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

Merge pull request #17 from mlemanczyk/codex/analyze-evenperfectbitscanner-for-optimizations

Browse Source 
Optimize GPU residue scanning and divisor path
This commit is contained in:
Marek Lemańczyk authored and GitHub committed 2025-09-13 16:48:52 +02:00
2 parents 515a1bb342 58eb65ca9e
commit 731927f10f
7 files changed
+149 -46
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

EvenPerfectBitScanner.Tests/ProgramTests.cs
+3 -2
View File
@@ -73,14 +73,14 @@ public class ProgramTests
var useDivisorField = typeof(Program).GetField("_useDivisor", BindingFlags.NonPublic | BindingFlags.Static)!; var useDivisorField = typeof(Program).GetField("_useDivisor", BindingFlags.NonPublic | BindingFlags.Static)!;
var divisorField = typeof(Program).GetField("_divisor", BindingFlags.NonPublic | BindingFlags.Static)!; var divisorField = typeof(Program).GetField("_divisor", BindingFlags.NonPublic | BindingFlags.Static)!;
var testerField = typeof(Program).GetField("_divisorTester", BindingFlags.NonPublic | BindingFlags.Static)!; var testerField = typeof(Program).GetField("_divisorTester", BindingFlags.NonPublic | BindingFlags.Static)!;
var candidatesField = typeof(Program).GetField("_divisorCandidates", BindingFlags.NonPublic | BindingFlags.Static)!; var candidatesField = typeof(MersenneNumberDivisorGpuTester).GetField("_divisorCandidates", BindingFlags.NonPublic | BindingFlags.Static)!;
var primeField = typeof(Program).GetField("PrimeTesters", BindingFlags.NonPublic | BindingFlags.Static)!; var primeField = typeof(Program).GetField("PrimeTesters", BindingFlags.NonPublic | BindingFlags.Static)!;
var residueField = typeof(Program).GetField("PResidue", BindingFlags.NonPublic | BindingFlags.Static)!; var residueField = typeof(Program).GetField("PResidue", BindingFlags.NonPublic | BindingFlags.Static)!;
var forceCpuProp = typeof(GpuContextPool).GetProperty("ForceCpu"); var forceCpuProp = typeof(GpuContextPool).GetProperty("ForceCpu");
useDivisorField.SetValue(null, true); useDivisorField.SetValue(null, true);
divisorField.SetValue(null, 0UL); divisorField.SetValue(null, 0UL);
testerField.SetValue(null, null); testerField.SetValue(null, new MersenneNumberDivisorGpuTester());
primeField.SetValue(null, new ThreadLocal<PrimeTester>(() => new PrimeTester(), trackAllValues: true)); primeField.SetValue(null, new ThreadLocal<PrimeTester>(() => new PrimeTester(), trackAllValues: true));
residueField.SetValue(null, new ThreadLocal<ModResidueTracker>(() => new ModResidueTracker(ResidueModel.Identity, 2UL, true), trackAllValues: true)); residueField.SetValue(null, new ThreadLocal<ModResidueTracker>(() => new ModResidueTracker(ResidueModel.Identity, 2UL, true), trackAllValues: true));
candidatesField.SetValue(null, new (ulong, uint)[] { (7UL, 3U), (23UL, 11U) }); candidatesField.SetValue(null, new (ulong, uint)[] { (7UL, 3U), (23UL, 11U) });
@@ -90,6 +90,7 @@ public class ProgramTests
{ {
Program.IsEvenPerfectCandidate(11UL, 32).Should().BeFalse(); Program.IsEvenPerfectCandidate(11UL, 32).Should().BeFalse();
Program.IsEvenPerfectCandidate(5UL, 32).Should().BeTrue(); Program.IsEvenPerfectCandidate(5UL, 32).Should().BeTrue();
