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

Clarify windowed pow2 invariants and tidy Montgomery pooling

Browse Source
This commit is contained in:
Marek Lemańczyk committed 2025-10-13 14:30:06 +02:00
1 parent 79695b0c56
commit 3f5904c340
1 file changed
+18 -12
Download Patch File Download Diff File Expand all files Collapse all files
PerfectNumbers.Core/ULongExtensions.cs
+18 -12
View File
@@ -79,7 +79,7 @@ public static class ULongExtensions
public static ulong MultiplyShiftRightShiftFirst(ulong value, ulong multiplier, int shift) public static ulong MultiplyShiftRightShiftFirst(ulong value, ulong multiplier, int shift)
{ {
ulong high = value >> shift; ulong high = value >> shift;
ulong mask = (1UL << shift) - 1UL; ulong mask = (1UL << shift) - 1UL;
ulong low = value & mask; ulong low = value & mask;
UInt128 highContribution = (UInt128)high * multiplier; UInt128 highContribution = (UInt128)high * multiplier;
@@ -330,7 +330,7 @@ public static class ULongExtensions
// return 1UL % modulus; // return 1UL % modulus;
// } // }
// Even though the by-divisor pipeline only feeds very large exponents, the legacy callers and unit tests // 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. // still hit this small-exponent fast path, so keep the fallback in place.
if (exponent <= Pow2WindowFallbackThreshold) if (exponent <= Pow2WindowFallbackThreshold)
{ {
@@ -426,15 +426,15 @@ public static class ULongExtensions
// return 0UL; // return 0UL;
// } // }
// Exponents fed into the GPU kernels are positive on the scanning path, yet the zero case remains for // The GPU pipeline only emits positive exponents on production workloads, so skip the zero guard to avoid an extra branch.
// completeness when unit tests probe boundary behavior. // Diagnostic harnesses must normalize their inputs before invoking this helper.
if (exponent == 0UL) // if (exponent == 0UL)
{ // {
return 1UL; // return 1UL;
} // }
// The GPU kernels mostly process large exponents as well, but the binary fallback is still useful when // The GPU kernels mostly process large exponents as well, but the binary fallback still serves
// unit tests probe edge cases around the threshold. // diagnostic harnesses that probe edge cases around the threshold.
if (exponent <= Pow2WindowFallbackThreshold) if (exponent <= Pow2WindowFallbackThreshold)
{ {
return Pow2ModBinaryGpuFallback(exponent, modulus); return Pow2ModBinaryGpuFallback(exponent, modulus);
@@ -548,10 +548,16 @@ public static class ULongExtensions
ulong result = divisor.MontgomeryOne; ulong result = divisor.MontgomeryOne;
ulong nPrime = divisor.NPrime; ulong nPrime = divisor.NPrime;
Span<ulong> oddPowerStorage = oddPowerCount <= PerfectNumberConstants.MaxOddPowersCount ? stackalloc ulong[PerfectNumberConstants.MaxOddPowersCount] : default; Span<ulong> oddPowerStorage = oddPowerCount <= PerfectNumberConstants.MaxOddPowersCount
? stackalloc ulong[PerfectNumberConstants.MaxOddPowersCount]
: default;
ulong[]? pooledOddPowers = null; ulong[]? pooledOddPowers = null;
if (oddPowerCount < PerfectNumberConstants.PooledArrayThreshold) if (oddPowerCount <= PerfectNumberConstants.MaxOddPowersCount)
{
oddPowerStorage = oddPowerStorage[..oddPowerCount];
}
else if (oddPowerCount < PerfectNumberConstants.PooledArrayThreshold)
{ {
oddPowerStorage = new ulong[oddPowerCount]; oddPowerStorage = new ulong[oddPowerCount];
} }