
         _remainderDeltaKernelCache.GetOrAdd(accelerator, acc => acc.LoadAutoGroupedStreamKernel<Index1D, ArrayView<ulong>, ArrayView<byte>, byte>(ComputeRemainderDeltasKernel));
@@ -64,30 +65,34 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
 
     public void ConfigureFromMaxPrime(ulong maxPrime)
     {
-        // Configuration happens once during startup, so no synchronization is required to cache the divisor limit.
-        _divisorLimit = ComputeDivisorLimitFromMaxPrime(maxPrime);
-        _isConfigured = true;
+        lock (_sync)
+        {
+            _divisorLimit = ComputeDivisorLimitFromMaxPrime(maxPrime);
+            _isConfigured = true;
+        }
     }
 
-    public bool IsPrime(ulong prime, out bool divisorsExhausted, TimeSpan? timeLimit = null)
+    public bool IsPrime(ulong prime, out bool divisorsExhausted)
     {
-        // The CLI always configures the tester during startup, so the historical guard is commented out to avoid branching.
-        // if (!_isConfigured)
-        // {
-        //     throw new InvalidOperationException("ConfigureFromMaxPrime must be called before using the tester.");
-        // }
+        ulong allowedMax;
+        int batchCapacity;
 
-        // Configuration is immutable, so the cached values can be read without locking.
-        ulong allowedMax = ComputeAllowedMaxDivisor(prime, _divisorLimit);
-        int batchCapacity = _gpuBatchSize;
+        lock (_sync)
+        {
+            if (!_isConfigured)
+            {
+                throw new InvalidOperationException("ConfigureFromMaxPrime must be called before using the tester.");
+            }
 
-        // The GPU path always receives an allowed maximum well above the minimal divisor, so the legacy
-        // guard stays commented out to document the invariant without branching.
-        // if (allowedMax < 3UL)
-        // {
-        //     divisorsExhausted = true;
-        //     return true;
-        // }
+            allowedMax = ComputeAllowedMaxDivisor(prime, _divisorLimit);
+            batchCapacity = _gpuBatchSize;
+        }
+
+        if (allowedMax < 3UL)
+        {
+            divisorsExhausted = true;
+            return true;
+        }
 
         bool composite;
         bool coveredRange;
@@ -104,7 +109,6 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
             composite = CheckDivisors(
             prime,
             allowedMax,
-            timeLimit,
             accelerator,
             kernel,
             resources.DivisorsBuffer,
@@ -123,7 +127,6 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
             resources.CycleCandidates,
             resources.CycleLengths,
             resources.CycleCapacity,
-            resources,
             out lastProcessed,
             out coveredRange,
             out processedCount
@@ -152,14 +155,18 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
             throw new ArgumentException("allowedMaxValues span must be at least as long as primes span.", nameof(allowedMaxValues));
         }
 
-        // The GPU pipeline calls ConfigureFromMaxPrime before preparing candidates, so the legacy guard
-        // remains commented out to document the invariant without branching.
-        // if (!_isConfigured)
-        // {
-        //     throw new InvalidOperationException("ConfigureFromMaxPrime must be called before using the tester.");
-        // }
+        ulong divisorLimit;
+
+        lock (_sync)
+        {
+            if (!_isConfigured)
+            {
+                throw new InvalidOperationException("ConfigureFromMaxPrime must be called before using the tester.");
+            }
+
+            divisorLimit = _divisorLimit;
+        }
 
-        ulong divisorLimit = _divisorLimit;
         for (int index = 0; index < primes.Length; index++)
         {
             allowedMaxValues[index] = ComputeAllowedMaxDivisor(primes[index], divisorLimit);
@@ -169,7 +176,6 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
     private static bool CheckDivisors(
         ulong prime,
         ulong allowedMax,
-        TimeSpan? timeLimit,
         Accelerator accelerator,
         Action<Index1D, ArrayView<MontgomeryDivisorData>, ArrayView<ulong>, ArrayView<byte>> kernel,
         MemoryBuffer1D<MontgomeryDivisorData, Stride1D.Dense> divisorsBuffer,
@@ -188,7 +194,6 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
         ulong[] cycleCandidates,
         ulong[] cycleLengths,
         int cycleCapacity,
-        BatchResources resources,
         out ulong lastProcessed,
         out bool coveredRange,
         out ulong processedCount)
@@ -198,15 +203,6 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
         bool processedAll = false;
         processedCount = 0UL;
         lastProcessed = 0UL;
-        coveredRange = false;
-
-        PrimeTestTimeLimit limitGuard;
-        if (!PrimeTestTimeLimit.TryCreate(timeLimit, out limitGuard))
-        {
-            return false;
-        }
-
-        bool enforceLimit = limitGuard.IsActive;
 
         Span<MontgomeryDivisorData> divisorDataSpan;
         Span<ulong> divisorSpan;
@@ -216,8 +212,7 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
         ArrayView1D<ulong, Stride1D.Dense> exponentView;
         ArrayView1D<byte, Stride1D.Dense> hitsView;
 
-        // Batch resources expose a positive cycle capacity, so skip the redundant clamp here.
-        // cycleCapacity = Math.Max(1, cycleCapacity);
+        cycleCapacity = Math.Max(1, cycleCapacity);
 
         int chunkCapacity = Math.Min(batchCapacity, cycleCapacity);
         int maxThreadsPerGroup = (int)accelerator.MaxNumThreadsPerGroup;
@@ -226,22 +221,20 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
             chunkCapacity = Math.Min(chunkCapacity, maxThreadsPerGroup);
         }
 
-        // cycleCapacity already bounds chunkCapacity above zero, so the fallback would never execute.
-        // if (chunkCapacity <= 0)
-        // {
-        //     chunkCapacity = cycleCapacity;
-        // }
+        if (chunkCapacity <= 0)
+        {
+            chunkCapacity = cycleCapacity;
+        }
 
         ulong[] filteredDivisors = ArrayPool<ulong>.Shared.Rent(chunkCapacity);
         ulong[] divisorGaps = ArrayPool<ulong>.Shared.Rent(chunkCapacity);
 
         UInt128 twoP128 = (UInt128)prime << 1;
-        // Primes above one yield a non-zero 2p value, so the legacy zero guard remains commented out.
-        // if (twoP128 == UInt128.Zero)
-        // {
-        //     coveredRange = true;
-        //     return false;
-        // }
+        if (twoP128 == UInt128.Zero)
+        {
+            coveredRange = true;
+            return false;
+        }
 
         UInt128 allowedMax128 = allowedMax;
         UInt128 firstDivisor128 = twoP128 + UInt128.One;
@@ -252,13 +245,11 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
         }
 
         UInt128 maxK128 = (allowedMax128 - UInt128.One) / twoP128;
-        // Once the first divisor falls within the allowed range, at least one multiplier k is admissible,
-        // so keep the defensive zero check commented out.
-        // if (maxK128 == UInt128.Zero)
-        // {
-        //     coveredRange = true;
-        //     return false;
-        // }
+        if (maxK128 == UInt128.Zero)
+        {
+            coveredRange = true;
+            return false;
+        }
 
         ulong maxK = maxK128 > ulong.MaxValue ? ulong.MaxValue : (ulong)maxK128;
         ulong currentK = 1UL;
@@ -287,12 +278,6 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
         {
             while (currentK <= maxK)
             {
-                if (enforceLimit && limitGuard.HasExpired())
-                {
-                    processedAll = false;
-                    return false;
-                }
-
                 int chunkCount = Math.Min(chunkCapacity, batchCapacity);
                 ulong remainingK = maxK - currentK + 1UL;
                 if ((ulong)chunkCount > remainingK)
@@ -300,12 +285,10 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
                     chunkCount = (int)remainingK;
                 }
 
-                // chunkCapacity and the remaining multiplier range are strictly positive here, so the zero
-                // chunk fallback never triggers on production scans.
-                // if (chunkCount <= 0)
-                // {
-                //     break;
-                // }
+                if (chunkCount <= 0)
+                {
+                    break;
+                }
 
                 Span<ulong> candidateSpan = cycleCandidates.AsSpan(0, chunkCount);
                 Span<ulong> gapSpan = divisorGaps.AsSpan(0, chunkCount);
@@ -343,10 +326,10 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
 
                 gapView.CopyFromCPU(ref MemoryMarshal.GetReference(gapSpan), chunkCount);
 
-                ComputeRemaindersOnGpu(chunkCount, remainder10, 10, gapView, deltaView, remainder10View, remainderDeltaKernel, remainderScanKernel, resources.GetRemainderScanConfig);
-                ComputeRemaindersOnGpu(chunkCount, remainder8, 8, gapView, deltaView, remainder8View, remainderDeltaKernel, remainderScanKernel, resources.GetRemainderScanConfig);
-                ComputeRemaindersOnGpu(chunkCount, remainder5, 5, gapView, deltaView, remainder5View, remainderDeltaKernel, remainderScanKernel, resources.GetRemainderScanConfig);
-                ComputeRemaindersOnGpu(chunkCount, remainder3, 3, gapView, deltaView, remainder3View, remainderDeltaKernel, remainderScanKernel, resources.GetRemainderScanConfig);
+                ComputeRemaindersOnGpu(chunkCount, remainder10, 10, gapView, deltaView, remainder10View, remainderDeltaKernel, remainderScanKernel);
+                ComputeRemaindersOnGpu(chunkCount, remainder8, 8, gapView, deltaView, remainder8View, remainderDeltaKernel, remainderScanKernel);
+                ComputeRemaindersOnGpu(chunkCount, remainder5, 5, gapView, deltaView, remainder5View, remainderDeltaKernel, remainderScanKernel);
+                ComputeRemaindersOnGpu(chunkCount, remainder3, 3, gapView, deltaView, remainder3View, remainderDeltaKernel, remainderScanKernel);
 
                 BuildCandidateMaskOnGpu(chunkCount, remainder10View, remainder8View, remainder5View, remainder3View, lastIsSevenFlag, maskView, maskSpan, candidateMaskKernel);
 
@@ -413,8 +396,6 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
                     ulong divisorValue = filteredDivisorsSpan[i];
                     divisorSpan[i] = divisorValue;
                     exponentSpan[i] = prime;
-                    // Divisors never repeat while scanning the monotonically increasing sequence, so regenerate
-                    // the Montgomery data for each entry rather than storing cache lines that no future batch would hit.
                     divisorDataSpan[i] = MontgomeryDivisorData.FromModulus(divisorValue);
                 }
 
@@ -464,18 +445,18 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
         ArrayView1D<byte, Stride1D.Dense> deltaView,
         ArrayView1D<byte, Stride1D.Dense> remainderView,
         Action<Index1D, ArrayView<ulong>, ArrayView<byte>, byte> deltaKernel,
-        Action<KernelConfig, Index1D, ArrayView<byte>, ArrayView<byte>, byte, byte> scanKernel,
-        Func<int, KernelConfig> configProvider)
+        Action<KernelConfig, Index1D, ArrayView<byte>, ArrayView<byte>, byte, byte> scanKernel)
     {
-        // GPU batches always contain at least one divisor gap, so the empty-count guard remains commented
-        // to eliminate redundant branching.
-        // if (count <= 0)
-        // {
-        //     return;
-        // }
+        if (count <= 0)
+        {
+            return;
+        }
 
         deltaKernel(count, gapView, deltaView, modulus);
-        KernelConfig kernelConfig = configProvider(count);
+        SharedMemoryConfig sharedMemoryConfig = SharedMemoryConfig.RequestDynamic<int>(count);
+        Index1D gridDim = new Index1D(1);
+        Index1D groupDim = new Index1D(count);
+        KernelConfig kernelConfig = new KernelConfig(gridDim, groupDim, sharedMemoryConfig);
 
         scanKernel(kernelConfig, new Index1D(count), deltaView, remainderView, baseRemainder, modulus);
     }
@@ -492,13 +473,11 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
         Span<byte> destination,
         Action<Index1D, ArrayView<byte>, ArrayView<byte>, ArrayView<byte>, ArrayView<byte>, byte, ArrayView<byte>> maskKernel)
     {
-        // Candidate batches always provide at least one entry, so the zero-count guard stays commented out
-        // to avoid clearing spans unnecessarily.
-        // if (count <= 0)
-        // {
-        //     destination.Clear();
-        //     return;
-        // }
+        if (count <= 0)
+        {
+            destination.Clear();
+            return;
+        }
 
         maskKernel(count, remainder10View, remainder8View, remainder5View, remainder3View, lastIsSevenFlag, maskView);
         maskView.CopyToCPU(ref MemoryMarshal.GetReference(destination), count);
@@ -506,12 +485,10 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
 
     private static byte FetchLastRemainder(ArrayView1D<byte, Stride1D.Dense> remainderView, int count)
     {
-        // Kernels always process at least one remainder in this path, so the empty-span guard remains
-        // commented to document the invariant.
-        // if (count <= 0)
-        // {
-        //     return 0;
-        // }
+        if (count <= 0)
+        {
+            return 0;
+        }
 
         byte value = 0;
         remainderView.SubView(count - 1, 1).CopyToCPU(ref value, 1);
@@ -524,11 +501,10 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
         if (result >= modulus)
         {
             result -= modulus;
-            // The step deltas remain below the modulus, so a single subtraction suffices on this path.
-            // if (result >= modulus)
-            // {
-            //     result %= modulus;
-            // }
+            if (result >= modulus)
+            {
+                result %= modulus;
+            }
         }
 
         return (byte)result;
@@ -536,60 +512,46 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
 
     private static ulong ComputeDivisorLimitFromMaxPrime(ulong maxPrime)
     {
-        // The production configuration seeds the scan with exponents above the first odd prime, so the defensive guard remains
-        // commented to document the invariant without branching.
-        // if (maxPrime <= 1UL)
-        // {
-        //     return 0UL;
-        // }
+        if (maxPrime <= 1UL)
+        {
+            return 0UL;
+        }
 
-        // Real workloads push maxPrime well past the 64-bit shift boundary, so keep the historical implementation commented out
-        // and return ulong.MaxValue directly.
-        // if (maxPrime - 1UL >= 64UL)
-        // {
-        //     return ulong.MaxValue;
-        // }
-        //
-        // return (1UL << (int)(maxPrime - 1UL)) - 1UL;
+        if (maxPrime - 1UL >= 64UL)
+        {
+            return ulong.MaxValue;
+        }
 
-        return ulong.MaxValue;
+        return (1UL << (int)(maxPrime - 1UL)) - 1UL;
     }
 
     private static ulong ComputeAllowedMaxDivisor(ulong prime, ulong divisorLimit)
     {
-        // The by-divisor search emits candidate exponents starting at the first odd prime, so the defensive guard remains
-        // commented to document that invariant.
-        // if (prime <= 1UL)
-        // {
-        //     return 0UL;
-        // }
+        if (prime <= 1UL)
+        {
+            return 0UL;
+        }
 
-        // Production exponents are always large enough that the shift-based clamp would overflow, so skip the legacy branch and
-        // return the configured divisor limit directly.
-        // if (prime - 1UL >= 64UL)
-        // {
-        //     return divisorLimit;
-        // }
-        //
-        // return Math.Min((1UL << (int)(prime - 1UL)) - 1UL, divisorLimit);
+        if (prime - 1UL >= 64UL)
+        {
+            return divisorLimit;
+        }
 
-        return divisorLimit;
+        return Math.Min((1UL << (int)(prime - 1UL)) - 1UL, divisorLimit);
     }
 
     private static void CheckKernel(Index1D index, ArrayView<MontgomeryDivisorData> divisors, ArrayView<ulong> exponents, ArrayView<byte> hits)
     {
         MontgomeryDivisorData divisor = divisors[index];
         ulong modulus = divisor.Modulus;
-        // The divisor staging path filters out non-prime and even moduli before they reach the device, so this guard
-        // would never fire for EvenPerfectBitScanner workloads.
-        // if (modulus <= 1UL || (modulus & 1UL) == 0UL)
-        // {
-        //     hits[index] = 0;
-        //     return;
-        // }
+        if (modulus <= 1UL || (modulus & 1UL) == 0UL)
+        {
+            hits[index] = 0;
+            return;
+        }
 
         ulong exponent = exponents[index];
-        hits[index] = exponent.Pow2ModWindowedGpu(modulus) == 1UL ? (byte)1 : (byte)0;
+        hits[index] = exponent.Pow2MontgomeryModWindowed(divisor, keepMontgomery: false) == 1UL ? (byte)1 : (byte)0;
     }
 
     public sealed class DivisorScanSession : IMersenneNumberDivisorByDivisorTester.IDivisorScanSession
@@ -629,20 +591,16 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
             }
 
             int length = primes.Length;
-            // Production batches always contain at least one exponent, so the empty-span guard remains commented out to remove
-            // the redundant branch.
-            // if (length == 0)
-            // {
-            //     return;
-            // }
+            if (length == 0)
+            {
+                return;
+            }
 
-            // The divisor staging path filters non-prime and even moduli before kernels launch, so keep the defensive
-            // modulus guard commented out to avoid branching in the hot loop.
-            // if (divisorData.Modulus <= 1UL || (divisorData.Modulus & 1UL) == 0UL)
-            // {
-            //     hits.Clear();
-            //     return;
-            // }
+            if (divisorData.Modulus <= 1UL || (divisorData.Modulus & 1UL) == 0UL)
+            {
+                hits.Clear();
+                return;
+            }
 
             ArrayView1D<ulong, Stride1D.Dense> exponentsView = _exponentsBuffer.View;
             ArrayView1D<ulong, Stride1D.Dense> resultsView = _resultsBuffer.View;
@@ -801,34 +759,33 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
         mask[globalIndex] = 1;
     }
 
-    private static void ComputeExponentKernel(Index1D index, MontgomeryDivisorData divisor, ArrayView<ulong> exponents, ArrayView<ulong> results)
+    private static void ComputeMontgomeryExponentKernel(Index1D index, MontgomeryDivisorData divisor, ArrayView<ulong> exponents, ArrayView<ulong> results)
     {
         ulong modulus = divisor.Modulus;
-        // Kernel inputs are constrained to odd prime moduli, so this defensive branch would never execute on the configured
-        // scanning path.
-        // if (modulus <= 1UL || (modulus & 1UL) == 0UL)
-        // {
-        //     results[index] = 0UL;
-        //     return;
-        // }
+        if (modulus <= 1UL || (modulus & 1UL) == 0UL)
+        {
+            results[index] = 0UL;
+            return;
+        }
 
         ulong exponent = exponents[index];
 
-        results[index] = exponent.Pow2ModWindowedGpu(modulus);
+        results[index] = exponent.Pow2MontgomeryModWindowed(divisor, keepMontgomery: true);
     }
 
     public ulong DivisorLimit
     {
         get
         {
-            // The by-divisor scanner always configures the tester before exposing the limit, so the guard remains
-            // commented out to document the invariant without forcing a runtime branch.
-            // if (!_isConfigured)
-            // {
-            //     throw new InvalidOperationException("ConfigureFromMaxPrime must be called before using the tester.");
-            // }
+            lock (_sync)
+            {
+                if (!_isConfigured)
+                {
+                    throw new InvalidOperationException("ConfigureFromMaxPrime must be called before using the tester.");
+                }
 
-            return _divisorLimit;
+                return _divisorLimit;
+            }
         }
     }
 
@@ -845,17 +802,15 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
 
     private void ReturnBatchResources(Accelerator accelerator, BatchResources resources) => _resourcePools.GetOrAdd(accelerator, static _ => []).Add(resources);
 
-    private sealed class BatchResources
+    private sealed class BatchResources : IDisposable
     {
         internal BatchResources(MersenneNumberDivisorByDivisorGpuTester owner, Accelerator accelerator, int capacity)
         {
             int cycleCapacity = DivisorCycleCache.Shared.PreferredBatchSize;
-            // The shared cache always reports a positive preferred size on production workloads, so the fallback remains
-            // commented out to avoid branching during resource initialization.
-            // if (cycleCapacity <= 0)
-            // {
-            //     cycleCapacity = 1;
-            // }
+            if (cycleCapacity <= 0)
+            {
+                cycleCapacity = 1;
+            }
 
             int actualCapacity = Math.Max(capacity, cycleCapacity);
 
@@ -907,24 +862,10 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
         internal readonly Action<KernelConfig, Index1D, ArrayView<byte>, ArrayView<byte>, byte, byte> RemainderScanKernel;
         internal readonly Action<Index1D, ArrayView<byte>, ArrayView<byte>, ArrayView<byte>, ArrayView<byte>, byte, ArrayView<byte>> CandidateMaskKernel;
 
-        private readonly Dictionary<int, KernelConfig> _remainderScanConfigs = new();
-
         internal readonly int Capacity;
 
         internal readonly int CycleCapacity;
 
-        internal KernelConfig GetRemainderScanConfig(int count)
-        {
-            if (!_remainderScanConfigs.TryGetValue(count, out KernelConfig config))
-            {
-                SharedMemoryConfig sharedMemoryConfig = SharedMemoryConfig.RequestDynamic<int>(count);
-                config = new KernelConfig(new Index1D(1), new Index1D(count), sharedMemoryConfig);
-                _remainderScanConfigs[count] = config;
-            }
-
-            return config;
-        }
-
         public void Dispose()
         {
             DivisorsBuffer.Dispose();
@@ -953,33 +894,37 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
 
     public ulong GetAllowedMaxDivisor(ulong prime)
     {
-        // The CLI configures the GPU tester once at startup, so callers never request divisor limits before
-        // initialization completes.
-        // if (!_isConfigured)
-        // {
-        //     throw new InvalidOperationException("ConfigureFromMaxPrime must be called before using the tester.");
-        // }
-
-        return ComputeAllowedMaxDivisor(prime, _divisorLimit);
-    }
-
-    public IMersenneNumberDivisorByDivisorTester.IDivisorScanSession CreateDivisorSession()
-    {
-        // Divisor sessions are handed out only after configuration, so this guard would never trigger on the
-        // production path.
-        // if (!_isConfigured)
-        // {
-        //     throw new InvalidOperationException("ConfigureFromMaxPrime must be called before using the tester.");
-        // }
-
-        if (_sessionPool.TryTake(out DivisorScanSession? session))
+        lock (_sync)
         {
-            session.Reset();
-            return session;
-        }
+            if (!_isConfigured)
+            {
+                throw new InvalidOperationException("ConfigureFromMaxPrime must be called before using the tester.");
+            }
 