Program.IsEvenPerfectCandidate(127UL, 32).Should().BeTrue();
} }
finally finally
{ {
PerfectNumbers.Core.Tests/PrimeTester/PrimeTesterIsPrimeTests.cs
+1
View File
@@ -12,6 +12,7 @@ public class PrimeTesterIsPrimeTests
[InlineData(2UL, true)] [InlineData(2UL, true)]
[InlineData(3UL, true)] [InlineData(3UL, true)]
[InlineData(4UL, false)] [InlineData(4UL, false)]
[InlineData(5UL, true)]
[InlineData(7UL, true)] [InlineData(7UL, true)]
[InlineData(97UL, true)] [InlineData(97UL, true)]
[InlineData(121UL, false)] [InlineData(121UL, false)]
PerfectNumbers.Core/Gpu/GpuKernelPool.cs
+85 -14
View File
@@ -1,4 +1,5 @@
using System.Collections.Concurrent; using System.Collections.Concurrent;
using System.Runtime.CompilerServices;
using ILGPU; using ILGPU;
using ILGPU.Runtime; using ILGPU.Runtime;
using static PerfectNumbers.Core.Gpu.GpuContextPool; using static PerfectNumbers.Core.Gpu.GpuContextPool;
@@ -29,7 +30,7 @@ public sealed class KernelContainer
public Action<Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, public Action<Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong,
ulong, ulong, ulong, ulong, ArrayView<ulong>, ArrayView1D<uint, Stride1D.Dense>>? Incremental; ulong, ulong, ulong, ulong, ArrayView<ulong>, ArrayView1D<uint, Stride1D.Dense>>? Incremental;
public Action<Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, public Action<Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong,
ulong, ulong, ulong, ulong, ArrayView<ulong>, ArrayView1D<uint, Stride1D.Dense>>? Pow2Mod; ResidueAutomatonArgs, ArrayView<ulong>, ArrayView1D<uint, Stride1D.Dense>, ArrayView1D<uint, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>>? Pow2Mod;
public Action<Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, public Action<Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong,
ResidueAutomatonArgs, ArrayView<int>, ArrayView1D<uint, Stride1D.Dense>>? IncrementalOrder; ResidueAutomatonArgs, ArrayView<int>, ArrayView1D<uint, Stride1D.Dense>>? IncrementalOrder;
public Action<Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, public Action<Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong,
@@ -37,6 +38,8 @@ public sealed class KernelContainer
// Optional device buffer with small divisor cycles (<= 4M). Index = divisor, value = cycle length. // Optional device buffer with small divisor cycles (<= 4M). Index = divisor, value = cycle length.
public MemoryBuffer1D<uint, Stride1D.Dense>? SmallCycles; public MemoryBuffer1D<uint, Stride1D.Dense>? SmallCycles;
public MemoryBuffer1D<uint, Stride1D.Dense>? SmallPrimes;
public MemoryBuffer1D<ulong, Stride1D.Dense>? SmallPrimesPow2;
public static T InitOnce<T>(ref T? slot, Func<T> factory) where T : class public static T InitOnce<T>(ref T? slot, Func<T> factory) where T : class
{ {
@@ -77,14 +80,14 @@ public ref struct GpuKernelLease(IDisposable limiter, GpuContextLease gpu, Kerne
} }
} }
public readonly Action<Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ulong, ulong, ulong, ulong, ArrayView<ulong>, ArrayView1D<uint, Stride1D.Dense>> Pow2ModKernel public readonly Action<Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ResidueAutomatonArgs, ArrayView<ulong>, ArrayView1D<uint, Stride1D.Dense>, ArrayView1D<uint, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>> Pow2ModKernel
