Use field-based exponent tracking for steppers

PerfectNumbers.Core/ExponentRemainderStepperCpu.cs

@@ -10,7 +10,7 @@ internal struct ExponentRemainderStepperCpu
     private readonly ulong _modulus;
     private readonly ulong _nPrime;
     private readonly ulong _montgomeryOne;
-    private ulong _previousExponent;
+    public ulong PreviousExponent;
     private ulong _currentMontgomery;
 
     public ExponentRemainderStepperCpu(in MontgomeryDivisorData divisor)
@@ -19,12 +19,10 @@ internal struct ExponentRemainderStepperCpu
         _modulus = divisor.Modulus;
         _nPrime = divisor.NPrime;
         _montgomeryOne = divisor.MontgomeryOne;
-        _previousExponent = 0UL;
+        PreviousExponent = 0UL;
         _currentMontgomery = divisor.MontgomeryOne;
     }
 
-    public ulong PreviousExponent => _previousExponent;
-
     public bool MatchesDivisor(in MontgomeryDivisorData divisor)
     {
         return _modulus == divisor.Modulus && _nPrime == divisor.NPrime && _montgomeryOne == divisor.MontgomeryOne;
@@ -36,7 +34,7 @@ internal struct ExponentRemainderStepperCpu
     /// </summary>
     public void Reset()
     {
-        _previousExponent = 0UL;
+        PreviousExponent = 0UL;
         _currentMontgomery = _montgomeryOne;
     }
 
@@ -71,13 +69,13 @@ internal struct ExponentRemainderStepperCpu
     /// <returns>The canonical residue produced by <c>2^exponent mod divisor</c>.</returns>
     public ulong ComputeNextCpu(ulong exponent)
     {
-        ulong delta = exponent - _previousExponent;
+        ulong delta = exponent - PreviousExponent;
         // TODO: Once divisor cycle lengths are mandatory, pull the delta multiplier from the
         // single-block divisor-cycle snapshot so we can skip the powmod entirely and reuse the
         // cached Montgomery residue ladder highlighted in MersenneDivisorCycleLengthGpuBenchmarks.
         ulong multiplier = delta.Pow2MontgomeryModWindowedCpu(_divisor, keepMontgomery: true);
         _currentMontgomery = _currentMontgomery.MontgomeryMultiply(multiplier, _modulus, _nPrime);
-        _previousExponent = exponent;
+        PreviousExponent = exponent;
         return ReduceCurrent();
     }
 
@@ -90,12 +88,12 @@ internal struct ExponentRemainderStepperCpu
     /// <returns><see langword="true"/> when the canonical residue equals one.</returns>
     public bool ComputeNextIsUnity(ulong exponent)
     {
-        ulong delta = exponent - _previousExponent;
+        ulong delta = exponent - PreviousExponent;
         // TODO: Reuse the divisor-cycle derived Montgomery delta once the cache exposes single-cycle
         // lookups so this branch also avoids recomputing powmods when the snapshot lacks the divisor.
         ulong multiplier = delta.Pow2MontgomeryModWindowedCpu(_divisor, keepMontgomery: true);
         _currentMontgomery = _currentMontgomery.MontgomeryMultiply(multiplier, _modulus, _nPrime);
-        _previousExponent = exponent;
+        PreviousExponent = exponent;
         return _currentMontgomery == _montgomeryOne;
     }
 
@@ -110,7 +108,7 @@ internal struct ExponentRemainderStepperCpu
     public bool TryInitializeFromMontgomeryResult(ulong exponent, ulong montgomeryResult, out bool isUnity)
     {
         _currentMontgomery = montgomeryResult;
-        _previousExponent = exponent;
+        PreviousExponent = exponent;
         isUnity = _currentMontgomery == _montgomeryOne;
         return true;
     }
@@ -129,7 +127,7 @@ internal struct ExponentRemainderStepperCpu
         // of relying on the caller-provided delta so incremental scans remain in sync with the snapshot
         // without computing additional cycles or mutating cache state.
         _currentMontgomery = _currentMontgomery.MontgomeryMultiply(montgomeryDelta, _modulus, _nPrime);
-        _previousExponent = exponent;
+        PreviousExponent = exponent;
         isUnity = _currentMontgomery == _montgomeryOne;
         return true;
     }
@@ -137,7 +135,7 @@ internal struct ExponentRemainderStepperCpu
     private void InitializeCpuState(ulong exponent)
     {
         _currentMontgomery = exponent.Pow2MontgomeryModWindowedCpu(_divisor, keepMontgomery: true);
-        _previousExponent = exponent;
+        PreviousExponent = exponent;
     }
 
     private ulong ReduceCurrent() => _currentMontgomery.MontgomeryMultiply(1UL, _modulus, _nPrime);

PerfectNumbers.Core/Gpu/ExponentRemainderStepperGpu.cs

@@ -13,23 +13,17 @@ internal struct ExponentRemainderStepperGpu
     private const int WindowSizeMax = 8;
 
     private readonly ulong _modulus;
-    private ulong _previousExponent;
+    public ulong PreviousExponent;
     private ulong _currentResidue;
 
