Heuristic tables split into 2 groups
This commit is contained in:
mlemanczyk committed
1 parent
c413ee1162
commit
183b147135
26 files changed
+1172
-940
No files matched your search
@@ -0,0 +1,34 @@
|
||||
using ILGPU;
|
||||
using ILGPU.Runtime;
|
||||
|
||||
namespace PerfectNumbers.Core.Gpu.Accelerators;
|
||||
|
||||
public static class AcceleratorExtensions
|
||||
{
|
||||
public static MemoryBuffer1D<ulong, Stride1D.Dense> CopySpanToDevice(this Accelerator accelerator, AcceleratorStream stream, in ulong[] span)
|
||||
{
|
||||
MemoryBuffer1D<ulong, Stride1D.Dense>? buffer;
|
||||
buffer = accelerator.Allocate1D(stream, span);
|
||||
|
||||
return buffer;
|
||||
}
|
||||
|
||||
public static MemoryBuffer1D<ulong, Stride1D.Dense> CopyUintSpanToDevice(this Accelerator accelerator, AcceleratorStream stream, ReadOnlySpan<uint> span)
|
||||
{
|
||||
MemoryBuffer1D<ulong, Stride1D.Dense>? buffer;
|
||||
buffer = accelerator.Allocate1D<ulong>(span.Length);
|
||||
|
||||
if (!span.IsEmpty)
|
||||
{
|
||||
var converted = new ulong[span.Length];
|
||||
for (int i = 0; i < span.Length; i++)
|
||||
{
|
||||
converted[i] = span[i];
|
||||
}
|
||||
|
||||
buffer.View.CopyFromCPU(stream, converted);
|
||||
}
|
||||
|
||||
return buffer;
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,147 @@
|
||||
using System.Runtime.CompilerServices;
|
||||
using ILGPU;
|
||||
using ILGPU.Runtime;
|
||||
|
||||
namespace PerfectNumbers.Core.Gpu.Accelerators;
|
||||
|
||||
public sealed class HeuristicCombinedPrimeTesterAccelerator
|
||||
{
|
||||
#region Pool
|
||||
|
||||
[ThreadStatic]
|
||||
private static Queue<HeuristicCombinedPrimeTesterAccelerator>? Pool;
|
||||
|
||||
public static void WarmUp(int acceleratorIndex, AcceleratorStream stream)
|
||||
{
|
||||
Console.WriteLine($"Preparing heuristic accelerator {acceleratorIndex}...");
|
||||
HeuristicGpuCombinedTables.WarmUp(acceleratorIndex, stream);
|
||||
// _ = GpuKernelPool.GetOrAddKernels(accelerator, stream, KernelType.None);
|
||||
// KernelContainer kernels = GpuKernelPool.GetOrAddKernels(accelerator, stream);
|
||||
// GpuStaticTableInitializer.EnsureStaticTables(accelerator, kernels, stream);
|
||||
Console.WriteLine($"Heuristic accelerator {acceleratorIndex} is ready");
|
||||
}
|
||||
|
||||
public static HeuristicCombinedPrimeTesterAccelerator Rent(int minBufferCapacity = 1)
|
||||
{
|
||||
var pool = Pool ??= new();
|
||||
if (!pool.TryDequeue(out var lease))
|
||||
{
|
||||
lease = new HeuristicCombinedPrimeTesterAccelerator(minBufferCapacity);
|
||||
}
|
||||
else
|
||||
{
|
||||
lease.EnsureCapacity(minBufferCapacity);
|
||||
}
|
||||
|
||||
return lease;
|
||||
}
|
||||
|
||||
public static void Return(HeuristicCombinedPrimeTesterAccelerator lease)
|
||||
{
|
||||
var pool = Pool ??= new();
|
||||
pool.Enqueue(lease);
|
||||
}
|
||||
|
||||
public static void DisposeAll()
|
||||
{
|
||||
if (Pool != null)
|
||||
{
|
||||
while (Pool.TryDequeue(out var lease))
|
||||
{
|
||||
lease.DisposeResources();
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
public static void Clear()
|
||||
{
|
||||
if (Pool != null)
|
||||
{
|
||||
var retained = new List<HeuristicCombinedPrimeTesterAccelerator>();
|
||||
while (Pool.TryDequeue(out var lease))
|
||||
{
|
||||
lease.DisposeResources();
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
#endregion
|
||||
|
||||
public readonly HeuristicGpuCombinedDivisorTables DivisorTables;
|
||||
internal readonly Context Context;
|
||||
public readonly Accelerator Accelerator;
|
||||
public readonly Kernel HeuristicCombinedTrialDivisionKernel;
|
||||
|
||||
public MemoryBuffer1D<ulong, Stride1D.Dense>? Input = null;
|
||||
public MemoryBuffer1D<byte, Stride1D.Dense>? Output = null;
|
||||
public MemoryBuffer1D<int, Stride1D.Dense>? HeuristicFlag = null!;
|
||||
public int BufferCapacity;
|
||||
|
||||
private HeuristicGpuCombinedTables _sharedTables;
|
||||
public MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicCombinedDivisorsEnding1 => _sharedTables.HeuristicCombinedDivisorsEnding1;
|
||||
public MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicCombinedDivisorSquaresEnding1 => _sharedTables.HeuristicCombinedDivisorSquaresEnding1;
|
||||
public MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicCombinedDivisorsEnding3 => _sharedTables.HeuristicCombinedDivisorsEnding3;
|
||||
public MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicCombinedDivisorSquaresEnding3 => _sharedTables.HeuristicCombinedDivisorSquaresEnding3;
|
||||
public MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicCombinedDivisorsEnding7 => _sharedTables.HeuristicCombinedDivisorsEnding7;
|
||||
public MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicCombinedDivisorSquaresEnding7 => _sharedTables.HeuristicCombinedDivisorSquaresEnding7;
|
||||
public MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicCombinedDivisorsEnding9 => _sharedTables.HeuristicCombinedDivisorsEnding9;
|
||||
public MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicCombinedDivisorSquaresEnding9 => _sharedTables.HeuristicCombinedDivisorSquaresEnding9;
|
||||
|
||||
private static readonly Accelerator[] _accelerators = AcceleratorPool.Shared.Accelerators;
|
||||
internal readonly int AcceleratorIndex;
|
||||
|
||||
public HeuristicCombinedPrimeTesterAccelerator(int minBufferCapacity)
|
||||
{
|
||||
var acceleratorIndex = AcceleratorPool.Shared.Rent();
|
||||
AcceleratorIndex = acceleratorIndex;
|
||||
Accelerator accelerator = _accelerators[acceleratorIndex];
|
||||
Context = accelerator.Context;
|
||||
|
||||
Accelerator = accelerator;
|
||||
EnsureCapacity(minBufferCapacity);
|
||||
|
||||
HeuristicCombinedTrialDivisionKernel = KernelUtil.GetKernel(accelerator.LoadAutoGroupedStreamKernel<Index1D, byte, ArrayView<int>, ulong, ulong, HeuristicGpuCombinedDivisorTables>(PrimeTesterKernels.HeuristicTrialCombinedDivisionKernel));
|
||||
|
||||
AcceleratorStream stream = AcceleratorStreamPool.Rent(acceleratorIndex);
|
||||
|
||||
// GpuStaticTableInitializer.EnsureStaticTables(accelerator, kernels, stream);
|
||||
|
||||
_sharedTables = HeuristicGpuCombinedTables.GetStaticTables(acceleratorIndex);
|
||||
DivisorTables = _sharedTables.CreateHeuristicDivisorTables();
|
||||
|
||||
stream.Synchronize();
|
||||
AcceleratorStreamPool.Return(acceleratorIndex, stream);
|
||||
}
|
||||
|
||||
public void EnsureCapacity(int minCapacity)
|
||||
{
|
||||
if (minCapacity <= BufferCapacity)
|
||||
{
|
||||
return;
|
||||
}
|
||||
|
||||
Input?.Dispose();
|
||||
Output?.Dispose();
|
||||
HeuristicFlag?.Dispose();
|
||||
|
||||
Input = Accelerator.Allocate1D<ulong>(minCapacity);
|
||||
Output = Accelerator.Allocate1D<byte>(minCapacity);
|
||||
HeuristicFlag = Accelerator.Allocate1D<int>(minCapacity);
|
||||
|
||||
BufferCapacity = minCapacity;
|
||||
}
|
||||
|
||||
