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

Stream handling simplified to avoid dictionary lookups in favor of array lookups with known indexes.

Browse Source
This commit is contained in:
mlemanczyk committed 2025-11-16 15:01:51 +01:00
1 parent cd7636ddbe
commit a84445ef2a
15 files changed
+67 -73
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

PerfectNumbers.Core/Gpu/Accelerators/AcceleratorStreamPool.cs
+10 -10
View File
@@ -7,8 +7,7 @@ public static class AcceleratorStreamPool
{
private static readonly Accelerator[] _accelerators = AcceleratorPool.Shared.Accelerators;
private static readonly SemaphoreSlim[] _locks = new SemaphoreSlim[PerfectNumberConstants.RollingAccelerators];
private static readonly ConcurrentQueue<AcceleratorStream>[] _streams = new ConcurrentQueue<AcceleratorStream> [PerfectNumberConstants.RollingAccelerators];
private static readonly AcceleratorStream[] _streams = new AcceleratorStream[PerfectNumberConstants.RollingAccelerators];
private static SemaphoreSlim CreateLock() => new(PerfectNumberConstants.ThreadsByAccelerator);
@@ -17,26 +16,27 @@ public static class AcceleratorStreamPool
public static AcceleratorStream Rent(int acceleratorIndex)
{
var streamLock = _locks[acceleratorIndex];
var queue = _streams[acceleratorIndex];
streamLock.Wait();
var queue = _streams[acceleratorIndex];
if (queue != null)
{
return queue;
}
return queue.TryDequeue(out var stream)
? stream
: _accelerators[acceleratorIndex].CreateStream();
queue = _accelerators[acceleratorIndex].CreateStream();
_streams[acceleratorIndex] = queue;
return queue;
}
public static void Return(int acceleratorIndex, AcceleratorStream stream)
public static void Return(int acceleratorIndex)
{
var streamLock = _locks[acceleratorIndex];
var queue = _streams[acceleratorIndex];
queue.Enqueue(stream);
streamLock.Release();
}
public static void WarmUp(int acceleratorIndex)
{
_streams[acceleratorIndex] = new();
_locks[acceleratorIndex] = CreateLock();
}
}
PerfectNumbers.Core/Gpu/Accelerators/HeuristicGpuTables.cs
+16 -16
View File
@@ -45,25 +45,25 @@ internal sealed class HeuristicGpuTables
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 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 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 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);
// 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;
PerfectNumbers.Core/Gpu/Accelerators/HeuristicPrimeTesterAccelerator.cs
+2 -2
View File
@@ -136,7 +136,7 @@ internal sealed class HeuristicPrimeTesterAccelerator
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, ArrayView1D<ulong, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>>(PrimeTesterKernels.HeuristicTrialDivisionKernel));
AcceleratorStream stream = AcceleratorStreamPool.Rent(acceleratorIndex);
@@ -146,7 +146,7 @@ internal sealed class HeuristicPrimeTesterAccelerator
_heuristicDivisorTables = _sharedTables.CreateHeuristicDivisorTables();
stream.Synchronize();
AcceleratorStreamPool.Return(acceleratorIndex, stream);
AcceleratorStreamPool.Return(acceleratorIndex);
}
public void EnsureCapacity(int minCapacity)