Isolate CPU divisor pools per thread

This commit is contained in:
Marek Lemańczyk committed 2025-10-11 21:17:12 +02:00
1 parent cd69c08fb4
commit 8e9390bc5e
7 files changed
+61 -34

No files matched your search

+14 -14
View File
@@ -63,7 +63,7 @@ public sealed class DivisorCycleCache
int[]? rentedMissing = null;
Span<int> missingBuffer = length <= StackBufferThreshold
? stackalloc int[length]
: new Span<int>(rentedMissing = ArrayPool<int>.Shared.Rent(length), 0, length);
: new Span<int>(rentedMissing = ThreadLocalArrayPool<int>.Shared.Rent(length), 0, length);
int missingCount = 0;
@@ -106,7 +106,7 @@ public sealed class DivisorCycleCache
if (rentedMissing is not null)
{
ArrayPool<int>.Shared.Return(rentedMissing, clearArray: false);
ThreadLocalArrayPool<int>.Shared.Return(rentedMissing, clearArray: false);
}
}
@@ -128,12 +128,12 @@ public sealed class DivisorCycleCache
ulong[]? rentedDivisors = null;
Span<ulong> workDivisors = count <= StackBufferThreshold
? stackalloc ulong[count]
: new Span<ulong>(rentedDivisors = ArrayPool<ulong>.Shared.Rent(count), 0, count);
: new Span<ulong>(rentedDivisors = ThreadLocalArrayPool<ulong>.Shared.Rent(count), 0, count);
ulong[]? rentedResults = null;
Span<ulong> workResults = count <= StackBufferThreshold
? stackalloc ulong[count]
: new Span<ulong>(rentedResults = ArrayPool<ulong>.Shared.Rent(count), 0, count);
: new Span<ulong>(rentedResults = ThreadLocalArrayPool<ulong>.Shared.Rent(count), 0, count);
for (int i = 0; i < count; i++)
{
@@ -149,8 +149,8 @@ public sealed class DivisorCycleCache
if (rentedDivisors is not null)
{
ArrayPool<ulong>.Shared.Return(rentedDivisors, clearArray: false);
ArrayPool<ulong>.Shared.Return(rentedResults!, clearArray: false);
ThreadLocalArrayPool<ulong>.Shared.Return(rentedDivisors, clearArray: false);
ThreadLocalArrayPool<ulong>.Shared.Return(rentedResults!, clearArray: false);
}
}
@@ -172,22 +172,22 @@ public sealed class DivisorCycleCache
ulong[]? rentedPow = null;
Span<ulong> powSpan = length <= StackBufferThreshold
? stackalloc ulong[length]
: new Span<ulong>(rentedPow = ArrayPool<ulong>.Shared.Rent(length), 0, length);
: new Span<ulong>(rentedPow = ThreadLocalArrayPool<ulong>.Shared.Rent(length), 0, length);
ulong[]? rentedOrder = null;
Span<ulong> orderSpan = length <= StackBufferThreshold
? stackalloc ulong[length]
: new Span<ulong>(rentedOrder = ArrayPool<ulong>.Shared.Rent(length), 0, length);
: new Span<ulong>(rentedOrder = ThreadLocalArrayPool<ulong>.Shared.Rent(length), 0, length);
ulong[]? rentedResult = null;
Span<ulong> resultSpan = length <= StackBufferThreshold
? stackalloc ulong[length]
: new Span<ulong>(rentedResult = ArrayPool<ulong>.Shared.Rent(length), 0, length);
: new Span<ulong>(rentedResult = ThreadLocalArrayPool<ulong>.Shared.Rent(length), 0, length);
byte[]? rentedStatus = null;
Span<byte> statusSpan = length <= StackBufferThreshold
? stackalloc byte[length]
: new Span<byte>(rentedStatus = ArrayPool<byte>.Shared.Rent(length), 0, length);
: new Span<byte>(rentedStatus = ThreadLocalArrayPool<byte>.Shared.Rent(length), 0, length);
ref ulong divisorRef = ref MemoryMarshal.GetReference(divisors);
divisorBuffer.View.CopyFromCPU(ref divisorRef, length);
@@ -234,10 +234,10 @@ public sealed class DivisorCycleCache
resultSpan.CopyTo(destination);
if (rentedPow is not null)
