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

Serialize GPU leases with per-accelerator locks

Browse Source
This commit is contained in:
Marek Lemańczyk committed 2025-09-20 18:19:21 +02:00
1 parent 2683dfcff1
commit abe695788b
6 files changed
+284 -182
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

PerfectNumbers.Core/Gpu/GpuKernelPool.cs
+152 -75
View File
@@ -1,5 +1,6 @@
using System.Collections.Concurrent;
using System.Runtime.CompilerServices;
using System.Threading;
using ILGPU;
using ILGPU.Runtime;
using static PerfectNumbers.Core.Gpu.GpuContextPool;
@@ -26,16 +27,16 @@ public readonly struct ResidueAutomatonArgs
public sealed class KernelContainer
{
// Serializes first-time initialization of kernels/buffers per accelerator.
public Action<Index1D, ulong, ulong, ArrayView<GpuUInt128>, ArrayView<ulong>>? Order;
public Action<Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong,
public Action<AcceleratorStream, Index1D, ulong, ulong, ArrayView<GpuUInt128>, ArrayView<ulong>>? Order;
public Action<AcceleratorStream, Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong,
ulong, ulong, ulong, ulong, ArrayView<ulong>, ArrayView1D<uint, Stride1D.Dense>>? Incremental;
public Action<Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong,
public Action<AcceleratorStream, Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong,
ResidueAutomatonArgs, ArrayView<ulong>, ArrayView1D<uint, Stride1D.Dense>,
ArrayView1D<uint, Stride1D.Dense>, ArrayView1D<uint, Stride1D.Dense>,
ArrayView1D<ulong, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>>? Pow2Mod;
public Action<Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong,
ResidueAutomatonArgs, ArrayView<int>, ArrayView1D<uint, Stride1D.Dense>>? IncrementalOrder;
public Action<Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong,
public Action<AcceleratorStream, Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong,
ResidueAutomatonArgs, ArrayView<int>, ArrayView1D<uint, Stride1D.Dense>>? IncrementalOrder;
public Action<AcceleratorStream, Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong,
ResidueAutomatonArgs, ArrayView<int>, ArrayView1D<uint, Stride1D.Dense>>? Pow2ModOrder;
// Optional device buffer with small divisor cycles (<= 4M). Index = divisor, value = cycle length.
@@ -77,72 +78,119 @@ public readonly struct ResiduePrimeViews(
public ref struct GpuKernelLease(IDisposable limiter, GpuContextLease gpu, KernelContainer kernels)
{
private bool disposedValue;
private IDisposable? _limiter = limiter;
private GpuContextLease? _gpu = gpu;
private KernelContainer _kernels = kernels;
private bool disposedValue;
private IDisposable? _limiter = limiter;
private GpuContextLease? _gpu = gpu;
private KernelContainer _kernels = kernels;
private AcceleratorStream? _stream;
private object? _executionLock = gpu.ExecutionLock;
public readonly Accelerator Accelerator => _gpu?.Accelerator ?? throw new NullReferenceException("GPU context is already released");
public readonly Accelerator Accelerator => _gpu?.Accelerator ?? throw new NullReferenceException("GPU context is already released");
public readonly Action<Index1D, ulong, ulong, ArrayView<GpuUInt128>, ArrayView<ulong>> OrderKernel
{
get
{
var accel = Accelerator; // avoid capturing 'this' in lambda
lock (_gpu!.Value.KernelInitLock)
{
return KernelContainer.InitOnce(ref _kernels.Order, () => accel.LoadAutoGroupedStreamKernel<Index1D, ulong, ulong, ArrayView<GpuUInt128>, ArrayView<ulong>>(OrderKernelScan));
}
}
}
public AcceleratorStream Stream
{
get
{
if (_stream is { } stream)
{
return stream;
}
public readonly Action<Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ResidueAutomatonArgs, ArrayView<ulong>, ArrayView1D<uint, Stride1D.Dense>, ArrayView1D<uint, Stride1D.Dense>, ArrayView1D<uint, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>> Pow2ModKernel
stream = Accelerator.CreateStream();
