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

Tiny cleanup

Browse Source
This commit is contained in:
Marek Lemańczyk committed 2025-09-24 03:09:58 +02:00
1 parent d472d48a84
commit a4bda0f0ee
1 file changed
+254 -254
Download Patch File Download Diff File Expand all files Collapse all files
PerfectNumbers.Core/Gpu/MersenneNumberDivisorByDivisorGpuTester.cs
+254 -254
View File
@@ -24,9 +24,9 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester
public readonly ulong MontgomeryTwo = montgomeryTwo; public readonly ulong MontgomeryTwo = montgomeryTwo;
} }
private readonly ConcurrentDictionary<Accelerator, Action<Index1D, ulong, ArrayView<MontgomeryDivisorData>, ArrayView<byte>>> _kernelCache = new(); private readonly ConcurrentDictionary<Accelerator, Action<Index1D, ulong, ArrayView<MontgomeryDivisorData>, ArrayView<byte>>> _kernelCache = new();
private readonly ConcurrentDictionary<Accelerator, Action<Index1D, MontgomeryDivisorData, ArrayView<ulong>, ArrayView<byte>>> _kernelByPrimeCache = new(); private readonly ConcurrentDictionary<Accelerator, Action<Index1D, MontgomeryDivisorData, ArrayView<ulong>, ArrayView<byte>>> _kernelByPrimeCache = new();
private readonly ConcurrentBag<DivisorScanSession> _sessionPool = new(); private readonly ConcurrentBag<DivisorScanSession> _sessionPool = new();
private Action<Index1D, ulong, ArrayView<MontgomeryDivisorData>, ArrayView<byte>> GetKernel(Accelerator accelerator) => private Action<Index1D, ulong, ArrayView<MontgomeryDivisorData>, ArrayView<byte>> GetKernel(Accelerator accelerator) =>
_kernelCache.GetOrAdd(accelerator, acc => acc.LoadAutoGroupedStreamKernel<Index1D, ulong, ArrayView<MontgomeryDivisorData>, ArrayView<byte>>(CheckKernel)); _kernelCache.GetOrAdd(accelerator, acc => acc.LoadAutoGroupedStreamKernel<Index1D, ulong, ArrayView<MontgomeryDivisorData>, ArrayView<byte>>(CheckKernel));
@@ -57,125 +57,125 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester
public bool IsPrime(ulong prime, out bool divisorsExhausted) public bool IsPrime(ulong prime, out bool divisorsExhausted)
{ {
ulong allowedMax; ulong allowedMax;
bool useCycles; bool useCycles;
int batchCapacity; int batchCapacity;
lock (_sync) lock (_sync)
{ {
if (!_isConfigured) if (!_isConfigured)
{ {
throw new InvalidOperationException("ConfigureFromMaxPrime must be called before using the tester."); throw new InvalidOperationException("ConfigureFromMaxPrime must be called before using the tester.");
} }
allowedMax = ComputeAllowedMaxDivisor(prime, _divisorLimit); allowedMax = ComputeAllowedMaxDivisor(prime, _divisorLimit);
useCycles = _useDivisorCycles; useCycles = _useDivisorCycles;
batchCapacity = _gpuBatchSize; batchCapacity = _gpuBatchSize;
} }
if (allowedMax < 3UL) if (allowedMax < 3UL)
{ {
divisorsExhausted = true; divisorsExhausted = true;
return true; return true;
} }
bool composite; bool composite;
bool coveredRange; bool coveredRange;
ulong processedCount; ulong processedCount;
ulong lastProcessed; ulong lastProcessed;
var gpuLease = GpuContextPool.RentPreferred(preferCpu: false); var gpuLease = GpuContextPool.RentPreferred(preferCpu: false);
var accelerator = gpuLease.Accelerator; var accelerator = gpuLease.Accelerator;
var kernel = GetKernel(accelerator); var kernel = GetKernel(accelerator);
var divisorsBuffer = accelerator.Allocate1D<MontgomeryDivisorData>(batchCapacity); var divisorsBuffer = accelerator.Allocate1D<MontgomeryDivisorData>(batchCapacity);
var hitsBuffer = accelerator.Allocate1D<byte>(batchCapacity); var hitsBuffer = accelerator.Allocate1D<byte>(batchCapacity);
