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

Performance improvements

Browse Source
This commit is contained in:
mlemanczyk committed 2025-11-13 00:51:32 +01:00
1 parent 5bc65464d0
commit 43863d03fb
11 files changed
+127 -86
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

PerfectNumbers.Core/Gpu/Accelerators/AcceleratorStreamPool.cs
+22 -6
View File
@@ -1,24 +1,26 @@
using System.Collections.Concurrent;
using ILGPU.Runtime;
using Open.Collections;
namespace PerfectNumbers.Core.Gpu.Accelerators;
public static class AcceleratorStreamPool
{
private static readonly ConcurrentDictionary<Accelerator, ConcurrentQueue<AcceleratorStream>> _streams = new(20_480, PerfectNumberConstants.RollingAccelerators);
private static readonly Dictionary<Accelerator, ConcurrentQueue<AcceleratorStream>> _streams = new(PerfectNumberConstants .RollingAccelerators);
// private static readonly ConcurrentDictionary<Accelerator, AcceleratorStream> _streams = new(AcceleratorPool.Shared.Accelerators.ToDictionary(static x => x, CreateStream));
private static SemaphoreSlim CreateLock() => new(PerfectNumberConstants.ThreadsByStream);
private static AcceleratorStream CreateStream(Accelerator accelerator) => accelerator.CreateStream();
private static readonly ConcurrentDictionary<Accelerator, SemaphoreSlim> _locks = new(AcceleratorPool.Shared.Accelerators.ToDictionary(x => x, static x => CreateLock()));
private static readonly Dictionary<Accelerator, SemaphoreSlim> _locks = new(AcceleratorPool.Shared.Accelerators.ToDictionary(x => x, static x => CreateLock()));
public static AcceleratorStream Rent(Accelerator accelerator)
{
var streamLock = _locks.GetOrAdd(accelerator, static _ => new SemaphoreSlim(PerfectNumberConstants.ThreadsByStream));
var streamLock = _locks[accelerator];
// var streamLock = _locks.GetOrAdd(accelerator, static _ => CreateLock());
streamLock.Wait();
var queue = _streams.GetOrAdd(accelerator, static _ => new());
var queue = _streams[accelerator];
return queue.TryDequeue(out var stream)
? stream
@@ -28,9 +30,23 @@ public static class AcceleratorStreamPool
public static void Return(AcceleratorStream stream)
{
Accelerator accelerator = stream.Accelerator;
var streamLock = _locks.GetOrAdd(accelerator, static _ => CreateLock());
var queue = _streams.GetOrAdd(accelerator, static _ => new());
var streamLock = _locks[accelerator];
var queue = _streams[accelerator];
queue.Enqueue(stream);
streamLock.Release();
}
public static void WarmUp(Accelerator accelerator)
{
_streams.Add(accelerator, new());
_ = _locks.TryAdd(accelerator, CreateLock());
// ConcurrentQueue<AcceleratorStream> streamQueue = new();
// var stream = accelerator.CreateStream();
// // var stream = accelerator.CreateStream();
// streamQueue.Enqueue(stream);
// _streams[accelerator] = streamQueue;
// _locks[accelerator] = CreateLock();
// stream.Synchronize();
}
}
PerfectNumbers.Core/Gpu/Accelerators/Pow2MontgomeryAccelerator.cs
+4
View File
@@ -72,6 +72,8 @@ public sealed class Pow2MontgomeryAccelerator
{
Console.WriteLine($"Preparing accelerator {i}...");
var accelerator = accelerators[i];
AcceleratorStreamPool.WarmUp(accelerator);
// AcceleratorStream stream = AcceleratorStreamPool.Rent(accelerator);
AcceleratorStream stream = accelerator.CreateStream();
LastDigitGpuTables.EnsureStaticTables(accelerator, stream);
// SharedHeuristicGpuTables.EnsureStaticTables(accelerator, stream);
@@ -80,6 +82,8 @@ public sealed class Pow2MontgomeryAccelerator
// GpuStaticTableInitializer.EnsureStaticTables(accelerator, kernels, stream);
stream.Synchronize();
stream.Dispose();
// AcceleratorStreamPool.Return(stream);
}
}