_stream = stream;
return stream;
}
}
public readonly ExecutionScope EnterExecutionScope()
{
return new ExecutionScope(_executionLock);
}
public readonly Action<AcceleratorStream, Index1D, ulong, ulong, ArrayView<GpuUInt128>, ArrayView<ulong>> OrderKernel
{
get
{
var accel = Accelerator; // avoid capturing 'this' in lambda
lock (_gpu!.Value.KernelInitLock)
{
return KernelContainer.InitOnce(ref _kernels.Order, () =>
{
var loaded = accel.LoadAutoGroupedStreamKernel<Index1D, ulong, ulong, ArrayView<GpuUInt128>, ArrayView<ulong>>(OrderKernelScan);
var kernel = KernelUtil.GetKernel(loaded);
return kernel.CreateLauncherDelegate<Action<AcceleratorStream, Index1D, ulong, ulong, ArrayView<GpuUInt128>, ArrayView<ulong>>>();
});
}
}
}
public readonly Action<AcceleratorStream, Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ResidueAutomatonArgs, ArrayView<ulong>, ArrayView1D<uint, Stride1D.Dense>, ArrayView1D<uint, Stride1D.Dense>, ArrayView1D<uint, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>> Pow2ModKernel
{
get
{
var accel = Accelerator;
lock (_gpu!.Value.KernelInitLock)
{
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<uint, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>>(Pow2ModKernelScan));
return KernelContainer.InitOnce(ref _kernels.Pow2Mod, () =>
{
var loaded = accel.LoadAutoGroupedStreamKernel<Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ResidueAutomatonArgs, ArrayView<ulong>, ArrayView1D<uint, Stride1D.Dense>, ArrayView1D<uint, Stride1D.Dense>, ArrayView1D<uint, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>>(Pow2ModKernelScan);
var kernel = KernelUtil.GetKernel(loaded);
return kernel.CreateLauncherDelegate<Action<AcceleratorStream, Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ResidueAutomatonArgs, ArrayView<ulong>, ArrayView1D<uint, Stride1D.Dense>, ArrayView1D<uint, Stride1D.Dense>, ArrayView1D<uint, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>>>();
});
}
}
}
public readonly Action<Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ulong, ulong, ulong, ulong, ArrayView<ulong>, ArrayView1D<uint, Stride1D.Dense>> IncrementalKernel
{
get
{
var accel = Accelerator;
lock (_gpu!.Value.KernelInitLock)
{
return KernelContainer.InitOnce(ref _kernels.Incremental, () => accel.LoadAutoGroupedStreamKernel<Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ulong, ulong, ulong, ulong, ArrayView<ulong>, ArrayView1D<uint, Stride1D.Dense>>(IncrementalKernelScan));
}
}
}
public readonly Action<AcceleratorStream, Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ulong, ulong, ulong, ulong, ArrayView<ulong>, ArrayView1D<uint, Stride1D.Dense>> IncrementalKernel
{
get
{
var accel = Accelerator;
lock (_gpu!.Value.KernelInitLock)
{
return KernelContainer.InitOnce(ref _kernels.Incremental, () =>
{
var loaded = accel.LoadAutoGroupedStreamKernel<Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ulong, ulong, ulong, ulong, ArrayView<ulong>, ArrayView1D<uint, Stride1D.Dense>>(IncrementalKernelScan);
var kernel = KernelUtil.GetKernel(loaded);
return kernel.CreateLauncherDelegate<Action<AcceleratorStream, Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ulong, ulong, ulong, ulong, ArrayView<ulong>, ArrayView1D<uint, Stride1D.Dense>>>();
});
}
}
}
public readonly Action<Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ResidueAutomatonArgs, ArrayView<int>, ArrayView1D<uint, Stride1D.Dense>> IncrementalOrderKernel
{
get
{
var accel = Accelerator;
lock (_gpu!.Value.KernelInitLock)
{
return KernelContainer.InitOnce(ref _kernels.IncrementalOrder, () => accel.LoadAutoGroupedStreamKernel<Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ResidueAutomatonArgs, ArrayView<int>, ArrayView1D<uint, Stride1D.Dense>>(IncrementalOrderKernelScan));
}
}
}
public readonly Action<AcceleratorStream, Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ResidueAutomatonArgs, ArrayView<int>, ArrayView1D<uint, Stride1D.Dense>> IncrementalOrderKernel