-        return new DivisorScanSession(this);
+            return ComputeAllowedMaxDivisor(prime, _divisorLimit);
+        }
     }
+
+    public DivisorScanSession CreateDivisorSession()
+    {
+        lock (_sync)
+        {
+            if (!_isConfigured)
+            {
+                throw new InvalidOperationException("ConfigureFromMaxPrime must be called before using the tester.");
+            }
+
+            if (_sessionPool.TryTake(out DivisorScanSession? session))
+            {
+                session.Reset();
+                return session;
+            }
+
+            return new DivisorScanSession(this);
+        }
+    }
+
+    IMersenneNumberDivisorByDivisorTester.IDivisorScanSession IMersenneNumberDivisorByDivisorTester.CreateDivisorSession() => CreateDivisorSession();
 }
 
 

PerfectNumbers.Core/Gpu/MersenneNumberDivisorCandidateGpuEvaluator.cs

@@ -1,273 +0,0 @@
-using System;
-using System.Buffers;
-using System.Runtime.InteropServices;
-using ILGPU;
-using ILGPU.Runtime;
-
-namespace PerfectNumbers.Core.Gpu;
-
-internal sealed class MersenneNumberDivisorCandidateGpuEvaluator
-{
-    private const int RemainderCount = 6;
-
-    private readonly GpuContextPool.GpuContextLease _lease;
-    private readonly Accelerator _accelerator;
-    private readonly Action<Index1D, CandidateFilterArgs, ArrayView<ulong>, ArrayView<byte>> _kernel;
-    private MemoryBuffer1D<ulong, Stride1D.Dense> _candidateBuffer;
-    private MemoryBuffer1D<byte, Stride1D.Dense> _maskBuffer;
-    private ulong[] _hostCandidates;
-    private byte[] _hostMask;
-    private int _capacity;
-    private bool _disposed;
-
-    internal MersenneNumberDivisorCandidateGpuEvaluator(int batchSize)
-    {
-        _lease = GpuContextPool.RentPreferred(preferCpu: false);
-        _accelerator = _lease.Accelerator;
-        _kernel = _accelerator.LoadAutoGroupedStreamKernel<Index1D, CandidateFilterArgs, ArrayView<ulong>, ArrayView<byte>>(EvaluateCandidatesKernel);
-        _capacity = Math.Max(1, batchSize);
-        _candidateBuffer = _accelerator.Allocate1D<ulong>(_capacity);
-        _maskBuffer = _accelerator.Allocate1D<byte>(_capacity);
-        _hostCandidates = ArrayPool<ulong>.Shared.Rent(_capacity);
-        _hostMask = ArrayPool<byte>.Shared.Rent(_capacity);
-    }
-
-    public void Dispose()
-    {
-        if (_disposed)
-        {
-            return;
-        }
-
-        _disposed = true;
-        _candidateBuffer.Dispose();
-        _maskBuffer.Dispose();
-        ArrayPool<ulong>.Shared.Return(_hostCandidates, clearArray: false);
-        ArrayPool<byte>.Shared.Return(_hostMask, clearArray: false);
-        _lease.Dispose();
-    }
-
-    internal void EvaluateCandidates(
-        ulong startDivisor,
-        ulong step,
-        ulong limit,
-        ReadOnlySpan<byte> remainders,
-        ReadOnlySpan<byte> steps,
-        bool lastIsSeven,
-        Span<ulong> candidates,
-        Span<byte> mask)
-    {
-        if (_disposed)
-        {
-            throw new ObjectDisposedException(nameof(MersenneNumberDivisorCandidateGpuEvaluator));
-        }
-
-        if (remainders.Length != RemainderCount || steps.Length != RemainderCount)
-        {
-            throw new ArgumentException("Remainder and step tables must contain exactly six entries.");
-        }
-
-        int count = candidates.Length;
-        // Candidate batches always carry at least one divisor on the by-divisor path, so the empty-count
-        // guard remains commented out to eliminate the branch.
-        // if (count == 0)
-        // {
-        //     return;
-        // }
-
-        if (count > _capacity)
-        {
-            EnsureCapacity(count);
-        }
-
-        Span<ulong> hostCandidateSpan = _hostCandidates.AsSpan(0, count);
-        Span<byte> hostMaskSpan = _hostMask.AsSpan(0, count);
-
-        ArrayView1D<ulong, Stride1D.Dense> candidateView = _candidateBuffer.View.SubView(0, count);
-        ArrayView1D<byte, Stride1D.Dense> maskView = _maskBuffer.View.SubView(0, count);
-
-        var args = new CandidateFilterArgs(
-            startDivisor,
-            step,
-            limit,
-            remainders[0],
-            remainders[1],
-            remainders[2],
-            remainders[3],
-            remainders[4],
-            remainders[5],
-            steps[0],
-            steps[1],
-            steps[2],
-            steps[3],
-            steps[4],
-            steps[5],
-            lastIsSeven ? (byte)1 : (byte)0);
-
-        _kernel(count, args, candidateView, maskView);
-
-        candidateView.CopyToCPU(ref MemoryMarshal.GetReference(hostCandidateSpan), count);
-        maskView.CopyToCPU(ref MemoryMarshal.GetReference(hostMaskSpan), count);
-
-        hostCandidateSpan.CopyTo(candidates);
-        hostMaskSpan.CopyTo(mask);
-    }
-
-    private void EnsureCapacity(int required)
-    {
-        if (required <= _capacity)
-        {
-            return;
-        }
-
-        int newCapacity = _capacity;
-        // The evaluator always starts with a positive capacity, so the zero-capacity fallback remains commented.
-        // if (newCapacity == 0)
-        // {
-        //     newCapacity = 1;
-        // }
-        while (newCapacity < required)
-        {
-            newCapacity <<= 1;
-        }
-
-        _candidateBuffer.Dispose();
-        _maskBuffer.Dispose();
-        ArrayPool<ulong>.Shared.Return(_hostCandidates, clearArray: false);
-        ArrayPool<byte>.Shared.Return(_hostMask, clearArray: false);
-
-        _capacity = newCapacity;
-        _candidateBuffer = _accelerator.Allocate1D<ulong>(_capacity);
-        _maskBuffer = _accelerator.Allocate1D<byte>(_capacity);
-        _hostCandidates = ArrayPool<ulong>.Shared.Rent(_capacity);
-        _hostMask = ArrayPool<byte>.Shared.Rent(_capacity);
-    }
-
-    private static void EvaluateCandidatesKernel(
-        Index1D index,
-        CandidateFilterArgs args,
-        ArrayView<ulong> candidates,
-        ArrayView<byte> mask)
-    {
-        int idx = index;
-        if (idx >= mask.Length)
-        {
-            return;
-        }
-
-        ulong offset = (ulong)idx;
-        ulong candidate = args.StartDivisor + args.Step * offset;
-        candidates[idx] = candidate;
-
-        if (candidate > args.Limit || (args.Step != 0UL && candidate < args.StartDivisor))
-        {
-            mask[idx] = 0;
-            return;
-        }
-
-        int remainderValue10 = ComputeRemainder(args.Remainder10, args.Step10, idx, 10);
-        int remainderValue8 = ComputeRemainder(args.Remainder8, args.Step8, idx, 8);
-        int remainderValue5 = ComputeRemainder(args.Remainder5, args.Step5, idx, 5);
-        int remainderValue3 = ComputeRemainder(args.Remainder3, args.Step3, idx, 3);
-        int remainderValue7 = ComputeRemainder(args.Remainder7, args.Step7, idx, 7);
-        int remainderValue11 = ComputeRemainder(args.Remainder11, args.Step11, idx, 11);
-
-        bool lastIsSeven = args.LastIsSevenFlag != 0;
-        bool admissible = lastIsSeven
-            ? (remainderValue10 == 3 || remainderValue10 == 7 || remainderValue10 == 9)
-            : (remainderValue10 == 1 || remainderValue10 == 3 || remainderValue10 == 9);
-
-        if (!admissible)
-        {
-            mask[idx] = 0;
-            return;
-        }
-
-        if ((remainderValue8 != 1 && remainderValue8 != 7)
-            || remainderValue3 == 0
-            || remainderValue5 == 0
-            || remainderValue7 == 0
-            || remainderValue11 == 0)
-        {
-            mask[idx] = 0;
-            return;
-        }
-
-        mask[idx] = 1;
-    }
-
-    private static int ComputeRemainder(byte remainder, byte step, int index, int modulus)
-    {
-        // Modulus lookups only cover {10, 8, 5, 3, 7, 11}, so the zero-modulus guard stays commented out.
-        // if (modulus == 0)
-        // {
-        //     return 0;
-        // }
-
-        long value = remainder;
-        value += (long)step * index;
-        int result = (int)(value % modulus);
-        if (result < 0)
-        {
-            result += modulus;
-        }
-
-        return result;
-    }
-
-    private readonly struct CandidateFilterArgs
-    {
-        public CandidateFilterArgs(
-            ulong startDivisor,
-            ulong step,
-            ulong limit,
-            byte remainder10,
-            byte remainder8,
-            byte remainder5,
-            byte remainder3,
-            byte remainder7,
-            byte remainder11,
-            byte step10,
-            byte step8,
-            byte step5,
-            byte step3,
-            byte step7,
-            byte step11,
-            byte lastIsSevenFlag)
-        {
-            StartDivisor = startDivisor;
-            Step = step;
-            Limit = limit;
-            Remainder10 = remainder10;
-            Remainder8 = remainder8;
-            Remainder5 = remainder5;
-            Remainder3 = remainder3;
-            Remainder7 = remainder7;
-            Remainder11 = remainder11;
-            Step10 = step10;
-            Step8 = step8;
-            Step5 = step5;
-            Step3 = step3;
-            Step7 = step7;
-            Step11 = step11;
-            LastIsSevenFlag = lastIsSevenFlag;
-        }
-
-        public ulong StartDivisor { get; }
-        public ulong Step { get; }
-        public ulong Limit { get; }
-        public byte Remainder10 { get; }
-        public byte Remainder8 { get; }
-        public byte Remainder5 { get; }
-        public byte Remainder3 { get; }
-        public byte Remainder7 { get; }
-        public byte Remainder11 { get; }
-        public byte Step10 { get; }
-        public byte Step8 { get; }
-        public byte Step5 { get; }
-        public byte Step3 { get; }
-        public byte Step7 { get; }
-        public byte Step11 { get; }
-        public byte LastIsSevenFlag { get; }
-    }
-}

PerfectNumbers.Core/Gpu/MersenneNumberDivisorMontgomeryGpuBuilder.cs

@@ -1,190 +0,0 @@
-using System;
-using System.Buffers;
-using System.Runtime.InteropServices;
-using PerfectNumbers.Core;
-using ILGPU;
-using ILGPU.Runtime;
-
-namespace PerfectNumbers.Core.Gpu;
-
-internal sealed class MersenneNumberDivisorMontgomeryGpuBuilder
-{
-    private readonly GpuContextPool.GpuContextLease _lease;
-    private readonly Accelerator _accelerator;
-    private readonly Action<Index1D, ArrayView<ulong>, ArrayView<MontgomeryDivisorData>> _kernel;
-    private MemoryBuffer1D<ulong, Stride1D.Dense> _modulusBuffer;
-    private MemoryBuffer1D<MontgomeryDivisorData, Stride1D.Dense> _resultBuffer;
-    private ulong[] _hostModulusBuffer;
-    private MontgomeryDivisorData[] _hostResultBuffer;
-    private int _capacity;
-    private bool _disposed;
-
-    internal MersenneNumberDivisorMontgomeryGpuBuilder(int batchSize)
-    {
-        _lease = GpuContextPool.RentPreferred(preferCpu: false);
-        _accelerator = _lease.Accelerator;
-        _kernel = _accelerator.LoadAutoGroupedStreamKernel<Index1D, ArrayView<ulong>, ArrayView<MontgomeryDivisorData>>(BuildKernel);
-        _capacity = Math.Max(1, batchSize);
-        _modulusBuffer = _accelerator.Allocate1D<ulong>(_capacity);
-        _resultBuffer = _accelerator.Allocate1D<MontgomeryDivisorData>(_capacity);
-        _hostModulusBuffer = ArrayPool<ulong>.Shared.Rent(_capacity);
-        _hostResultBuffer = ArrayPool<MontgomeryDivisorData>.Shared.Rent(_capacity);
-    }
-
-    public void Dispose()
-    {
-        if (_disposed)
-        {
-            return;
-        }
-
-        _disposed = true;
-        _modulusBuffer.Dispose();
-        _resultBuffer.Dispose();
-        ArrayPool<ulong>.Shared.Return(_hostModulusBuffer, clearArray: false);
-        ArrayPool<MontgomeryDivisorData>.Shared.Return(_hostResultBuffer, clearArray: false);
-        _lease.Dispose();
-    }
-
-    internal void Build(ReadOnlySpan<ulong> moduli, Span<MontgomeryDivisorData> destination)
-    {
-        if (_disposed)
-        {
-            throw new ObjectDisposedException(nameof(MersenneNumberDivisorMontgomeryGpuBuilder));
-        }
-
-        int length = moduli.Length;
-        // Montgomery batches always include at least one modulus on the by-divisor path, so the empty-length
-        // guard stays commented out to avoid branching.
-        // if (length == 0)
-        // {
-        //     return;
-        // }
-
-        EnsureCapacity(length);
-
-        Span<ulong> hostModulusSpan = _hostModulusBuffer.AsSpan(0, length);
-        moduli.CopyTo(hostModulusSpan);
-
-        ArrayView1D<ulong, Stride1D.Dense> modulusView = _modulusBuffer.View.SubView(0, length);
-        modulusView.CopyFromCPU(ref MemoryMarshal.GetReference(hostModulusSpan), length);
-
-        ArrayView1D<MontgomeryDivisorData, Stride1D.Dense> resultView = _resultBuffer.View.SubView(0, length);
-        _kernel(length, modulusView, resultView);
-
-        Span<MontgomeryDivisorData> hostResultSpan = _hostResultBuffer.AsSpan(0, length);
-        resultView.CopyToCPU(ref MemoryMarshal.GetReference(hostResultSpan), length);
-
-        hostResultSpan.CopyTo(destination);
-    }
-
-    private void EnsureCapacity(int required)
-    {
-        if (required <= _capacity)
-        {
-            return;
-        }
-
-        int newCapacity = _capacity;
-        // The builder always starts with a positive capacity, so the zero-capacity fallback remains commented.
-        // if (newCapacity == 0)
-        // {
-        //     newCapacity = 1;
-        // }
-        while (newCapacity < required)
-        {
-            newCapacity <<= 1;
-        }
-
-        _modulusBuffer.Dispose();
-        _resultBuffer.Dispose();
-        ArrayPool<ulong>.Shared.Return(_hostModulusBuffer, clearArray: false);
-        ArrayPool<MontgomeryDivisorData>.Shared.Return(_hostResultBuffer, clearArray: false);
-
-        _capacity = newCapacity;
-        _modulusBuffer = _accelerator.Allocate1D<ulong>(_capacity);
-        _resultBuffer = _accelerator.Allocate1D<MontgomeryDivisorData>(_capacity);
-        _hostModulusBuffer = ArrayPool<ulong>.Shared.Rent(_capacity);
-        _hostResultBuffer = ArrayPool<MontgomeryDivisorData>.Shared.Rent(_capacity);
-    }
-
-    private static void BuildKernel(Index1D index, ArrayView<ulong> moduli, ArrayView<MontgomeryDivisorData> destination)
-    {
-        int idx = index;
-        // The builder launches kernels with an extent equal to the modulus count, so the bounds guard remains commented
-        // out to avoid redundant branching on every thread.
-        // if (idx >= moduli.Length)
-        // {
-        //     return;
-        // }
-
-        ulong modulus = moduli[idx];
-        // All staged divisors are odd primes on the scanner path, so keep the defensive modulus guard commented out to
-        // remove the branch from the kernel.
-        // if (modulus <= 1UL || (modulus & 1UL) == 0UL)
-        // {
-        //     destination[idx] = default;
-        //     return;
-        // }
-
-        ulong nPrime = ComputeMontgomeryNPrime(modulus);
-        ulong montgomeryOne = ComputeMontgomeryResidue64(modulus);
-        ulong montgomeryTwo = montgomeryOne.MulMod64(2UL, modulus);
-        ulong montgomeryTwoSquared = MontgomeryMultiplyGpu(montgomeryTwo, montgomeryTwo, modulus, nPrime);
-
-        destination[idx] = new MontgomeryDivisorData(
-            modulus,
-            nPrime,
-            montgomeryOne,
-            montgomeryTwo,
-            montgomeryTwoSquared);
-    }
-
-    private static ulong ComputeMontgomeryResidue64(ulong modulus)
-    {
-        ulong remainder = ulong.MaxValue % modulus;
-        remainder++;
-        if (remainder >= modulus)
-        {
-            remainder -= modulus;
-        }
-
-        return remainder;
-    }
-
-    private static ulong MontgomeryMultiplyGpu(ulong a, ulong b, ulong modulus, ulong nPrime)
-    {
-        GpuUInt128 product = new(a);
-        product.Mul64(new GpuUInt128(b));
-        ulong tLow = product.Low;
-        ulong tHigh = product.High;
-
-        ulong m = unchecked(tLow * nPrime);
-        GpuUInt128 mProduct = new(m);
-        mProduct.Mul64(new GpuUInt128(modulus));
-        ulong mLow = mProduct.Low;
-        ulong mHigh = mProduct.High;
-
-        ulong carry = unchecked((tLow + mLow) < tLow ? 1UL : 0UL);
-        ulong result = tHigh + mHigh + carry;
-        if (result >= modulus)
-        {
-            result -= modulus;
-        }
-
-        return result;
-    }
-
-    private static ulong ComputeMontgomeryNPrime(ulong modulus)
-    {
-        ulong inv = modulus;
-        inv *= unchecked(2UL - modulus * inv);
-        inv *= unchecked(2UL - modulus * inv);
-        inv *= unchecked(2UL - modulus * inv);
-        inv *= unchecked(2UL - modulus * inv);
-        inv *= unchecked(2UL - modulus * inv);
-        inv *= unchecked(2UL - modulus * inv);
-        inv *= unchecked(2UL - modulus * inv);
-        return unchecked(0UL - inv);
-    }
-}

PerfectNumbers.Core/Gpu/MersenneNumberDivisorRemainderGpuStepper.cs

@@ -1,146 +0,0 @@
-using System;
-using System.Buffers;
-using System.Runtime.InteropServices;
-using ILGPU;
-using ILGPU.Runtime;
-
-namespace PerfectNumbers.Core.Gpu;
-
-internal sealed class MersenneNumberDivisorRemainderGpuStepper
-{
-    private const int RemainderCount = 6;
-
-    private readonly GpuContextPool.GpuContextLease _lease;
-    private readonly Accelerator _accelerator;
-    private readonly Action<Index1D, int, ArrayView<byte>, ArrayView<byte>, ArrayView<byte>> _kernel;
-    private MemoryBuffer1D<byte, Stride1D.Dense> _remainderBuffer;
-    private MemoryBuffer1D<byte, Stride1D.Dense> _stepBuffer;
-    private MemoryBuffer1D<byte, Stride1D.Dense> _resultBuffer;
-    private byte[] _hostRemainders;
-    private byte[] _hostSteps;
-    private readonly int _length;
-    private int _capacity;
-    private bool _disposed;
-
-    internal MersenneNumberDivisorRemainderGpuStepper(ReadOnlySpan<byte> steps)
-    {
-        if (steps.Length != RemainderCount)
-        {
-            throw new ArgumentException("Step table must contain exactly six entries.");
-        }
-
-        _lease = GpuContextPool.RentPreferred(preferCpu: false);
-        _accelerator = _lease.Accelerator;
-        _kernel = _accelerator.LoadAutoGroupedStreamKernel<Index1D, int, ArrayView<byte>, ArrayView<byte>, ArrayView<byte>>(AdvanceRemaindersKernel);
-
-        _length = steps.Length;
-        _capacity = Math.Max(1, _length);
-        _remainderBuffer = _accelerator.Allocate1D<byte>(_capacity);
-        _stepBuffer = _accelerator.Allocate1D<byte>(_capacity);
-        _resultBuffer = _accelerator.Allocate1D<byte>(_capacity);
-
-        _hostRemainders = ArrayPool<byte>.Shared.Rent(_capacity);
-        _hostSteps = ArrayPool<byte>.Shared.Rent(_capacity);
-
-        steps.CopyTo(_hostSteps.AsSpan(0, _length));
-
-        ArrayView1D<byte, Stride1D.Dense> stepView = _stepBuffer.View.SubView(0, _length);
-        stepView.CopyFromCPU(ref MemoryMarshal.GetReference(_hostSteps.AsSpan(0, _length)), _length);
-    }
-
-    public void Dispose()
-    {
-        if (_disposed)
-        {
-            throw new InvalidOperationException("MersenneNumberDivisorRemainderGpuStepper was disposed more than once.");
-        }
-
-        _disposed = true;
-        _remainderBuffer.Dispose();
-        _stepBuffer.Dispose();
-        _resultBuffer.Dispose();
-        ArrayPool<byte>.Shared.Return(_hostRemainders, clearArray: false);
-        ArrayPool<byte>.Shared.Return(_hostSteps, clearArray: false);
-        _lease.Dispose();
-    }
-
-    internal void Advance(int advanceCount, Span<byte> remainders)
-    {
-        if (_disposed)
-        {
-            throw new ObjectDisposedException(nameof(MersenneNumberDivisorRemainderGpuStepper));
-        }
-
-        // GPU remainder batches always advance by a positive chunk count and operate on fixed-size tables, so
-        // the defensive early-return remains commented out.
-        // if (advanceCount <= 0 || remainders.Length == 0)
-        // {
-        //     return;
-        // }
-
-        // Production code always calls the stepper with a fixed-length table, so the historical length guard stays commented.
-        // if (remainders.Length != _length)
-        // {
-        //     throw new ArgumentException("Remainder buffer length must match the configured length.");
-        // }
-
-        Span<byte> hostRemainderSpan = _hostRemainders.AsSpan(0, _length);
-        remainders.CopyTo(hostRemainderSpan);
-
-        ArrayView1D<byte, Stride1D.Dense> remainderView = _remainderBuffer.View.SubView(0, _length);
-        remainderView.CopyFromCPU(ref MemoryMarshal.GetReference(hostRemainderSpan), _length);
-
-        ArrayView1D<byte, Stride1D.Dense> stepView = _stepBuffer.View.SubView(0, _length);
-        ArrayView1D<byte, Stride1D.Dense> resultView = _resultBuffer.View.SubView(0, _length);
-
-        _kernel(_length, advanceCount, remainderView, stepView, resultView);
-
-        resultView.CopyToCPU(ref MemoryMarshal.GetReference(hostRemainderSpan), _length);
-        hostRemainderSpan.CopyTo(remainders);
-    }
-
-    private static void AdvanceRemaindersKernel(Index1D index, int advanceCount, ArrayView<byte> remainders, ArrayView<byte> steps, ArrayView<byte> results)
-    {
-        int idx = index;
-        // Kernel launches match the remainder table length, so the bounds guard stays commented to avoid redundant branching.
-        // if (idx >= remainders.Length)
-        // {
-        //     return;
-        // }
-
-        int modulus = GetModulus(idx);
-        // The modulus selector only emits positive bases (10, 8, 5, 3, 7, 11) for the configured indices, so the zero-modulus
-        // fallback remains commented out.
-        // if (modulus == 0)
-        // {
-        //     results[idx] = remainders[idx];
-        //     return;
-        // }
-
-        int remainder = remainders[idx] % modulus;
-        int step = steps[idx] % modulus;
-
-        long delta = (long)step * advanceCount;
-        int adjusted = (int)((remainder + delta) % modulus);
-        if (adjusted < 0)
-        {
-            adjusted += modulus;
-        }
-
-        results[idx] = (byte)adjusted;
-    }
-
-    private static int GetModulus(int index)
-    {
-        return index switch
-        {
-            0 => 10,
-            1 => 8,
-            2 => 5,
-            3 => 3,
-            4 => 7,
-            5 => 11,
-            _ => 0,
-        };
-    }
-}