{
ArrayPool<ulong>.Shared.Return(rentedPow, clearArray: false);
ArrayPool<ulong>.Shared.Return(rentedOrder!, clearArray: false);
ArrayPool<ulong>.Shared.Return(rentedResult!, clearArray: false);
ArrayPool<byte>.Shared.Return(rentedStatus!, clearArray: false);
ThreadLocalArrayPool<ulong>.Shared.Return(rentedPow, clearArray: false);
ThreadLocalArrayPool<ulong>.Shared.Return(rentedOrder!, clearArray: false);
ThreadLocalArrayPool<ulong>.Shared.Return(rentedResult!, clearArray: false);
ThreadLocalArrayPool<byte>.Shared.Return(rentedStatus!, clearArray: false);
}
execution.Dispose();
+5 -5
View File
@@ -567,7 +567,7 @@ public class MersenneDivisorCycles
}
int count = factorCounts.Count;
ulong[] primes = ArrayPool<ulong>.Shared.Rent(count);
ulong[] primes = ThreadLocalArrayPool<ulong>.Shared.Rent(count);
try
{
int index = 0;
@@ -607,7 +607,7 @@ public class MersenneDivisorCycles
}
finally
{
ArrayPool<ulong>.Shared.Return(primes, clearArray: false);
ThreadLocalArrayPool<ulong>.Shared.Return(primes, clearArray: false);
}
}
@@ -793,7 +793,7 @@ public class MersenneDivisorCycles
tasks[taskIndex] = Task.Run(() =>
{
ulong rangeLength = localEnd - localStart + 1UL;
(ulong divisor, ulong cycle)[] localCycles = ArrayPool<(ulong, ulong)>.Shared.Rent(checked((int)rangeLength));
(ulong divisor, ulong cycle)[] localCycles = ThreadLocalArrayPool<(ulong, ulong)>.Shared.Rent(checked((int)rangeLength));
int localCycleIndex = 0;
try
@@ -841,7 +841,7 @@ public class MersenneDivisorCycles
}
finally
{
ArrayPool<(ulong, ulong)>.Shared.Return(localCycles, clearArray: false);
ThreadLocalArrayPool<(ulong, ulong)>.Shared.Return(localCycles, clearArray: false);
}
});
}
@@ -856,7 +856,7 @@ public class MersenneDivisorCycles
ulong start = 3UL;
var locker = new object();
var pool = ArrayPool<ulong>.Shared;
var pool = ThreadLocalArrayPool<ulong>.Shared;
ulong batchSizeUL = (ulong)batchSize, d, end;
int count = (int)Math.Min(batchSizeUL, maxDivisor), i, idx;
ulong[] divisors, outCycles, validDivisors;
+3 -3
View File
@@ -49,12 +49,12 @@ namespace PerfectNumbers.Core
FactorEntry[]? source = Factors;
if (source is null || Count == 0)
{
FactorEntry[] local = ArrayPool<FactorEntry>.Shared.Rent(1);
FactorEntry[] local = ThreadLocalArrayPool<FactorEntry>.Shared.Rent(1);
local[0] = new FactorEntry(prime);
return Rent(local, 1UL, true, 1);
}
FactorEntry[] extended = ArrayPool<FactorEntry>.Shared.Rent(Count + 1);
FactorEntry[] extended = ThreadLocalArrayPool<FactorEntry>.Shared.Rent(Count + 1);
Array.Copy(source, 0, extended, 0, Count);
extended[Count] = new FactorEntry(prime);
Span<FactorEntry> span = extended.AsSpan(0, Count + 1);
@@ -70,7 +70,7 @@ namespace PerfectNumbers.Core
if (factors is not null)
{
Factors = null;
ArrayPool<FactorEntry>.Shared.Return(factors, clearArray: false);
ThreadLocalArrayPool<FactorEntry>.Shared.Return(factors, clearArray: false);
}
_next = s_poolHead;
@@ -81,12 +81,12 @@ namespace PerfectNumbers.Core
{
if (_primes is not null)
{
ArrayPool<ulong>.Shared.Return(_primes, clearArray: false);
ThreadLocalArrayPool<ulong>.Shared.Return(_primes, clearArray: false);
}
if (_exponents is not null)
{
ArrayPool<byte>.Shared.Return(_exponents, clearArray: false);
ThreadLocalArrayPool<byte>.Shared.Return(_exponents, clearArray: false);
}
_arraysFromPool = false;