{ {
get get
{ {
var accel = Accelerator; var accel = Accelerator;
lock (_gpu!.Value.KernelInitLock) lock (_gpu!.Value.KernelInitLock)
{ {
return KernelContainer.InitOnce(ref _kernels.Pow2Mod, () => accel.LoadAutoGroupedStreamKernel<Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ulong, ulong, ulong, ulong, ArrayView<ulong>, ArrayView1D<uint, Stride1D.Dense>>(Pow2ModKernelScan)); return KernelContainer.InitOnce(ref _kernels.Pow2Mod, () => accel.LoadAutoGroupedStreamKernel<Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ResidueAutomatonArgs, ArrayView<ulong>, ArrayView1D<uint, Stride1D.Dense>, ArrayView1D<uint, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>>(Pow2ModKernelScan));
} }
} }
} }
@@ -274,33 +277,30 @@ public ref struct GpuKernelLease(IDisposable limiter, GpuContextLease gpu, Kerne
} }
private static void Pow2ModKernelScan(Index1D index, ulong exponent, GpuUInt128 twoP, GpuUInt128 kStart, byte lastIsSeven, ulong _, private static void Pow2ModKernelScan(Index1D index, ulong exponent, GpuUInt128 twoP, GpuUInt128 kStart, byte lastIsSeven, ulong _,
ulong q0m10, ulong q0m8, ulong q0m3, ulong q0m5, ArrayView<ulong> orders, ResidueAutomatonArgs ra, ArrayView<ulong> orders,
ArrayView1D<uint, Stride1D.Dense> smallCycles) ArrayView1D<uint, Stride1D.Dense> smallCycles, ArrayView1D<uint, Stride1D.Dense> smallPrimes,
ArrayView1D<ulong, Stride1D.Dense> smallPrimesPow2)
{ {
ulong idx = (ulong)index.X; ulong idx = (ulong)index.X;
ulong step10 = ((exponent % 10UL) << 1) % 10UL; ulong r10 = ra.Q0M10 + (ra.Step10 * idx) % 10UL; r10 -= (r10 >= 10UL) ? 10UL : 0UL;
ulong r10 = q0m10 + (step10 * idx) % 10UL; r10 -= (r10 >= 10UL) ? 10UL : 0UL;
bool shouldCheck = r10 != 5UL; bool shouldCheck = r10 != 5UL;
if (shouldCheck) if (shouldCheck)
{ {
ulong step8 = ((exponent & 7UL) << 1) & 7UL; ulong r8 = (ra.Q0M8 + ((ra.Step8 * idx) & 7UL)) & 7UL;
ulong r8 = (q0m8 + ((step8 * idx) & 7UL)) & 7UL;
if (r8 != 1UL && r8 != 7UL) if (r8 != 1UL && r8 != 7UL)
{ {
shouldCheck = false; shouldCheck = false;
} }
else else
{ {
ulong step3 = ((exponent % 3UL) << 1) % 3UL; ulong r3 = ra.Q0M3 + (ra.Step3 * (idx % 3UL)); r3 -= (r3 >= 3UL) ? 3UL : 0UL;
ulong r3 = q0m3 + (step3 * (idx % 3UL)); r3 -= (r3 >= 3UL) ? 3UL : 0UL;
if (r3 == 0UL) if (r3 == 0UL)
{ {
shouldCheck = false; shouldCheck = false;
} }
else else
{ {
ulong step5Loc = ((exponent % 5UL) << 1) % 5UL; ulong r5 = ra.Q0M5 + (ra.Step5 * (idx % 5UL)); if (r5 >= 5UL) r5 -= 5UL;
ulong r5 = q0m5 + (step5Loc * (idx % 5UL)); if (r5 >= 5UL) r5 -= 5UL;
if (r5 == 0UL) if (r5 == 0UL)
{ {
shouldCheck = false; shouldCheck = false;
@@ -316,7 +316,6 @@ public ref struct GpuKernelLease(IDisposable limiter, GpuContextLease gpu, Kerne
GpuUInt128 k = kStart + (GpuUInt128)idx; GpuUInt128 k = kStart + (GpuUInt128)idx;
GpuUInt128 q = twoP.Mul64(k) + GpuUInt128.One; GpuUInt128 q = twoP.Mul64(k) + GpuUInt128.One;
// Small-cycles in-kernel early rejection from device table
if (q.High == 0UL && q.Low <= (ulong)(smallCycles.Length - 1)) if (q.High == 0UL && q.Low <= (ulong)(smallCycles.Length - 1))
{ {
uint cyc = smallCycles[(int)q.Low]; uint cyc = smallCycles[(int)q.Low];