PerfectNumbers.Core/Gpu/MersenneNumberDivisorResidueGpuEvaluator.cs

@@ -1,128 +0,0 @@
-using System;
-using System.Buffers;
-using System.Runtime.InteropServices;
-using PerfectNumbers.Core;
-using ILGPU;
-using ILGPU.Runtime;
-
-namespace PerfectNumbers.Core.Gpu;
-
-internal sealed class MersenneNumberDivisorResidueGpuEvaluator
-{
-    private readonly GpuContextPool.GpuContextLease _lease;
-    private readonly Accelerator _accelerator;
-    private readonly Action<Index1D, MontgomeryDivisorData, ArrayView<ulong>, ArrayView<ulong>> _kernel;
-    private MemoryBuffer1D<ulong, Stride1D.Dense> _exponentBuffer;
-    private MemoryBuffer1D<ulong, Stride1D.Dense> _resultBuffer;
-    private ulong[] _hostBuffer;
-    private int _capacity;
-    private bool _disposed;
-
-    internal MersenneNumberDivisorResidueGpuEvaluator(int batchSize)
-    {
-        _lease = GpuContextPool.RentPreferred(preferCpu: false);
-        _accelerator = _lease.Accelerator;
-        _kernel = _accelerator.LoadAutoGroupedStreamKernel<Index1D, MontgomeryDivisorData, ArrayView<ulong>, ArrayView<ulong>>(ComputeResiduesKernel);
-        _capacity = Math.Max(1, batchSize);
-        _exponentBuffer = _accelerator.Allocate1D<ulong>(_capacity);
-        _resultBuffer = _accelerator.Allocate1D<ulong>(_capacity);
-        _hostBuffer = ArrayPool<ulong>.Shared.Rent(_capacity);
-    }
-
-    public void Dispose()
-    {
-        if (_disposed)
-        {
-            throw new InvalidOperationException("MersenneNumberDivisorResidueGpuEvaluator was disposed more than once.");
-        }
-
-        _disposed = true;
-        _exponentBuffer.Dispose();
-        _resultBuffer.Dispose();
-        ArrayPool<ulong>.Shared.Return(_hostBuffer, clearArray: false);
-        _lease.Dispose();
-    }
-
-    internal void ComputeResidues(ReadOnlySpan<ulong> exponents, in MontgomeryDivisorData divisorData, Span<ulong> destination)
-    {
-        if (_disposed)
-        {
-            throw new ObjectDisposedException(nameof(MersenneNumberDivisorResidueGpuEvaluator));
-        }
-
-        int length = exponents.Length;
-        // Residue batches always include at least one exponent when scheduled from the by-divisor scanner, so
-        // the empty-length guard remains commented out to avoid branching.
-        // if (length == 0)
-        // {
-        //     return;
-        // }
-
-        EnsureCapacity(length);
-
-        Span<ulong> hostSpan = _hostBuffer.AsSpan(0, length);
-        exponents.CopyTo(hostSpan);
-
-        ArrayView1D<ulong, Stride1D.Dense> exponentView = _exponentBuffer.View.SubView(0, length);
-        exponentView.CopyFromCPU(ref MemoryMarshal.GetReference(hostSpan), length);
-
-        ArrayView1D<ulong, Stride1D.Dense> resultView = _resultBuffer.View.SubView(0, length);
-        _kernel(length, divisorData, exponentView, resultView);
-        resultView.CopyToCPU(ref MemoryMarshal.GetReference(hostSpan), length);
-
-        hostSpan.CopyTo(destination);
-    }
-
-    private void EnsureCapacity(int required)
-    {
-        if (required <= _capacity)
-        {
-            return;
-        }
-
-        int newCapacity = _capacity;
-        // The evaluator always starts with a positive capacity, so the zero-capacity fallback stays commented.
-        // if (newCapacity == 0)
-        // {
-        //     newCapacity = 1;
-        // }
-
-        while (newCapacity < required)
-        {
-            newCapacity <<= 1;
-        }
-
-        _exponentBuffer.Dispose();
-        _resultBuffer.Dispose();
-        ArrayPool<ulong>.Shared.Return(_hostBuffer, clearArray: false);
-
-        _capacity = newCapacity;
-        _exponentBuffer = _accelerator.Allocate1D<ulong>(_capacity);
-        _resultBuffer = _accelerator.Allocate1D<ulong>(_capacity);
-        _hostBuffer = ArrayPool<ulong>.Shared.Rent(_capacity);
-    }
-
-    private static void ComputeResiduesKernel(Index1D index, MontgomeryDivisorData divisorData, ArrayView<ulong> exponents, ArrayView<ulong> results)
-    {
-        int idx = index;
-        int length = (int)exponents.Length;
-        // Batched launches match the buffer length on production workloads, so the bounds guard stays commented out to remove
-        // the redundant comparison.
-        // if (idx >= length)
-        // {
-        //     return;
-        // }
-
-        ulong exponent = exponents[idx];
-        ulong modulus = divisorData.Modulus;
-        // Residue batches only include valid odd moduli, so the defensive check stays commented to avoid re-validating
-        // the invariant on every iteration.
-        // if (modulus <= 1UL || (modulus & 1UL) == 0UL)
-        // {
-        //     results[idx] = 0UL;
-        //     return;
-        // }
-
-        results[idx] = exponent.Pow2ModWindowedGpu(modulus);
-    }
-}

PerfectNumbers.Core/Gpu/MersenneNumberLucasLehmerGpuTester.cs

@@ -327,9 +327,8 @@ public class MersenneNumberLucasLehmerGpuTester
             // TODO: Replace StreamWriter with the pooled TextFileWriter pipeline so persisting NTT
             // parameters reuses the zero-allocation buffered writes highlighted in the scanner I/O
             // benchmarks instead of allocating a new encoder per append.
-            var writer = new StreamWriter(ParameterFilePath, append: true);
+            using var writer = new StreamWriter(ParameterFilePath, append: true);
             writer.WriteLine($"{length} {modulus.High} {modulus.Low} {primitiveRoot.High} {primitiveRoot.Low}");
-            writer.Dispose();
         }
     }
 
@@ -339,7 +338,7 @@ public class MersenneNumberLucasLehmerGpuTester
         ulong foundRoot = 0UL;
         bool success = false;
         var processorCount = Environment.ProcessorCount;
-        var cts = new System.Threading.CancellationTokenSource();
+        using var cts = new System.Threading.CancellationTokenSource();
         object sync = new();
 
         // TODO: Move this Parallel.For to the shared low-overhead work scheduler once the NTT parameter
@@ -382,13 +381,11 @@ public class MersenneNumberLucasLehmerGpuTester
         {
             modulus = new GpuUInt128(0UL, foundCandidate);
             primitiveRoot = new GpuUInt128(0UL, foundRoot);
-            cts.Dispose();
             return true;
         }
 
         modulus = new GpuUInt128(0UL, 18446744069414584321UL);
         primitiveRoot = new GpuUInt128(0UL, 7UL);
-        cts.Dispose();
         return false;
     }
 

PerfectNumbers.Core/Gpu/MontgomeryOddPowerGpu.cs

@@ -1,88 +0,0 @@
-using System;
-using System.Runtime.InteropServices;
-using ILGPU.Runtime;
-
-namespace PerfectNumbers.Core.Gpu;
-
-internal static class MontgomeryOddPowerGpu
-{
-    public static bool TryCompute(in MontgomeryDivisorData divisor, int oddPowerCount, Span<ulong> destination)
-    {
-        if (oddPowerCount <= 0)
-        {
-            return true;
-        }
-
-        Span<MontgomeryDivisorData> divisorSpan = stackalloc MontgomeryDivisorData[1];
-        divisorSpan[0] = divisor;
-        Span<int> countSpan = stackalloc int[1];
-        countSpan[0] = oddPowerCount;
-        Span<ulong> resultSpan = stackalloc ulong[PerfectNumberConstants.MaxOddPowersCount];
-
-        if (!TryComputeBatch(divisorSpan, countSpan, resultSpan, PerfectNumberConstants.MaxOddPowersCount))
-        {
-            return false;
-        }
-
-        resultSpan[..oddPowerCount].CopyTo(destination);
-        return true;
-    }
-
-    public static bool TryComputeBatch(
-        ReadOnlySpan<MontgomeryDivisorData> divisors,
-        ReadOnlySpan<int> oddPowerCounts,
-        Span<ulong> results,
-        int stride)
-    {
-        int divisorCount = divisors.Length;
-        if (divisorCount == 0)
-        {
-            return true;
-        }
-
-        if (oddPowerCounts.Length < divisorCount)
-        {
-            throw new ArgumentException(
-                "oddPowerCounts span must contain at least as many elements as divisors span.",
-                nameof(oddPowerCounts));
-        }
-
-        if (results.Length < divisorCount * stride)
-        {
-            throw new ArgumentException(
-                "The result span is shorter than the required batched odd power storage.",
-                nameof(results));
-        }
-
-        if (GpuContextPool.ForceCpu)
-        {
-            return false;
-        }
-
-        var lease = GpuKernelPool.GetKernel(useGpuOrder: true);
-        var execution = lease.EnterExecutionScope();
-        var accelerator = lease.Accelerator;
-        var stream = lease.Stream;
-
-        var divisorBuffer = accelerator.Allocate1D<MontgomeryDivisorData>(divisorCount);
-        var countBuffer = accelerator.Allocate1D<int>(divisorCount);
-        var resultBuffer = accelerator.Allocate1D<ulong>(divisorCount * stride);
-
-        divisorBuffer.View.CopyFromCPU(ref MemoryMarshal.GetReference(divisors), divisorCount);
-        countBuffer.View.CopyFromCPU(ref MemoryMarshal.GetReference(oddPowerCounts), divisorCount);
-        resultBuffer.MemSetToZero();
-
-        var kernel = lease.MontgomeryOddPowerKernel;
-        kernel(stream, divisorCount, divisorBuffer.View, countBuffer.View, resultBuffer.View, stride);
-        stream.Synchronize();
-
-        resultBuffer.View.CopyToCPU(ref MemoryMarshal.GetReference(results), divisorCount * stride);
-
-        resultBuffer.Dispose();
-        countBuffer.Dispose();
-        divisorBuffer.Dispose();
-        execution.Dispose();
-        lease.Dispose();
-        return true;
-    }
-}

PerfectNumbers.Core/Gpu/NttGpuMath.cs

@@ -47,7 +47,7 @@ public static class NttGpuMath
 
     private static readonly ConcurrentDictionary<Accelerator, Action<Index1D, ArrayView<GpuUInt128>, int>> BitReverseKernelCache = new(); // TODO: Precompute bit-reversal kernels during boot so we can remove the concurrent lookup overhead flagged in the benchmarks.
 
-    private sealed class SquareCacheEntry
+    private sealed class SquareCacheEntry : IDisposable
     {
         public MemoryBuffer1D<GpuUInt128, Stride1D.Dense> Buffer { get; }
         public GpuUInt128 Root { get; }

PerfectNumbers.Core/Gpu/PrimeOrderGpuHeuristics.cs

@@ -4,10 +4,12 @@ using System.Collections.Concurrent;
 using ILGPU;
 using ILGPU.Runtime;
 using System.Numerics;
+using System.Globalization;
 using PerfectNumbers.Core;
 using System.Runtime.InteropServices;
 using System.Runtime.CompilerServices;
 using ILGPU.Runtime.OpenCL;
+
 namespace PerfectNumbers.Core.Gpu;
 
 internal enum GpuPow2ModStatus
@@ -27,6 +29,25 @@ internal static partial class PrimeOrderGpuHeuristics
     private static readonly ConcurrentDictionary<Accelerator, OrderKernelLauncher> OrderKernelCache = new();
     private static readonly ConcurrentDictionary<Accelerator, SmallPrimeDeviceCache> SmallPrimeDeviceCaches = new();
 
+    private static void ReportPow2FailureCore(string context, string modulusDescription, GpuPow2ModStatus status)
+    {
+        string message = $"GPU pow2mod failed with status {status} for divisor {modulusDescription} during {context}.";
+        Console.Error.WriteLine(message);
+        throw new InvalidOperationException(message);
+    }
+
+    internal static void ReportPow2Failure(string context, ulong modulus, GpuPow2ModStatus status)
+    {
+        string modulusDescription = modulus.ToString(CultureInfo.InvariantCulture);
+        ReportPow2FailureCore(context, modulusDescription, status);
+    }
+
+    internal static void ReportPow2Failure(string context, UInt128 modulus, GpuPow2ModStatus status)
+    {
+        string modulusDescription = modulus.ToString(CultureInfo.InvariantCulture);
+        ReportPow2FailureCore(context, modulusDescription, status);
+    }
+
     public readonly struct OrderKernelConfig(ulong previousOrder, byte hasPreviousOrder, uint smallFactorLimit, int maxPowChecks, int mode)
     {
         public readonly ulong PreviousOrder = previousOrder;
@@ -415,9 +436,9 @@ internal static partial class PrimeOrderGpuHeuristics
         ulong? previousOrder,
         PrimeOrderCalculator.PrimeOrderSearchConfig config,
         in MontgomeryDivisorData divisorData,
-        out ulong order)
+        out PrimeOrderCalculator.PrimeOrderResult result)
     {
-        order = 0UL;
+        result = default;
 
         var lease = GpuKernelPool.GetKernel(useGpuOrder: true);
         var execution = lease.EnterExecutionScope();
@@ -497,17 +518,18 @@ internal static partial class PrimeOrderGpuHeuristics
             throw new InvalidOperationException("GPU Pollard Rho stack overflow; increase HeuristicStackCapacity.");
         }
 
+        ulong order = 0UL;
         resultBuffer.View.CopyToCPU(ref order, 1);
 
-        if (kernelStatus == PrimeOrderKernelStatus.FactoringFailure)
-        {
-            order = 0UL;
-        }
+        PrimeOrderCalculator.PrimeOrderStatus finalStatus = kernelStatus == PrimeOrderKernelStatus.Found
+            ? PrimeOrderCalculator.PrimeOrderStatus.Found
+            : PrimeOrderCalculator.PrimeOrderStatus.HeuristicUnresolved;
 
+        result = new PrimeOrderCalculator.PrimeOrderResult(finalStatus, order);
         DisposeResources();
         lease.Dispose();
 
-        return order != 0UL;
+        return true;
 
         void DisposeResources()
         {
@@ -805,15 +827,13 @@ internal static partial class PrimeOrderGpuHeuristics
         for (int i = 0; i < factorCount; i++)
         {
             ulong factor = factors[i];
-            // The factorization routine emits strictly greater-than-one factors for phi on this path, so the
-            // defensive continue would never execute.
-            // if (factor <= 1UL)
-            // {
-            //     continue;
-            // }
+            if (factor <= 1UL)
+            {
+                continue;
+            }
 
             ulong reduced = phi / factor;
-            if (reduced.Pow2ModWindowedGpu(divisor.Modulus) == 1UL)
+            if (reduced.Pow2MontgomeryModWindowed(divisor, keepMontgomery: false) == 1UL)
             {
                 return false;
             }
@@ -829,7 +849,7 @@ internal static partial class PrimeOrderGpuHeuristics
         if (residue == 1UL || residue == 7UL)
         {
             ulong half = phi >> 1;
-            if ((half).Pow2ModWindowedGpu(divisor.Modulus) == 1UL)
+            if (Pow2EqualsOneKernel(half, divisor))
             {
                 order = half;
             }
@@ -850,11 +870,10 @@ internal static partial class PrimeOrderGpuHeuristics
         {
             ulong primeFactor = factors[i];
             int exponent = exponents[i];
-            // Factorization never yields unity on the scanning path, so skip the redundant guard.
-            // if (primeFactor <= 1UL)
-            // {
-            //     continue;
-            // }
+            if (primeFactor <= 1UL)
+            {
+                continue;
+            }
 
             for (int iteration = 0; iteration < exponent; iteration++)
             {
@@ -864,7 +883,7 @@ internal static partial class PrimeOrderGpuHeuristics
                 }
 
                 ulong reduced = order / primeFactor;
-                if ((reduced).Pow2ModWindowedGpu(divisor.Modulus) == 1UL)
+                if (Pow2EqualsOneKernel(reduced, divisor))
                 {
                     order = reduced;
                     continue;
@@ -890,14 +909,12 @@ internal static partial class PrimeOrderGpuHeuristics
         ArrayView1D<ulong, Stride1D.Dense> compositeStack,
         ArrayView1D<byte, Stride1D.Dense> statusOut)
     {
-        // Order values handed to the kernel are strictly positive on the production path, so the zero guard stays commented
-        // out to remove the redundant branch.
-        // if (order == 0UL)
-        // {
-        //     return false;
-        // }
+        if (order == 0UL)
+        {
+            return false;
+        }
 
-        if ((order).Pow2ModWindowedGpu(divisor.Modulus) != 1UL)
+        if (!Pow2EqualsOneKernel(order, divisor))
         {
             return false;
         }
@@ -927,11 +944,10 @@ internal static partial class PrimeOrderGpuHeuristics
         {
             ulong primeFactor = factors[i];
             int exponent = exponents[i];
-            // Factorization never yields one on the by-divisor path, so the defensive continue stays commented out.
-            // if (primeFactor <= 1UL)
-            // {
-            //     continue;
-            // }
+            if (primeFactor <= 1UL)
+            {
+                continue;
+            }
 
             ulong reduced = order;
             for (int iteration = 0; iteration < exponent; iteration++)
@@ -942,7 +958,7 @@ internal static partial class PrimeOrderGpuHeuristics
                 }
 
                 reduced /= primeFactor;
-                if ((reduced).Pow2ModWindowedGpu(divisor.Modulus) == 1UL)
+                if (Pow2EqualsOneKernel(reduced, divisor))
                 {
                     return false;
                 }
@@ -1004,32 +1020,28 @@ internal static partial class PrimeOrderGpuHeuristics
         long candidateCapacity = candidates.Length;
         int candidateLimit = candidateCapacity < HeuristicCandidateLimit ? (int)candidateCapacity : HeuristicCandidateLimit;
         int candidateCount = BuildCandidatesKernel(order, workFactors, workExponents, factorCount, candidates, stackIndex, stackExponent, stackProduct, candidateLimit);
-        // Candidate enumeration always yields at least one order on the configured workloads, so the zero-count guard remains
-        // commented out to eliminate the redundant branch.
-        // if (candidateCount == 0)
-        // {
-        //     return false;
-        // }
+        if (candidateCount == 0)
+        {
+            return false;
+        }
 