[MethodImpl(MethodImplOptions.AggressiveInlining)]
|
||||
public void Return() => Return(this);
|
||||
|
||||
public void DisposeResources()
|
||||
{
|
||||
Input?.Dispose();
|
||||
Output?.Dispose();
|
||||
HeuristicFlag?.Dispose();
|
||||
// ReleaseSharedTables(_sharedTables);
|
||||
// These resources are shared between GPU leases
|
||||
// Accelerator.Dispose();
|
||||
// AcceleratorContext.Dispose();
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,69 @@
|
||||
using ILGPU;
|
||||
using ILGPU.Runtime;
|
||||
|
||||
namespace PerfectNumbers.Core.Gpu.Accelerators;
|
||||
|
||||
internal sealed class HeuristicGpuCombinedTables
|
||||
{
|
||||
private static readonly Accelerator[] _accelerators = AcceleratorPool.Shared.Accelerators;
|
||||
|
||||
private static readonly HeuristicGpuCombinedTables[] _sharedTables = new HeuristicGpuCombinedTables[PerfectNumberConstants.RollingAccelerators];
|
||||
|
||||
internal static void WarmUp(int acceleratorIndex, AcceleratorStream stream) => _sharedTables[acceleratorIndex] = new(_accelerators[acceleratorIndex], stream);
|
||||
|
||||
internal static HeuristicGpuCombinedTables GetStaticTables(int acceleratorIndex) => _sharedTables[acceleratorIndex];
|
||||
|
||||
internal HeuristicGpuCombinedTables(Accelerator accelerator, AcceleratorStream stream)
|
||||
{
|
||||
var combinedEnding1 = HeuristicCombinedPrimeTester.CombinedDivisorsEnding1;
|
||||
HeuristicCombinedDivisorsEnding1 = accelerator.CopySpanToDevice(stream, combinedEnding1);
|
||||
var combinedSquaresEnding1 = HeuristicCombinedPrimeTester.CombinedDivisorsEnding1Squares;
|
||||
HeuristicCombinedDivisorSquaresEnding1 = accelerator.CopySpanToDevice(stream, combinedSquaresEnding1);
|
||||
|
||||
var combinedEnding3 = HeuristicCombinedPrimeTester.CombinedDivisorsEnding3;
|
||||
HeuristicCombinedDivisorsEnding3 = accelerator.CopySpanToDevice(stream, combinedEnding3);
|
||||
var combinedSquaresEnding3 = HeuristicCombinedPrimeTester.CombinedDivisorsEnding3Squares;
|
||||
HeuristicCombinedDivisorSquaresEnding3 = accelerator.CopySpanToDevice(stream, combinedSquaresEnding3);
|
||||
|
||||
var combinedEnding7 = HeuristicCombinedPrimeTester.CombinedDivisorsEnding7;
|
||||
HeuristicCombinedDivisorsEnding7 = accelerator.CopySpanToDevice(stream, combinedEnding7);
|
||||
var combinedSquaresEnding7 = HeuristicCombinedPrimeTester.CombinedDivisorsEnding7Squares;
|
||||
HeuristicCombinedDivisorSquaresEnding7 = accelerator.CopySpanToDevice(stream, combinedSquaresEnding7);
|
||||
|
||||
var combinedEnding9 = HeuristicCombinedPrimeTester.CombinedDivisorsEnding9;
|
||||
HeuristicCombinedDivisorsEnding9 = accelerator.CopySpanToDevice(stream, combinedEnding9);
|
||||
var combinedSquaresEnding9 = HeuristicCombinedPrimeTester.CombinedDivisorsEnding9Squares;
|
||||
HeuristicCombinedDivisorSquaresEnding9 = accelerator.CopySpanToDevice(stream, combinedSquaresEnding9);
|
||||
}
|
||||
|
||||
internal readonly MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicCombinedDivisorsEnding1;
|
||||
internal readonly MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicCombinedDivisorSquaresEnding1;
|
||||
internal readonly MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicCombinedDivisorsEnding3;
|
||||
internal readonly MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicCombinedDivisorSquaresEnding3;
|
||||
internal readonly MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicCombinedDivisorsEnding7;
|
||||
internal readonly MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicCombinedDivisorSquaresEnding7;
|
||||
internal readonly MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicCombinedDivisorsEnding9;
|
||||
internal readonly MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicCombinedDivisorSquaresEnding9;
|
||||
|
||||
internal HeuristicGpuCombinedDivisorTables CreateHeuristicDivisorTables() => new(
|
||||
HeuristicCombinedDivisorsEnding1.View,
|
||||
HeuristicCombinedDivisorsEnding3.View,
|
||||
HeuristicCombinedDivisorsEnding7.View,
|
||||
HeuristicCombinedDivisorsEnding9.View,
|
||||
HeuristicCombinedDivisorSquaresEnding1.View,
|
||||
HeuristicCombinedDivisorSquaresEnding3.View,
|
||||
HeuristicCombinedDivisorSquaresEnding7.View,
|
||||
HeuristicCombinedDivisorSquaresEnding9.View);
|
||||
|
||||
internal void Dispose()
|
||||
{
|
||||
HeuristicCombinedDivisorsEnding1.Dispose();
|
||||
HeuristicCombinedDivisorSquaresEnding1.Dispose();
|
||||
HeuristicCombinedDivisorsEnding3.Dispose();
|
||||
HeuristicCombinedDivisorSquaresEnding3.Dispose();
|
||||
HeuristicCombinedDivisorsEnding7.Dispose();
|
||||
HeuristicCombinedDivisorSquaresEnding7.Dispose();
|
||||
HeuristicCombinedDivisorsEnding9.Dispose();
|
||||
HeuristicCombinedDivisorSquaresEnding9.Dispose();
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,79 @@
|
||||
using ILGPU;
|
||||
using ILGPU.Runtime;
|
||||
|
||||
namespace PerfectNumbers.Core.Gpu.Accelerators;
|
||||
|
||||
internal sealed class HeuristicGpuGroupABTables
|
||||
{
|
||||
private static readonly Accelerator[] _accelerators = AcceleratorPool.Shared.Accelerators;
|
||||
|
||||
private static readonly HeuristicGpuGroupABTables[] _sharedTables = new HeuristicGpuGroupABTables[PerfectNumberConstants.RollingAccelerators];
|
||||
|
||||
private static readonly SemaphoreSlim[] _locks = [..Enumerable.Range(1, PerfectNumberConstants.RollingAccelerators).Select(_ => new SemaphoreSlim(1))];
|
||||
|
||||
internal static HeuristicGpuGroupABTables EnsureStaticTables(int acceleratorIndex, AcceleratorStream stream)
|
||||
{
|
||||
var @lock = _locks[acceleratorIndex];
|
||||
@lock.Wait();
|
||||
if (_sharedTables[acceleratorIndex] is {} existing)
|
||||
{
|
||||
@lock.Release();
|
||||
return existing;
|
||||
}
|
||||
|
||||
var accelerator = _accelerators[acceleratorIndex];
|
||||
existing = new(accelerator, stream);
|
||||
_sharedTables[acceleratorIndex] = existing;
|
||||
@lock.Release();
|
||||
return existing;
|
||||
}
|
||||
|
||||
internal HeuristicGpuGroupABTables(Accelerator accelerator, AcceleratorStream stream)
|
||||
{
|
||||
var heuristicGroupA = HeuristicPrimeSieves.GroupADivisorsStorage;
|
||||
HeuristicGroupADivisors = accelerator.CopySpanToDevice(stream, heuristicGroupA);
|
||||
|
||||
var heuristicGroupASquares = HeuristicPrimeSieves.GroupADivisorSquaresStorage;
|
||||
HeuristicGroupADivisorSquares = accelerator.CopySpanToDevice(stream, heuristicGroupASquares);
|
||||
|
||||
HeuristicGroupBDivisorsEnding1 = accelerator.CopyUintSpanToDevice(stream, DivisorGenerator.SmallPrimesLastOneWithoutLastThree);
|
||||
HeuristicGroupBDivisorSquaresEnding1 = accelerator.CopySpanToDevice(stream, DivisorGenerator.SmallPrimesPow2LastOneWithoutLastThree);
|
||||
|
||||
HeuristicGroupBDivisorsEnding7 = accelerator.CopyUintSpanToDevice(stream, DivisorGenerator.SmallPrimesLastSevenWithoutLastThree);
|
||||