     [MethodImpl(MethodImplOptions.AggressiveInlining)]
     public ExponentRemainderStepperGpu(in MontgomeryDivisorData divisor)
     {
         _modulus = divisor.Modulus;
-        _previousExponent = 0UL;
+        PreviousExponent = 0UL;
         _currentResidue = 1UL;
     }
 
-    public ulong PreviousExponent
-    {
-        [MethodImpl(MethodImplOptions.AggressiveInlining)]
-        get => _previousExponent;
-    }
-
     [MethodImpl(MethodImplOptions.AggressiveInlining)]
     public bool MatchesDivisor(in MontgomeryDivisorData divisor)
     {
@@ -43,7 +37,7 @@ internal struct ExponentRemainderStepperGpu
     [MethodImpl(MethodImplOptions.AggressiveInlining)]
     public void Reset()
     {
-        _previousExponent = 0UL;
+        PreviousExponent = 0UL;
         _currentResidue = 1UL;
     }
 
@@ -103,26 +97,26 @@ internal struct ExponentRemainderStepperGpu
     private void InitializeStateGpu(ulong exponent)
     {
         _currentResidue = Pow2ModWindowed(exponent, _modulus);
-        _previousExponent = exponent;
+        PreviousExponent = exponent;
     }
 
     [MethodImpl(MethodImplOptions.AggressiveInlining)]
     private void AdvanceStateGpu(ulong exponent)
     {
-        ulong delta = exponent - _previousExponent;
+        ulong delta = exponent - PreviousExponent;
         ulong multiplier = Pow2ModWindowed(delta, _modulus);
-        _currentResidue = _currentResidue.MulMod64Gpu(multiplier, _modulus);
+        _currentResidue = MultiplyMod(_currentResidue, multiplier, _modulus);
-        _previousExponent = exponent;
+        PreviousExponent = exponent;
     }
 
     [MethodImpl(MethodImplOptions.AggressiveInlining)]
     internal static ulong Pow2ModWindowed(ulong exponent, ulong modulus)
     {
-        if (exponent == 0UL)
+        // EvenPerfectBitScanner never provides zero exponents on the GPU path, so the guard stays commented out.
-        {
+        // if (exponent == 0UL)
-            // The GPU by-divisor flow never routes modulus ≤ 1 here because divisors follow q = 2kp + 1 ≥ 3.
+        // {
-            return 1UL;
+        //     return 1UL;
-        }
+        // }
 
         int bitLength = GetBitLength(exponent);
         int windowSize = GetWindowSize(bitLength);
@@ -132,7 +126,7 @@ internal struct ExponentRemainderStepperGpu
         while (index >= 0)
         {
             ulong currentBit = (exponent >> index) & 1UL;
-            ulong squared = result.MulMod64Gpu(result, modulus);
+            ulong squared = MultiplyMod(result, result, modulus);
             bool processWindow = currentBit != 0UL;
 
             result = processWindow ? result : squared;
@@ -147,7 +141,7 @@ internal struct ExponentRemainderStepperGpu
 
             for (int square = 0; square < windowLength; square++)
             {
-                result = result.MulMod64Gpu(result, modulus);
+                result = MultiplyMod(result, result, modulus);
             }
 
             if (!processWindow)
@@ -158,13 +152,21 @@ internal struct ExponentRemainderStepperGpu
             ulong mask = (1UL << windowLength) - 1UL;
             ulong windowValue = (exponent >> windowStart) & mask;
             ulong multiplier = ComputeWindowedOddPower(windowValue, modulus);
-            result = result.MulMod64Gpu(multiplier, modulus);
+            result = MultiplyMod(result, multiplier, modulus);
             index = windowStart - 1;
         }
 
         return result;
     }
 
+    [MethodImpl(MethodImplOptions.AggressiveInlining)]
+    private static ulong MultiplyMod(ulong left, ulong right, ulong modulus)
+    {
+        // GpuUInt128.MulMod is the fastest GPU-compatible multiply-reduce helper available to this kernel path.
+        var product = new GpuUInt128(left);
+        return product.MulMod(right, modulus);
+    }
+
     [MethodImpl(MethodImplOptions.AggressiveInlining)]
     private static ulong ComputeWindowedOddPower(ulong windowValue, ulong modulus)
     {
@@ -172,18 +174,19 @@ internal struct ExponentRemainderStepperGpu
         ulong baseValue = 2UL;
         if (windowValue == 1UL)
         {
+            // Single-bit windows occur frequently in production primes, so keep the fast return for window value 1.
             return baseValue;
         }
 
         ulong result = baseValue;
         ulong remaining = (windowValue - 1UL) >> 1;
-        ulong squareBase = baseValue.MulMod64Gpu(baseValue, modulus);
+        ulong squareBase = MultiplyMod(baseValue, baseValue, modulus);
 
         while (remaining != 0UL)
         {
-            result = ((remaining & 1UL) != 0UL) ? result.MulMod64Gpu(squareBase, modulus) : result;
+            result = ((remaining & 1UL) != 0UL) ? MultiplyMod(result, squareBase, modulus) : result;
             remaining >>= 1;
-            squareBase = (remaining != 0UL) ? squareBase.MulMod64Gpu(squareBase, modulus) : squareBase;
+            squareBase = (remaining != 0UL) ? MultiplyMod(squareBase, squareBase, modulus) : squareBase;
         }
 
         return result;