         SortCandidatesKernel(prime, previousOrder, hasPreviousOrder != 0, candidates, candidateCount);
 
         int powBudget = maxPowChecks <= 0 ? candidateCount : maxPowChecks;
-        // The heuristics always reserve at least one check, so the zero-budget fallback remains commented out.
-        // if (powBudget <= 0)
-        // {
-        //     powBudget = candidateCount;
-        // }
+        if (powBudget <= 0)
+        {
+            powBudget = candidateCount;
+        }
 
         int powUsed = 0;
 
         for (int i = 0; i < candidateCount && powUsed < powBudget; i++)
         {
             ulong candidate = candidates[i];
-            // Candidate generation skips 1 by construction, so the defensive continue stays commented out.
-            // if (candidate <= 1UL)
-            // {
-            //     continue;
-            // }
+            if (candidate <= 1UL)
+            {
+                continue;
+            }
 
             if (powUsed >= powBudget)
             {
@@ -1037,7 +1049,7 @@ internal static partial class PrimeOrderGpuHeuristics
             }
 
             powUsed++;
-            if ((candidate).Pow2ModWindowedGpu(divisor.Modulus) != 1UL)
+            if (!Pow2EqualsOneKernel(candidate, divisor))
             {
                 continue;
             }
@@ -1317,11 +1329,10 @@ internal static partial class PrimeOrderGpuHeuristics
         {
             ulong primeFactor = factors[i];
             int exponent = exponents[i];
-            // Factorization never yields one for candidate orders, so keep the guard commented out to avoid re-checking inputs.
-            // if (primeFactor <= 1UL)
-            // {
-            //     continue;
-            // }
+            if (primeFactor <= 1UL)
+            {
+                continue;
+            }
 
             ulong reduced = candidate;
             for (int iteration = 0; iteration < exponent; iteration++)
@@ -1338,7 +1349,7 @@ internal static partial class PrimeOrderGpuHeuristics
                 }
 
                 powUsed++;
-                if ((reduced).Pow2ModWindowedGpu(divisor.Modulus) == 1UL)
+                if (Pow2EqualsOneKernel(reduced, divisor))
                 {
                     return false;
                 }
@@ -1360,21 +1371,17 @@ internal static partial class PrimeOrderGpuHeuristics
         ArrayView1D<ulong, Stride1D.Dense> compositeStack,
         ArrayView1D<byte, Stride1D.Dense> statusOut)
     {
-        // Composite candidates sent here are always greater than one during production scans, so the trivial-return branch stays
-        // commented out.
-        // if (initial <= 1UL)
-        // {
-        //     return true;
-        // }
+        if (initial <= 1UL)
+        {
+            return true;
+        }
 
         int stackCapacity = (int)compositeStack.Length;
-        // Pollard-Rho always reserves a positive stack on production workloads, so the overflow guard remains commented out
-        // to avoid branching here.
-        // if (stackCapacity <= 0)
-        // {
-        //     statusOut[0] = (byte)PrimeOrderKernelStatus.PollardOverflow;
-        //     return false;
-        // }
+        if (stackCapacity <= 0)
+        {
+            statusOut[0] = (byte)PrimeOrderKernelStatus.PollardOverflow;
+            return false;
+        }
 
         int stackTop = 0;
         compositeStack[stackTop] = initial;
@@ -1384,12 +1391,10 @@ internal static partial class PrimeOrderGpuHeuristics
         {
             stackTop--;
             ulong composite = compositeStack[stackTop];
-            // Composite values enqueued for factoring are always greater than one on production workloads, so the guard stays
-            // commented out to avoid extra branching.
-            // if (composite <= 1UL)
-            // {
-            //     continue;
-            // }
+            if (composite <= 1UL)
+            {
+                continue;
+            }
 
             if (!PeelSmallPrimesKernel(
                     composite,
@@ -1597,7 +1602,9 @@ internal static partial class PrimeOrderGpuHeuristics
             bLocal >>= BitOperations.TrailingZeroCount(bLocal);
             if (aLocal > bLocal)
             {
-                (aLocal, bLocal) = (bLocal, aLocal);
+                ulong temp = aLocal;
+                aLocal = bLocal;
+                bLocal = temp;
             }
 
             bLocal -= aLocal;
@@ -1608,10 +1615,15 @@ internal static partial class PrimeOrderGpuHeuristics
         }
     }
 
+    private static bool Pow2EqualsOneKernel(ulong exponent, in MontgomeryDivisorData divisor)
+    {
+        return exponent.Pow2MontgomeryModWindowed(divisor, keepMontgomery: false) == 1UL;
+    }
+
     private static ulong CalculateByDoublingKernel(ulong prime)
     {
         ulong order = 1UL;
-        ulong value = 2UL; // Candidate primes exceed two, so the modulo would always return two.
+        ulong value = 2UL % prime;
 
         while (value != 1UL)
         {
@@ -1629,11 +1641,10 @@ internal static partial class PrimeOrderGpuHeuristics
 
     private static GpuPow2ModStatus TryPow2ModBatchInternal(ReadOnlySpan<UInt128> exponents, UInt128 prime, Span<UInt128> remainders)
     {
-        // Residue batches are never empty on production workloads, so skip the zero-length success shortcut.
-        // if (exponents.Length == 0)
-        // {
-        //     return GpuPow2ModStatus.Success;
-        // }
+        if (exponents.Length == 0)
+        {
+            return GpuPow2ModStatus.Success;
+        }
 
         if (remainders.Length < exponents.Length)
         {
@@ -1709,30 +1720,27 @@ internal static partial class PrimeOrderGpuHeuristics
     private static void ComputePow2ModCpu(ReadOnlySpan<ulong> exponents, ulong prime, in MontgomeryDivisorData divisorData, Span<ulong> results)
     {
         int length = exponents.Length;
-        ulong modulus = divisorData.Modulus;
-        _ = prime; // modulus and divisorData both describe the same q; retain the parameter to preserve the public signature.
         for (int i = 0; i < length; i++)
         {
-            results[i] = exponents[i].Pow2ModWindowedCpu(modulus);
+            results[i] = Pow2ModCpu(exponents[i], prime, divisorData);
         }
     }
 
     private static bool TryComputeOnGpuWide(ReadOnlySpan<UInt128> exponents, UInt128 prime, Span<UInt128> results)
     {
         int length = exponents.Length;
-        // Production launches always carry work, so keep the empty-batch shortcut commented out.
-        // if (length == 0)
-        // {
-        //     return true;
-        // }
+        if (length == 0)
+        {
+            return true;
+        }
 
         GpuUInt128[]? rentedExponents = null;
         GpuUInt128[]? rentedResults = null;
         var lease = GpuKernelPool.GetKernel(useGpuOrder: true);
-        var execution = lease.EnterExecutionScope();
 
         try
         {
+            var execution = lease.EnterExecutionScope();
             Accelerator accelerator = lease.Accelerator;
             AcceleratorStream stream = lease.Stream;
             var kernel = GetPow2ModWideKernel(accelerator);
@@ -1769,6 +1777,7 @@ internal static partial class PrimeOrderGpuHeuristics
 
             exponentBuffer.Dispose();
             remainderBuffer.Dispose();
+            execution.Dispose();
             return true;
         }
         catch (Exception)
@@ -1778,8 +1787,6 @@ internal static partial class PrimeOrderGpuHeuristics
         }
         finally
         {
-            // Small batches use the stack-allocated fast path, leaving the rented arrays null. Retain the guards so pooled
-            // buffers are only returned when they were actually leased.
             if (rentedExponents is not null)
             {
                 ArrayPool<GpuUInt128>.Shared.Return(rentedExponents, clearArray: false);
@@ -1790,7 +1797,6 @@ internal static partial class PrimeOrderGpuHeuristics
                 ArrayPool<GpuUInt128>.Shared.Return(rentedResults, clearArray: false);
             }
 
-            execution.Dispose();
             lease.Dispose();
         }
     }
@@ -1804,6 +1810,16 @@ internal static partial class PrimeOrderGpuHeuristics
         }
     }
 
+    private static ulong Pow2ModCpu(ulong exponent, ulong modulus, in MontgomeryDivisorData divisorData)
+    {
+        if (modulus <= 1UL)
+        {
+            return 0UL;
+        }
+
+        return exponent.Pow2MontgomeryModWindowed(divisorData, keepMontgomery: false);
+    }
+
     private static Action<AcceleratorStream, Index1D, ArrayView1D<ulong, Stride1D.Dense>, MontgomeryDivisorData, ArrayView1D<ulong, Stride1D.Dense>> GetPow2ModKernel(Accelerator accelerator)
     {
         return Pow2ModKernelCache.GetOrAdd(accelerator, static accel =>
@@ -1816,17 +1832,8 @@ internal static partial class PrimeOrderGpuHeuristics
 
     private static void Pow2ModKernel(Index1D index, ArrayView1D<ulong, Stride1D.Dense> exponents, MontgomeryDivisorData divisor, ArrayView1D<ulong, Stride1D.Dense> remainders)
     {
-        ulong modulus = divisor.Modulus;
-        // The by-divisor heuristics call this kernel only with odd prime moduli, so the defensive branch stays commented to
-        // document the invariant without spending a branch in the kernel.
-        // if (modulus <= 1UL || (modulus & 1UL) == 0UL)
-        // {
-        //     remainders[index] = 0UL;
-        //     return;
-        // }
-
         ulong exponent = exponents[index];
-        remainders[index] = exponent.Pow2ModWindowedGpu(modulus);
+        remainders[index] = exponent.Pow2MontgomeryModWindowed(divisor, keepMontgomery: false);
     }
 
     private static Action<AcceleratorStream, Index1D, ArrayView1D<GpuUInt128, Stride1D.Dense>, GpuUInt128, ArrayView1D<GpuUInt128, Stride1D.Dense>> GetPow2ModWideKernel(Accelerator accelerator)

PerfectNumbers.Core/IMersenneNumberDivisorByDivisorTester.cs

@@ -12,16 +12,14 @@ public interface IMersenneNumberDivisorByDivisorTester
 
     ulong GetAllowedMaxDivisor(ulong prime);
 
-    bool IsPrime(ulong prime, out bool divisorsExhausted, TimeSpan? timeLimit = null);
+    bool IsPrime(ulong prime, out bool divisorsExhausted);
 
     IDivisorScanSession CreateDivisorSession();
 
     void PrepareCandidates(ReadOnlySpan<ulong> primes, Span<ulong> allowedMaxValues);
 
-    public interface IDivisorScanSession
+    public interface IDivisorScanSession : IDisposable
     {
         void CheckDivisor(ulong divisor, in MontgomeryDivisorData divisorData, ulong divisorCycle, ReadOnlySpan<ulong> primes, Span<byte> hits);
-
-        void Dispose();
     }
 }

PerfectNumbers.Core/MersenneDivisorCycles.cs

@@ -1,4 +1,3 @@
-using System;
 using System.Buffers;
 using System.Buffers.Binary;
 using System.Collections.Generic;
@@ -19,12 +18,6 @@ public class MersenneDivisorCycles
     // Lightweight read-mostly cache for small divisors (<= 4,000,000). 0 => unknown
     private ulong[]? _smallCycles;
 
-    [ThreadStatic]
-    private static Dictionary<ulong, int>? s_factorCountsPool;
-
-    [ThreadStatic]
-    private static Dictionary<ulong, int>? s_factorScratchPool;
-
     public static MersenneDivisorCycles Shared { get; } = new MersenneDivisorCycles();
 
     public MersenneDivisorCycles()
@@ -74,27 +67,23 @@ public class MersenneDivisorCycles
             return true;
         }
 
-        // Exponent and divisor validation happens during scheduling, so the defensive guard remains commented out to keep the
-        // hot path branch-free.
-        // if (exponent <= 1UL || divisor <= 1UL || (divisor & 1UL) == 0UL)
-        // {
-        //     return false;
-        // }
+        if (exponent <= 1UL || divisor <= 1UL || (divisor & 1UL) == 0UL)
+        {
+            return false;
+        }
 
-        // Production scans never revisit divisors in the small-cycle snapshot, so skip the legacy lookup here. Factoring and
-        // diagnostics can query TryMatchSmallCycleForFactoring when they need those cached values.
-        // if (divisor <= PerfectNumberConstants.MaxQForDivisorCycles)
-        // {
-        //     var small = Shared._smallCycles;
-        //     if (small is not null)
-        //     {
-        //         ulong cached = small[(int)divisor];
-        //         if (cached != 0UL)
-        //         {
-        //             return cached == exponent;
-        //         }
-        //     }
-        // }
+        if (divisor <= PerfectNumberConstants.MaxQForDivisorCycles)
+        {
+            var small = Shared._smallCycles;
+            if (small is not null)
+            {
+                ulong cached = small[(int)divisor];
+                if (cached != 0UL)
+                {
+                    return cached == exponent;
+                }
+            }
+        }
 
         if (TryCalculateCycleLengthHeuristic(divisor, divisorData, out ulong heuristicCycle) && heuristicCycle != 0UL)
         {
@@ -102,7 +91,6 @@ public class MersenneDivisorCycles
         }
 
         Dictionary<ulong, FactorCacheEntry>? cache = factorCache;
-        // Each by-divisor scan session starts with a null cache and fills it lazily, so keep the allocation guard in place.
         if (cache is null)
         {
             cache = new Dictionary<ulong, FactorCacheEntry>(8);
@@ -114,7 +102,7 @@ public class MersenneDivisorCycles
             return cycleLength == exponent;
         }
 
-        return exponent.Pow2ModWindowedCpu(divisorData.Modulus) == 1UL;
+        return exponent.Pow2MontgomeryModWindowed(divisorData, keepMontgomery: false) == 1UL;
     }
 
     public static bool CycleEqualsExponentForMersenneCandidate(ulong divisor, in MontgomeryDivisorData divisorData, ulong exponent)
@@ -124,70 +112,35 @@ public class MersenneDivisorCycles
             return false;
         }
 
-        // Candidate divisors on the production path sit well above the precomputed snapshot, so skip the small-cycle lookup
-        // here and leave factoring scenarios to call TryMatchSmallCycleForFactoring when necessary.
-        // if (divisor <= PerfectNumberConstants.MaxQForDivisorCycles)
-        // {
-        //     MersenneDivisorCycles shared = Shared;
-        //     ulong[]? small = shared._smallCycles;
-        //     if (small is not null)
-        //     {
-        //         ulong cached = small[(int)divisor];
-        //         if (cached != 0UL)
-        //         {
-        //             return cached == exponent;
-        //         }
-        //     }
-        // }
-
-        return exponent.Pow2ModWindowedCpu(divisorData.Modulus) == 1UL;
-    }
-
-    public static bool TryMatchSmallCycleForFactoring(ulong divisor, ulong exponent, out ulong cycleLength)
-    {
-        if (divisor > PerfectNumberConstants.MaxQForDivisorCycles)
+        if (divisor <= PerfectNumberConstants.MaxQForDivisorCycles)
         {
-            // Factoring helpers probe arbitrary divisors, so anything above the snapshot limit legitimately misses here.
-            cycleLength = 0UL;
-            return false;
+            MersenneDivisorCycles shared = Shared;
+            ulong[]? small = shared._smallCycles;
+            if (small is not null)
+            {
+                ulong cached = small[(int)divisor];
+                if (cached != 0UL)
+                {
+                    return cached == exponent;
+                }
+            }
         }
 
-        ulong[]? small = Shared._smallCycles;
-        if (small is null)
-        {
-            // Snapshot recycling during diagnostics can temporarily clear the table; surface the miss so callers recompute.
-            cycleLength = 0UL;
-            return false;
-        }
-
-        ulong cached = small[(int)divisor];
-        if (cached == 0UL)
-        {
-            // Only a subset of small divisors carry cached cycles; signal the miss so factoring falls back to the GPU helper.
-            cycleLength = 0UL;
-            return false;
-        }
-
-        cycleLength = cached;
-        return cached == exponent;
+        return exponent.Pow2MontgomeryModWindowed(divisorData, keepMontgomery: false) == 1UL;
     }
 
     private static bool IsValidMersenneDivisorCandidate(ulong divisor, ulong exponent)
     {
-        // Production workloads pre-filter invalid inputs, so skip the defensive guards here. Factoring and diagnostics should
-        // validate inputs before calling into the candidate path.
-        // if (exponent <= 1UL || divisor <= 1UL || (divisor & 1UL) == 0UL)
-        // {
-        //     return false;
-        // }
+        if (exponent <= 1UL || divisor <= 1UL || (divisor & 1UL) == 0UL)
+        {
+            return false;
+        }
 
         UInt128 step = (UInt128)exponent << 1;
-        // The overflow bail-out is handled by factoring-specific call sites; keep the historical branch commented to document the
-        // legacy behavior.
-        // if (step == UInt128.Zero)
-        // {
-        //     return false;
-        // }
+        if (step == UInt128.Zero)
+        {
+            return false;
+        }
 
         UInt128 adjusted = (UInt128)divisor - UInt128.One;
         return adjusted % step == UInt128.Zero;
@@ -196,7 +149,7 @@ public class MersenneDivisorCycles
     public static IEnumerable<(ulong divisor, ulong cycleLength)> EnumerateStream(Stream compressor)
     {
         // Binary pairs: (ulong divisor, ulong cycle)
-        var reader = new BinaryReader(compressor, Encoding.UTF8, leaveOpen: true);
+        using var reader = new BinaryReader(compressor, Encoding.UTF8, leaveOpen: true);
         while (true)
         {
             ulong d, c;
@@ -207,7 +160,6 @@ public class MersenneDivisorCycles
             }
             catch (EndOfStreamException)
             {
-                reader.Dispose();
                 yield break;
             }
 
@@ -282,14 +234,14 @@ public class MersenneDivisorCycles
             return (UInt128)cycle;
         }
 
-        UInt128 wideOrder = PrimeOrderCalculator.Calculate(
+        PrimeOrderCalculator.PrimeOrderResultWide wideResult = PrimeOrderCalculator.Calculate(
             divisor,
             previousOrder: null,
             PrimeOrderCalculator.PrimeOrderSearchConfig.HeuristicDefault,
             PrimeOrderCalculator.PrimeOrderHeuristicDevice.Cpu);
-        if (wideOrder != UInt128.Zero)
+        if (wideResult.Order != UInt128.Zero)
         {
-            return wideOrder;
+            return wideResult.Order;
         }
 
         // Otherwise, find order of 2 mod divisor
@@ -313,93 +265,72 @@ public class MersenneDivisorCycles
     }
 
     public static bool TryCalculateCycleLengthForExponent(
-        ulong divisor,        
+        ulong divisor,
         ulong exponent,
         in MontgomeryDivisorData divisorData,
         Dictionary<ulong, FactorCacheEntry>? factorCache,
-        out ulong cycleLength)
+        out ulong cycleLength,
+        bool allowGpuPow2 = false)
     {
         cycleLength = 0UL;
 
-        // The by-divisor scanner only supplies odd candidates of the form 2*k*exponent + 1,
-        // so this power-of-two fast-path never triggers on that call path.
-        // The historical fast-path is left here for reference but commented out because it is unreachable
-        // on the EvenPerfectBitScanner pipeline.
-        // if ((divisor & (divisor - 1UL)) == 0UL)
-        // {
-        //     cycleLength = 1UL;
-        //     return true;
-        // }
+        if ((divisor & (divisor - 1UL)) == 0UL)
+        {
+            cycleLength = 1UL;
+            return true;
+        }
 
-        // EvenPerfectBitScanner filters exponents <= 3 before invoking this helper and each
-        // divisor candidate is at least 2 * exponent + 1, so this guard never matches.
-        // The fallback calculation is therefore unreachable on the active call path.
-        // if (divisor <= 3UL || exponent <= 1UL)
-        // {
-        //     cycleLength = CalculateCycleLength(divisor, divisorData);
-        //     return true;
-        // }
+        if (divisor <= 3UL || exponent <= 1UL)
+        {
+            cycleLength = CalculateCycleLength(divisor, divisorData);
+            return true;
+        }
 
-        ulong computedOrder = PrimeOrderCalculator.Calculate(
+        PrimeOrderCalculator.PrimeOrderResult orderResult = PrimeOrderCalculator.Calculate(
             divisor,
             previousOrder: null,
             divisorData,
             PrimeOrderCalculator.PrimeOrderSearchConfig.HeuristicDefault,
-            PrimeOrderCalculator.PrimeOrderHeuristicDevice.Cpu);
-        if (computedOrder != 0UL)
+            PrimeOrderCalculator.PrimeOrderHeuristicDevice.Cpu,
+            allowGpuPow2);
+        if (orderResult.Order != 0UL)
         {
-            cycleLength = computedOrder;
+            cycleLength = orderResult.Order;
             return true;
         }
 
         ulong phi = divisor - 1UL;
-        // Divisors reach this helper only when they exceed the cached table, which means
-        // divisor >= 2 * exponent + 1 and phi cannot underflow to zero here.
-        // The defensive underflow guard is kept for posterity but disabled for the EvenPerfectBitScanner path.
-        // if (phi == 0UL)
-        // {
-        //     return false;
-        // }
+        if (phi == 0UL)
+        {
+            return false;
+        }
 
         int twoCount = BitOperations.TrailingZeroCount(phi);
         ulong reducedPhi = phi >> twoCount;
-        // Every divisor candidate generated by the by-divisor scan satisfies q ≡ 1 (mod 2 * exponent),
-        // ensuring reducedPhi keeps a factor of exponent and remains non-zero in this context.
-        // The divisibility check is therefore unreachable for the current caller and is commented out.
-        // if (reducedPhi == 0UL || (reducedPhi % exponent) != 0UL)
-        // {
-        //     return false;
-        // }
+        if (reducedPhi == 0UL || (reducedPhi % exponent) != 0UL)
+        {
+            return false;
+        }
 
         ulong k = reducedPhi / exponent;
-        // reducedPhi always contains at least one factor of exponent, so dividing by exponent leaves k >= 1.
-        // if (k == 0UL)
-        // {
-        //     return false;
-        // }
+        if (k == 0UL)
+        {
+            return false;
+        }
 