ulong[] divisors = ArrayPool<ulong>.Shared.Rent(batchCapacity); ulong[] divisors = ArrayPool<ulong>.Shared.Rent(batchCapacity);
byte[] hits = ArrayPool<byte>.Shared.Rent(batchCapacity); byte[] hits = ArrayPool<byte>.Shared.Rent(batchCapacity);
MontgomeryDivisorData[] divisorData = ArrayPool<MontgomeryDivisorData>.Shared.Rent(batchCapacity); MontgomeryDivisorData[] divisorData = ArrayPool<MontgomeryDivisorData>.Shared.Rent(batchCapacity);
try try
{ {
composite = CheckDivisors( composite = CheckDivisors(
prime, prime,
allowedMax, allowedMax,
useCycles, useCycles,
accelerator, accelerator,
kernel, kernel,
divisorsBuffer, divisorsBuffer,
hitsBuffer, hitsBuffer,
divisors, divisors,
hits, hits,
divisorData, divisorData,
out lastProcessed, out lastProcessed,
out coveredRange, out coveredRange,
out processedCount out processedCount
); );
} }
finally finally
{ {
ArrayPool<byte>.Shared.Return(hits, clearArray: true); ArrayPool<byte>.Shared.Return(hits, clearArray: true);
ArrayPool<ulong>.Shared.Return(divisors, clearArray: true); ArrayPool<ulong>.Shared.Return(divisors, clearArray: true);
ArrayPool<MontgomeryDivisorData>.Shared.Return(divisorData, clearArray: true); ArrayPool<MontgomeryDivisorData>.Shared.Return(divisorData, clearArray: true);
hitsBuffer.Dispose(); hitsBuffer.Dispose();
divisorsBuffer.Dispose(); divisorsBuffer.Dispose();
gpuLease.Dispose(); gpuLease.Dispose();
} }
if (processedCount > 0UL) if (processedCount > 0UL)
{ {
lock (_sync) lock (_sync)
{ {
UpdateStatusUnsafe(lastProcessed, processedCount); UpdateStatusUnsafe(lastProcessed, processedCount);
} }
} }
if (composite) if (composite)
{ {
divisorsExhausted = true; divisorsExhausted = true;
return false; return false;
} }
divisorsExhausted = coveredRange; divisorsExhausted = coveredRange;
return true; return true;
} }
private bool CheckDivisors( private bool CheckDivisors(
ulong prime, ulong prime,
ulong allowedMax, ulong allowedMax,
bool useCycles, bool useCycles,
Accelerator accelerator, Accelerator accelerator,
Action<Index1D, ulong, ArrayView<MontgomeryDivisorData>, ArrayView<byte>> kernel, Action<Index1D, ulong, ArrayView<MontgomeryDivisorData>, ArrayView<byte>> kernel,
MemoryBuffer1D<MontgomeryDivisorData, Stride1D.Dense> divisorsBuffer, MemoryBuffer1D<MontgomeryDivisorData, Stride1D.Dense> divisorsBuffer,
MemoryBuffer1D<byte, Stride1D.Dense> hitsBuffer, MemoryBuffer1D<byte, Stride1D.Dense> hitsBuffer,
ulong[] divisors, ulong[] divisors,
byte[] hits, byte[] hits,
MontgomeryDivisorData[] divisorData, MontgomeryDivisorData[] divisorData,
out ulong lastProcessed, out ulong lastProcessed,
out bool coveredRange, out bool coveredRange,
out ulong processedCount) out ulong processedCount)
{ {
lastProcessed = 0UL; lastProcessed = 0UL;
processedCount = 0UL; processedCount = 0UL;
if (allowedMax < 3UL) if (allowedMax < 3UL)
{ {
coveredRange = true; coveredRange = true;
return false; return false;
} }
int batchCapacity = (int)divisorsBuffer.Length; int batchCapacity = (int)divisorsBuffer.Length;
bool composite = false; bool composite = false;
bool processedAll = false; bool processedAll = false;
ulong cycle, divisor = 3UL, next; ulong cycle, divisor = 3UL, next;
int batchSize, i; int batchSize, i;
bool reachedEndInBatch; bool reachedEndInBatch;
Span<MontgomeryDivisorData> divisorDataSpan; Span<MontgomeryDivisorData> divisorDataSpan;
Span<ulong> divisorSpan; Span<ulong> divisorSpan;
Span<byte> hitsSpan; Span<byte> hitsSpan;
MersenneDivisorCycles divisorCycles = MersenneDivisorCycles.Shared; MersenneDivisorCycles divisorCycles = MersenneDivisorCycles.Shared;
