mlemanczyk/even-perfect-numbers-scanner
1
0 Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Route prime-order pow2 kernel through plain modular helper

Browse Source
This commit is contained in:
Marek Lemańczyk committed 2025-10-13 05:08:34 +02:00
1 parent 7ef0906a9b
commit 716ffcb243
7 files changed
+78 -79
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

PerfectNumbers.Core/Cpu/MersenneNumberDivisorByDivisorCpuTester.cs
+8 -9
View File
@@ -652,7 +652,7 @@ public sealed class MersenneNumberDivisorByDivisorCpuTester : IMersenneNumberDiv
private bool _disposed;
private ulong[]? _primeDeltas;
private ulong[]? _cycleRemainders;
private ulong[]? _montgomeryDeltas;
private ulong[]? _residues;
private int _bufferCapacity;
private MersenneNumberDivisorResidueGpuEvaluator? _gpuResidueEvaluator;
@@ -765,13 +765,12 @@ public sealed class MersenneNumberDivisorByDivisorCpuTester : IMersenneNumberDiv
ComputeCycleRemainders(divisorCycle, primes, deltaSpan, remainderSpan);
EnsureGpuEvaluator();
Span<ulong> residueSpan = _montgomeryDeltas!.AsSpan(0, length);
Span<ulong> residueSpan = _residues!.AsSpan(0, length);
_gpuResidueEvaluator!.ComputeResidues(primes, divisorData, residueSpan);
ulong montgomeryOne = divisorData.MontgomeryOne;
for (int i = 0; i < length; i++)
{
hits[i] = remainderSpan[i] == 0UL && residueSpan[i] == montgomeryOne ? (byte)1 : (byte)0;
hits[i] = remainderSpan[i] == 0UL && residueSpan[i] == 1UL ? (byte)1 : (byte)0;
}
}
@@ -801,7 +800,7 @@ public sealed class MersenneNumberDivisorByDivisorCpuTester : IMersenneNumberDiv
return;
}
if (requiredLength <= _bufferCapacity && _primeDeltas is not null && _cycleRemainders is not null && _montgomeryDeltas is not null)
if (requiredLength <= _bufferCapacity && _primeDeltas is not null && _cycleRemainders is not null && _residues is not null)
{
return;
}
@@ -821,7 +820,7 @@ public sealed class MersenneNumberDivisorByDivisorCpuTester : IMersenneNumberDiv
_primeDeltas = ArrayPool<ulong>.Shared.Rent(newCapacity);
_cycleRemainders = ArrayPool<ulong>.Shared.Rent(newCapacity);
_montgomeryDeltas = ArrayPool<ulong>.Shared.Rent(newCapacity);
_residues = ArrayPool<ulong>.Shared.Rent(newCapacity);
_bufferCapacity = newCapacity;
}
@@ -839,10 +838,10 @@ public sealed class MersenneNumberDivisorByDivisorCpuTester : IMersenneNumberDiv
_cycleRemainders = null;
}
if (_montgomeryDeltas is not null)
if (_residues is not null)
{
ArrayPool<ulong>.Shared.Return(_montgomeryDeltas, clearArray: false);
_montgomeryDeltas = null;
ArrayPool<ulong>.Shared.Return(_residues, clearArray: false);
_residues = null;
}
_bufferCapacity = 0;
PerfectNumbers.Core/Gpu/MersenneNumberDivisorByDivisorGpuTester.cs
+4 -4
View File
@@ -35,7 +35,7 @@ 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>>(ComputeMontgomeryExponentKernel));
_kernelExponentCache.GetOrAdd(accelerator, acc => acc.LoadAutoGroupedStreamKernel<Index1D, MontgomeryDivisorData, ArrayView<ulong>, ArrayView<ulong>>(ComputeExponentKernel));
private Action<Index1D, ArrayView<ulong>, ArrayView<byte>, byte> GetRemainderDeltaKernel(Accelerator accelerator) =>
_remainderDeltaKernelCache.GetOrAdd(accelerator, acc => acc.LoadAutoGroupedStreamKernel<Index1D, ArrayView<ulong>, ArrayView<byte>, byte>(ComputeRemainderDeltasKernel));
@@ -569,7 +569,7 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
}
ulong exponent = exponents[index];
hits[index] = exponent.Pow2MontgomeryModWindowedGpu(divisor, keepMontgomery: false) == 1UL ? (byte)1 : (byte)0;
hits[index] = exponent.Pow2ModBinaryGpu(modulus) == 1UL ? (byte)1 : (byte)0;
}
public sealed class DivisorScanSession : IMersenneNumberDivisorByDivisorTester.IDivisorScanSession
@@ -777,7 +777,7 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
mask[globalIndex] = 1;
}
private static void ComputeMontgomeryExponentKernel(Index1D index, MontgomeryDivisorData divisor, ArrayView<ulong> exponents, ArrayView<ulong> results)
private static void ComputeExponentKernel(Index1D index, MontgomeryDivisorData divisor, ArrayView<ulong> exponents, ArrayView<ulong> results)
{
ulong modulus = divisor.Modulus;
if (modulus <= 1UL || (modulus & 1UL) == 0UL)