-        Dictionary<ulong, int> factorCounts = AcquireFactorCountDictionary(8);
-        try
+        Dictionary<ulong, int> factorCounts = new(8);
+        if (twoCount > 0)
         {
-            // The divisor candidates are odd, so phi is even and this branch always records at least one factor of two.
-            // if (twoCount > 0)
-            // {
-            //     factorCounts[2UL] = twoCount;
-            // }
             factorCounts[2UL] = twoCount;
-
-            if (!AccumulateFactors(exponent, factorCache, factorCounts, cacheResult: true) || !AccumulateFactors(k, factorCache, factorCounts, cacheResult: false))
-            {
-                return false;
-            }
-
-            cycleLength = ReduceOrder(divisorData, phi, factorCounts, exponent);
-            return true;
         }
-        finally
+
+        if (!AccumulateFactors(exponent, factorCache, factorCounts, cacheResult: true) || !AccumulateFactors(k, factorCache, factorCounts, cacheResult: false))
         {
-            ReleaseFactorCountDictionary(factorCounts);
+            return false;
         }
+
+        cycleLength = ReduceOrder(divisorData, phi, factorCounts, allowGpuPow2);
+        return true;
     }
 
     private static bool AccumulateFactors(
@@ -413,119 +344,61 @@ public class MersenneDivisorCycles
             return true;
         }
 
-        FactorCacheEntry cachedFactorization;
-        PrimeFactorPartialResult? lease = null;
-        bool fromCache;
-
-        try
+        if (!TryGetFactorization(value, cache, cacheResult, out FactorCacheEntry factorization))
         {
-            if (!TryGetFactorization(value, cache, cacheResult, out cachedFactorization, out lease, out fromCache))
-            {
-                return false;
-            }
-
-            if (fromCache)
-            {
-                ulong[] primes = cachedFactorization.Primes;
-                byte[] exponents = cachedFactorization.Exponents;
-                int length = cachedFactorization.Count;
-
-                for (int i = 0; i < length; i++)
-                {
-                    AddFactor(counts, primes[i], exponents[i]);
-                }
-
-                return true;
-            }
-
-            if (lease is null || lease.Count == 0)
-            {
-                return true;
-            }
-
-            ReadOnlySpan<ulong> primesSpan = lease.Primes;
-            ReadOnlySpan<byte> exponentsSpan = lease.Exponents;
-            int spanLength = lease.Count;
-
-            for (int i = 0; i < spanLength; i++)
-            {
-                AddFactor(counts, primesSpan[i], exponentsSpan[i]);
-            }
-
-            return true;
+            return false;
         }
-        finally
+
+        ulong[] primes = factorization.Primes;
+        byte[] exponents = factorization.Exponents;
+        int length = factorization.Count;
+
+        for (int i = 0; i < length; i++)
         {
-            lease?.Dispose();
+            AddFactor(counts, primes[i], exponents[i]);
         }
+
+        return true;
     }
 
     private static bool TryGetFactorization(
         ulong value,
         Dictionary<ulong, FactorCacheEntry>? cache,
         bool cacheResult,
-        out FactorCacheEntry factorization,
-        out PrimeFactorPartialResult? lease,
-        out bool fromCache)
+        out FactorCacheEntry factorization)
     {
-        factorization = default;
-        lease = null;
-        fromCache = false;
-
-        if (value <= 1UL)
-        {
-            return true;
-        }
-
         if (cache is not null && cache.TryGetValue(value, out factorization))
         {
-            fromCache = true;
             return true;
         }
 
-        Dictionary<ulong, int> scratch = AcquireFactorScratchDictionary(8);
-        try
+        Dictionary<ulong, int> scratch = new(8);
+        if (!TryFactorIntoCountsInternal(value, scratch))
         {
-            if (!TryFactorIntoCountsInternal(value, scratch))
-            {
-                factorization = default;
-                return false;
-            }
-
-            int count = scratch.Count;
-            if (count == 0)
-            {
-                return true;
-            }
-
-            PrimeFactorPartialResult result = PrimeFactorPartialResult.Rent(count);
-            lease = result;
-
-            Span<ulong> primes = result.PrimeWriteSpan;
-            Span<byte> exponents = result.ExponentWriteSpan;
-            int index = 0;
-            foreach (KeyValuePair<ulong, int> entry in scratch)
-            {
-                primes[index] = entry.Key;
-                exponents[index] = checked((byte)entry.Value);
-                index++;
-            }
-
-            result.CommitCount(index);
-            result.Sort();
-
-            if (cache is not null && cacheResult)
-            {
-                FactorCacheEntry cacheEntry = result.ToCacheEntry();
-                cache[value] = cacheEntry;
-            }
-
-            return true;
+            factorization = default!;
+            return false;
         }
-        finally
+
+        int count = scratch.Count;
+        ulong[] primes = new ulong[count];
+        byte[] exponents = new byte[count];
+        int index = 0;
+        foreach (KeyValuePair<ulong, int> entry in scratch)
         {
-            ReleaseFactorScratchDictionary(scratch);
+            primes[index] = entry.Key;
+            exponents[index] = checked((byte)entry.Value);
+            index++;
         }
+
+        Array.Sort(primes, exponents);
+        factorization = new FactorCacheEntry(primes, exponents, count);
+
+        if (cache is not null && cacheResult)
+        {
+            cache[value] = factorization;
+        }
+
+        return true;
     }
 
     private static bool TryFactorIntoCountsInternal(ulong value, Dictionary<ulong, int> counts)
@@ -600,22 +473,15 @@ public class MersenneDivisorCycles
         }
     }
 
-    private static ulong ReduceOrder(
-        in MontgomeryDivisorData divisorData,
-        ulong initialOrder,
-        Dictionary<ulong, int> factorCounts,
-        ulong targetOrder)
+    private static ulong ReduceOrder(in MontgomeryDivisorData divisorData, ulong initialOrder, Dictionary<ulong, int> factorCounts, bool allowGpuPow2)
     {
-        if (factorCounts.Count == 0 || initialOrder == targetOrder)
+        if (factorCounts.Count == 0)
         {
             return initialOrder;
         }
 
         int count = factorCounts.Count;
-        bool rentFromPool = count >= PerfectNumberConstants.PooledArrayThreshold;
-        ulong[] primes = rentFromPool
-            ? ThreadLocalArrayPool<ulong>.Shared.Rent(count)
-            : new ulong[count];
+        ulong[] primes = ArrayPool<ulong>.Shared.Rent(count);
         try
         {
             int index = 0;
@@ -641,16 +507,9 @@ public class MersenneDivisorCycles
                     }
 
                     ulong candidate = order / prime;
-                    if (candidate.Pow2MontgomeryModWindowedCpu(divisorData, keepMontgomery: false) == 1UL)
+                    if (Pow2EqualsOne(candidate, divisorData, allowGpuPow2))
                     {
                         order = candidate;
-                        if (order == targetOrder)
-                        {
-                            // Early-exit once the order matches the tested exponent;
-                            // further factor processing would not change the outcome for the by-divisor scan.
-                            return order;
-                        }
-
                         continue;
                     }
 
@@ -662,13 +521,30 @@ public class MersenneDivisorCycles
         }
         finally
         {
-            if (rentFromPool)
-            {
-                ThreadLocalArrayPool<ulong>.Shared.Return(primes, clearArray: false);
-            }
+            ArrayPool<ulong>.Shared.Return(primes, clearArray: false);
         }
     }
 
+    [MethodImpl(MethodImplOptions.AggressiveInlining)]
+    private static bool Pow2EqualsOne(ulong exponent, in MontgomeryDivisorData divisorData, bool allowGpuPow2)
+    {
+        if (allowGpuPow2)
+        {
+            GpuPow2ModStatus status = PrimeOrderGpuHeuristics.TryPow2Mod(exponent, divisorData.Modulus, out ulong remainder, divisorData);
+            if (status == GpuPow2ModStatus.Success)
+            {
+                return remainder == 1UL;
+            }
+
+            PrimeOrderGpuHeuristics.ReportPow2Failure(
+                "divisor cycle reduction",
+                divisorData.Modulus,
+                status);
+        }
+
+        return exponent.Pow2MontgomeryModWindowed(divisorData, keepMontgomery: false) == 1UL;
+    }
+
     private static ulong PollardRho64(ulong n)
     {
         if ((n & 1UL) == 0UL)
@@ -738,53 +614,6 @@ public class MersenneDivisorCycles
         }
     }
 
-    // Thread-local dictionary pools keep the factoring helpers reusable so the GC stays out of the by-divisor CPU loop.
-    private static Dictionary<ulong, int> AcquireFactorCountDictionary(int capacityHint)
-    {
-        Dictionary<ulong, int>? dictionary = s_factorCountsPool;
-        if (dictionary is null)
-        {
-            return new Dictionary<ulong, int>(capacityHint);
-        }
-
-        s_factorCountsPool = null;
-        dictionary.Clear();
-        dictionary.EnsureCapacity(capacityHint);
-        return dictionary;
-    }
-
-    private static void ReleaseFactorCountDictionary(Dictionary<ulong, int> dictionary)
-    {
-        dictionary.Clear();
-        if (s_factorCountsPool is null)
-        {
-            s_factorCountsPool = dictionary;
-        }
-    }
-
-    private static Dictionary<ulong, int> AcquireFactorScratchDictionary(int capacityHint)
-    {
-        Dictionary<ulong, int>? dictionary = s_factorScratchPool;
-        if (dictionary is null)
-        {
-            return new Dictionary<ulong, int>(capacityHint);
-        }
-
-        s_factorScratchPool = null;
-        dictionary.Clear();
-        dictionary.EnsureCapacity(capacityHint);
-        return dictionary;
-    }
-
-    private static void ReleaseFactorScratchDictionary(Dictionary<ulong, int> dictionary)
-    {
-        dictionary.Clear();
-        if (s_factorScratchPool is null)
-        {
-            s_factorScratchPool = dictionary;
-        }
-    }
-
     private static ulong NextRandomUInt64()
     {
         Span<byte> buffer = stackalloc byte[8];
@@ -830,7 +659,7 @@ public class MersenneDivisorCycles
         // Start fresh to avoid mixing formats
         outputStream.SetLength(0L);
         outputStream.Position = 0L;
-        var writer = new BinaryWriter(outputStream, Encoding.UTF8, leaveOpen: false);
+        using var writer = new BinaryWriter(outputStream, Encoding.UTF8, leaveOpen: false);
         for (var taskIndex = 0; taskIndex < threads; taskIndex++)
         {
             ulong threadStart = start + (ulong)taskIndex * blockSize;
@@ -851,11 +680,7 @@ public class MersenneDivisorCycles
             tasks[taskIndex] = Task.Run(() =>
             {
                 ulong rangeLength = localEnd - localStart + 1UL;
-                int localLength = checked((int)rangeLength);
-                bool rentLocalCycles = localLength >= PerfectNumberConstants.PooledArrayThreshold;
-                (ulong divisor, ulong cycle)[] localCycles = rentLocalCycles
-                    ? ThreadLocalArrayPool<(ulong, ulong)>.Shared.Rent(localLength)
-                    : new (ulong divisor, ulong cycle)[localLength];
+                (ulong divisor, ulong cycle)[] localCycles = ArrayPool<(ulong, ulong)>.Shared.Rent(checked((int)rangeLength));
                 int localCycleIndex = 0;
 
                 try
@@ -903,16 +728,12 @@ public class MersenneDivisorCycles
                 }
                 finally
                 {
-                    if (rentLocalCycles)
-                    {
-                        ThreadLocalArrayPool<(ulong, ulong)>.Shared.Return(localCycles, clearArray: false);
-                    }
+                    ArrayPool<(ulong, ulong)>.Shared.Return(localCycles, clearArray: false);
                 }
             });
         }
 
         Task.WaitAll(tasks);
-        writer.Dispose();
     }
 
     public static void GenerateGpu(string path, ulong maxDivisor, int batchSize = 1_000_000, long skipCount = 0L, long nextPosition = 0L)
@@ -922,7 +743,7 @@ public class MersenneDivisorCycles
         ulong start = 3UL;
         var locker = new object();
 
-        var pool = ThreadLocalArrayPool<ulong>.Shared;
+        var pool = ArrayPool<ulong>.Shared;
         ulong batchSizeUL = (ulong)batchSize, d, end;
         int count = (int)Math.Min(batchSizeUL, maxDivisor), i, idx;
         ulong[] divisors, outCycles, validDivisors;
@@ -949,7 +770,7 @@ public class MersenneDivisorCycles
             outputStream.Position = 0L;
         }
 
-        var writer = new BinaryWriter(outputStream, Encoding.UTF8, leaveOpen: false);
+        using var writer = new BinaryWriter(outputStream, Encoding.UTF8, leaveOpen: false);
         try
         {
             while (skipCount > 0L && start <= maxDivisor)
@@ -1024,7 +845,6 @@ public class MersenneDivisorCycles
             pool.Return(outCycles, clearArray: false);
         }
 
-        writer.Dispose();
         stream.Dispose();
         lease.Dispose();
 
@@ -1126,36 +946,32 @@ public class MersenneDivisorCycles
         return CalculateCycleLengthFallback(divisor);
     }
 
-    internal static bool TryCalculateCycleLengthHeuristic(ulong divisor, in MontgomeryDivisorData divisorData, out ulong cycleLength)
+    internal static bool TryCalculateCycleLengthHeuristic(ulong divisor, in MontgomeryDivisorData divisorData, out ulong cycleLength, bool allowGpuPow2 = false)
     {
-        // The by-divisor scanner only supplies odd candidates of the form 2*k*exponent + 1,
-        // so this power-of-two fast-path never triggers on that call path.
-        // The early exit is retained for reference but commented out because it is unreachable on the scanner path.
-        // if ((divisor & (divisor - 1UL)) == 0UL)
-        // {
-        //     cycleLength = 1UL;
-        //     return true;
-        // }
+        if ((divisor & (divisor - 1UL)) == 0UL)
+        {
+            cycleLength = 1UL;
+            return true;
+        }
 
-        // By the divisor-by-divisor pipeline, candidates have the shape 2 * exponent * k + 1 with exponent >= 5,
-        // so the divisor never drops below 11 on this call path.
-        // if (divisor <= 3UL)
-        // {
-        //     cycleLength = CalculateCycleLengthFallback(divisor);
-        //     return true;
-        // }
+        if (divisor <= 3UL)
+        {
+            cycleLength = CalculateCycleLengthFallback(divisor);
+            return true;
+        }
 
         if (PrimeTester.IsPrimeInternal(divisor, CancellationToken.None))
         {
-            ulong computedOrder = PrimeOrderCalculator.Calculate(
+            PrimeOrderCalculator.PrimeOrderResult orderResult = PrimeOrderCalculator.Calculate(
                     divisor,
                     previousOrder: null,
                     divisorData,
                     PrimeOrderCalculator.PrimeOrderSearchConfig.HeuristicDefault,
-                    PrimeOrderCalculator.PrimeOrderHeuristicDevice.Cpu);
-            if (computedOrder != 0UL)
+                    PrimeOrderCalculator.PrimeOrderHeuristicDevice.Cpu,
+                    allowGpuPow2);
+            if (orderResult.Order != 0UL)
             {
-                cycleLength = computedOrder;
+                cycleLength = orderResult.Order;
                 return true;
             }
         }

PerfectNumbers.Core/MersenneNumberDivisorByDivisorTester.cs

@@ -18,8 +18,7 @@ public static class MersenneNumberDivisorByDivisorTester
             Action markComposite,
             Action clearComposite,
             Action<ulong, bool, bool, bool> printResult,
-            int threadCount,
-            TimeSpan? primeTestLimit)
+            int threadCount)
     {
         if (candidates.Count == 0)
         {
@@ -113,12 +112,9 @@ public static class MersenneNumberDivisorByDivisorTester
         ulong maxPrime = 0UL;
         int primeWriteIndex = 0;
 
-		// This implementation is terribly slow, while this method expect prime p given as --filter-p input already.
-		// We don't need to additionally check it.
-
-        // using IEnumerator<ulong> primeEnumerator = Prime.Numbers.GetEnumerator();
-        // bool hasPrime = primeEnumerator.MoveNext();
-        // ulong currentPrime = hasPrime ? primeEnumerator.Current : 0UL;
+        using IEnumerator<ulong> primeEnumerator = Prime.Numbers.GetEnumerator();
+        bool hasPrime = primeEnumerator.MoveNext();
+        ulong currentPrime = hasPrime ? primeEnumerator.Current : 0UL;
 
         for (int index = startIndex; index < candidateCount; index++)
         {
@@ -131,29 +127,25 @@ public static class MersenneNumberDivisorByDivisorTester
                 continue;
             }
 
-			// This implementation is terribly slow, while this method expect prime p given as --filter-p input already.
-			// We don't need to additionally check it.
-			
-			// while (hasPrime && currentPrime < candidate)
-			// {
-			//     hasPrime = primeEnumerator.MoveNext();
-			//     if (!hasPrime)
-			//     {
-			//         break;
-			//     }
+            while (hasPrime && currentPrime < candidate)
+            {
+                hasPrime = primeEnumerator.MoveNext();
+                if (!hasPrime)
+                {
+                    break;
+                }
 
-			//     currentPrime = primeEnumerator.Current;
-			// }
+                currentPrime = primeEnumerator.Current;
+            }
 
-			// if (!hasPrime || currentPrime != candidate)
-			// if (!Number.IsPrime(candidate))
-			// {
-			//     markComposite();
-			//     printResult(candidate, false, false, false);
-			//     continue;
-			// }
+            if (!hasPrime || currentPrime != candidate)
+            {
+                markComposite();
+                printResult(candidate, false, false, false);
+                continue;
+            }
 
-			primesSpan[primeWriteIndex++] = candidate;
+            primesSpan[primeWriteIndex++] = candidate;
 
             if (candidate > maxPrime)
             {
@@ -180,13 +172,11 @@ public static class MersenneNumberDivisorByDivisorTester
             return;
         }
 
-        // EvenPerfectBitScanner removes candidates <= 1 before reaching this point,
-        // so this guard never triggers during production runs.
-        // if (maxPrime <= 1UL)
-        // {
-        //     Console.WriteLine("The filter specified by --filter-p must contain at least one prime exponent greater than 1 for --mersenne=bydivisor.");
-        //     return;
-        // }
+        if (maxPrime <= 1UL)
+        {
+            Console.WriteLine("The filter specified by --filter-p must contain at least one prime exponent greater than 1 for --mersenne=bydivisor.");
+            return;
+        }
 
         tester.ConfigureFromMaxPrime(maxPrime);
 
@@ -246,7 +236,7 @@ public static class MersenneNumberDivisorByDivisorTester
 
         void ProcessPrime(ulong prime)
         {
-            bool isPrime = tester.IsPrime(prime, out bool divisorsExhausted, primeTestLimit);
+            bool isPrime = tester.IsPrime(prime, out bool divisorsExhausted);
 
             if (!isPrime)
             {
@@ -259,52 +249,21 @@ public static class MersenneNumberDivisorByDivisorTester
             printResult(prime, true, divisorsExhausted, true);
         }
 
-        if (workerCount > filteredPrimes.Count)
-        {
-            workerCount = filteredPrimes.Count;
-        }
-
-        if (workerCount <= 1)
+        if (workerCount == 1)
         {
             foreach (ulong prime in filteredPrimes)
             {
                 ProcessPrime(prime);
             }
-
-            return;
         }
-
-        (int Start, int End)[] workerRanges = new (int, int)[workerCount];
-        int totalPrimeCount = filteredPrimes.Count;
-        int baseGroupSize = totalPrimeCount / workerCount;
-        int remaining = totalPrimeCount % workerCount;
-        int groupStartIndex = 0;
-
-        for (int workerIndex = 0; workerIndex < workerCount; workerIndex++)
+        else
         {
-            int groupSize = baseGroupSize;
-            if (workerIndex < remaining)
+            ParallelOptions options = new()
             {
-                groupSize++;
-            }
+                MaxDegreeOfParallelism = workerCount
+            };
 
-            startIndex = groupStartIndex;
-            int endIndex = startIndex + groupSize;
-            workerRanges[workerIndex] = (startIndex, endIndex);
-            groupStartIndex = endIndex;
+            Parallel.ForEach(filteredPrimes, options, ProcessPrime);
         }
-
-        ParallelOptions parallelOptions = new()
-        {
-            MaxDegreeOfParallelism = workerCount
-        };
-
-        Parallel.ForEach(workerRanges, parallelOptions, range =>
-        {
-            for (int candidateIndex = range.Start; candidateIndex < range.End; candidateIndex++)
-            {
-                ProcessPrime(filteredPrimes[candidateIndex]);
-            }
-        });
     }
 }

PerfectNumbers.Core/MersennePrimeFactorTester.cs

@@ -371,6 +371,10 @@ public static class MersennePrimeFactorTester
             return 2UL;
         }
 
+        var rng = new Random(1234567); // TODO: Replace this RNG with the deterministic span-based
+                                       // sequence measured faster in PollardRhoBenchmarks so we avoid
+                                       // per-call Random allocations.
+
         ulong c, d, x, y, diff;
         while (true)
         {
@@ -379,8 +383,8 @@ public static class MersennePrimeFactorTester
                 return 0UL;
             }
 
-            c = (DeterministicRandom.NextUInt64() % (n - 1UL)) + 1UL;
-            x = (DeterministicRandom.NextUInt64() % (n - 2UL)) + 2UL;
+            c = (ulong)rng.Next(2, int.MaxValue);
+            x = (ulong)rng.Next(2, int.MaxValue);
             y = x;
             d = 1UL;
 
@@ -519,6 +523,9 @@ public static class MersennePrimeFactorTester
             return 2;
         }
 
+        var rng = new Random(1234567); // TODO: Replace with the deterministic UInt128 sequence from
+                                       // PollardRho128Benchmarks so we eliminate Random allocations.
+
         UInt128 c, d, x, y, diff;
         while (true)
         {
@@ -527,8 +534,8 @@ public static class MersennePrimeFactorTester
                 return 0;
             }
 