HeuristicGroupBDivisorSquaresEnding7 = accelerator.CopySpanToDevice(stream, DivisorGenerator.SmallPrimesPow2LastSevenWithoutLastThree);
|
||||
|
||||
HeuristicGroupBDivisorsEnding9 = accelerator.CopyUintSpanToDevice(stream, DivisorGenerator.SmallPrimesLastNineWithoutLastThree);
|
||||
HeuristicGroupBDivisorSquaresEnding9 = accelerator.CopySpanToDevice(stream, DivisorGenerator.SmallPrimesPow2LastNineWithoutLastThree);
|
||||
}
|
||||
|
||||
internal readonly MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicGroupADivisors;
|
||||
internal readonly MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicGroupADivisorSquares;
|
||||
internal readonly MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicGroupBDivisorsEnding1;
|
||||
internal readonly MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicGroupBDivisorSquaresEnding1;
|
||||
internal readonly MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicGroupBDivisorsEnding7;
|
||||
internal readonly MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicGroupBDivisorSquaresEnding7;
|
||||
internal readonly MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicGroupBDivisorsEnding9;
|
||||
internal readonly MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicGroupBDivisorSquaresEnding9;
|
||||
|
||||
internal HeuristicGpuGroupABDivisorTables CreateHeuristicDivisorTables() => new(
|
||||
HeuristicGroupADivisors.View,
|
||||
HeuristicGroupADivisorSquares.View,
|
||||
HeuristicGroupBDivisorsEnding1.View,
|
||||
HeuristicGroupBDivisorSquaresEnding1.View,
|
||||
HeuristicGroupBDivisorsEnding7.View,
|
||||
HeuristicGroupBDivisorSquaresEnding7.View,
|
||||
HeuristicGroupBDivisorsEnding9.View,
|
||||
HeuristicGroupBDivisorSquaresEnding9.View);
|
||||
|
||||
internal void Dispose()
|
||||
{
|
||||
HeuristicGroupADivisors.Dispose();
|
||||
HeuristicGroupADivisorSquares.Dispose();
|
||||
HeuristicGroupBDivisorsEnding1.Dispose();
|
||||
HeuristicGroupBDivisorSquaresEnding1.Dispose();
|
||||
HeuristicGroupBDivisorsEnding7.Dispose();
|
||||
HeuristicGroupBDivisorSquaresEnding7.Dispose();
|
||||
HeuristicGroupBDivisorsEnding9.Dispose();
|
||||
HeuristicGroupBDivisorSquaresEnding9.Dispose();
|
||||
}
|
||||
}
|
||||
@@ -1,150 +0,0 @@
|
||||
using ILGPU;
|
||||
using ILGPU.Runtime;
|
||||
|
||||
namespace PerfectNumbers.Core.Gpu.Accelerators;
|
||||
|
||||
internal sealed class HeuristicGpuTables
|
||||
{
|
||||
private static readonly Accelerator[] _accelerators = AcceleratorPool.Shared.Accelerators;
|
||||
|
||||
private static readonly HeuristicGpuTables[] _sharedTables = new HeuristicGpuTables[PerfectNumberConstants.RollingAccelerators];
|
||||
|
||||
private static readonly SemaphoreSlim[] _locks = [..Enumerable.Range(1, PerfectNumberConstants.RollingAccelerators).Select(_ => new SemaphoreSlim(1))];
|
||||
|
||||
internal static HeuristicGpuTables EnsureStaticTables(int acceleratorIndex, AcceleratorStream stream)
|
||||
{
|
||||
var @lock = _locks[acceleratorIndex];
|
||||
@lock.Wait();
|
||||
if (_sharedTables[acceleratorIndex] is {} existing)
|
||||
{
|
||||
@lock.Release();
|
||||
return existing;
|
||||
}
|
||||
|
||||
var accelerator = _accelerators[acceleratorIndex];
|
||||
existing = new(accelerator, stream);
|
||||
_sharedTables[acceleratorIndex] = existing;
|
||||
@lock.Release();
|
||||
return existing;
|
||||
}
|
||||
|
||||
internal HeuristicGpuTables(Accelerator accelerator, AcceleratorStream stream)
|
||||
{
|
||||
var heuristicGroupA = HeuristicPrimeSieves.GroupADivisorsStorage;
|
||||
HeuristicGroupADivisors = CopySpanToDevice(accelerator, stream, heuristicGroupA);
|
||||
|
||||
var heuristicGroupASquares = HeuristicPrimeSieves.GroupADivisorSquaresStorage;
|
||||
HeuristicGroupADivisorSquares = CopySpanToDevice(accelerator, stream, heuristicGroupASquares);
|
||||
|
||||
HeuristicGroupBDivisorsEnding1 = CopyUintSpanToDevice(accelerator, stream, DivisorGenerator.SmallPrimesLastOneWithoutLastThree);
|
||||
HeuristicGroupBDivisorSquaresEnding1 = CopySpanToDevice(accelerator, stream, DivisorGenerator.SmallPrimesPow2LastOneWithoutLastThree);
|
||||
|
||||
HeuristicGroupBDivisorsEnding7 = CopyUintSpanToDevice(accelerator, stream, DivisorGenerator.SmallPrimesLastSevenWithoutLastThree);
|
||||
HeuristicGroupBDivisorSquaresEnding7 = CopySpanToDevice(accelerator, stream, DivisorGenerator.SmallPrimesPow2LastSevenWithoutLastThree);
|
||||
|
||||
HeuristicGroupBDivisorsEnding9 = CopyUintSpanToDevice(accelerator, stream, DivisorGenerator.SmallPrimesLastNineWithoutLastThree);
|
||||
HeuristicGroupBDivisorSquaresEnding9 = CopySpanToDevice(accelerator, stream, DivisorGenerator.SmallPrimesPow2LastNineWithoutLastThree);
|
||||
|
||||
var combinedEnding1 = HeuristicCombinedPrimeTester.CombinedDivisorsEnding1;
|
||||
HeuristicCombinedDivisorsEnding1 = CopySpanToDevice(accelerator, stream, combinedEnding1);
|
||||
var combinedSquaresEnding1 = HeuristicCombinedPrimeTester.CombinedDivisorsEnding1Squares;
|
||||
HeuristicCombinedDivisorSquaresEnding1 = CopySpanToDevice(accelerator, stream, combinedSquaresEnding1);
|
||||
|
||||
var combinedEnding3 = HeuristicCombinedPrimeTester.CombinedDivisorsEnding3;
|
||||
HeuristicCombinedDivisorsEnding3 = CopySpanToDevice(accelerator, stream, combinedEnding3);
|
||||
var combinedSquaresEnding3 = HeuristicCombinedPrimeTester.CombinedDivisorsEnding3Squares;
|
||||
HeuristicCombinedDivisorSquaresEnding3 = CopySpanToDevice(accelerator, stream, combinedSquaresEnding3);
|
||||
|
||||
var combinedEnding7 = HeuristicCombinedPrimeTester.CombinedDivisorsEnding7;
|
||||
HeuristicCombinedDivisorsEnding7 = CopySpanToDevice(accelerator, stream, combinedEnding7);
|
||||
var combinedSquaresEnding7 = HeuristicCombinedPrimeTester.CombinedDivisorsEnding7Squares;
|
||||
HeuristicCombinedDivisorSquaresEnding7 = CopySpanToDevice(accelerator, stream, combinedSquaresEnding7);
|
||||
|
||||
var combinedEnding9 = HeuristicCombinedPrimeTester.CombinedDivisorsEnding9;
|
||||
HeuristicCombinedDivisorsEnding9 = CopySpanToDevice(accelerator, stream, combinedEnding9);
|
||||
var combinedSquaresEnding9 = HeuristicCombinedPrimeTester.CombinedDivisorsEnding9Squares;
|
||||
HeuristicCombinedDivisorSquaresEnding9 = CopySpanToDevice(accelerator, stream, combinedSquaresEnding9);
|
||||
}
|
||||
|
||||
internal readonly MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicGroupADivisors;
|
||||
internal readonly MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicGroupADivisorSquares;
|
||||
internal readonly MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicGroupBDivisorsEnding1;
|
||||
internal readonly MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicGroupBDivisorSquaresEnding1;
|
||||
internal readonly MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicGroupBDivisorsEnding7;
|
||||
internal readonly MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicGroupBDivisorSquaresEnding7;
|
||||