@@ -332,9 +331,54 @@ public ref struct GpuKernelLease(IDisposable limiter, GpuContextLease gpu, Kerne
return; return;
} }
int primesLen = (int)smallPrimes.Length;
for (int i = 0; i < primesLen; i++)
{
ulong square = smallPrimesPow2[i];
if (new GpuUInt128(0UL, square) > q)
{
break;
}
ulong prime = smallPrimes[i];
if (Mod128By64(q, prime) == 0UL)
{
orders[index] = 0UL;
return;
}
}
orders[index] = exponent; orders[index] = exponent;
} }
[MethodImpl(MethodImplOptions.AggressiveInlining)]
private static ulong Mod128By64(GpuUInt128 value, ulong modulus)
{
ulong rem = 0UL;
ulong part = value.High;
for (int i = 0; i < 64; i++)
{
rem = (rem << 1) | (part >> 63);
if (rem >= modulus)
{
rem -= modulus;
}
part <<= 1;
}
part = value.Low;
for (int i = 0; i < 64; i++)
{
rem = (rem << 1) | (part >> 63);
if (rem >= modulus)
{
rem -= modulus;
}
part <<= 1;
}
return rem;
}
// Device-friendly small cycle length for 2 mod divisor // Device-friendly small cycle length for 2 mod divisor
private static ulong CalculateCycleLengthSmall(ulong divisor) private static ulong CalculateCycleLengthSmall(ulong divisor)
{ {
@@ -527,6 +571,33 @@ public class GpuKernelPool
} }
} }
public static (ArrayView1D<uint, Stride1D.Dense> primes, ArrayView1D<ulong, Stride1D.Dense> primesPow2) EnsureSmallPrimesOnDevice(Accelerator accelerator)
{
var kernels = GetKernels(accelerator);
if (kernels.SmallPrimes is { } p && kernels.SmallPrimesPow2 is { } p2)
{
return (p.View, p2.View);
}
lock (kernels)
{
if (kernels.SmallPrimes is { } existing && kernels.SmallPrimesPow2 is { } existingPow2)
{
return (existing.View, existingPow2.View);
}
var hostPrimes = PrimesGenerator.SmallPrimes;
var hostPow2 = PrimesGenerator.SmallPrimesPow2;
var devicePrimes = accelerator.Allocate1D<uint>(hostPrimes.Length);
devicePrimes.View.CopyFromCPU(hostPrimes);
var devicePow2 = accelerator.Allocate1D<ulong>(hostPow2.Length);
devicePow2.View.CopyFromCPU(hostPow2);
kernels.SmallPrimes = devicePrimes;
kernels.SmallPrimesPow2 = devicePow2;
return (devicePrimes.View, devicePow2.View);
}
}
public static GpuKernelLease GetKernel(bool useGpuOrder) public static GpuKernelLease GetKernel(bool useGpuOrder)
{ {
var limiter = GpuPrimeWorkLimiter.Acquire(); var limiter = GpuPrimeWorkLimiter.Acquire();
PerfectNumbers.Core/Gpu/MersenneNumberDivisorGpuTester.cs
+6 -2
View File
@@ -8,6 +8,7 @@ namespace PerfectNumbers.Core.Gpu;
public sealed class MersenneNumberDivisorGpuTester public sealed class MersenneNumberDivisorGpuTester
{ {
private readonly ConcurrentDictionary<Accelerator, Action<Index1D, ulong, ulong, ArrayView<byte>>> _kernelCache = new(); private readonly ConcurrentDictionary<Accelerator, Action<Index1D, ulong, ulong, ArrayView<byte>>> _kernelCache = new();
private readonly ConcurrentDictionary<Accelerator, MemoryBuffer1D<byte, Stride1D.Dense>> _resultBuffers = new();
private Action<Index1D, ulong, ulong, ArrayView<byte>> GetKernel(Accelerator accelerator) => private Action<Index1D, ulong, ulong, ArrayView<byte>> GetKernel(Accelerator accelerator) =>
_kernelCache.GetOrAdd(accelerator, acc => acc.LoadAutoGroupedStreamKernel<Index1D, ulong, ulong, ArrayView<byte>>(Kernel)); _kernelCache.GetOrAdd(accelerator, acc => acc.LoadAutoGroupedStreamKernel<Index1D, ulong, ulong, ArrayView<byte>>(Kernel));