-            c = (DeterministicRandom.NextUInt128() % (n - 1)) + 1;
-            x = (DeterministicRandom.NextUInt128() % (n - 2)) + 2;
+            c = (UInt128)rng.NextInt64(2, long.MaxValue);
+            x = (UInt128)rng.NextInt64(2, long.MaxValue);
             y = x;
             d = 1;
 

PerfectNumbers.Core/MontgomeryDivisorData.cs

@@ -45,4 +45,32 @@ public readonly struct MontgomeryDivisorData(ulong modulus, ulong nPrime, ulong
             montgomeryTwo,
             montgomeryTwoSquared);
     }
+
+}
+
+internal static class MontgomeryDivisorDataCache
+{
+    [ThreadStatic]
+    private static bool s_hasCachedValue;
+
+    [ThreadStatic]
+    private static ulong s_cachedModulus;
+
+    [ThreadStatic]
+    private static MontgomeryDivisorData s_cachedData;
+
+    [MethodImpl(MethodImplOptions.AggressiveInlining)]
+    public static MontgomeryDivisorData Get(ulong modulus)
+    {
+        if (s_hasCachedValue && s_cachedModulus == modulus)
+        {
+            return s_cachedData;
+        }
+
+        MontgomeryDivisorData computed = MontgomeryDivisorData.FromModulus(modulus);
+        s_cachedModulus = modulus;
+        s_cachedData = computed;
+        s_hasCachedValue = true;
+        return computed;
+    }
 }

PerfectNumbers.Core/MutableUInt128.cs

@@ -1,243 +0,0 @@
-using System;
-using System.Numerics;
-using System.Runtime.CompilerServices;
-
-namespace PerfectNumbers.Core;
-
-/// <summary>
-/// Mutable helper for 128-bit arithmetic derived from the .NET runtime UInt128 implementation.
-/// Portions adapted from https://github.com/dotnet/runtime/blob/v8.0.0/src/libraries/System.Private.CoreLib/src/System/UInt128.cs.
-/// </summary>
-public struct MutableUInt128
-{
-    public ulong High;
-    public ulong Low;
-
-    public MutableUInt128(ulong value)
-    {
-        Low = value;
-    }
-
-    public MutableUInt128(ulong high, ulong low)
-    {
-        High = high;
-        Low = low;
-    }
-
-    public MutableUInt128(UInt128 value)
-    {
-        High = (ulong)(value >> 64);
-        Low = (ulong)value;
-    }
-
-    [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    public void Set(ulong value)
-    {
-        High = 0UL;
-        Low = value;
-    }
-
-    [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    public void Set(UInt128 value)
-    {
-        High = (ulong)(value >> 64);
-        Low = (ulong)value;
-    }
-
-    [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    public void Set(ulong high, ulong low)
-    {
-        High = high;
-        Low = low;
-    }
-
-    [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    public readonly UInt128 ToUInt128() => ((UInt128)High << 64) | Low;
-
-    [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    public void Add(ulong value)
-    {
-        ulong low = Low + value;
-        if (low < value)
-        {
-            High++;
-        }
-
-        Low = low;
-    }
-
-    [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    public void Add(MutableUInt128 value)
-    {
-        ulong temp = Low + value.Low;
-        Low = temp;
-        temp = temp < Low ? 1UL : 0UL;
-		temp += High;
-		temp += value.High;
-        High = temp;
-    }
-
-    [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    public void Sub(ulong value)
-    {
-        ulong temp = Low;
-        ulong result = temp - value;
-        temp = result > temp ? 1UL : 0UL;
-
-        Low = result;
-        High -= temp;
-    }
-
-    [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    public void Sub(MutableUInt128 value)
-    {
-        ulong temp = Low;
-        ulong result = temp - value.Low;
-        temp = temp < value.Low ? 1UL : 0UL;
-
-        Low = result;
-        High = High - value.High - temp;
-    }
-
-    [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    public void Mul(ulong value)
-    {
-        ulong currentHigh = High;
-        ulong currentLow = Low;
-
-        ulong temp = Mul64(currentLow, value, out currentLow);
-        Mul64(currentHigh, value, out currentHigh);
-        temp += currentHigh;
-
-        Low = currentLow;
-        High = temp;
-    }
-
-    [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    public static ulong Mul64(ulong left, ulong right, out ulong low)
-    {
-        const ulong Mask32 = 0xFFFF_FFFFUL;
-
-        ulong leftHigh = left >> 32;
-        ulong rightLow = right & Mask32;
-
-        ulong leftLow = left & Mask32;
-        ulong lowLow = leftLow * rightLow;
-        ulong temp = right >> 32;
-        leftLow *= temp;
-        rightLow *= leftHigh;
-        leftHigh *= temp;
-
-        temp = (lowLow >> 32) + (leftLow & Mask32) + (rightLow & Mask32);
-        leftHigh += (leftLow >> 32) + (rightLow >> 32) + (temp >> 32);
-
-        low = ((temp & Mask32) << 32) | (lowLow & Mask32);
-        return leftHigh;
-    }
-
-    [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    public void MulAdd(ulong multiplier, ulong addend)
-    {
-        const ulong Mask32 = 0xFFFF_FFFFUL;
-
-        ulong multiplierLow = multiplier & Mask32;
-        ulong multiplierHigh = multiplier >> 32;
-
-        ulong product11 = Low;
-        ulong product01 = product11 & Mask32;
-        product11 >>= 32;
-
-        ulong product00 = product01 * multiplierLow;
-        product01 *= multiplierHigh;
-        ulong product10 = product11 * multiplierLow;
-        product11 *= multiplierHigh;
-
-        ulong high = (product00 >> 32) + (product01 & Mask32) + (product10 & Mask32);
-        product11 = product11 + (product01 >> 32) + (product10 >> 32) + (high >> 32);
-        product10 = ((high & Mask32) << 32) | (product00 & Mask32);
-
-        high = High;
-        product00 = high & Mask32;
-        high >>= 32;
-
-        product01 = product00 * multiplierLow;
-        product00 *= multiplierHigh;
-        high *= multiplierLow;
-
-        high = (product01 >> 32) + (product00 & Mask32) + (high & Mask32);
-		product01 = ((high & Mask32) << 32) | (product01 & Mask32);
-
-		// Calculate low. We're assigning addend to local variable for a tiny bit better performance.
-		product00 = addend;
-		high = product10 + product00;
-        Low = high;
-
-		// Calculate high including carry
-		high = high < product00 ? 1UL : 0UL;
-		high += product11 + product01;
-		
-        // Discard overflow above 128 bits to match Multiply's truncation semantics.
-        High = high;
-    }
-
-    [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    public readonly ulong Mod(ulong modulus)
-	{
-		// This case will never happen in production code.
-		// if (modulus == 0UL)
-		// {
-		//     throw new DivideByZeroException();
-		// }
-
-		ulong high = High;
-		if (high == 0UL)
-        {
-            return Low % modulus;
-        }
-
-        MutableUInt128 remainder = default;
-        int highBitLength = 64 - BitOperations.LeadingZeroCount(high);
-        if (highBitLength > 0)
-        {
-            ProcessWord(ref remainder, high, highBitLength, modulus);
-        }
-
-        ProcessWord(ref remainder, Low, 64, modulus);
-        return remainder.Low;
-    }
-
-    [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    private static void ProcessWord(ref MutableUInt128 remainder, ulong word, int bitCount, ulong modulus)
-    {
-        for (int bit = bitCount - 1; bit >= 0; bit--)
-        {
-            remainder.ShiftLeft();
-            if (((word >> bit) & 1UL) != 0)
-            {
-                remainder.Add(1UL);
-            }
-
-            remainder.Reduce(modulus);
-        }
-    }
-
-    [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    private void ShiftLeft()
-    {
-        High = (High << 1) | (Low >> 63);
-        Low <<= 1;
-    }
-
-    [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    private void Reduce(ulong modulus)
-    {
-        if (High != 0UL || Low >= modulus)
-        {
-            Sub(modulus);
-            if (High != 0UL || Low >= modulus)
-            {
-                Sub(modulus);
-            }
-        }
-    }
-}

PerfectNumbers.Core/PartialFactorResult.cs

@@ -1,94 +0,0 @@
-using System;
-
-namespace PerfectNumbers.Core
-{
-    public sealed class PartialFactorResult
-    {
-        [ThreadStatic]
-        private static PartialFactorResult? s_poolHead;
-
-        public static readonly PartialFactorResult Empty = new()
-        {
-            Cofactor = 1UL,
-            FullyFactored = true,
-        };
-
-        public FactorEntry[]? Factors;
-        public ulong Cofactor;
-        public bool FullyFactored;
-        public int Count;
-
-        private PartialFactorResult? _next;
-        private bool _factorsFromPool;
-
-        private PartialFactorResult()
-        {
-        }
-
-        public static PartialFactorResult Rent(in FactorEntry[]? factors, ulong cofactor, bool fullyFactored, int count)
-        {
-            PartialFactorResult? instance = s_poolHead;
-            if (instance is null)
-            {
-                instance = new PartialFactorResult();
-            }
-            else
-            {
-                s_poolHead = instance._next;
-            }
-
-            FactorEntry[]? factorArray = factors;
-            instance._next = null;
-            instance.Factors = factorArray;
-            instance.Cofactor = cofactor;
-            instance.FullyFactored = fullyFactored;
-            instance.Count = count;
-            instance._factorsFromPool = factorArray is not null && factorArray.Length >= PerfectNumberConstants.PooledArrayThreshold;
-            return instance;
-        }
-
-        public PartialFactorResult WithAdditionalPrime(ulong prime)
-        {
-            FactorEntry[]? source = Factors;
-            if (source is null || Count == 0)
-            {
-                FactorEntry[] single = new FactorEntry[1];
-                single[0] = new FactorEntry(prime);
-                return Rent(single, 1UL, true, 1);
-            }
-
-            int newCount = Count + 1;
-            FactorEntry[] extended = newCount >= PerfectNumberConstants.PooledArrayThreshold
-                ? ThreadLocalArrayPool<FactorEntry>.Shared.Rent(newCount)
-                : new FactorEntry[newCount];
-
-            Array.Copy(source, 0, extended, 0, Count);
-            extended[Count] = new FactorEntry(prime);
-            Span<FactorEntry> span = extended.AsSpan(0, newCount);
-            span.Sort(static (a, b) => a.Value.CompareTo(b.Value));
-            return Rent(extended, 1UL, true, newCount);
-        }
-
-        public void Dispose()
-        {
-            if (this == Empty)
-            {
-                return;
-            }
-
-            FactorEntry[]? factors = Factors;
-            if (factors is not null)
-            {
-                Factors = null;
-                if (_factorsFromPool)
-                {
-                    ThreadLocalArrayPool<FactorEntry>.Shared.Return(factors, clearArray: false);
-                }
-            }
-
-            _factorsFromPool = false;
-            _next = s_poolHead;
-            s_poolHead = this;
-        }
-    }
-}

PerfectNumbers.Core/PerfectNumberConstants.cs

@@ -8,7 +8,6 @@ public static class PerfectNumberConstants
     public const int MaxQForDivisorCycles = 4_000_000;
     public static readonly uint PrimesLimit = 1_000_000; //(ulong)Array.MaxLength;// 1_000_000;
     public const int MaxOddPowersCount = 128;
-    public const int PooledArrayThreshold = 64;
     // 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.
     // TODO: Promote these magic numbers into a runtime profile derived from EvenPerfectBitScanner.Benchmarks so we can retune

PerfectNumbers.Core/PerfectNumbers.Core.csproj

@@ -22,7 +22,4 @@
     <InternalsVisibleTo Include="EvenPerfectBitScanner.Benchmarks" />
   </ItemGroup>
 
-  <ItemGroup>
-    <Compile Remove=".temp.cs" />
-  </ItemGroup>
 </Project>

PerfectNumbers.Core/PrimeFactorPartialResult.cs

@@ -1,150 +0,0 @@
-using System;
-using System.Buffers;
-
-namespace PerfectNumbers.Core
-{
-    internal sealed class PrimeFactorPartialResult
-    {
-        [ThreadStatic]
-        private static PrimeFactorPartialResult? s_poolHead;
-
-        private PrimeFactorPartialResult? _next;
-        private ulong[]? _primes;
-        private byte[]? _exponents;
-        private bool _arraysFromPool;
-        private int _capacity;
-
-        private PrimeFactorPartialResult()
-        {
-        }
-
-        public static PrimeFactorPartialResult Rent(int capacity)
-        {
-            PrimeFactorPartialResult? instance = s_poolHead;
-            if (instance is null)
-            {
-                instance = new PrimeFactorPartialResult();
-            }
-            else
-            {
-                s_poolHead = instance._next;
-            }
-
-            instance._next = null;
-            instance.EnsureCapacity(capacity);
-            instance.Count = 0;
-            return instance;
-        }
-
-        public int Count { get; private set; }
-
-        internal ReadOnlySpan<ulong> Primes => _primes is null ? ReadOnlySpan<ulong>.Empty : _primes.AsSpan(0, Count);
-
-        internal ReadOnlySpan<byte> Exponents => _exponents is null ? ReadOnlySpan<byte>.Empty : _exponents.AsSpan(0, Count);
-
-        internal Span<ulong> PrimeWriteSpan => _primes is null ? Span<ulong>.Empty : _primes.AsSpan(0, _capacity);
-
-        internal Span<byte> ExponentWriteSpan => _exponents is null ? Span<byte>.Empty : _exponents.AsSpan(0, _capacity);
-
-        internal void CommitCount(int count)
-        {
-            Count = count;
-        }
-
-        internal void Sort()
-        {
-            if (Count == 0 || _primes is null || _exponents is null)
-            {
-                return;
-            }
-
-            Array.Sort(_primes, _exponents, 0, Count);
-        }
-
-        internal MersenneDivisorCycles.FactorCacheEntry ToCacheEntry()
-        {
-            if (Count == 0 || _primes is null || _exponents is null)
-            {
-                return default;
-            }
-
-            ulong[] primes = new ulong[Count];
-            byte[] exponents = new byte[Count];
-            Array.Copy(_primes, 0, primes, 0, Count);
-            Array.Copy(_exponents, 0, exponents, 0, Count);
-            return new MersenneDivisorCycles.FactorCacheEntry(primes, exponents, Count);
-        }
-
-        public void Dispose()
-        {
-            if (_arraysFromPool)
-            {
-                if (_primes is not null)
-                {
-                    ThreadLocalArrayPool<ulong>.Shared.Return(_primes, clearArray: false);
-                }
-
-                if (_exponents is not null)
-                {
-                    ThreadLocalArrayPool<byte>.Shared.Return(_exponents, clearArray: false);
-                }
-
-                _arraysFromPool = false;
-                _primes = null;
-                _exponents = null;
-            }
-
-            _capacity = 0;
-            Count = 0;
-
-            _next = s_poolHead;
-            s_poolHead = this;
-        }
-
-        private void EnsureCapacity(int capacity)
-        {
-            if (capacity <= 0)
-            {
-                _capacity = 0;
-                return;
-            }
-
-            if (_primes is not null && _primes.Length >= capacity)
-            {
-                _capacity = capacity;
-                return;
-            }
-
-            if (_arraysFromPool)
-            {
-                if (_primes is not null)
-                {
-                    ThreadLocalArrayPool<ulong>.Shared.Return(_primes, clearArray: false);
-                }
-
-                if (_exponents is not null)
-                {
-                    ThreadLocalArrayPool<byte>.Shared.Return(_exponents, clearArray: false);
-                }
-
-                _arraysFromPool = false;
-                _primes = null;
-                _exponents = null;
-            }
-
-            if (capacity >= PerfectNumberConstants.PooledArrayThreshold)
-            {
-                _primes = ThreadLocalArrayPool<ulong>.Shared.Rent(capacity);
-                _exponents = ThreadLocalArrayPool<byte>.Shared.Rent(capacity);
-                _arraysFromPool = true;
-            }
-            else
-            {
-                _primes = new ulong[capacity];
-                _exponents = new byte[capacity];
-            }
-
-            _capacity = capacity;
-        }
-    }
-}

PerfectNumbers.Core/PrimeOrderCalculator.Wide.cs

@@ -9,15 +9,31 @@ namespace PerfectNumbers.Core;
 
 internal static partial class PrimeOrderCalculator
 {
-    public static UInt128 Calculate(
+    internal readonly struct PrimeOrderResultWide
+    {
+        public PrimeOrderResultWide(PrimeOrderStatus status, UInt128 order)
+        {
+            Status = status;
+            Order = order;
+        }
+
+        public PrimeOrderStatus Status { get; }
+
+        public UInt128 Order { get; }
+    }
+
+    public static PrimeOrderResultWide Calculate(UInt128 prime, UInt128? previousOrder, PrimeOrderSearchConfig config)
+        => Calculate(prime, previousOrder, config, PrimeOrderHeuristicDevice.Gpu);
+
+    public static PrimeOrderResultWide Calculate(
         UInt128 prime,
         UInt128? previousOrder,
         PrimeOrderSearchConfig config,
         PrimeOrderHeuristicDevice device)
     {
-        var scope = UsePow2Mode(device);
+        bool enableGpuPow2 = device == PrimeOrderHeuristicDevice.Gpu;
+        using var scope = UsePow2Mode(device, enableGpuPow2);
         MontgomeryDivisorData divisorData;
-        UInt128 result;
         if (prime <= ulong.MaxValue)
         {
             ulong? previous = null;
@@ -36,49 +52,41 @@ internal static partial class PrimeOrderCalculator
 
             ulong prime64 = (ulong)prime;
             divisorData = MontgomeryDivisorData.FromModulus(prime64);
-            ulong order64 = Calculate(prime64, previous, divisorData, config, device);
-            result = order64 == 0UL ? UInt128.Zero : (UInt128)order64;
+            PrimeOrderResult result = Calculate(prime64, previous, divisorData, config, device, enableGpuPow2);
+            return new PrimeOrderResultWide(result.Status, result.Order);
         }
         else
         {
             divisorData = default;
-            result = CalculateWideInternal(prime, previousOrder, divisorData, config);
         }
 
-        scope.Dispose();
-        return result;
+        return CalculateWideInternal(prime, previousOrder, divisorData, config);
     }
 
-    private static UInt128 CalculateWideInternal(UInt128 prime, UInt128? previousOrder, in MontgomeryDivisorData divisorData, PrimeOrderSearchConfig config)
+    private static PrimeOrderResultWide CalculateWideInternal(UInt128 prime, UInt128? previousOrder, in MontgomeryDivisorData divisorData, PrimeOrderSearchConfig config)
     {
         if (prime <= UInt128.One)
         {
-            return UInt128.One;
+            return new PrimeOrderResultWide(PrimeOrderStatus.Found, UInt128.One);
         }
 
         if (prime == (UInt128)3UL)
         {
-            return (UInt128)2UL;
+            return new PrimeOrderResultWide(PrimeOrderStatus.Found, (UInt128)2UL);
         }
 
         UInt128 phi = prime - UInt128.One;
-        var debugScope = UseDebugLogging(!IsGpuHeuristicDevice);
+        using var debugScope = UseDebugLogging(!IsGpuHeuristicDevice);
         PartialFactorResult128 phiFactors = PartialFactorWide(phi, config);
-        UInt128 result;
         if (phiFactors.Factors is null)
         {
-            result = FinishStrictlyWide(prime, divisorData);
-        }
-        else
-        {
-            result = RunHeuristicPipelineWide(prime, previousOrder, divisorData, config, phi, phiFactors);
+            return FinishStrictlyWide(prime, divisorData, config.Mode);
         }
 
-        debugScope.Dispose();
-        return result;
+        return RunHeuristicPipelineWide(prime, previousOrder, divisorData, config, phi, phiFactors);
     }
 
-    private static UInt128 RunHeuristicPipelineWide(
+    private static PrimeOrderResultWide RunHeuristicPipelineWide(
         UInt128 prime,
         UInt128? previousOrder,
         in MontgomeryDivisorData divisorData,
@@ -88,7 +96,7 @@ internal static partial class PrimeOrderCalculator
     {
         if (phiFactors.FullyFactored && TrySpecialMaxWide(phi, prime, divisorData, phiFactors))
         {
-            return phi;
+            return new PrimeOrderResultWide(PrimeOrderStatus.Found, phi);
         }
 
         UInt128 candidateOrder = InitializeStartingOrderWide(prime, phi, divisorData);
@@ -96,25 +104,27 @@ internal static partial class PrimeOrderCalculator
 
         if (TryConfirmOrderWide(prime, candidateOrder, divisorData, config))
         {
-            return candidateOrder;
+            return new PrimeOrderResultWide(PrimeOrderStatus.Found, candidateOrder);
         }
 
         if (config.Mode == PrimeOrderMode.Strict)
         {
-            return FinishStrictlyWide(prime, divisorData);
+            return FinishStrictlyWide(prime, divisorData, PrimeOrderMode.Strict);
         }
 
         if (TryHeuristicFinishWide(prime, candidateOrder, previousOrder, divisorData, config, out UInt128 order))
         {
-            return order;
+            return new PrimeOrderResultWide(PrimeOrderStatus.Found, order);
         }
 
-        return candidateOrder;
+        return FinishStrictlyWide(prime, divisorData, config.Mode);
     }
 
-    private static UInt128 FinishStrictlyWide(UInt128 prime, in MontgomeryDivisorData divisorData)
+    private static PrimeOrderResultWide FinishStrictlyWide(UInt128 prime, in MontgomeryDivisorData divisorData, PrimeOrderMode mode)
     {
-        return CalculateByFactorizationWide(prime, divisorData);
+        UInt128 strictOrder = CalculateByFactorizationWide(prime, divisorData);
+        PrimeOrderStatus status = mode == PrimeOrderMode.Strict ? PrimeOrderStatus.Found : PrimeOrderStatus.HeuristicUnresolved;
+        return new PrimeOrderResultWide(status, strictOrder);
     }
 
     private static bool TrySpecialMaxWide(UInt128 phi, UInt128 prime, in MontgomeryDivisorData divisorData, PartialFactorResult128 factors)
@@ -162,7 +172,7 @@ internal static partial class PrimeOrderCalculator
                 return order;
             }
 