internal readonly MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicGroupBDivisorsEnding9;
|
||||
internal readonly MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicGroupBDivisorSquaresEnding9;
|
||||
internal readonly MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicCombinedDivisorsEnding1;
|
||||
internal readonly MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicCombinedDivisorSquaresEnding1;
|
||||
internal readonly MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicCombinedDivisorsEnding3;
|
||||
internal readonly MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicCombinedDivisorSquaresEnding3;
|
||||
internal readonly MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicCombinedDivisorsEnding7;
|
||||
internal readonly MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicCombinedDivisorSquaresEnding7;
|
||||
internal readonly MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicCombinedDivisorsEnding9;
|
||||
internal readonly MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicCombinedDivisorSquaresEnding9;
|
||||
|
||||
internal HeuristicGpuDivisorTables CreateHeuristicDivisorTables() => new(
|
||||
HeuristicCombinedDivisorsEnding1.View,
|
||||
HeuristicCombinedDivisorsEnding3.View,
|
||||
HeuristicCombinedDivisorsEnding7.View,
|
||||
HeuristicCombinedDivisorsEnding9.View,
|
||||
HeuristicCombinedDivisorSquaresEnding1.View,
|
||||
HeuristicCombinedDivisorSquaresEnding3.View,
|
||||
HeuristicCombinedDivisorSquaresEnding7.View,
|
||||
HeuristicCombinedDivisorSquaresEnding9.View,
|
||||
HeuristicGroupADivisors.View,
|
||||
HeuristicGroupADivisorSquares.View,
|
||||
HeuristicGroupBDivisorsEnding1.View,
|
||||
HeuristicGroupBDivisorSquaresEnding1.View,
|
||||
HeuristicGroupBDivisorsEnding7.View,
|
||||
HeuristicGroupBDivisorSquaresEnding7.View,
|
||||
HeuristicGroupBDivisorsEnding9.View,
|
||||
HeuristicGroupBDivisorSquaresEnding9.View);
|
||||
|
||||
private static MemoryBuffer1D<ulong, Stride1D.Dense> CopySpanToDevice(Accelerator accelerator, AcceleratorStream stream, in ulong[] span)
|
||||
{
|
||||
MemoryBuffer1D<ulong, Stride1D.Dense>? buffer;
|
||||
buffer = accelerator.Allocate1D(stream, span);
|
||||
|
||||
return buffer;
|
||||
}
|
||||
|
||||
private static MemoryBuffer1D<ulong, Stride1D.Dense> CopyUintSpanToDevice(Accelerator accelerator, AcceleratorStream stream, ReadOnlySpan<uint> span)
|
||||
{
|
||||
MemoryBuffer1D<ulong, Stride1D.Dense>? buffer;
|
||||
buffer = accelerator.Allocate1D<ulong>(span.Length);
|
||||
|
||||
if (!span.IsEmpty)
|
||||
{
|
||||
var converted = new ulong[span.Length];
|
||||
for (int i = 0; i < span.Length; i++)
|
||||
{
|
||||
converted[i] = span[i];
|
||||
}
|
||||
|
||||
buffer.View.CopyFromCPU(stream, converted);
|
||||
}
|
||||
|
||||
return buffer;
|
||||
}
|
||||
|
||||
internal void Dispose()
|
||||
{
|
||||
HeuristicGroupADivisors.Dispose();
|
||||
HeuristicGroupADivisorSquares.Dispose();
|
||||
HeuristicGroupBDivisorsEnding1.Dispose();
|
||||
HeuristicGroupBDivisorSquaresEnding1.Dispose();
|
||||
HeuristicGroupBDivisorsEnding7.Dispose();
|
||||
HeuristicGroupBDivisorSquaresEnding7.Dispose();
|
||||
HeuristicGroupBDivisorsEnding9.Dispose();
|
||||
HeuristicGroupBDivisorSquaresEnding9.Dispose();
|
||||
HeuristicCombinedDivisorsEnding1.Dispose();
|
||||
HeuristicCombinedDivisorSquaresEnding1.Dispose();
|
||||
HeuristicCombinedDivisorsEnding3.Dispose();
|
||||
HeuristicCombinedDivisorSquaresEnding3.Dispose();
|
||||
HeuristicCombinedDivisorsEnding7.Dispose();
|
||||
HeuristicCombinedDivisorSquaresEnding7.Dispose();
|
||||
HeuristicCombinedDivisorsEnding9.Dispose();
|
||||
HeuristicCombinedDivisorSquaresEnding9.Dispose();
|
||||
}
|
||||
}
|
||||
+20
-35
@@ -1,15 +1,14 @@
|
||||
using System.Runtime.CompilerServices;
|
||||
using ILGPU;
|
||||
using ILGPU.Runtime;
|
||||
|
||||
namespace PerfectNumbers.Core.Gpu.Accelerators;
|
||||
|
||||
internal sealed class HeuristicPrimeTesterAccelerator
|
||||
internal sealed class HeuristicGroupABPrimeTesterAccelerator
|
||||
{
|
||||
#region Pool
|
||||
|
||||
[ThreadStatic]
|
||||
private static Queue<HeuristicPrimeTesterAccelerator>? Pool;
|
||||
private static Queue<HeuristicGroupABPrimeTesterAccelerator>? Pool;
|
||||
|
||||
internal static void WarmUp()
|
||||
{
|
||||
@@ -20,23 +19,22 @@ internal sealed class HeuristicPrimeTesterAccelerator
|
||||
Console.WriteLine($"Preparing heuristic accelerator {i}...");
|
||||
var accelerator = accelerators[i];
|
||||
AcceleratorStream stream = accelerator.CreateStream();
|
||||
Accelerators.HeuristicGpuTables.EnsureStaticTables(i, stream);
|
||||
_ = HeuristicGpuGroupABTables.EnsureStaticTables(i, stream);
|
||||
// _ = GpuKernelPool.GetOrAddKernels(accelerator, stream, KernelType.None);
|
||||
// KernelContainer kernels = GpuKernelPool.GetOrAddKernels(accelerator, stream);
|
||||
// GpuStaticTableInitializer.EnsureStaticTables(accelerator, kernels, stream);
|
||||
stream.Synchronize();
|
||||
stream.Dispose();
|
||||
accelerator.Synchronize();
|
||||
Console.WriteLine($"Heuristic accelerator {i} is ready");
|
||||
}
|
||||
}
|
||||
|
||||
internal static HeuristicPrimeTesterAccelerator Rent(int minBufferCapacity = 1)
|
||||
internal static HeuristicGroupABPrimeTesterAccelerator Rent(int minBufferCapacity = 1)
|
||||
{
|
||||
var pool = Pool ??= new();
|
||||
if (!pool.TryDequeue(out var lease))
|
||||
{
|
||||
lease = new HeuristicPrimeTesterAccelerator(minBufferCapacity);
|
||||
lease = new HeuristicGroupABPrimeTesterAccelerator(minBufferCapacity);
|
||||
}
|
||||
else
|
||||
{
|
||||
@@ -46,7 +44,7 @@ internal sealed class HeuristicPrimeTesterAccelerator
|
||||
return lease;
|
||||
}
|
||||
|
||||
internal static void Return(HeuristicPrimeTesterAccelerator lease)
|
||||
internal static void Return(HeuristicGroupABPrimeTesterAccelerator lease)
|
||||
{
|
||||
var pool = Pool ??= new();
|
||||
pool.Enqueue(lease);
|
||||
@@ -67,7 +65,7 @@ internal sealed class HeuristicPrimeTesterAccelerator
|
||||
{
|
||||
if (Pool != null)
|
||||
{
|
||||
var retained = new List<HeuristicPrimeTesterAccelerator>();
|
||||
var retained = new List<HeuristicGroupABPrimeTesterAccelerator>();
|
||||
while (Pool.TryDequeue(out var lease))
|
||||
{
|
||||
lease.DisposeResources();
|
||||
@@ -77,10 +75,9 @@ internal sealed class HeuristicPrimeTesterAccelerator
|
||||
|
||||
#endregion
|
||||
|
||||
private readonly HeuristicGpuDivisorTables _heuristicDivisorTables;
|
||||
public readonly HeuristicGpuGroupABDivisorTables DivisorTables;
|
||||
internal readonly Context Context;
|
||||
public readonly Accelerator Accelerator;
|
||||
public AcceleratorStream? Stream;
|
||||
|
||||
[ThreadStatic]
|
||||
public static Dictionary<Accelerator, Action<AcceleratorStream, Index1D, ArrayView<ulong>, ArrayView<uint>, ArrayView<uint>, ArrayView<uint>, ArrayView<uint>, ArrayView<ulong>, ArrayView<ulong>, ArrayView<ulong>, ArrayView<ulong>, ArrayView<byte>>>? SmallPrimeSieveKernel;