@@ -37,11 +38,10 @@ public sealed class MersenneNumberDivisorGpuTester
var gpu = GpuContextPool.RentPreferred(preferCpu: false); var gpu = GpuContextPool.RentPreferred(preferCpu: false);
var accelerator = gpu.Accelerator; var accelerator = gpu.Accelerator;
var kernel = GetKernel(accelerator); var kernel = GetKernel(accelerator);
var resultBuffer = accelerator.Allocate1D<byte>(1); var resultBuffer = _resultBuffers.GetOrAdd(accelerator, acc => acc.Allocate1D<byte>(1));
kernel(1, exponent, divisor, resultBuffer.View); kernel(1, exponent, divisor, resultBuffer.View);
accelerator.Synchronize(); accelerator.Synchronize();
bool divisible = resultBuffer.GetAsArray1D()[0] != 0; bool divisible = resultBuffer.GetAsArray1D()[0] != 0;
resultBuffer.Dispose();
gpu.Dispose(); gpu.Dispose();
return divisible; return divisible;
} }
@@ -93,6 +93,10 @@ public sealed class MersenneNumberDivisorGpuTester
} }
d = kMul2 * p + 1UL; d = kMul2 * p + 1UL;
if (p < 64UL && d == ((1UL << (int)p) - 1UL))
{
continue;
}
// kkMul2 now becomes cycle. We're reusing existing variable for the best performance // kkMul2 now becomes cycle. We're reusing existing variable for the best performance
kMul2 = divisorCycles.GetCycle(d); kMul2 = divisorCycles.GetCycle(d);
if (p % kMul2 != 0UL) if (p % kMul2 != 0UL)
PerfectNumbers.Core/Gpu/MersenneNumberIncrementalGpuTester.cs
+41 -17
View File
@@ -1,5 +1,6 @@
using System.Buffers; using System.Buffers;
using System.Runtime.InteropServices; using System.Runtime.InteropServices;
using ILGPU;
using ILGPU.Runtime; using ILGPU.Runtime;
namespace PerfectNumbers.Core.Gpu; namespace PerfectNumbers.Core.Gpu;
@@ -18,9 +19,20 @@ public class MersenneNumberIncrementalGpuTester(GpuKernelType kernelType, bool u
ulong divMul = (ulong)((((UInt128)1 << 64) - UInt128.One) / exponent) + 1UL; ulong divMul = (ulong)((((UInt128)1 << 64) - UInt128.One) / exponent) + 1UL;
byte last = lastIsSeven ? (byte)1 : (byte)0; // ILGPU kernels do not support bool parameters byte last = lastIsSeven ? (byte)1 : (byte)0; // ILGPU kernels do not support bool parameters
var kernel = (_kernelType == GpuKernelType.Pow2Mod) var pow2Kernel = gpuLease.Pow2ModKernel;
? gpuLease.Pow2ModKernel var incKernel = gpuLease.IncrementalKernel;
: gpuLease.IncrementalKernel; ulong step10 = ((exponent % 10UL) << 1) % 10UL;
ulong step8 = ((exponent & 7UL) << 1) & 7UL;
ulong step3 = ((exponent % 3UL) << 1) % 3UL;
ulong step5 = ((exponent % 5UL) << 1) % 5UL;
GpuUInt128 twoPGpu = (GpuUInt128)twoP;
var smallCyclesView = GpuKernelPool.EnsureSmallCyclesOnDevice(accelerator);
ArrayView1D<uint, Stride1D.Dense> smallPrimesView = default;
ArrayView1D<ulong, Stride1D.Dense> smallPrimesPow2View = default;
if (_kernelType == GpuKernelType.Pow2Mod)
{
(smallPrimesView, smallPrimesPow2View) = GpuKernelPool.EnsureSmallPrimesOnDevice(accelerator);
}
var orderBuffer = accelerator.Allocate1D<ulong>(batchSize); var orderBuffer = accelerator.Allocate1D<ulong>(batchSize);
// Avoid giant stack allocations that can trigger StackOverflow when batchSize is large. // Avoid giant stack allocations that can trigger StackOverflow when batchSize is large.