-            tempArray = ThreadLocalArrayPool<FactorEntry128>.Shared.Rent(length + 1);
+            tempArray = ArrayPool<FactorEntry128>.Shared.Rent(length + 1);
             Span<FactorEntry128> buffer = tempArray.AsSpan(0, length);
             factorSpan.CopyTo(buffer);
 
@@ -200,7 +210,7 @@ internal static partial class PrimeOrderCalculator
         {
             if (tempArray is not null)
             {
-                ThreadLocalArrayPool<FactorEntry128>.Shared.Return(tempArray, clearArray: false);
+                ArrayPool<FactorEntry128>.Shared.Return(tempArray, clearArray: false);
             }
         }
     }
@@ -664,7 +674,7 @@ internal static partial class PrimeOrderCalculator
             return new PartialFactorResult128(null, value, false, 0);
         }
 
-        FactorEntry128[]? rented = counts.Count > 0 ? ThreadLocalArrayPool<FactorEntry128>.Shared.Rent(counts.Count) : null;
+        FactorEntry128[]? rented = counts.Count > 0 ? ArrayPool<FactorEntry128>.Shared.Rent(counts.Count) : null;
         int index = 0;
         if (rented is not null)
         {
@@ -682,7 +692,7 @@ internal static partial class PrimeOrderCalculator
             span.Sort(static (a, b) => a.Value.CompareTo(b.Value));
             resultArray = new FactorEntry128[index];
             span.CopyTo(resultArray);
-            ThreadLocalArrayPool<FactorEntry128>.Shared.Return(rented, clearArray: false);
+            ArrayPool<FactorEntry128>.Shared.Return(rented, clearArray: false);
         }
 
         bool fullyFactored = cofactor == UInt128.One;
@@ -1021,6 +1031,11 @@ internal static partial class PrimeOrderCalculator
             {
                 return wideRemainder;
             }
+
+            PrimeOrderGpuHeuristics.ReportPow2Failure(
+                "wide prime order pow2 check",
+                modulus,
+                wideStatus);
         }
 
         return exponent.Pow2MontgomeryModWindowed(modulus);
@@ -1062,20 +1077,15 @@ internal static partial class PrimeOrderCalculator
 
     private readonly struct FactorEntry128
     {
-		public FactorEntry128(UInt128 value)
-		{
-			Value = value;
-		}
+        public FactorEntry128(UInt128 value, int exponent)
+        {
+            Value = value;
+            Exponent = exponent;
+        }
 
-		public FactorEntry128(UInt128 value, int exponent)
-		{
-			Value = value;
-			Exponent = exponent;
-		}
+        public UInt128 Value { get; }
 
-        public readonly UInt128 Value;
-
-		public readonly int Exponent = 1;
+        public int Exponent { get; }
     }
 
     private readonly struct PartialFactorResult128
@@ -1088,13 +1098,13 @@ internal static partial class PrimeOrderCalculator
             Count = count;
         }
 
-		public readonly FactorEntry128[]? Factors;
+        public FactorEntry128[]? Factors { get; }
 
-		public readonly UInt128 Cofactor;
+        public UInt128 Cofactor { get; }
 
-		public readonly bool FullyFactored;
+        public bool FullyFactored { get; }
 
-		public readonly int Count;
+        public int Count { get; }
 
         public static PartialFactorResult128 Empty => new(null, UInt128.One, true, 0);
 
@@ -1102,13 +1112,14 @@ internal static partial class PrimeOrderCalculator
         {
             if (Factors is null)
             {
-                FactorEntry128[] local = [new FactorEntry128(prime)];
-				return new PartialFactorResult128(local, UInt128.One, true, 1);
+                FactorEntry128[] local = new FactorEntry128[1];
+                local[0] = new FactorEntry128(prime, 1);
+                return new PartialFactorResult128(local, UInt128.One, true, 1);
             }
 
             FactorEntry128[] extended = new FactorEntry128[Count + 1];
             Array.Copy(Factors, extended, Count);
-            extended[Count] = new FactorEntry128(prime);
+            extended[Count] = new FactorEntry128(prime, 1);
             return new PartialFactorResult128(extended, UInt128.One, true, Count + 1);
         }
     }

PerfectNumbers.Core/PrimeTestTimeLimit.cs

@@ -1,72 +0,0 @@
-using System;
-using System.Diagnostics;
-
-namespace PerfectNumbers.Core;
-
-internal readonly struct PrimeTestTimeLimit
-{
-    private readonly long _deadlineTimestamp;
-
-    private PrimeTestTimeLimit(bool isActive, long deadlineTimestamp)
-    {
-        IsActive = isActive;
-        _deadlineTimestamp = deadlineTimestamp;
-    }
-
-    internal bool IsActive { get; }
-
-    internal static bool TryCreate(TimeSpan? timeLimit, out PrimeTestTimeLimit limit)
-    {
-        if (!timeLimit.HasValue)
-        {
-            limit = new PrimeTestTimeLimit(isActive: false, deadlineTimestamp: long.MaxValue);
-            return true;
-        }
-
-        TimeSpan limitValue = timeLimit.GetValueOrDefault();
-        if (limitValue <= TimeSpan.Zero)
-        {
-            limit = new PrimeTestTimeLimit(isActive: true, deadlineTimestamp: long.MinValue);
-            return false;
-        }
-
-        double seconds = limitValue.TotalSeconds;
-        if (double.IsInfinity(seconds))
-        {
-            limit = new PrimeTestTimeLimit(isActive: true, deadlineTimestamp: long.MaxValue);
-            return true;
-        }
-
-        double scaled = seconds * Stopwatch.Frequency;
-        long limitTicks = scaled >= long.MaxValue ? long.MaxValue : (long)scaled;
-        long startTimestamp = Stopwatch.GetTimestamp();
-        long deadline = startTimestamp <= long.MaxValue - limitTicks ? startTimestamp + limitTicks : long.MaxValue;
-        if (deadline < startTimestamp)
-        {
-            deadline = long.MaxValue;
-        }
-
-        limit = new PrimeTestTimeLimit(isActive: true, deadlineTimestamp: deadline);
-        return true;
-    }
-
-    internal bool HasExpired()
-    {
-        if (!IsActive)
-        {
-            return false;
-        }
-
-        if (_deadlineTimestamp == long.MaxValue)
-        {
-            return false;
-        }
-
-        if (_deadlineTimestamp == long.MinValue)
-        {
-            return true;
-        }
-
-        return Stopwatch.GetTimestamp() >= _deadlineTimestamp;
-    }
-}

PerfectNumbers.Core/PrimeTester.cs

@@ -176,7 +176,7 @@ public sealed class PrimeTester(bool useInternal = false)
             DevicePrimes.View.CopyFromCPU(primes);
         }
 
-        internal sealed class ScratchBuffers
+        internal sealed class ScratchBuffers : IDisposable
         {
             public MemoryBuffer1D<ulong, Stride1D.Dense> Input { get; private set; }
             public MemoryBuffer1D<byte, Stride1D.Dense> Output { get; private set; }

PerfectNumbers.Core/TextFileWriter.cs

@@ -3,7 +3,7 @@ namespace PerfectNumbers.Core;
 /// <summary>
 /// Provides thread-safe file writing for both text and binary data.
 /// </summary>
-public sealed class TextFileWriter
+public sealed class TextFileWriter : IDisposable
 {
     private readonly FileStream _stream;
     private StreamWriter? _textWriter;

PerfectNumbers.Core/ThreadLocalArrayPool.cs

@@ -1,27 +0,0 @@
-using System;
-using System.Buffers;
-using System.Runtime.CompilerServices;
-
-namespace PerfectNumbers.Core;
-
-internal static class ThreadLocalArrayPool<T>
-{
-    [ThreadStatic]
-    private static ArrayPool<T>? s_threadPool;
-
-    public static ArrayPool<T> Shared
-    {
-        [MethodImpl(MethodImplOptions.AggressiveInlining)]
-        get
-        {
-            ArrayPool<T>? pool = s_threadPool;
-            if (pool is null)
-            {
-                pool = ArrayPool<T>.Create();
-                s_threadPool = pool;
-            }
-
-            return pool;
-        }
-    }
-}

PerfectNumbers.Core/UInt128Extensions.cs

@@ -218,7 +218,7 @@ public static class UInt128Extensions
     }
 
     [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    private static bool IsBitSet(GpuUInt128 value, int bitIndex)
+    private static bool IsBitSet(in GpuUInt128 value, int bitIndex)
     {
         if (bitIndex >= 64)
         {
@@ -229,16 +229,18 @@ public static class UInt128Extensions
     }
 
     [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    private static ulong ExtractWindowValue(GpuUInt128 exponent, int windowStart, int windowBitCount)
+    private static ulong ExtractWindowValue(in GpuUInt128 exponent, int windowStart, int windowBitCount)
     {
-		ulong mask = (1UL << windowBitCount) - 1UL;
         if (windowStart != 0)
         {
-            exponent.ShiftRight(windowStart);
-            return exponent.Low & mask;
+            GpuUInt128 shifted = exponent;
+            shifted.ShiftRight(windowStart);
+            ulong mask = (1UL << windowBitCount) - 1UL;
+            return shifted.Low & mask;
         }
 
-        return exponent.Low & mask;
+        ulong directMask = (1UL << windowBitCount) - 1UL;
+        return exponent.Low & directMask;
     }
 
     [MethodImpl(MethodImplOptions.AggressiveInlining)]

PerfectNumbers.Core/ULongExtensions.cs

@@ -1,5 +1,4 @@
 using System;
-using System.Buffers;
 using System.Numerics;
 using System.Runtime.CompilerServices;
 using ILGPU.Algorithms;
@@ -12,17 +11,13 @@ public static class ULongExtensions
     private const int Pow2WindowSize = 8;
     private const ulong Pow2WindowFallbackThreshold = 32UL;
     private const int Pow2WindowOddPowerCount = 1 << (Pow2WindowSize - 1);
-    private const int Pow2WindowSizeGpu = 5;
-    private const int Pow2WindowOddPowerCountGpu = 1 << (Pow2WindowSizeGpu - 1);
 
     public static ulong CalculateOrder(this ulong q)
     {
-        // Order computations only run on admissible divisors, so the small-prime guard remains commented out to keep the
-        // path branch-free.
-        // if (q <= 2UL)
-        // {
-        //     return 0UL;
-        // }
+        if (q <= 2UL)
+        {
+            return 0UL;
+        }
 
         ulong order = q - 1UL, prime, temp;
         uint[] smallPrimes = PrimesGenerator.SmallPrimes;
@@ -292,220 +287,26 @@ public static class ULongExtensions
     }
 
     [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    public static ulong MulMod64(this ulong a, ulong b, ulong modulus)
-    {
-        // Scanner inputs always provide odd prime moduli, so the legacy modulus guard stays commented out to avoid extra
-        // branching.
-        // if (modulus <= 1UL)
-        // {
-        //     return 0UL;
-        // }
+    // TODO: Replace this fallback with the UInt128 Montgomery helper measured fastest in
+    // MulMod64Benchmarks so CPU callers stop paying for triple modulo operations.
+    public static ulong MulMod64(this ulong a, ulong b, ulong modulus) => (ulong)(UInt128)(((a % modulus) * (b % modulus)) % modulus);
 
-        ulong reducedA = a >= modulus ? a % modulus : a;
-        ulong reducedB = b >= modulus ? b % modulus : b;
-        GpuUInt128 state = new(reducedA);
-        return state.MulMod(reducedB, modulus);
+    [MethodImpl(MethodImplOptions.AggressiveInlining)]
+    public static ulong MulMod64GpuCompatible(this ulong a, ulong b, ulong modulus)
+    {
+        // TODO: Remove this GPU-compatible shim from production once callers migrate to MulMod64,
+        // which the benchmarks show is roughly 6-7× faster on dense 64-bit inputs.
+        GpuUInt128 state = new(a % modulus);
+        return state.MulMod(b, modulus);
     }
 
     [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    public static ulong Pow2ModWindowedCpu(this ulong exponent, ulong modulus)
+    public static ulong MulMod64GpuCompatibleDeferred(this ulong a, ulong b, ulong modulus)
     {
-        // The by-divisor pipeline only schedules odd prime moduli, so the defensive check below
-        // would never trigger on that path. Leaving it commented documents the invariant without
-        // forcing a branch in the hot loop.
-        // if (modulus <= 1UL)
-        // {
-        //     return 0UL;
-        // }
-
-        // Exponents flow from the strictly positive candidate list on the by-divisor scanner, so
-        // this guard would be redundant there.
-        // if (exponent == 0UL)
-        // {
-        //     return 1UL % modulus;
-        // }
-
-        // Even though the by-divisor pipeline only feeds very large exponents, the legacy diagnostic callers
-        // still hit this small-exponent fast path, so keep the fallback in place.
-        if (exponent <= Pow2WindowFallbackThreshold)
-        {
-            return Pow2ModBinaryFallback(exponent, modulus);
-        }
-
-        int bitLength = GetPortableBitLength(exponent);
-        int windowSize = GetWindowSize(bitLength);
-        int oddPowerCount = 1 << (windowSize - 1);
-
-        Span<ulong> oddPowerStorage = oddPowerCount <= PerfectNumberConstants.MaxOddPowersCount
-            ? stackalloc ulong[PerfectNumberConstants.MaxOddPowersCount]
-            : new ulong[oddPowerCount];
-        oddPowerStorage = oddPowerStorage[..oddPowerCount];
-        InitializePlainOddPowers(modulus, oddPowerStorage);
-
-        // With modulus guaranteed to exceed one on the scanning path, the modulo operation would
-        // always yield one, so keep the literal to highlight the intended base value.
-        ulong result = 1UL;
-        int index = bitLength - 1;
-
-        while (index >= 0)
-        {
-            if (((exponent >> index) & 1UL) == 0UL)
-            {
-                result = result.MulMod64(result, modulus);
-                index--;
-                continue;
-            }
-
-            int windowStart = index - windowSize + 1;
-            if (windowStart < 0)
-            {
-                windowStart = 0;
-            }
-
-            while (((exponent >> windowStart) & 1UL) == 0UL)
-            {
-                windowStart++;
-            }
-
-            int windowLength = index - windowStart + 1;
-            for (int square = 0; square < windowLength; square++)
-            {
-                result = result.MulMod64(result, modulus);
-            }
-
-            ulong windowValue = (exponent >> windowStart) & ((1UL << windowLength) - 1UL);
-            int tableIndex = (int)((windowValue - 1UL) >> 1);
-            result = result.MulMod64(oddPowerStorage[tableIndex], modulus);
-
-            index = windowStart - 1;
-        }
-
-        return result;
-    }
-
-    [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    private static ulong Pow2ModBinaryFallback(ulong exponent, ulong modulus)
-    {
-        // The scanning path keeps modulus strictly greater than one, but other utility callers still
-        // rely on this fallback for tiny exponents, so leave the modulo in place for safety.
-        ulong remainingExponent = exponent;
-        // Production moduli are always odd primes greater than two, so normalize the seed values directly.
-        ulong result = 1UL;
-        ulong baseValue = 2UL;
-
-        while (remainingExponent != 0UL)
-        {
-            if ((remainingExponent & 1UL) != 0UL)
-            {
-                result = result.MulMod64(baseValue, modulus);
-            }
-
-            remainingExponent >>= 1;
-            if (remainingExponent == 0UL)
-            {
-                break;
-            }
-
-            baseValue = baseValue.MulMod64(baseValue, modulus);
-        }
-
-        return result;
-    }
-
-    [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    public static ulong Pow2ModWindowedGpu(this ulong exponent, ulong modulus)
-    {
-        // The GPU pipeline feeds only odd prime moduli, so the removed guard would never fire for
-        // EvenPerfectBitScanner workloads.
-        // if (modulus <= 1UL)
-        // {
-        //     return 0UL;
-        // }
-
-        // The GPU pipeline only emits positive exponents on production workloads, so skip the zero guard to avoid an extra branch.
-        // Diagnostic harnesses must normalize their inputs before invoking this helper.
-        // if (exponent == 0UL)
-        // {
-        //     return 1UL;
-        // }
-
-        // The GPU kernels mostly process large exponents as well, but the binary fallback still serves
-        // diagnostic harnesses that probe edge cases around the threshold.
-        if (exponent <= Pow2WindowFallbackThreshold)
-        {
-            return Pow2ModBinaryGpuFallback(exponent, modulus);
-        }
-
-        int bitLength = GetPortableBitLength(exponent);
-        int windowSize = GetGpuWindowSize(bitLength);
-        int oddPowerCount = 1 << (windowSize - 1);
-
-        PlainOddPowerTableGpu oddPowers = new PlainOddPowerTableGpu(modulus, oddPowerCount);
-
-        ulong result = 1UL;
-        int index = bitLength - 1;
-
-        while (index >= 0)
-        {
-            if (((exponent >> index) & 1UL) == 0UL)
-            {
-                result = result.MulMod64(result, modulus);
-                index--;
-                continue;
-            }
-
-            int windowStart = index - windowSize + 1;
-            if (windowStart < 0)
-            {
-                windowStart = 0;
-            }
-
-            while (((exponent >> windowStart) & 1UL) == 0UL)
-            {
-                windowStart++;
-            }
-
-            int windowLength = index - windowStart + 1;
-            for (int square = 0; square < windowLength; square++)
-            {
-                result = result.MulMod64(result, modulus);
-            }
-
-            ulong windowValue = (exponent >> windowStart) & ((1UL << windowLength) - 1UL);
-            int tableIndex = (int)((windowValue - 1UL) >> 1);
-            result = result.MulMod64(oddPowers.Get(tableIndex), modulus);
-
-            index = windowStart - 1;
-        }
-
-        return result;
-    }
-
-    [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    private static ulong Pow2ModBinaryGpuFallback(ulong exponent, ulong modulus)
-    {
-        ulong remainingExponent = exponent;
-        // Production moduli are always odd primes greater than two, so normalize the seed values directly.
-        ulong result = 1UL;
-        ulong baseValue = 2UL;
-
-        while (remainingExponent != 0UL)
-        {
-            if ((remainingExponent & 1UL) != 0UL)
-            {
-                result = result.MulMod64(baseValue, modulus);
-            }
-
-            remainingExponent >>= 1;
-            if (remainingExponent == 0UL)
-            {
-                break;
-            }
-
-            baseValue = baseValue.MulMod64(baseValue, modulus);
-        }
-
-        return result;
+        // TODO: Move this deferred helper to the benchmark suite; the baseline MulMod64 avoids the
+        // 5-40× slowdown seen across real-world operand distributions.
+        GpuUInt128 state = new(a);
+        return state.MulModWithNativeModulo(b, modulus);
     }
 
     [MethodImpl(MethodImplOptions.AggressiveInlining)]
@@ -525,14 +326,12 @@ public static class ULongExtensions
     }
 
     [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    public static ulong Pow2MontgomeryModWindowedCpu(this ulong exponent, in MontgomeryDivisorData divisor, bool keepMontgomery)
+    public static ulong Pow2MontgomeryModWindowed(this ulong exponent, in MontgomeryDivisorData divisor, bool keepMontgomery)
     {
         ulong modulus = divisor.Modulus;
         if (exponent == 0UL)
         {
-            return keepMontgomery
-                ? divisor.MontgomeryOne
-                : 1UL; // Montgomery moduli stay above one on the scanning path, so the modulo would always yield one.
+            return keepMontgomery ? divisor.MontgomeryOne : 1UL % modulus;
         }
 
         if (exponent <= Pow2WindowFallbackThreshold)
@@ -546,101 +345,7 @@ public static class ULongExtensions
 
         ulong result = divisor.MontgomeryOne;
         ulong nPrime = divisor.NPrime;
-
-        Span<ulong> oddPowerStorage = oddPowerCount <= PerfectNumberConstants.MaxOddPowersCount
-            ? stackalloc ulong[PerfectNumberConstants.MaxOddPowersCount]
-            : default;
-        ulong[]? pooledOddPowers = null;
-
-        if (oddPowerCount <= PerfectNumberConstants.MaxOddPowersCount)
-        {
-            oddPowerStorage = oddPowerStorage[..oddPowerCount];
-        }
-        else if (oddPowerCount < PerfectNumberConstants.PooledArrayThreshold)
-        {
-            oddPowerStorage = new ulong[oddPowerCount];
-        }
-        else
-        {
-            pooledOddPowers = ArrayPool<ulong>.Shared.Rent(oddPowerCount);
-            oddPowerStorage = pooledOddPowers;
-        }
-
-        Span<ulong> oddPowers = oddPowerStorage[..oddPowerCount];
-        InitializeMontgomeryOddPowersCpu(divisor, modulus, nPrime, oddPowers);
-
-        int index = bitLength - 1;
-        while (index >= 0)
-        {
-            if (((exponent >> index) & 1UL) == 0UL)
-            {
-                result = result.MontgomeryMultiply(result, modulus, nPrime);
-                index--;
-                continue;
-            }
-
-            int windowStart = index - windowSize + 1;
-            if (windowStart < 0)
-            {
-                windowStart = 0;
-            }
-
-            while (((exponent >> windowStart) & 1UL) == 0UL)
-            {
-                windowStart++;
-            }
-
-            int windowLength = index - windowStart + 1;
-            for (int square = 0; square < windowLength; square++)
-            {
-                result = result.MontgomeryMultiply(result, modulus, nPrime);
-            }
-
-            ulong mask = (1UL << windowLength) - 1UL;
-            ulong windowValue = (exponent >> windowStart) & mask;
-            int tableIndex = (int)((windowValue - 1UL) >> 1);
-            ulong multiplier = oddPowers[tableIndex];
-            result = result.MontgomeryMultiply(multiplier, modulus, nPrime);
-
-            index = windowStart - 1;
-        }
-
-        if (pooledOddPowers is not null)
-        {
-            ArrayPool<ulong>.Shared.Return(pooledOddPowers);
-        }
-
-        if (keepMontgomery)
-        {
-            return result;
-        }
-
-        return result.MontgomeryMultiply(1UL, modulus, nPrime);
-    }
-
-    [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    public static ulong Pow2MontgomeryModWindowedGpu(this ulong exponent, in MontgomeryDivisorData divisor, bool keepMontgomery)
-    {
-        ulong modulus = divisor.Modulus;
-        if (exponent == 0UL)
-        {
-            return keepMontgomery
-                ? divisor.MontgomeryOne
-                : 1UL; // Montgomery moduli stay above one on the scanning path, so the modulo would always yield one.
-        }
-
-        if (exponent <= Pow2WindowFallbackThreshold)
-        {
-            return Pow2MontgomeryModSingleBit(exponent, divisor, keepMontgomery);
-        }
-
-        int bitLength = GetPortableBitLength(exponent);
-        int windowSize = GetWindowSize(bitLength);
-        int oddPowerCount = 1 << (windowSize - 1);
-
-        ulong result = divisor.MontgomeryOne;
-        ulong nPrime = divisor.NPrime;
-        ulong[] oddPowers = InitializeMontgomeryOddPowersGpu(divisor, modulus, nPrime, oddPowerCount);
+        ulong[] oddPowers = InitializeMontgomeryOddPowers(divisor, modulus, nPrime, oddPowerCount);
 
         int index = bitLength - 1;
         while (index >= 0)
@@ -756,219 +461,7 @@ public static class ULongExtensions
     }
 