|
||||
@@ -100,12 +97,12 @@ internal sealed class HeuristicPrimeTesterAccelerator
|
||||
|
||||
public readonly Kernel HeuristicTrialDivisionKernel;
|
||||
|
||||
public MemoryBuffer1D<ulong, Stride1D.Dense> Input = null!;
|
||||
public MemoryBuffer1D<byte, Stride1D.Dense> Output = null!;
|
||||
public MemoryBuffer1D<int, Stride1D.Dense> HeuristicFlag = null!;
|
||||
public MemoryBuffer1D<ulong, Stride1D.Dense>? Input = null;
|
||||
public MemoryBuffer1D<byte, Stride1D.Dense>? Output = null;
|
||||
public MemoryBuffer1D<int, Stride1D.Dense>? HeuristicFlag = null!;
|
||||
public int BufferCapacity;
|
||||
|
||||
private HeuristicGpuTables _sharedTables;
|
||||
private HeuristicGpuGroupABTables _sharedTables;
|
||||
public MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicGroupADivisors => _sharedTables.HeuristicGroupADivisors;
|
||||
public MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicGroupADivisorSquares => _sharedTables.HeuristicGroupADivisorSquares;
|
||||
public MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicGroupBDivisorsEnding1 => _sharedTables.HeuristicGroupBDivisorsEnding1;
|
||||
@@ -114,36 +111,28 @@ internal sealed class HeuristicPrimeTesterAccelerator
|
||||
public MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicGroupBDivisorSquaresEnding7 => _sharedTables.HeuristicGroupBDivisorSquaresEnding7;
|
||||
public MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicGroupBDivisorsEnding9 => _sharedTables.HeuristicGroupBDivisorsEnding9;
|
||||
public MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicGroupBDivisorSquaresEnding9 => _sharedTables.HeuristicGroupBDivisorSquaresEnding9;
|
||||
public MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicCombinedDivisorsEnding1 => _sharedTables.HeuristicCombinedDivisorsEnding1;
|
||||
public MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicCombinedDivisorSquaresEnding1 => _sharedTables.HeuristicCombinedDivisorSquaresEnding1;
|
||||
public MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicCombinedDivisorsEnding3 => _sharedTables.HeuristicCombinedDivisorsEnding3;
|
||||
public MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicCombinedDivisorSquaresEnding3 => _sharedTables.HeuristicCombinedDivisorSquaresEnding3;
|
||||
public MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicCombinedDivisorsEnding7 => _sharedTables.HeuristicCombinedDivisorsEnding7;
|
||||
public MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicCombinedDivisorSquaresEnding7 => _sharedTables.HeuristicCombinedDivisorSquaresEnding7;
|
||||
public MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicCombinedDivisorsEnding9 => _sharedTables.HeuristicCombinedDivisorsEnding9;
|
||||
public MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicCombinedDivisorSquaresEnding9 => _sharedTables.HeuristicCombinedDivisorSquaresEnding9;
|
||||
|
||||
public HeuristicGpuDivisorTables HeuristicGpuTables => _heuristicDivisorTables;
|
||||
|
||||
private static readonly Accelerator[] _accelerators = AcceleratorPool.Shared.Accelerators;
|
||||
internal readonly int AcceleratorIndex;
|
||||
|
||||
internal HeuristicPrimeTesterAccelerator(int minBufferCapacity)
|
||||
internal HeuristicGroupABPrimeTesterAccelerator(int minBufferCapacity)
|
||||
{
|
||||
var acceleratorIndex = AcceleratorPool.Shared.Rent();
|
||||
AcceleratorIndex = acceleratorIndex;
|
||||
Accelerator accelerator = _accelerators[acceleratorIndex];
|
||||
Context = accelerator.Context;
|
||||
|
||||
Accelerator = accelerator;
|
||||
EnsureCapacity(minBufferCapacity);
|
||||
|
||||
HeuristicTrialDivisionKernel = KernelUtil.GetKernel(accelerator.LoadAutoGroupedStreamKernel<Index1D, ArrayView<int>, ulong, ulong, HeuristicGpuDivisorTableKind, HeuristicGpuDivisorTables>(PrimeTesterKernels.HeuristicTrialDivisionKernel));
|
||||
HeuristicTrialDivisionKernel = KernelUtil.GetKernel(accelerator.LoadAutoGroupedStreamKernel<Index1D, ArrayView<int>, ulong, ulong, HeuristicGpuDivisorTableKind, HeuristicGpuGroupABDivisorTables>(PrimeTesterKernels.HeuristicTrialGroupABDivisionKernel));
|
||||
|
||||
AcceleratorStream stream = AcceleratorStreamPool.Rent(acceleratorIndex);
|
||||
|
||||
// GpuStaticTableInitializer.EnsureStaticTables(accelerator, kernels, stream);
|
||||
|
||||
_sharedTables = Accelerators.HeuristicGpuTables.EnsureStaticTables(acceleratorIndex, stream);
|
||||
_heuristicDivisorTables = _sharedTables.CreateHeuristicDivisorTables();
|
||||
_sharedTables = HeuristicGpuGroupABTables.EnsureStaticTables(acceleratorIndex, stream);
|
||||
DivisorTables = _sharedTables.CreateHeuristicDivisorTables();
|
||||
|
||||
stream.Synchronize();
|
||||
AcceleratorStreamPool.Return(acceleratorIndex, stream);
|
||||
@@ -156,8 +145,8 @@ internal sealed class HeuristicPrimeTesterAccelerator
|
||||
return;
|
||||
}
|
||||
|
||||
Input.Dispose();
|
||||
Output.Dispose();
|
||||
Input?.Dispose();
|
||||
Output?.Dispose();
|
||||
HeuristicFlag?.Dispose();
|
||||
|
||||
Input = Accelerator.Allocate1D<ulong>(minCapacity);
|
||||
@@ -169,8 +158,6 @@ internal sealed class HeuristicPrimeTesterAccelerator
|
||||
|
||||
public void Dispose()
|
||||
{
|
||||
Stream?.Dispose();
|
||||
Stream = null;
|
||||
Return(this);
|
||||
}
|
||||
|
||||
@@ -179,8 +166,6 @@ internal sealed class HeuristicPrimeTesterAccelerator
|
||||
Input?.Dispose();
|
||||
Output?.Dispose();
|
||||
HeuristicFlag?.Dispose();
|
||||
Stream?.Dispose();
|
||||
Stream = null;
|
||||
// ReleaseSharedTables(_sharedTables);
|
||||
// These resources are shared between GPU leases
|
||||
// Accelerator.Dispose();
|
||||
@@ -68,6 +68,7 @@ public sealed class PrimeOrderCalculatorAccelerator
|
||||
// Don't take this from the pool as quick uploads of data to the accelerator consumes much of GPU's memory and throws.
|
||||
AcceleratorStream stream = accelerator.CreateStream();
|
||||
LastDigitGpuTables.WarmUp(i, stream);
|
||||
HeuristicCombinedPrimeTesterAccelerator.WarmUp(i, stream);
|
||||
// SharedHeuristicGpuTables.EnsureStaticTables(accelerator, stream);
|
||||
// _ = GpuKernelPool.GetOrAddKernels(accelerator, stream, KernelType.None);
|
||||
// KernelContainer kernels = GpuKernelPool.GetOrAddKernels(accelerator, stream);
|
||||
@@ -76,6 +77,7 @@ public sealed class PrimeOrderCalculatorAccelerator
|
||||
stream.Dispose();
|
||||
accelerator.Synchronize();
|
||||
}
|
||||
|
||||
}
|
||||
|
||||
#endregion
|
||||
|
||||
@@ -0,0 +1,45 @@
|
||||
using System.Runtime.CompilerServices;
|
||||
using ILGPU;
|
||||
using ILGPU.Runtime;
|
||||
|
||||
namespace PerfectNumbers.Core.Gpu;
|
||||
|
||||
public readonly struct HeuristicGpuCombinedDivisorTables(
|
||||
ArrayView1D<ulong, Stride1D.Dense> combinedDivisorsEnding1,
|
||||
ArrayView1D<ulong, Stride1D.Dense> combinedDivisorsEnding3,
|
||||