{
get
{
var accel = Accelerator;
lock (_gpu!.Value.KernelInitLock)
{
return KernelContainer.InitOnce(ref _kernels.IncrementalOrder, () =>
{
var loaded = accel.LoadAutoGroupedStreamKernel<Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ResidueAutomatonArgs, ArrayView<int>, ArrayView1D<uint, Stride1D.Dense>>(IncrementalOrderKernelScan);
var kernel = KernelUtil.GetKernel(loaded);
return kernel.CreateLauncherDelegate<Action<AcceleratorStream, Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ResidueAutomatonArgs, ArrayView<int>, ArrayView1D<uint, Stride1D.Dense>>>();
});
}
}
}
public readonly Action<Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ResidueAutomatonArgs, ArrayView<int>, ArrayView1D<uint, Stride1D.Dense>> Pow2ModOrderKernel
{
get
{
var accel = Accelerator;
lock (_gpu!.Value.KernelInitLock)
{
return KernelContainer.InitOnce(ref _kernels.Pow2ModOrder, () => accel.LoadAutoGroupedStreamKernel<Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ResidueAutomatonArgs, ArrayView<int>, ArrayView1D<uint, Stride1D.Dense>>(Pow2ModOrderKernelScan));
}
}
}
public readonly Action<AcceleratorStream, Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ResidueAutomatonArgs, ArrayView<int>, ArrayView1D<uint, Stride1D.Dense>> Pow2ModOrderKernel
{
get
{
var accel = Accelerator;
lock (_gpu!.Value.KernelInitLock)
{
return KernelContainer.InitOnce(ref _kernels.Pow2ModOrder, () =>
{
var loaded = accel.LoadAutoGroupedStreamKernel<Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ResidueAutomatonArgs, ArrayView<int>, ArrayView1D<uint, Stride1D.Dense>>(Pow2ModOrderKernelScan);
var kernel = KernelUtil.GetKernel(loaded);
return kernel.CreateLauncherDelegate<Action<AcceleratorStream, Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ResidueAutomatonArgs, ArrayView<int>, ArrayView1D<uint, Stride1D.Dense>>>();
});
}
}
}
// TODO: Plumb the small-cycles device buffer into all kernels that can benefit
// (some already accept it). Consider a compact type (byte/ushort) for memory footprint.
@@ -273,18 +321,23 @@ public ref struct GpuKernelLease(IDisposable limiter, GpuContextLease gpu, Kerne
orders[index] = exponent;
}
private void Dispose(bool disposing)
{
if (!disposedValue)
{
if (disposing)
{
_gpu?.Dispose();
_gpu = null;
private void Dispose(bool disposing)
{
if (!disposedValue)
{
if (disposing)
{
_stream?.Dispose();
_stream = null;
_limiter?.Dispose();
_limiter = null;
}
_gpu?.Dispose();
_gpu = null;
_executionLock = null;
_limiter?.Dispose();
_limiter = null;
}
// free unmanaged resources (unmanaged objects) and override finalizer
// set large fields to null
@@ -563,13 +616,35 @@ public ref struct GpuKernelLease(IDisposable limiter, GpuContextLease gpu, Kerne
Atomic.Or(ref found[0], 1);
}
public void Dispose()
{
// Do not change this code. Put cleanup code in 'Dispose(bool disposing)' method
Dispose(disposing: true);
// Enabled this when the struct is converted to a class
// GC.SuppressFinalize(this);
}
public void Dispose()
{
// Do not change this code. Put cleanup code in 'Dispose(bool disposing)' method
Dispose(disposing: true);
// Enabled this when the struct is converted to a class
// GC.SuppressFinalize(this);
}
public readonly struct ExecutionScope : IDisposable
{
private readonly object? _lock;
public ExecutionScope(object? sync)
{
_lock = sync;
if (_lock is not null)
{
Monitor.Enter(_lock);
}
}
public void Dispose()
{
if (_lock is not null)
{
Monitor.Exit(_lock);
}
}
}
}
public class GpuKernelPool
@@ -667,10 +742,12 @@ public class GpuKernelPool
public static void Run(Action<Accelerator, AcceleratorStream> action)
{
var lease = GetKernel(useGpuOrder: true);
var execution = lease.EnterExecutionScope();
var accelerator = lease.Accelerator;
var stream = accelerator.CreateStream();
action(accelerator, stream);
stream.Dispose();
execution.Dispose();
lease.Dispose();
}