@@ -39,27 +51,39 @@ public class MersenneNumberIncrementalGpuTester(GpuKernelType kernelType, bool u
Span<ulong> orders = orderArray.AsSpan(0, currentSize); Span<ulong> orders = orderArray.AsSpan(0, currentSize);
// Precompute residue automaton bases for this batch // Precompute residue automaton bases for this batch
UInt128 q0 = twoP * kStart + UInt128.One; UInt128 q0 = twoP * kStart + UInt128.One;
ulong q0m10 = q0.Mod10(); if (_kernelType == GpuKernelType.Pow2Mod)
ulong q0m8 = q0.Mod8(); {
ulong q0m3 = q0.Mod3(); q0.Mod10_8_5_3(out ulong q0m10, out ulong q0m8, out ulong q0m5, out ulong q0m3);
ulong q0m5 = q0.Mod5(); var ra = new ResidueAutomatonArgs(q0m10, step10, q0m8, step8, q0m3, step3, q0m5, step5);
ulong step10 = ((exponent % 10UL) << 1) % 10UL; pow2Kernel(currentSize, exponent, twoPGpu, (GpuUInt128)kStart, last, divMul,
ulong step8 = ((exponent & 7UL) << 1) & 7UL; ra, orderBuffer.View, smallCyclesView, smallPrimesView, smallPrimesPow2View);
ulong step3 = ((exponent % 3UL) << 1) % 3UL; }
ulong step5 = ((exponent % 5UL) << 1) % 5UL; else
{
var smallCyclesView = GpuKernelPool.EnsureSmallCyclesOnDevice(accelerator); q0.Mod10_8_5_3(out ulong q0m10, out ulong q0m8, out ulong q0m5, out ulong q0m3);
kernel(currentSize, exponent, (GpuUInt128)twoP, (GpuUInt128)kStart, last, divMul, incKernel(currentSize, exponent, twoPGpu, (GpuUInt128)kStart, last, divMul,
q0m10, q0m8, q0m3, q0m5, orderBuffer.View, smallCyclesView); q0m10, q0m8, q0m3, q0m5, orderBuffer.View, smallCyclesView);
}
accelerator.Synchronize(); accelerator.Synchronize();
orderBuffer.View.CopyToCPU(ref MemoryMarshal.GetReference(orders), currentSize); orderBuffer.View.CopyToCPU(ref MemoryMarshal.GetReference(orders), currentSize);
q = twoP * kStart + UInt128.One; if (_kernelType == GpuKernelType.Pow2Mod)
for (i = 0; i < currentSize && Volatile.Read(ref isPrime); i++, q += twoP) {
for (i = 0; i < currentSize; i++)
{ {
if (orders[i] != 0UL) if (orders[i] != 0UL)
{ {
if (q.IsPrimeCandidate()) Volatile.Write(ref isPrime, false);
break;
}
}
}
else
{
q = twoP * kStart + UInt128.One;
for (i = 0; i < currentSize && Volatile.Read(ref isPrime); i++, q += twoP)
{
if (orders[i] != 0UL && q.IsPrimeCandidate())
{ {
Volatile.Write(ref isPrime, false); Volatile.Write(ref isPrime, false);
} }
PerfectNumbers.Core/Gpu/MersenneNumberResidueGpuTester.cs
+12 -10
View File
@@ -16,12 +16,18 @@ public class MersenneNumberResidueGpuTester(bool useGpuOrder)
{ {
var gpuLease = GpuKernelPool.GetKernel(_useGpuOrder); var gpuLease = GpuKernelPool.GetKernel(_useGpuOrder);
var accelerator = gpuLease.Accelerator; var accelerator = gpuLease.Accelerator;
// Ensure device has small cycles table for in-kernel lookup // Ensure device has small cycles and primes tables for in-kernel lookup