ArrayView1D<ulong, Stride1D.Dense> combinedDivisorsEnding7,
|
||||
ArrayView1D<ulong, Stride1D.Dense> combinedDivisorsEnding9,
|
||||
ArrayView1D<ulong, Stride1D.Dense> combinedDivisorSquaresEnding1,
|
||||
ArrayView1D<ulong, Stride1D.Dense> combinedDivisorSquaresEnding3,
|
||||
ArrayView1D<ulong, Stride1D.Dense> combinedDivisorSquaresEnding7,
|
||||
ArrayView1D<ulong, Stride1D.Dense> combinedDivisorSquaresEnding9)
|
||||
{
|
||||
public readonly ArrayView1D<ulong, Stride1D.Dense> CombinedDivisorsEnding1 = combinedDivisorsEnding1;
|
||||
public readonly ArrayView1D<ulong, Stride1D.Dense> CombinedDivisorsEnding3 = combinedDivisorsEnding3;
|
||||
public readonly ArrayView1D<ulong, Stride1D.Dense> CombinedDivisorsEnding7 = combinedDivisorsEnding7;
|
||||
public readonly ArrayView1D<ulong, Stride1D.Dense> CombinedDivisorsEnding9 = combinedDivisorsEnding9;
|
||||
public readonly ArrayView1D<ulong, Stride1D.Dense> CombinedDivisorSquaresEnding1 = combinedDivisorSquaresEnding1;
|
||||
public readonly ArrayView1D<ulong, Stride1D.Dense> CombinedDivisorSquaresEnding3 = combinedDivisorSquaresEnding3;
|
||||
public readonly ArrayView1D<ulong, Stride1D.Dense> CombinedDivisorSquaresEnding7 = combinedDivisorSquaresEnding7;
|
||||
public readonly ArrayView1D<ulong, Stride1D.Dense> CombinedDivisorSquaresEnding9 = combinedDivisorSquaresEnding9;
|
||||
|
||||
[MethodImpl(MethodImplOptions.AggressiveInlining)]
|
||||
public ArrayView1D<ulong, Stride1D.Dense> SelectDivisors(byte ending) => ending switch
|
||||
{
|
||||
1 => CombinedDivisorsEnding1,
|
||||
3 => CombinedDivisorsEnding3,
|
||||
7 => CombinedDivisorsEnding7,
|
||||
9 => CombinedDivisorsEnding9,
|
||||
_ => ArrayView1D<ulong, Stride1D.Dense>.Empty,
|
||||
};
|
||||
|
||||
[MethodImpl(MethodImplOptions.AggressiveInlining)]
|
||||
public ArrayView1D<ulong, Stride1D.Dense> SelectDivisorSquares(byte ending) => ending switch
|
||||
{
|
||||
1 => CombinedDivisorSquaresEnding1,
|
||||
3 => CombinedDivisorSquaresEnding3,
|
||||
7 => CombinedDivisorSquaresEnding7,
|
||||
9 => CombinedDivisorSquaresEnding9,
|
||||
_ => ArrayView1D<ulong, Stride1D.Dense>.Empty,
|
||||
};
|
||||
}
|
||||
@@ -1,139 +0,0 @@
|
||||
using System.Runtime.CompilerServices;
|
||||
using ILGPU;
|
||||
using ILGPU.Runtime;
|
||||
|
||||
namespace PerfectNumbers.Core.Gpu;
|
||||
|
||||
public readonly struct HeuristicGpuDivisorTables
|
||||
{
|
||||
public readonly ArrayView1D<ulong, Stride1D.Dense> CombinedDivisorsEnding1;
|
||||
public readonly ArrayView1D<ulong, Stride1D.Dense> CombinedDivisorsEnding3;
|
||||
public readonly ArrayView1D<ulong, Stride1D.Dense> CombinedDivisorsEnding7;
|
||||
public readonly ArrayView1D<ulong, Stride1D.Dense> CombinedDivisorsEnding9;
|
||||
public readonly ArrayView1D<ulong, Stride1D.Dense> CombinedDivisorSquaresEnding1;
|
||||
public readonly ArrayView1D<ulong, Stride1D.Dense> CombinedDivisorSquaresEnding3;
|
||||
public readonly ArrayView1D<ulong, Stride1D.Dense> CombinedDivisorSquaresEnding7;
|
||||
public readonly ArrayView1D<ulong, Stride1D.Dense> CombinedDivisorSquaresEnding9;
|
||||
public readonly ArrayView1D<ulong, Stride1D.Dense> GroupADivisors;
|
||||
public readonly ArrayView1D<ulong, Stride1D.Dense> GroupADivisorSquares;
|
||||
public readonly ArrayView1D<ulong, Stride1D.Dense> GroupBDivisorsEnding1;
|
||||
public readonly ArrayView1D<ulong, Stride1D.Dense> GroupBDivisorSquaresEnding1;
|
||||
public readonly ArrayView1D<ulong, Stride1D.Dense> GroupBDivisorsEnding7;
|
||||
public readonly ArrayView1D<ulong, Stride1D.Dense> GroupBDivisorSquaresEnding7;
|
||||
public readonly ArrayView1D<ulong, Stride1D.Dense> GroupBDivisorsEnding9;
|
||||
public readonly ArrayView1D<ulong, Stride1D.Dense> GroupBDivisorSquaresEnding9;
|
||||
|
||||
public HeuristicGpuDivisorTables(
|
||||
ArrayView1D<ulong, Stride1D.Dense> combinedDivisorsEnding1,
|
||||
ArrayView1D<ulong, Stride1D.Dense> combinedDivisorsEnding3,
|
||||
ArrayView1D<ulong, Stride1D.Dense> combinedDivisorsEnding7,
|
||||
ArrayView1D<ulong, Stride1D.Dense> combinedDivisorsEnding9,
|
||||
ArrayView1D<ulong, Stride1D.Dense> combinedDivisorSquaresEnding1,
|
||||
ArrayView1D<ulong, Stride1D.Dense> combinedDivisorSquaresEnding3,
|
||||
ArrayView1D<ulong, Stride1D.Dense> combinedDivisorSquaresEnding7,
|
||||
ArrayView1D<ulong, Stride1D.Dense> combinedDivisorSquaresEnding9,
|
||||
ArrayView1D<ulong, Stride1D.Dense> groupADivisors,
|
||||
ArrayView1D<ulong, Stride1D.Dense> groupADivisorSquares,
|
||||
ArrayView1D<ulong, Stride1D.Dense> groupBDivisorsEnding1,
|
||||
ArrayView1D<ulong, Stride1D.Dense> groupBDivisorSquaresEnding1,
|
||||
ArrayView1D<ulong, Stride1D.Dense> groupBDivisorsEnding7,
|
||||
ArrayView1D<ulong, Stride1D.Dense> groupBDivisorSquaresEnding7,
|
||||
ArrayView1D<ulong, Stride1D.Dense> groupBDivisorsEnding9,
|
||||
ArrayView1D<ulong, Stride1D.Dense> groupBDivisorSquaresEnding9)
|
||||
{
|
||||
CombinedDivisorsEnding1 = combinedDivisorsEnding1;
|
||||
CombinedDivisorsEnding3 = combinedDivisorsEnding3;
|
||||
CombinedDivisorsEnding7 = combinedDivisorsEnding7;
|
||||
CombinedDivisorsEnding9 = combinedDivisorsEnding9;
|
||||
CombinedDivisorSquaresEnding1 = combinedDivisorSquaresEnding1;
|
||||
CombinedDivisorSquaresEnding3 = combinedDivisorSquaresEnding3;
|
||||
CombinedDivisorSquaresEnding7 = combinedDivisorSquaresEnding7;
|
||||
CombinedDivisorSquaresEnding9 = combinedDivisorSquaresEnding9;
|
||||
GroupADivisors = groupADivisors;
|
||||
GroupADivisorSquares = groupADivisorSquares;
|
||||
GroupBDivisorsEnding1 = groupBDivisorsEnding1;
|
||||
GroupBDivisorSquaresEnding1 = groupBDivisorSquaresEnding1;
|
||||
GroupBDivisorsEnding7 = groupBDivisorsEnding7;
|
||||
GroupBDivisorSquaresEnding7 = groupBDivisorSquaresEnding7;
|
||||
GroupBDivisorsEnding9 = groupBDivisorsEnding9;
|
||||
GroupBDivisorSquaresEnding9 = groupBDivisorSquaresEnding9;
|
||||
}
|
||||
|
||||
[MethodImpl(MethodImplOptions.AggressiveInlining)]
|
||||
public ArrayView1D<ulong, Stride1D.Dense> SelectCombined(byte ending)
|
||||
{
|
||||
return ending switch
|
||||
{
|
||||
1 => CombinedDivisorsEnding1,
|
||||
3 => CombinedDivisorsEnding3,
|
||||
7 => CombinedDivisorsEnding7,
|
||||
9 => CombinedDivisorsEnding9,
|
||||
_ => ArrayView1D<ulong, Stride1D.Dense>.Empty,
|
||||
};
|
||||
}
|
||||
|
||||
[MethodImpl(MethodImplOptions.AggressiveInlining)]
|
||||
public ArrayView1D<ulong, Stride1D.Dense> SelectCombinedSquares(byte ending)
|
||||
{
|
||||
return ending switch
|
||||
{
|
||||
1 => CombinedDivisorSquaresEnding1,
|
||||
3 => CombinedDivisorSquaresEnding3,
|
||||
7 => CombinedDivisorSquaresEnding7,
|
||||
9 => CombinedDivisorSquaresEnding9,
|
||||
_ => ArrayView1D<ulong, Stride1D.Dense>.Empty,
|
||||
};
|
||||
}
|
||||
|
||||
[MethodImpl(MethodImplOptions.AggressiveInlining)]
|
||||
public ArrayView1D<ulong, Stride1D.Dense> SelectGroupB(byte ending)
|
||||
{
|
||||
return ending switch
|
||||
{