     [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    private static int GetGpuWindowSize(int bitLength)
-    {
-        int window = GetWindowSize(bitLength);
-        return window > Pow2WindowSizeGpu ? Pow2WindowSizeGpu : window;
-    }
-
-    [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    private static void InitializePlainOddPowers(ulong modulus, Span<ulong> destination)
-    {
-        int oddPowerCount = destination.Length;
-        // The by-divisor kernels always request at least one odd power, but other helpers reuse this initializer
-        // for smaller workloads, so the early exit stays in place.
-        if (oddPowerCount == 0)
-        {
-            return;
-        }
-
-        ulong baseValue = 2UL; // Moduli arrive as odd primes greater than two, so the modulo would always return two.
-        destination[0] = baseValue;
-        if (oddPowerCount == 1)
-        {
-            return;
-        }
-
-        ulong square = baseValue.MulMod64(baseValue, modulus);
-        for (int i = 1; i < oddPowerCount; i++)
-        {
-            // Reuse baseValue to accumulate successive odd powers and avoid allocating another local.
-            baseValue = baseValue.MulMod64(square, modulus);
-            destination[i] = baseValue;
-        }
-    }
-
-    private struct PlainOddPowerTableGpu
-    {
-        private readonly int _count;
-        private ulong _value0;
-        private ulong _value1;
-        private ulong _value2;
-        private ulong _value3;
-        private ulong _value4;
-        private ulong _value5;
-        private ulong _value6;
-        private ulong _value7;
-        private ulong _value8;
-        private ulong _value9;
-        private ulong _value10;
-        private ulong _value11;
-        private ulong _value12;
-        private ulong _value13;
-        private ulong _value14;
-        private ulong _value15;
-
-        public PlainOddPowerTableGpu(ulong modulus, int requestedCount)
-        {
-            _count = requestedCount > Pow2WindowOddPowerCountGpu ? Pow2WindowOddPowerCountGpu : requestedCount;
-            _value0 = 0UL;
-            _value1 = 0UL;
-            _value2 = 0UL;
-            _value3 = 0UL;
-            _value4 = 0UL;
-            _value5 = 0UL;
-            _value6 = 0UL;
-            _value7 = 0UL;
-            _value8 = 0UL;
-            _value9 = 0UL;
-            _value10 = 0UL;
-            _value11 = 0UL;
-            _value12 = 0UL;
-            _value13 = 0UL;
-            _value14 = 0UL;
-            _value15 = 0UL;
-
-            // Window planning always requests at least one odd power, so keep the zero-count check commented out to avoid
-            // branching here.
-            // if (_count == 0)
-            // {
-            //     return;
-            // }
-
-            ulong baseValue = 2UL;
-            _value0 = baseValue;
-            if (_count == 1)
-            {
-                return;
-            }
-
-            ulong square = baseValue.MulMod64(baseValue, modulus);
-            ulong current = baseValue;
-            for (int i = 1; i < _count; i++)
-            {
-                current = current.MulMod64(square, modulus);
-                SetValue(i, current);
-            }
-        }
-
-        [MethodImpl(MethodImplOptions.AggressiveInlining)]
-        public ulong Get(int index)
-        {
-            // Windowed lookups never exceed the populated table bounds on production workloads, so leave the check commented
-            // out to remove the branch.
-            // if ((uint)index >= (uint)_count)
-            // {
-            //     return 0UL;
-            // }
-
-            return index switch
-            {
-                0 => _value0,
-                1 => _value1,
-                2 => _value2,
-                3 => _value3,
-                4 => _value4,
-                5 => _value5,
-                6 => _value6,
-                7 => _value7,
-                8 => _value8,
-                9 => _value9,
-                10 => _value10,
-                11 => _value11,
-                12 => _value12,
-                13 => _value13,
-                14 => _value14,
-                _ => _value15,
-            };
-        }
-
-        [MethodImpl(MethodImplOptions.AggressiveInlining)]
-        private void SetValue(int index, ulong value)
-        {
-            switch (index)
-            {
-                case 1:
-                    _value1 = value;
-                    break;
-                case 2:
-                    _value2 = value;
-                    break;
-                case 3:
-                    _value3 = value;
-                    break;
-                case 4:
-                    _value4 = value;
-                    break;
-                case 5:
-                    _value5 = value;
-                    break;
-                case 6:
-                    _value6 = value;
-                    break;
-                case 7:
-                    _value7 = value;
-                    break;
-                case 8:
-                    _value8 = value;
-                    break;
-                case 9:
-                    _value9 = value;
-                    break;
-                case 10:
-                    _value10 = value;
-                    break;
-                case 11:
-                    _value11 = value;
-                    break;
-                case 12:
-                    _value12 = value;
-                    break;
-                case 13:
-                    _value13 = value;
-                    break;
-                case 14:
-                    _value14 = value;
-                    break;
-                case 15:
-                    _value15 = value;
-                    break;
-            }
-        }
-    }
-
-    private static void InitializeMontgomeryOddPowersCpu(in MontgomeryDivisorData divisor, ulong modulus, ulong nPrime, Span<ulong> destination)
-    {
-        int oddPowerCount = destination.Length;
-        if (oddPowerCount == 0)
-        {
-            return;
-        }
-
-        bool computedOnGpu = oddPowerCount <= PerfectNumberConstants.MaxOddPowersCount
-            && MontgomeryOddPowerGpu.TryCompute(divisor, oddPowerCount, destination);
-        if (computedOnGpu)
-        {
-            return;
-        }
-
-        destination[0] = divisor.MontgomeryTwo;
-        if (oddPowerCount == 1)
-        {
-            return;
-        }
-
-        ulong square = divisor.MontgomeryTwoSquared;
-        ulong current = destination[0];
-        for (int i = 1; i < oddPowerCount; i++)
-        {
-            current = current.MontgomeryMultiply(square, modulus, nPrime);
-            destination[i] = current;
-        }
-    }
-
-    [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    private static ulong[] InitializeMontgomeryOddPowersGpu(in MontgomeryDivisorData divisor, ulong modulus, ulong nPrime, int oddPowerCount)
+    private static ulong[] InitializeMontgomeryOddPowers(in MontgomeryDivisorData divisor, ulong modulus, ulong nPrime, int oddPowerCount)
     {
         ulong[] oddPowers = new ulong[PerfectNumberConstants.MaxOddPowersCount];
         oddPowers[0] = divisor.MontgomeryTwo;
@@ -988,31 +481,17 @@ public static class ULongExtensions
     }
 
     [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    public static ulong Pow2MontgomeryModWithCycleCpu(this ulong exponent, ulong cycleLength, in MontgomeryDivisorData divisor)
+    public static ulong Pow2MontgomeryModWithCycle(this ulong exponent, ulong cycleLength, in MontgomeryDivisorData divisor)
     {
         ulong rotationCount = exponent % cycleLength;
-        return Pow2MontgomeryModWindowedCpu(rotationCount, divisor, keepMontgomery: false);
+        return Pow2MontgomeryModWindowed(rotationCount, divisor, keepMontgomery: false);
     }
 
     [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    public static ulong Pow2MontgomeryModFromCycleRemainderCpu(this ulong reducedExponent, in MontgomeryDivisorData divisor)
+    public static ulong Pow2MontgomeryModFromCycleRemainder(this ulong reducedExponent, in MontgomeryDivisorData divisor)
     {
-        return Pow2MontgomeryModWindowedCpu(reducedExponent, divisor, keepMontgomery: false);
+        return Pow2MontgomeryModWindowed(reducedExponent, divisor, keepMontgomery: false);
     }
-
-    [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    public static ulong Pow2MontgomeryModWithCycleGpu(this ulong exponent, ulong cycleLength, in MontgomeryDivisorData divisor)
-    {
-        ulong rotationCount = exponent % cycleLength;
-        return Pow2MontgomeryModWindowedGpu(rotationCount, divisor, keepMontgomery: false);
-    }
-
-    [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    public static ulong Pow2MontgomeryModFromCycleRemainderGpu(this ulong reducedExponent, in MontgomeryDivisorData divisor)
-    {
-        return Pow2MontgomeryModWindowedGpu(reducedExponent, divisor, keepMontgomery: false);
-    }
-
     [MethodImpl(MethodImplOptions.AggressiveInlining)]
     public static UInt128 PowMod(this ulong exponent, UInt128 modulus)
     {

apply_patch.cs

@@ -335,13 +335,10 @@ if (patchResults.Any(static r => !r.Success))
 
 static List<PatchFile> ParsePatchFromFile(string patchFilePath)
 {
-    var stream = File.OpenRead(patchFilePath);
-    var reader = new StreamReader(stream, Encoding.UTF8, detectEncodingFromByteOrderMarks: true, bufferSize: 16_384, leaveOpen: false);
-    var lineReader = new PatchLineReader(reader);
+    using var stream = File.OpenRead(patchFilePath);
+    using var reader = new StreamReader(stream, Encoding.UTF8, detectEncodingFromByteOrderMarks: true, bufferSize: 16_384, leaveOpen: false);
+    using var lineReader = new PatchLineReader(reader);
     var files = ParsePatch(lineReader);
-    lineReader.Dispose();
-    reader.Dispose();
-    stream.Dispose();
     return files;
 }
 
@@ -503,7 +500,7 @@ static void FinalizePatchFile(
     files.Add(new PatchFile(originalPath, modifiedPath, hunks ?? new List<PatchHunk>(), type));
 }
 
-private sealed class PatchLineReader
+private sealed class PatchLineReader : IDisposable
 {
     private const int ReadBufferSize = 4096;
     private const int InitialLineBufferSize = 256;

todo_list.txt

@@ -122,6 +122,7 @@ PerfectNumbers.Core\Cpu\MersenneNumberDivisorByDivisorCpuTester.cs:203:
 PerfectNumbers.Core\Cpu\MersenneNumberDivisorByDivisorCpuTester.cs:207:                // TODO: When divisorCycle is zero (cache miss), compute only that single cycle on the device
 PerfectNumbers.Core\Cpu\MersenneNumberDivisorByDivisorCpuTester.cs:214:                // TODO: Remove this no-cycle branch and require on-demand cycle computation (without
 PerfectNumbers.Core\Cpu\MersenneNumberDivisorByDivisorCpuTester.cs:240:            // TODO: Replace this linear increment with the batched divisor-cycle walker validated in the
+PerfectNumbers.Core\Cpu\MersenneNumberDivisorByDivisorCpuTester.cs:262:                : prime.Pow2MontgomeryMod(divisorData); // TODO: Switch to the upcoming ProcessEightBitWindows helper once the
 PerfectNumbers.Core\Cpu\MersenneNumberDivisorByDivisorCpuTester.cs:283:        // TODO: Replace this modulo with the ring-buffer style counter (subtract loop) used in the fast CLI
 PerfectNumbers.Core\Cpu\MersenneNumberDivisorByDivisorCpuTester.cs:356:                // TODO: Replace this guard with an on-demand CPU/GPU cycle computation path that skips
 PerfectNumbers.Core\Cpu\MersenneNumberDivisorByDivisorCpuTester.cs:396:            // TODO: Remove this no-cycle fallback once divisor cycles are mandatory; the per-prime pow2 ladder here
@@ -265,6 +266,7 @@ todo_list.txt:121:PerfectNumbers.Core\Cpu\MersenneNumberDivisorByDivisorCpuTeste
 todo_list.txt:122:PerfectNumbers.Core\Cpu\MersenneNumberDivisorByDivisorCpuTester.cs:207:                // TODO: When divisorCycle is zero (cache miss), compute only that single cycle on the device
 todo_list.txt:123:PerfectNumbers.Core\Cpu\MersenneNumberDivisorByDivisorCpuTester.cs:214:                // TODO: Remove this no-cycle branch and require on-demand cycle computation (without
 todo_list.txt:124:PerfectNumbers.Core\Cpu\MersenneNumberDivisorByDivisorCpuTester.cs:240:            // TODO: Replace this linear increment with the batched divisor-cycle walker validated in the
+todo_list.txt:125:PerfectNumbers.Core\Cpu\MersenneNumberDivisorByDivisorCpuTester.cs:262:                : prime.Pow2MontgomeryMod(divisorData); // TODO: Switch to the upcoming ProcessEightBitWindows helper once the
 todo_list.txt:126:PerfectNumbers.Core\Cpu\MersenneNumberDivisorByDivisorCpuTester.cs:283:        // TODO: Replace this modulo with the ring-buffer style counter (subtract loop) used in the fast CLI
 todo_list.txt:127:PerfectNumbers.Core\Cpu\MersenneNumberDivisorByDivisorCpuTester.cs:356:                // TODO: Replace this guard with an on-demand CPU/GPU cycle computation path that skips
 todo_list.txt:128:PerfectNumbers.Core\Cpu\MersenneNumberDivisorByDivisorCpuTester.cs:396:            // TODO: Remove this no-cycle fallback once divisor cycles are mandatory; the per-prime pow2 ladder here
@@ -405,6 +407,17 @@ PerfectNumbers.Core\ModResidueTracker.cs:257:                        // TODO: Us
 PerfectNumbers.Core\ModResidueTracker.cs:263:                // TODO: Consult MersenneDivisorCycles.Shared (or similar caches) here so we reuse precomputed cycle lengths
 PerfectNumbers.Core\ModResidueTracker.cs:295:                                // TODO: Route this through the planned UInt128 windowed powmod so the inner MulMod adopts the
 PerfectNumbers.Core\ModResidueTracker.cs:314:                // TODO: Replace this double-and-add fallback with the UInt128 intrinsic-backed multiplier once the
+PerfectNumbers.Core\Gpu\MersenneNumberDivisorByDivisorGpuTester.cs:49:        set => _useDivisorCycles = value; // TODO: Remove this flag once GPU runs unconditionally fetch cycle lengths so every divisor check benefits from the measured speedups.
+PerfectNumbers.Core\Gpu\MersenneNumberDivisorByDivisorGpuTester.cs:298:                // TODO: Stage MontgomeryDivisorData in a GPU-ready cache keyed by divisor so we can copy
+PerfectNumbers.Core\Gpu\MersenneNumberDivisorByDivisorGpuTester.cs:372:        // TODO: Replace the block lease with a direct single-cycle lookup so the GPU path
+PerfectNumbers.Core\Gpu\MersenneNumberDivisorByDivisorGpuTester.cs:395:                        // TODO: Drop this block hand-off once the cache exposes a direct
+PerfectNumbers.Core\Gpu\MersenneNumberDivisorByDivisorGpuTester.cs:409:                            // TODO: Replace this `%` with the cycle-cache helper that subtracts precomputed
+PerfectNumbers.Core\Gpu\MersenneNumberDivisorByDivisorGpuTester.cs:425:                        // TODO: Trigger an immediate GPU computation for the single missing cycle without storing
+PerfectNumbers.Core\Gpu\MersenneNumberDivisorByDivisorGpuTester.cs:476:                    // TODO: Preload ProcessEightBitWindows-friendly divisor metadata for GPU batches so this
+PerfectNumbers.Core\Gpu\MersenneNumberDivisorByDivisorGpuTester.cs:563:        hits[index] = exponent.Pow2MontgomeryMod(divisor) == 1UL ? (byte)1 : (byte)0; // TODO: Replace Pow2MontgomeryMod with the ProcessEightBitWindows helper once available so GPU by-divisor scans reuse the benchmarked fast pow2 ladder.
+PerfectNumbers.Core\Gpu\MersenneNumberDivisorByDivisorGpuTester.cs:632:                // TODO: Replace this guard with an on-demand cycle computation that runs on the device selected
+PerfectNumbers.Core\Gpu\MersenneNumberDivisorByDivisorGpuTester.cs:787:                        // TODO: Replace this `%` with the divisor-cycle remainder helper so we reuse cached
+PerfectNumbers.Core\Gpu\MersenneNumberDivisorByDivisorGpuTester.cs:858:        results[index] = exponent.Pow2MontgomeryModMontgomery(divisor); // TODO: Swap this Montgomery-domain pow2 with the windowed helper slated to replace the slower ladder in Pow2Minus1Mod benchmarks.
 PerfectNumbers.Core\MersenneNumberDivisorByDivisorTester.cs:43:                // TODO: Swap this per-run enumerator creation for the shared PrimeIterator so by-divisor scans reuse the
 PerfectNumbers.Core\MersenneNumberDivisorByDivisorTester.cs:219:                        // TODO: Swap Task.Factory.StartNew for the shared low-overhead work queue once the scanner's
 PerfectNumbers.Core\MersenneNumberDivisorByDivisorTester.cs:240:                                bool useDivisorCycles = tester.UseDivisorCycles; // TODO: Remove the conditional path once all testers always enable divisor cycles to keep every divisor scan on the faster cached-length track.

tools/GpuNumericTypeProbe.csx

@@ -54,11 +54,10 @@ static ProbeResult ProbeBigInteger(Accelerator accelerator)
     try
     {
         var kernel = accelerator.LoadAutoGroupedStreamKernel<Index1D, ArrayView1D<int, Stride1D.Dense>>(BigIntegerKernel);
-        var buffer = accelerator.Allocate1D<int>(1);
+        using var buffer = accelerator.Allocate1D<int>(1);
         var extent = new Index1D((int)buffer.Length);
         kernel(extent, buffer.View);
         accelerator.Synchronize();
-        buffer.Dispose();
         return ProbeResult.Success(
             "System.Numerics.BigInteger",
             "Addition, subtraction, multiplication, division, and modulo compile on the CPU accelerator.");
@@ -74,10 +73,9 @@ static ProbeResult ProbeEInteger(Accelerator accelerator)
     try
     {
         var kernel = accelerator.LoadAutoGroupedStreamKernel<Index1D, ArrayView1D<int, Stride1D.Dense>>(EIntegerKernel);
-        var buffer = accelerator.Allocate1D<int>(1);
+        using var buffer = accelerator.Allocate1D<int>(1);
         kernel(new Index1D((int)buffer.Length), buffer.View);
         accelerator.Synchronize();
-        buffer.Dispose();
         return ProbeResult.Success(
             "PeterO.Numbers.EInteger",
             "Addition, subtraction, multiplication, division, and remainder compile on the CPU accelerator.");
@@ -95,7 +93,7 @@ static ProbeResult ProbeERational(Accelerator accelerator)
     try
     {
         var kernel = accelerator.LoadAutoGroupedStreamKernel<Index1D, ArrayView1D<int, Stride1D.Dense>>(ERationalKernel);
-        var buffer = accelerator.Allocate1D<int>(1);
+        using var buffer = accelerator.Allocate1D<int>(1);
         kernel(new Index1D((int)buffer.Length), buffer.View);
         accelerator.Synchronize();
 
@@ -103,7 +101,6 @@ static ProbeResult ProbeERational(Accelerator accelerator)
         details += hasRemainder
             ? " ERational exposes a remainder API that requires further investigation."
             : " ERational does not expose a remainder/modulo API.";
-        buffer.Dispose();
         return ProbeResult.Success("PeterO.Numbers.ERational", details);
     }
     catch (Exception ex)
@@ -123,10 +120,9 @@ static ProbeResult ProbeOpenNumericPrimesIsPrime(Accelerator accelerator)
     try
     {
         var kernel = accelerator.LoadAutoGroupedStreamKernel<Index1D, ArrayView1D<int, Stride1D.Dense>>(OpenNumericPrimesIsPrimeKernel);
-        var buffer = accelerator.Allocate1D<int>(4);
+        using var buffer = accelerator.Allocate1D<int>(4);
         kernel(new Index1D((int)buffer.Length), buffer.View);
         accelerator.Synchronize();
-        buffer.Dispose();
         return ProbeResult.Success(
             "Open.Numeric.Primes.Prime.Numbers.IsPrime",
             "IsPrime compiles and runs on the CPU accelerator kernel.");

tools/HeuristicMultiplyShiftTiming/HeuristicMultiplyShiftTiming.csproj

@@ -8,7 +8,4 @@
   <ItemGroup>
     <ProjectReference Include="../../PerfectNumbers.Core/PerfectNumbers.Core.csproj" />
   </ItemGroup>
-  <ItemGroup>
-    <Compile Remove=".temp.cs" />
-  </ItemGroup>
-</Project>
+</Project>