ArrayView1D<MontgomeryDivisorData, Stride1D.Dense> divisorView; ArrayView1D<MontgomeryDivisorData, Stride1D.Dense> divisorView;
ArrayView1D<byte, Stride1D.Dense> hitsView; ArrayView1D<byte, Stride1D.Dense> hitsView;
@@ -184,45 +184,45 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester
batchSize = 0; batchSize = 0;
reachedEndInBatch = false; reachedEndInBatch = false;
while (batchSize < batchCapacity && divisor <= allowedMax) while (batchSize < batchCapacity && divisor <= allowedMax)
{ {
ulong currentDivisor = divisor; ulong currentDivisor = divisor;
ulong nextDivisor = currentDivisor + 2UL; ulong nextDivisor = currentDivisor + 2UL;
processedCount++; processedCount++;
if (useCycles) if (useCycles)
{ {
cycle = divisorCycles!.GetCycle(currentDivisor); cycle = divisorCycles!.GetCycle(currentDivisor);
if (cycle == 0UL || prime % cycle != 0UL) if (cycle == 0UL || prime % cycle != 0UL)
{ {
lastProcessed = currentDivisor; lastProcessed = currentDivisor;
if (nextDivisor <= currentDivisor) if (nextDivisor <= currentDivisor)
{ {
reachedEndInBatch = true; reachedEndInBatch = true;
break; break;
} }
if (nextDivisor > allowedMax) if (nextDivisor > allowedMax)
{ {
divisor = nextDivisor; divisor = nextDivisor;
reachedEndInBatch = true; reachedEndInBatch = true;
break; break;
} }
divisor = nextDivisor; divisor = nextDivisor;
continue; continue;
} }
} }
divisors[batchSize++] = currentDivisor; divisors[batchSize++] = currentDivisor;
lastProcessed = currentDivisor; lastProcessed = currentDivisor;
next = nextDivisor; next = nextDivisor;
// This will be only true when we exceed ulong.MaxValue // This will be only true when we exceed ulong.MaxValue
if (next <= divisor) if (next <= divisor)
{ {
reachedEndInBatch = true; reachedEndInBatch = true;
break; break;
} }
@@ -236,15 +236,15 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester
divisor = next; divisor = next;
} }
if (batchSize == 0) if (batchSize == 0)
{ {
if (reachedEndInBatch) if (reachedEndInBatch)
{ {
processedAll = true; processedAll = true;
} }
break; break;
} }
divisorDataSpan = divisorData.AsSpan(0, batchSize); divisorDataSpan = divisorData.AsSpan(0, batchSize);
divisorView = divisorsBuffer.View.SubView(0, batchSize); divisorView = divisorsBuffer.View.SubView(0, batchSize);
@@ -263,20 +263,20 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester
accelerator.Synchronize(); accelerator.Synchronize();
hitsView.CopyToCPU(ref MemoryMarshal.GetReference(hitsSpan), batchSize); hitsView.CopyToCPU(ref MemoryMarshal.GetReference(hitsSpan), batchSize);
for (i = 0; i < batchSize; i++) for (i = 0; i < batchSize; i++)
{ {
if (hitsSpan[i] != 0) if (hitsSpan[i] != 0)
{ {
composite = true; composite = true;
lastProcessed = divisorSpan[i]; lastProcessed = divisorSpan[i];
break; break;
} }
} }
if (!composite) if (!composite)
{ {
lastProcessed = divisorSpan[batchSize - 1]; lastProcessed = divisorSpan[batchSize - 1];
} }
if (composite) if (composite)
{ {
break; break;
@@ -289,10 +289,10 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester
} }
} }
coveredRange = composite || processedAll || divisor > allowedMax; coveredRange = composite || processedAll || divisor > allowedMax;
return composite; return composite;
} }
private static ulong ComputeDivisorLimitFromMaxPrime(ulong maxPrime) private static ulong ComputeDivisorLimitFromMaxPrime(ulong maxPrime)
{ {
@@ -309,44 +309,44 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester
return (1UL << (int)(maxPrime - 1UL)) - 1UL; return (1UL << (int)(maxPrime - 1UL)) - 1UL;