|
||||
1 => GroupBDivisorsEnding1,
|
||||
7 => GroupBDivisorsEnding7,
|
||||
9 => GroupBDivisorsEnding9,
|
||||
_ => ArrayView1D<ulong, Stride1D.Dense>.Empty,
|
||||
};
|
||||
}
|
||||
|
||||
[MethodImpl(MethodImplOptions.AggressiveInlining)]
|
||||
public ArrayView1D<ulong, Stride1D.Dense> SelectGroupBSquares(byte ending)
|
||||
{
|
||||
return ending switch
|
||||
{
|
||||
1 => GroupBDivisorSquaresEnding1,
|
||||
7 => GroupBDivisorSquaresEnding7,
|
||||
9 => GroupBDivisorSquaresEnding9,
|
||||
_ => ArrayView1D<ulong, Stride1D.Dense>.Empty,
|
||||
};
|
||||
}
|
||||
|
||||
[MethodImpl(MethodImplOptions.AggressiveInlining)]
|
||||
public ArrayView1D<ulong, Stride1D.Dense> SelectDivisors(HeuristicGpuDivisorTableKind kind, byte ending)
|
||||
{
|
||||
return kind switch
|
||||
{
|
||||
HeuristicGpuDivisorTableKind.GroupA => GroupADivisors,
|
||||
HeuristicGpuDivisorTableKind.GroupBEnding1 => GroupBDivisorsEnding1,
|
||||
HeuristicGpuDivisorTableKind.GroupBEnding7 => GroupBDivisorsEnding7,
|
||||
HeuristicGpuDivisorTableKind.GroupBEnding9 => GroupBDivisorsEnding9,
|
||||
HeuristicGpuDivisorTableKind.Combined => SelectCombined(ending),
|
||||
_ => ArrayView1D<ulong, Stride1D.Dense>.Empty,
|
||||
};
|
||||
}
|
||||
|
||||
[MethodImpl(MethodImplOptions.AggressiveInlining)]
|
||||
public ArrayView1D<ulong, Stride1D.Dense> SelectDivisorSquares(HeuristicGpuDivisorTableKind kind, byte ending)
|
||||
{
|
||||
return kind switch
|
||||
{
|
||||
HeuristicGpuDivisorTableKind.GroupA => GroupADivisorSquares,
|
||||
HeuristicGpuDivisorTableKind.GroupBEnding1 => GroupBDivisorSquaresEnding1,
|
||||
HeuristicGpuDivisorTableKind.GroupBEnding7 => GroupBDivisorSquaresEnding7,
|
||||
HeuristicGpuDivisorTableKind.GroupBEnding9 => GroupBDivisorSquaresEnding9,
|
||||
HeuristicGpuDivisorTableKind.Combined => SelectCombinedSquares(ending),
|
||||
_ => ArrayView1D<ulong, Stride1D.Dense>.Empty,
|
||||
};
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,63 @@
|
||||
using System.Runtime.CompilerServices;
|
||||
using ILGPU;
|
||||
using ILGPU.Runtime;
|
||||
|
||||
namespace PerfectNumbers.Core.Gpu;
|
||||
|
||||
public readonly struct HeuristicGpuGroupABDivisorTables(
|
||||
ArrayView1D<ulong, Stride1D.Dense> groupADivisors,
|
||||
ArrayView1D<ulong, Stride1D.Dense> groupADivisorSquares,
|
||||
ArrayView1D<ulong, Stride1D.Dense> groupBDivisorsEnding1,
|
||||
ArrayView1D<ulong, Stride1D.Dense> groupBDivisorSquaresEnding1,
|
||||
ArrayView1D<ulong, Stride1D.Dense> groupBDivisorsEnding7,
|
||||
ArrayView1D<ulong, Stride1D.Dense> groupBDivisorSquaresEnding7,
|
||||
ArrayView1D<ulong, Stride1D.Dense> groupBDivisorsEnding9,
|
||||
ArrayView1D<ulong, Stride1D.Dense> groupBDivisorSquaresEnding9)
|
||||
{
|
||||
public readonly ArrayView1D<ulong, Stride1D.Dense> GroupADivisors = groupADivisors;
|
||||
public readonly ArrayView1D<ulong, Stride1D.Dense> GroupADivisorSquares = groupADivisorSquares;
|
||||
public readonly ArrayView1D<ulong, Stride1D.Dense> GroupBDivisorsEnding1 = groupBDivisorsEnding1;
|
||||
public readonly ArrayView1D<ulong, Stride1D.Dense> GroupBDivisorSquaresEnding1 = groupBDivisorSquaresEnding1;
|
||||
public readonly ArrayView1D<ulong, Stride1D.Dense> GroupBDivisorsEnding7 = groupBDivisorsEnding7;
|
||||
public readonly ArrayView1D<ulong, Stride1D.Dense> GroupBDivisorSquaresEnding7 = groupBDivisorSquaresEnding7;
|
||||
public readonly ArrayView1D<ulong, Stride1D.Dense> GroupBDivisorsEnding9 = groupBDivisorsEnding9;
|
||||
public readonly ArrayView1D<ulong, Stride1D.Dense> GroupBDivisorSquaresEnding9 = groupBDivisorSquaresEnding9;
|
||||
|
||||
[MethodImpl(MethodImplOptions.AggressiveInlining)]
|
||||
public ArrayView1D<ulong, Stride1D.Dense> SelectGroupB(byte ending) => ending switch
|
||||
{
|
||||
1 => GroupBDivisorsEnding1,
|
||||
7 => GroupBDivisorsEnding7,
|
||||
9 => GroupBDivisorsEnding9,
|
||||
_ => ArrayView1D<ulong, Stride1D.Dense>.Empty,
|
||||
};
|
||||
|
||||
[MethodImpl(MethodImplOptions.AggressiveInlining)]
|
||||
public ArrayView1D<ulong, Stride1D.Dense> SelectGroupBSquares(byte ending) => ending switch
|
||||
{
|
||||
1 => GroupBDivisorSquaresEnding1,
|
||||
7 => GroupBDivisorSquaresEnding7,
|
||||
9 => GroupBDivisorSquaresEnding9,
|
||||
_ => ArrayView1D<ulong, Stride1D.Dense>.Empty,
|
||||
};
|
||||
|
||||
[MethodImpl(MethodImplOptions.AggressiveInlining)]
|
||||
public ArrayView1D<ulong, Stride1D.Dense> SelectDivisors(HeuristicGpuDivisorTableKind kind) => kind switch
|
||||
{
|
||||
HeuristicGpuDivisorTableKind.GroupA => GroupADivisors,
|
||||
HeuristicGpuDivisorTableKind.GroupBEnding1 => GroupBDivisorsEnding1,
|
||||
HeuristicGpuDivisorTableKind.GroupBEnding7 => GroupBDivisorsEnding7,
|
||||
HeuristicGpuDivisorTableKind.GroupBEnding9 => GroupBDivisorsEnding9,
|
||||
_ => ArrayView1D<ulong, Stride1D.Dense>.Empty,
|
||||
};
|
||||
|
||||
[MethodImpl(MethodImplOptions.AggressiveInlining)]
|
||||
public ArrayView1D<ulong, Stride1D.Dense> SelectDivisorSquares(HeuristicGpuDivisorTableKind kind) => kind switch
|
||||
{
|
||||
HeuristicGpuDivisorTableKind.GroupA => GroupADivisorSquares,
|
||||
HeuristicGpuDivisorTableKind.GroupBEnding1 => GroupBDivisorSquaresEnding1,
|
||||
HeuristicGpuDivisorTableKind.GroupBEnding7 => GroupBDivisorSquaresEnding7,
|
||||
HeuristicGpuDivisorTableKind.GroupBEnding9 => GroupBDivisorSquaresEnding9,
|
||||
_ => ArrayView1D<ulong, Stride1D.Dense>.Empty,
|
||||
};
|
||||
}
|
||||
@@ -11,7 +11,6 @@ public enum HeuristicGpuDivisorTableKind : byte
|
||||
GroupBEnding1 = 1,
|
||||
GroupBEnding7 = 7,
|
||||
GroupBEnding9 = 9,
|
||||
Combined = 255,
|
||||
}
|
||||
|
||||
internal static class PrimeTesterKernels
|
||||
@@ -78,17 +77,49 @@ internal static class PrimeTesterKernels
|
||||
results[index] = result;
|
||||
}
|
||||
|
||||
public static void HeuristicTrialDivisionKernel(
|
||||
public static void HeuristicTrialCombinedDivisionKernel(
|
||||
Index1D index,
|
||||
byte nMod10,
|
||||
ArrayView<int> resultFlag,
|
||||
ulong n,
|
||||
ulong maxDivisorSquare,
|
||||
HeuristicGpuCombinedDivisorTables tables)
|
||||
{
|
||||
// byte nMod10 = (byte)(n % 10UL);
|
||||
ArrayView1D<ulong, Stride1D.Dense> divisors = tables.SelectDivisors(nMod10);
|
||||
ArrayView1D<ulong, Stride1D.Dense> divisorSquares = tables.SelectDivisorSquares(nMod10);
|
||||
|
||||
int divisorLength = (int)divisorSquares.Length;
|
||||
int threadIndex = index;
|
||||
if (threadIndex >= divisorLength)
|
||||
{
|
||||
return;
|
||||
}
|
||||
|
||||
ulong divisorSquare = divisorSquares[threadIndex];
|
||||
if (divisorSquare > maxDivisorSquare)
|
||||
{
|
||||
return;
|
||||
}
|
||||
|
||||
ulong divisor = divisors[threadIndex];
|
||||
divisor = n % divisor;
|
||||
if (divisor == 0UL)
|
||||
{
|
||||
resultFlag[0] = 1;
|
||||
}
|
||||
}
|
||||
|
||||
public static void HeuristicTrialGroupABDivisionKernel(
|
||||
Index1D index,
|
||||
ArrayView<int> resultFlag,