@@ -788,7 +788,7 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
ulong exponent = exponents[index];
results[index] = exponent.Pow2MontgomeryModWindowedGpu(divisor, keepMontgomery: true);
results[index] = exponent.Pow2ModBinaryGpu(modulus);
}
public ulong DivisorLimit
PerfectNumbers.Core/Gpu/MersenneNumberDivisorResidueGpuEvaluator.cs
+8 -1
View File
@@ -109,6 +109,13 @@ internal sealed class MersenneNumberDivisorResidueGpuEvaluator : IDisposable
}
ulong exponent = exponents[idx];
results[idx] = exponent.Pow2MontgomeryModWindowedGpu(divisorData, keepMontgomery: true);
ulong modulus = divisorData.Modulus;
if (modulus <= 1UL || (modulus & 1UL) == 0UL)
{
results[idx] = 0UL;
return;
}
results[idx] = exponent.Pow2ModBinaryGpu(modulus);
}
}
PerfectNumbers.Core/Gpu/PrimeOrderGpuHeuristics.cs
+9 -52
View File
@@ -811,7 +811,7 @@ internal static partial class PrimeOrderGpuHeuristics
}
ulong reduced = phi / factor;
if (reduced.Pow2MontgomeryModWindowedGpu(divisor, keepMontgomery: false) == 1UL)
if (reduced.Pow2ModBinaryGpu(divisor.Modulus) == 1UL)
{
return false;
}
@@ -1595,7 +1595,7 @@ internal static partial class PrimeOrderGpuHeuristics
private static bool Pow2EqualsOneKernel(ulong exponent, in MontgomeryDivisorData divisor)
{
return exponent.Pow2MontgomeryModWindowedGpu(divisor, keepMontgomery: false) == 1UL;
return exponent.Pow2ModBinaryGpu(divisor.Modulus) == 1UL;
}
private static ulong CalculateByDoublingKernel(ulong prime)
@@ -1810,58 +1810,15 @@ internal static partial class PrimeOrderGpuHeuristics
private static void Pow2ModKernel(Index1D index, ArrayView1D<ulong, Stride1D.Dense> exponents, MontgomeryDivisorData divisor, ArrayView1D<ulong, Stride1D.Dense> remainders)
{
ulong exponent = exponents[index];
ulong modulus = divisor.Modulus;
// ulong nPrime = divisor.NPrime;
// ulong result = divisor.MontgomeryOne;
// ulong baseValue = divisor.MontgomeryTwo;
// ulong remainingExponent = exponent;
if (modulus <= 1UL || (modulus & 1UL) == 0UL)
{
remainders[index] = 0UL;
return;
}
// if (remainingExponent == 0UL)
// {
// remainders[index] = result.MontgomeryMultiply(1UL, modulus, nPrime);
// return;
// }
// while (remainingExponent != 0UL)
// {
// if ((remainingExponent & 1UL) != 0UL)
// {
// result = result.MontgomeryMultiply(baseValue, modulus, nPrime);
// }
// remainingExponent >>= 1;
// if (remainingExponent == 0UL)
// {
// break;
// }
// // Reusing baseValue to hold base^2 so we avoid allocating another register for the squared term.
// baseValue = baseValue.MontgomeryMultiply(baseValue, modulus, nPrime);
// }
// remainders[index] = result.MontgomeryMultiply(1UL, modulus, nPrime);
// Plain modular exponentiation without Montgomery reduction kept for reference.
ulong plainResult = 1UL;// % modulus;
ulong plainBase = 2UL;// % modulus;
while (exponent != 0UL)
{
if ((exponent & 1UL) != 0UL)
{
plainResult = (plainResult * plainBase) % modulus;
}
exponent >>= 1;
if (exponent == 0UL)
{
break;
}
plainBase = (plainBase * plainBase) % modulus;
}
remainders[index] = plainResult;
ulong exponent = exponents[index];
remainders[index] = exponent.Pow2ModBinaryGpu(modulus);
}
private static Action<AcceleratorStream, Index1D, ArrayView1D<GpuUInt128, Stride1D.Dense>, GpuUInt128, ArrayView1D<GpuUInt128, Stride1D.Dense>> GetPow2ModWideKernel(Accelerator accelerator)
PerfectNumbers.Core/ULongExtensions.cs
+31
View File
@@ -310,6 +310,37 @@ public static class ULongExtensions
return state.MulModWithNativeModulo(b, modulus);
}
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public static ulong Pow2ModBinaryGpu(this ulong exponent, ulong modulus)
{
if (modulus <= 1UL)
{
return 0UL;
}
ulong remainingExponent = exponent;
ulong result = 1UL % modulus;
ulong baseValue = 2UL % modulus;
while (remainingExponent != 0UL)
{
if ((remainingExponent & 1UL) != 0UL)
{
result = result.MulMod64GpuCompatible(baseValue, modulus);
}
remainingExponent >>= 1;
if (remainingExponent == 0UL)
{
break;
}
baseValue = baseValue.MulMod64GpuCompatible(baseValue, modulus);
}
return result;
}
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public static ulong MontgomeryMultiply(this ulong a, ulong b, ulong modulus, ulong nPrime)
{