} }
private static ulong ComputeAllowedMaxDivisor(ulong prime, ulong divisorLimit) private static ulong ComputeAllowedMaxDivisor(ulong prime, ulong divisorLimit)
{ {
if (prime <= 1UL) if (prime <= 1UL)
{ {
return 0UL; return 0UL;
} }
if (prime - 1UL >= 64UL) if (prime - 1UL >= 64UL)
{ {
return divisorLimit; return divisorLimit;
} }
return Math.Min((1UL << (int)(prime - 1UL)) - 1UL, divisorLimit); return Math.Min((1UL << (int)(prime - 1UL)) - 1UL, divisorLimit);
} }
private void UpdateStatusUnsafe(ulong lastProcessed, ulong processedCount) private void UpdateStatusUnsafe(ulong lastProcessed, ulong processedCount)
{ {
if (processedCount == 0UL) if (processedCount == 0UL)
{ {
return; return;
} }
ulong interval = (ulong)PerfectNumberConstants.ConsoleInterval; ulong interval = (ulong)PerfectNumberConstants.ConsoleInterval;
if (interval == 0UL) if (interval == 0UL)
{ {
_lastStatusDivisor = 0UL; _lastStatusDivisor = 0UL;
return; return;
} }
ulong total = _lastStatusDivisor + processedCount; ulong total = _lastStatusDivisor + processedCount;
ulong messages = total / interval; ulong messages = total / interval;
_lastStatusDivisor = total % interval; _lastStatusDivisor = total % interval;
for (ulong i = 0; i < messages; i++) for (ulong i = 0; i < messages; i++)
{ {
Console.WriteLine($"...processed by-divisor candidate = {lastProcessed}"); Console.WriteLine($"...processed by-divisor candidate = {lastProcessed}");
} }
} }
private static void CheckKernel(Index1D index, ulong prime, ArrayView<MontgomeryDivisorData> divisors, ArrayView<byte> hits) private static void CheckKernel(Index1D index, ulong prime, ArrayView<MontgomeryDivisorData> divisors, ArrayView<byte> hits)
{ {
@@ -361,41 +361,41 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester
hits[index] = Pow2Mod(prime, divisor) == 1UL ? (byte)1 : (byte)0; hits[index] = Pow2Mod(prime, divisor) == 1UL ? (byte)1 : (byte)0;
} }
public sealed class DivisorScanSession : IDisposable public sealed class DivisorScanSession : IDisposable
{ {
private readonly MersenneNumberDivisorByDivisorGpuTester _owner; private readonly MersenneNumberDivisorByDivisorGpuTester _owner;
private readonly GpuContextPool.GpuContextLease _lease; private readonly GpuContextPool.GpuContextLease _lease;
private readonly Accelerator _accelerator; private readonly Accelerator _accelerator;
private readonly Action<Index1D, MontgomeryDivisorData, ArrayView<ulong>, ArrayView<byte>> _kernel; private readonly Action<Index1D, MontgomeryDivisorData, ArrayView<ulong>, ArrayView<byte>> _kernel;
private readonly MemoryBuffer1D<ulong, Stride1D.Dense> _primesBuffer; private readonly MemoryBuffer1D<ulong, Stride1D.Dense> _primesBuffer;
private readonly MemoryBuffer1D<byte, Stride1D.Dense> _hitsBuffer; private readonly MemoryBuffer1D<byte, Stride1D.Dense> _hitsBuffer;
private readonly ulong[] _primesHost; private readonly ulong[] _primesHost;
private readonly byte[] _hitsHost; private readonly byte[] _hitsHost;
private readonly int[] _indexHost; private readonly int[] _indexHost;
private bool _disposed; private bool _disposed;
internal DivisorScanSession(MersenneNumberDivisorByDivisorGpuTester owner) internal DivisorScanSession(MersenneNumberDivisorByDivisorGpuTester owner)
{ {
_owner = owner; _owner = owner;
_lease = GpuContextPool.RentPreferred(preferCpu: false); _lease = GpuContextPool.RentPreferred(preferCpu: false);
_accelerator = _lease.Accelerator; _accelerator = _lease.Accelerator;
_kernel = owner.GetKernelByPrime(_accelerator); _kernel = owner.GetKernelByPrime(_accelerator);