var smallCyclesView = GpuKernelPool.EnsureSmallCyclesOnDevice(accelerator); var smallCyclesView = GpuKernelPool.EnsureSmallCyclesOnDevice(accelerator);
var (smallPrimesView, smallPrimesPow2View) = GpuKernelPool.EnsureSmallPrimesOnDevice(accelerator);
int batchSize = GpuConstants.ScanBatchSize; int batchSize = GpuConstants.ScanBatchSize;
UInt128 kStart = 1UL; UInt128 kStart = 1UL;
byte last = lastIsSeven ? (byte)1 : (byte)0; byte last = lastIsSeven ? (byte)1 : (byte)0;
var kernel = gpuLease.Pow2ModKernel; var kernel = gpuLease.Pow2ModKernel;
ulong step10 = ((exponent % 10UL) << 1) % 10UL;
ulong step8 = ((exponent & 7UL) << 1) & 7UL;
ulong step3 = ((exponent % 3UL) << 1) % 3UL;
ulong step5 = ((exponent % 5UL) << 1) % 5UL;
GpuUInt128 twoPGpu = (GpuUInt128)twoP;
var orderBuffer = accelerator.Allocate1D<ulong>(batchSize); var orderBuffer = accelerator.Allocate1D<ulong>(batchSize);
ulong[] orderArray = ArrayPool<ulong>.Shared.Rent(batchSize); ulong[] orderArray = ArrayPool<ulong>.Shared.Rent(batchSize);
@@ -47,25 +53,21 @@ public class MersenneNumberResidueGpuTester(bool useGpuOrder)
q = twoP * kStart + UInt128.One; q = twoP * kStart + UInt128.One;
q.Mod10_8_5_3(out ulong q0m10, out ulong q0m8, out ulong q0m5, out ulong q0m3); q.Mod10_8_5_3(out ulong q0m10, out ulong q0m8, out ulong q0m5, out ulong q0m3);
ulong step10 = ((exponent % 10UL) << 1) % 10UL; var ra = new ResidueAutomatonArgs(q0m10, step10, q0m8, step8, q0m3, step3, q0m5, step5);
ulong step8 = ((exponent & 7UL) << 1) & 7UL;
ulong step3 = ((exponent % 3UL) << 1) % 3UL;
ulong step5 = ((exponent % 5UL) << 1) % 5UL;
kernel(currentSize, exponent, (GpuUInt128)twoP, (GpuUInt128)kStart, last, 0UL, kernel(currentSize, exponent, twoPGpu, (GpuUInt128)kStart, last, 0UL,
q0m10, q0m8, q0m3, q0m5, orderBuffer.View, smallCyclesView); ra, orderBuffer.View, smallCyclesView, smallPrimesView, smallPrimesPow2View);
accelerator.Synchronize(); accelerator.Synchronize();
orderBuffer.View.CopyToCPU(ref MemoryMarshal.GetReference(orders), currentSize); orderBuffer.View.CopyToCPU(ref MemoryMarshal.GetReference(orders), currentSize);
accelerator.Synchronize();
if (!Volatile.Read(ref isPrime)) if (!Volatile.Read(ref isPrime))
{ {
break; break;
} }
for (i = 0; i < currentSize; i++, q += twoP) for (i = 0; i < currentSize; i++)
{ {
if (orders[i] != 0UL && q.IsPrimeCandidate()) if (orders[i] != 0UL)
{ {
Volatile.Write(ref isPrime, false); Volatile.Write(ref isPrime, false);
break; break;
PerfectNumbers.Core/PrimeTester.cs
+1 -1
View File
@@ -42,7 +42,7 @@ public sealed class PrimeTester(bool useInternal = false)
{ {
result = false; result = false;
} }
else if (n % 5 == 0) else if (n > 5UL && (n % 5UL) == 0UL)
{ {
result = false; result = false;
} }