Trim GPU exponent batches using cached divisor cycles

This commit is contained in:
Marek Lemańczyk committed 2025-09-25 20:23:25 +02:00
1 parent 35691ff710
commit 9b7611fe62
5 files changed
+353 -111

No files matched your search

+32
View File
@@ -246,6 +246,38 @@ public static class ULongExtensions
return result.MontgomeryMultiply(1UL, modulus, nPrime);
}
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public static ulong Pow2MontgomeryModMontgomery(this ulong exponent, in MontgomeryDivisorData divisor)
{
ulong modulus = divisor.Modulus;
if (modulus <= 1UL || (modulus & 1UL) == 0UL)
{
return 0UL;
}
ulong result = divisor.MontgomeryOne;
ulong baseVal = divisor.MontgomeryTwo;
ulong remainingExponent = exponent;
while (remainingExponent > 0UL)
{
if ((remainingExponent & 1UL) != 0UL)
{
result = result.MontgomeryMultiply(baseVal, modulus, divisor.NPrime);
}
remainingExponent >>= 1;
if (remainingExponent == 0UL)
{
break;
}
baseVal = baseVal.MontgomeryMultiply(baseVal, modulus, divisor.NPrime);
}
return result;
}
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public static ulong Pow2MontgomeryModWithCycle(this ulong exponent, ulong cycleLength, in MontgomeryDivisorData divisor)
{