_primesBuffer = _accelerator.Allocate1D<ulong>(owner._gpuBatchSize); _primesBuffer = _accelerator.Allocate1D<ulong>(owner._gpuBatchSize);
_hitsBuffer = _accelerator.Allocate1D<byte>(owner._gpuBatchSize); _hitsBuffer = _accelerator.Allocate1D<byte>(owner._gpuBatchSize);
_primesHost = ArrayPool<ulong>.Shared.Rent(owner._gpuBatchSize); _primesHost = ArrayPool<ulong>.Shared.Rent(owner._gpuBatchSize);
_hitsHost = ArrayPool<byte>.Shared.Rent(owner._gpuBatchSize); _hitsHost = ArrayPool<byte>.Shared.Rent(owner._gpuBatchSize);
_indexHost = ArrayPool<int>.Shared.Rent(owner._gpuBatchSize); _indexHost = ArrayPool<int>.Shared.Rent(owner._gpuBatchSize);
} }
internal void Reset() internal void Reset()
{ {
_disposed = false; _disposed = false;
} }
public void CheckDivisor(ulong divisor, ReadOnlySpan<ulong> primes, Span<byte> hits) public void CheckDivisor(ulong divisor, ReadOnlySpan<ulong> primes, Span<byte> hits)
{ {
if (_disposed) if (_disposed)
{ {
throw new ObjectDisposedException(nameof(DivisorScanSession)); throw new ObjectDisposedException(nameof(DivisorScanSession));
} }
@@ -486,24 +486,24 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester
offset += batchSize; offset += batchSize;
} }
} }
public void Dispose() public void Dispose()
{ {
if (_disposed) if (_disposed)
{ {
return; return;
} }
_disposed = true; _disposed = true;
_owner.ReturnSession(this); _owner.ReturnSession(this);
} }
} }
internal void ReturnSession(DivisorScanSession session) internal void ReturnSession(DivisorScanSession session)
{ {
_sessionPool.Add(session); _sessionPool.Add(session);
} }
private static ulong Pow2Mod(ulong exponent, in MontgomeryDivisorData divisor) private static ulong Pow2Mod(ulong exponent, in MontgomeryDivisorData divisor)
{ {
@@ -540,7 +540,7 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester
{ {
ulong tLow = unchecked(a * b); ulong tLow = unchecked(a * b);
ulong m = unchecked(tLow * nPrime); ulong m = unchecked(tLow * nPrime);
// We're reusing tLow variable as the result to perform just a little bit better // We're reusing tLow variable as the result to perform just a little bit better
tLow = unchecked(MultiplyHigh(a, b) + MultiplyHigh(m, modulus) + (unchecked(tLow + m * modulus) < tLow ? 1UL : 0UL)); tLow = unchecked(MultiplyHigh(a, b) + MultiplyHigh(m, modulus) + (unchecked(tLow + m * modulus) < tLow ? 1UL : 0UL));
if (tLow >= modulus) if (tLow >= modulus)
@@ -624,32 +624,32 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester
{ {
lock (_sync) lock (_sync)
{ {
if (!_isConfigured) if (!_isConfigured)
{ {
throw new InvalidOperationException("ConfigureFromMaxPrime must be called before using the tester."); throw new InvalidOperationException("ConfigureFromMaxPrime must be called before using the tester.");
} }
return ComputeAllowedMaxDivisor(prime, _divisorLimit); return ComputeAllowedMaxDivisor(prime, _divisorLimit);
} }
} }
public DivisorScanSession CreateDivisorSession() public DivisorScanSession CreateDivisorSession()
{ {
lock (_sync) lock (_sync)
{ {
if (!_isConfigured) if (!_isConfigured)
{ {
throw new InvalidOperationException("ConfigureFromMaxPrime must be called before using the tester."); throw new InvalidOperationException("ConfigureFromMaxPrime must be called before using the tester.");
} }
if (_sessionPool.TryTake(out DivisorScanSession? session)) if (_sessionPool.TryTake(out DivisorScanSession? session))
{ {
session.Reset(); session.Reset();
return session; return session;
} }
return new DivisorScanSession(this); return new DivisorScanSession(this);
} }
} }
} }