|
||||
ulong n,
|
||||
ulong maxDivisorSquare,
|
||||
HeuristicGpuDivisorTableKind tableKind,
|
||||
HeuristicGpuDivisorTables tables)
|
||||
HeuristicGpuGroupABDivisorTables tables)
|
||||
{
|
||||
byte nMod10 = (byte)(n % 10UL);
|
||||
ArrayView1D<ulong, Stride1D.Dense> divisors = tables.SelectDivisors(tableKind, nMod10);
|
||||
ArrayView1D<ulong, Stride1D.Dense> divisorSquares = tables.SelectDivisorSquares(tableKind, nMod10);
|
||||
ArrayView1D<ulong, Stride1D.Dense> divisors = tables.SelectDivisors(tableKind);
|
||||
ArrayView1D<ulong, Stride1D.Dense> divisorSquares = tables.SelectDivisorSquares(tableKind);
|
||||
|
||||
int divisorLength = (int)divisorSquares.Length;
|
||||
int threadIndex = index;
|
||||
|
||||
@@ -8,6 +8,7 @@ using System.Threading;
|
||||
using PerfectNumbers.Core;
|
||||
using ILGPU;
|
||||
using ILGPU.Runtime;
|
||||
using PerfectNumbers.Core.Gpu.Accelerators;
|
||||
|
||||
namespace PerfectNumbers.Core.Gpu;
|
||||
|
||||
@@ -106,9 +107,12 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
|
||||
ulong lastProcessed;
|
||||
|
||||
// GpuPrimeWorkLimiter();
|
||||
int acceleratorIndex = AcceleratorPool.Shared.Rent();
|
||||
var accelerator = _accelerators[acceleratorIndex];
|
||||
var stream = accelerator.CreateStream();
|
||||
// int acceleratorIndex = AcceleratorPool.Shared.Rent();
|
||||
// var accelerator = _accelerators[acceleratorIndex];
|
||||
var gpu = HeuristicCombinedPrimeTesterAccelerator.Rent(batchCapacity);
|
||||
var acceleratorIndex = gpu.AcceleratorIndex;
|
||||
var accelerator = gpu.Accelerator;
|
||||
var stream = AcceleratorStreamPool.Rent(acceleratorIndex);
|
||||
|
||||
// Monitor.Enter(gpuLease.ExecutionLock);
|
||||
|
||||
@@ -116,6 +120,7 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
|
||||
var resources = RentBatchResources(accelerator, batchCapacity);
|
||||
|
||||
composite = CheckDivisors(
|
||||
gpu,
|
||||
prime,
|
||||
allowedMax,
|
||||
kernel,
|
||||
@@ -137,8 +142,9 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
|
||||
out coveredRange,
|
||||
out processedCount);
|
||||
|
||||
stream.Dispose();
|
||||
AcceleratorStreamPool.Return(acceleratorIndex, stream);
|
||||
ReturnBatchResources(resources);
|
||||
HeuristicCombinedPrimeTesterAccelerator.Return(gpu);
|
||||
// Monitor.Exit(gpuLease.ExecutionLock);
|
||||
// GpuPrimeWorkLimiter.Release();
|
||||
|
||||
@@ -182,6 +188,7 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
|
||||
}
|
||||
|
||||
private static bool CheckDivisors(
|
||||
HeuristicCombinedPrimeTesterAccelerator gpu,
|
||||
ulong prime,
|
||||
ulong allowedMax,
|
||||
Action<AcceleratorStream, Index1D, ArrayView<GpuDivisorPartialData>, ArrayView<int>, ArrayView<int>, ArrayView<ulong>, ArrayView<ulong>, ArrayView<byte>, ArrayView<int>> kernel,
|
||||
@@ -205,7 +212,7 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
|
||||
{
|
||||
int batchCapacity = (int)divisorDataBuffer.Length;
|
||||
bool composite = false;
|
||||
bool processedAll = false;
|
||||
bool processedAll;
|
||||
processedCount = 0UL;
|
||||
lastProcessed = 0UL;
|
||||
|
||||
@@ -303,7 +310,7 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
|
||||
{
|
||||
ulong divisorValue = (ulong)nextDivisor128;
|
||||
MontgomeryDivisorData montgomeryData = MontgomeryDivisorData.FromModulus(divisorValue);
|
||||
ulong divisorCycle = ResolveDivisorCycle(divisorValue, prime, in montgomeryData);
|
||||
ulong divisorCycle = ResolveDivisorCycle(gpu, divisorValue, prime, in montgomeryData);
|
||||
if (divisorCycle == prime)
|
||||
{
|
||||
processedCount += (ulong)(i + 1);
|
||||
@@ -375,9 +382,9 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
|
||||
|
||||
|
||||
[MethodImpl(MethodImplOptions.AggressiveInlining)]
|
||||
private static ulong ResolveDivisorCycle(ulong divisor, ulong prime, in MontgomeryDivisorData divisorData)
|
||||
private static ulong ResolveDivisorCycle(HeuristicCombinedPrimeTesterAccelerator gpu, ulong divisor, ulong prime, in MontgomeryDivisorData divisorData)
|
||||
{
|
||||
if (!MersenneDivisorCycles.TryCalculateCycleLengthForExponentCpu(divisor, prime, divisorData, out ulong computedCycle, out bool primeOrderFailed) || computedCycle == 0UL)
|
||||
if (!MersenneDivisorCycles.TryCalculateCycleLengthForExponentCpu(gpu, divisor, prime, divisorData, out ulong computedCycle, out bool primeOrderFailed) || computedCycle == 0UL)
|
||||
{
|
||||
return MersenneDivisorCycles.CalculateCycleLength(divisor, divisorData, skipPrimeOrderHeuristic: primeOrderFailed);
|
||||
}
|
||||
@@ -498,6 +505,7 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
|
||||
private static readonly Accelerator[] _accelerators = AcceleratorPool.Shared.Accelerators;
|
||||
|
||||
private readonly MersenneNumberDivisorByDivisorGpuTester _owner;
|
||||
private readonly HeuristicCombinedPrimeTesterAccelerator _primeTesterAccelerator;
|
||||
private readonly Accelerator _accelerator;
|
||||
private readonly AcceleratorStream _stream;
|
||||
private Action<AcceleratorStream, Index1D, ArrayView<GpuDivisorPartialData>, ArrayView<int>, ArrayView<int>, ArrayView<ulong>, ArrayView<ulong>, ArrayView<byte>, ArrayView<int>> _kernel = null!;
|
||||
@@ -516,8 +524,8 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
|
||||
{
|
||||
_owner = owner;
|
||||
// GpuPrimeWorkLimiter.Acquire();
|
||||
int acceleratorIndex = AcceleratorPool.Shared.Rent();
|
||||
Accelerator accelerator = _accelerators[acceleratorIndex];
|
||||
_primeTesterAccelerator = HeuristicCombinedPrimeTesterAccelerator.Rent();
|
||||
Accelerator accelerator = _primeTesterAccelerator.Accelerator;
|
||||
_accelerator = accelerator;
|
||||
_stream = accelerator.CreateStream();
|
||||
_capacity = Math.Max(1, owner._gpuBatchSize);
|
||||
@@ -600,7 +608,8 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
|
||||
|
||||
if (cycle == 0UL)
|
||||
{
|
||||
if (!MersenneDivisorCycles.TryCalculateCycleLengthForExponentCpu(divisor, firstPrime, divisorData, out ulong computedCycle, out bool primeOrderFailed) || computedCycle == 0UL)
|
||||
var gpu = _primeTesterAccelerator;
|
||||
if (!MersenneDivisorCycles.TryCalculateCycleLengthForExponentCpu(gpu, divisor, firstPrime, divisorData, out ulong computedCycle, out bool primeOrderFailed) || computedCycle == 0UL)
|
||||
{
|
||||
cycle = MersenneDivisorCycles.CalculateCycleLength(divisor, divisorData, skipPrimeOrderHeuristic: primeOrderFailed);
|
||||
}
|
||||
@@ -657,10 +666,11 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
|
||||
ref byte hitRef = ref MemoryMarshal.GetReference(hitSlice);
|
||||
hitView.CopyToCPU(stream, ref hitRef, length);
|
||||
stream.Synchronize();
|
||||
|
||||
// Monitor.Exit(_lease.ExecutionLock);
|
||||
}
|
||||
|
||||
public void Dispose()
|
||||
public void Return()
|
||||
{
|
||||
_owner._sessionPool.Add(this);
|
||||
}
|
||||
|
||||
Reference in new issue
Block a user