@@ -119,12 +119,12 @@ namespace PerfectNumbers.Core
{
if (_primes is not null)
{
ArrayPool<ulong>.Shared.Return(_primes, clearArray: false);
ThreadLocalArrayPool<ulong>.Shared.Return(_primes, clearArray: false);
}
if (_exponents is not null)
{
ArrayPool<byte>.Shared.Return(_exponents, clearArray: false);
ThreadLocalArrayPool<byte>.Shared.Return(_exponents, clearArray: false);
}
_arraysFromPool = false;
@@ -134,8 +134,8 @@ namespace PerfectNumbers.Core
if (capacity >= PerfectNumberConstants.PooledArrayThreshold)
{
_primes = ArrayPool<ulong>.Shared.Rent(capacity);
_exponents = ArrayPool<byte>.Shared.Rent(capacity);
_primes = ThreadLocalArrayPool<ulong>.Shared.Rent(capacity);
_exponents = ThreadLocalArrayPool<byte>.Shared.Rent(capacity);
_arraysFromPool = true;
}
else
@@ -162,7 +162,7 @@ internal static partial class PrimeOrderCalculator
return order;
}
tempArray = ArrayPool<FactorEntry128>.Shared.Rent(length + 1);
tempArray = ThreadLocalArrayPool<FactorEntry128>.Shared.Rent(length + 1);
Span<FactorEntry128> buffer = tempArray.AsSpan(0, length);
factorSpan.CopyTo(buffer);
@@ -200,7 +200,7 @@ internal static partial class PrimeOrderCalculator
{
if (tempArray is not null)
{
ArrayPool<FactorEntry128>.Shared.Return(tempArray, clearArray: false);
ThreadLocalArrayPool<FactorEntry128>.Shared.Return(tempArray, clearArray: false);
}
}
}
@@ -664,7 +664,7 @@ internal static partial class PrimeOrderCalculator
return new PartialFactorResult128(null, value, false, 0);
}
FactorEntry128[]? rented = counts.Count > 0 ? ArrayPool<FactorEntry128>.Shared.Rent(counts.Count) : null;
FactorEntry128[]? rented = counts.Count > 0 ? ThreadLocalArrayPool<FactorEntry128>.Shared.Rent(counts.Count) : null;
int index = 0;
if (rented is not null)
{
@@ -682,7 +682,7 @@ internal static partial class PrimeOrderCalculator
span.Sort(static (a, b) => a.Value.CompareTo(b.Value));
resultArray = new FactorEntry128[index];
span.CopyTo(resultArray);
ArrayPool<FactorEntry128>.Shared.Return(rented, clearArray: false);
ThreadLocalArrayPool<FactorEntry128>.Shared.Return(rented, clearArray: false);
}
bool fullyFactored = cofactor == UInt128.One;
+2 -2
View File
@@ -1058,7 +1058,7 @@ internal static partial class PrimeOrderCalculator
return PartialFactorResult.Rent(null, cofactor, cofactor == 1UL, 0);
}
FactorEntry[] factors = ArrayPool<FactorEntry>.Shared.Rent(counts.Count);
FactorEntry[] factors = ThreadLocalArrayPool<FactorEntry>.Shared.Rent(counts.Count);
int index = 0;
foreach (KeyValuePair<ulong, int> entry in counts)
{
@@ -1092,7 +1092,7 @@ internal static partial class PrimeOrderCalculator
return PartialFactorResult.Rent(null, cofactor, cofactor == 1UL, 0);
}
FactorEntry[] array = ArrayPool<FactorEntry>.Shared.Rent(actualCount);
FactorEntry[] array = ThreadLocalArrayPool<FactorEntry>.Shared.Rent(actualCount);
int arrayIndex = 0;
for (int i = 0; i < factorCount; i++)
{
@@ -0,0 +1,27 @@
using System;
using System.Buffers;
using System.Runtime.CompilerServices;
namespace PerfectNumbers.Core;
internal static class ThreadLocalArrayPool<T>
{
[ThreadStatic]
private static ArrayPool<T>? s_threadPool;
public static ArrayPool<T> Shared
{
[MethodImpl(MethodImplOptions.AggressiveInlining)]
get
{
ArrayPool<T>? pool = s_threadPool;
if (pool is null)
{
pool = ArrayPool<T>.Create();
s_threadPool = pool;
}
return pool;
}
}
}