Rented pools made thread static

This commit is contained in:
mlemanczyk committed 2025-10-11 14:14:28 +02:00
1 parent 1cbf49c8b5
commit 12140bca0a
1 file changed
+62 -51
+62 -51
View File
@@ -233,59 +233,69 @@ internal static partial class PrimeOrderCalculator
return order; return order;
} }
private static ulong ExponentLoweringCpu(ulong order, ulong prime, PartialFactorResult factors, in MontgomeryDivisorData divisorData) [ThreadStatic]
{ private static ArrayPool<FactorEntry>? _factorPool;
FactorEntry[]? tempArray = null; private static ArrayPool<FactorEntry> FactorPool => _factorPool ??= ArrayPool<FactorEntry>.Create();
try
{
ReadOnlySpan<FactorEntry> factorSpan = factors.Factors;
int length = factors.Count;
tempArray = ArrayPool<FactorEntry>.Shared.Rent(length + 1);
Span<FactorEntry> buffer = tempArray;
factorSpan.CopyTo(buffer);
bool isPrime = Open.Numeric.Primes.Prime.Numbers.IsPrime(factors.Cofactor); [ThreadStatic]
// bool isPrime = PrimeTester.IsPrimeInternal(factors.Cofactor, CancellationToken.None); private static ArrayPool<ulong>? _ulongPool;
if (!factors.FullyFactored && factors.Cofactor > 1UL && isPrime) private static ArrayPool<ulong> ULongPool => _ulongPool ??= ArrayPool<ulong>.Create();
{
buffer[length] = new FactorEntry(factors.Cofactor);
length++;
}
buffer.Slice(0, length).Sort(static (a, b) => a.Value.CompareTo(b.Value)); private static ulong ExponentLoweringCpu(ulong order, ulong prime, PartialFactorResult factors, in MontgomeryDivisorData divisorData)
{
ArrayPool<FactorEntry> factorPool = FactorPool;
for (int i = 0; i < length; i++) FactorEntry[]? tempArray = null;
{ try
ulong primeFactor = buffer[i].Value; {
int exponent = buffer[i].Exponent; ReadOnlySpan<FactorEntry> factorSpan = factors.Factors;
for (int iteration = 0; iteration < exponent; iteration++) int length = factors.Count;
{ tempArray = factorPool.Rent(length + 1);
if ((order % primeFactor) != 0UL) Span<FactorEntry> buffer = tempArray;
{ factorSpan.CopyTo(buffer);
break;
}
ulong reduced = order / primeFactor; bool isPrime = Open.Numeric.Primes.Prime.Numbers.IsPrime(factors.Cofactor);
if (Pow2EqualsOneCpu(reduced, prime, divisorData)) // bool isPrime = PrimeTester.IsPrimeInternal(factors.Cofactor, CancellationToken.None);
{ if (!factors.FullyFactored && factors.Cofactor > 1UL && isPrime)
order = reduced; {
continue; buffer[length] = new FactorEntry(factors.Cofactor);
} length++;
}
break; buffer.Slice(0, length).Sort(static (a, b) => a.Value.CompareTo(b.Value));
}
}
return order; for (int i = 0; i < length; i++)
} {
finally ulong primeFactor = buffer[i].Value;
{ int exponent = buffer[i].Exponent;
if (tempArray is not null) for (int iteration = 0; iteration < exponent; iteration++)
{ {
ArrayPool<FactorEntry>.Shared.Return(tempArray, clearArray: false); if ((order % primeFactor) != 0UL)
} {
} break;
} }
ulong reduced = order / primeFactor;
if (Pow2EqualsOneCpu(reduced, prime, divisorData))
{
order = reduced;
continue;
}
break;
}
}
return order;
}
finally
{
if (tempArray is not null)
{
factorPool.Return(tempArray, clearArray: false);
}
}
}
private static bool TryConfirmOrderCpu(ulong prime, ulong order, in MontgomeryDivisorData divisorData, in PrimeOrderSearchConfig config) private static bool TryConfirmOrderCpu(ulong prime, ulong order, in MontgomeryDivisorData divisorData, in PrimeOrderSearchConfig config)
{ {
@@ -431,6 +441,7 @@ internal static partial class PrimeOrderCalculator
const int MaxGpuBatchSize = 1024; const int MaxGpuBatchSize = 1024;
const int StackGpuBatchSize = 1024; const int StackGpuBatchSize = 1024;
Span<ulong> stackGpuRemainders = stackalloc ulong[StackGpuBatchSize]; Span<ulong> stackGpuRemainders = stackalloc ulong[StackGpuBatchSize];
ArrayPool<ulong> uLongPool = ULongPool;
while (index < candidateCount && powUsed < powBudget) while (index < candidateCount && powUsed < powBudget)
{ {
int remaining = candidateCount - index; int remaining = candidateCount - index;
@@ -462,7 +473,7 @@ internal static partial class PrimeOrderCalculator
} }
else else
{ {
gpuPool = ArrayPool<ulong>.Shared.Rent(batchSize); gpuPool = uLongPool.Rent(batchSize);
Span<ulong> pooledRemainders = gpuPool.AsSpan(0, batchSize); Span<ulong> pooledRemainders = gpuPool.AsSpan(0, batchSize);
status = PrimeOrderGpuHeuristics.TryPow2ModBatch(batch, prime, pooledRemainders, divisorData); status = PrimeOrderGpuHeuristics.TryPow2ModBatch(batch, prime, pooledRemainders, divisorData);
if (status == GpuPow2ModStatus.Success) if (status == GpuPow2ModStatus.Success)
@@ -472,7 +483,7 @@ internal static partial class PrimeOrderCalculator
} }
else else
{ {
ArrayPool<ulong>.Shared.Return(gpuPool, clearArray: false); uLongPool.Return(gpuPool, clearArray: false);
gpuPool = null; gpuPool = null;
} }
} }
@@ -510,7 +521,7 @@ internal static partial class PrimeOrderCalculator
if (gpuPool is not null) if (gpuPool is not null)
{ {
ArrayPool<ulong>.Shared.Return(gpuPool, clearArray: false); uLongPool.Return(gpuPool, clearArray: false);
} }
result = candidate; result = candidate;
@@ -519,7 +530,7 @@ internal static partial class PrimeOrderCalculator
if (gpuPool is not null) if (gpuPool is not null)
{ {
ArrayPool<ulong>.Shared.Return(gpuPool, clearArray: false); uLongPool.Return(gpuPool, clearArray: false);
} }
index += batchSize; index += batchSize;