131 lines
6.7 KiB
C#
131 lines
6.7 KiB
C#
using System.Collections.Concurrent;
|
|
using ILGPU;
|
|
using ILGPU.Runtime;
|
|
|
|
namespace PerfectNumbers.Core.Gpu;
|
|
|
|
public sealed class GpuKernelLease
|
|
{
|
|
private static readonly ConcurrentQueue<GpuKernelLease> Pool = new();
|
|
|
|
public KernelContainer Kernels;
|
|
public AcceleratorStream Stream;
|
|
|
|
#pragma warning disable CS8618 // _gpu and Stream will be set by Rent method
|
|
private GpuKernelLease(KernelContainer kernels)
|
|
#pragma warning restore CS8618
|
|
{
|
|
Kernels = kernels;
|
|
}
|
|
|
|
internal static GpuKernelLease Rent(AcceleratorStream stream, KernelContainer kernels)
|
|
{
|
|
if (!Pool.TryDequeue(out var lease))
|
|
{
|
|
lease = new GpuKernelLease(kernels);
|
|
}
|
|
|
|
lease.Kernels = kernels;
|
|
lease.Stream = stream;
|
|
return lease;
|
|
}
|
|
|
|
public Action<AcceleratorStream, Index1D, ulong, ulong, ArrayView<GpuUInt128>, ArrayView<ulong>> OrderKernel
|
|
{
|
|
get
|
|
{
|
|
return GpuKernelPool.InitOnce(ref Kernels!.Order, () =>
|
|
{
|
|
var loaded = SharedGpuContext.Accelerator.LoadAutoGroupedStreamKernel<Index1D, ulong, ulong, ArrayView<GpuUInt128>, ArrayView<ulong>>(OrderKernels.OrderKernelScan);
|
|
var kernel = KernelUtil.GetKernel(loaded);
|
|
return kernel.CreateLauncherDelegate<Action<AcceleratorStream, Index1D, ulong, ulong, ArrayView<GpuUInt128>, ArrayView<ulong>>>();
|
|
});
|
|
}
|
|
}
|
|
|
|
public Action<AcceleratorStream, Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ResidueAutomatonArgs, ArrayView<ulong>, ArrayView1D<ulong, Stride1D.Dense>, ArrayView1D<uint, Stride1D.Dense>, ArrayView1D<uint, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>> Pow2ModKernel
|
|
{
|
|
get
|
|
{
|
|
return GpuKernelPool.InitOnce(ref Kernels!.Pow2Mod, () =>
|
|
{
|
|
var loaded = SharedGpuContext.Accelerator.LoadAutoGroupedStreamKernel<Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ResidueAutomatonArgs, ArrayView<ulong>, ArrayView1D<ulong, Stride1D.Dense>, ArrayView1D<uint, Stride1D.Dense>, ArrayView1D<uint, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>>(Pow2ModKernels.Pow2ModKernelScan);
|
|
var kernel = KernelUtil.GetKernel(loaded);
|
|
return kernel.CreateLauncherDelegate<Action<AcceleratorStream, Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ResidueAutomatonArgs, ArrayView<ulong>, ArrayView1D<ulong, Stride1D.Dense>, ArrayView1D<uint, Stride1D.Dense>, ArrayView1D<uint, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>>>();
|
|
});
|
|
}
|
|
}
|
|
|
|
public Action<AcceleratorStream, Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ulong, ulong, ulong, ulong, ArrayView<ulong>, ArrayView1D<ulong, Stride1D.Dense>> IncrementalKernel
|
|
{
|
|
get
|
|
{
|
|
return GpuKernelPool.InitOnce(ref Kernels!.Incremental, () =>
|
|
{
|
|
var loaded = SharedGpuContext.Accelerator.LoadAutoGroupedStreamKernel<Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ulong, ulong, ulong, ulong, ArrayView<ulong>, ArrayView1D<ulong, Stride1D.Dense>>(IncrementalKernels.IncrementalKernelScan);
|
|
var kernel = KernelUtil.GetKernel(loaded);
|
|
return kernel.CreateLauncherDelegate<Action<AcceleratorStream, Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ulong, ulong, ulong, ulong, ArrayView<ulong>, ArrayView1D<ulong, Stride1D.Dense>>>();
|
|
});
|
|
}
|
|
}
|
|
|
|
public Action<AcceleratorStream, Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ResidueAutomatonArgs, ArrayView<int>, ArrayView1D<ulong, Stride1D.Dense>> IncrementalOrderKernel
|
|
{
|
|
get
|
|
{
|
|
return GpuKernelPool.InitOnce(ref Kernels!.IncrementalOrder, () =>
|
|
{
|
|
var loaded = SharedGpuContext.Accelerator.LoadAutoGroupedStreamKernel<Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ResidueAutomatonArgs, ArrayView<int>, ArrayView1D<ulong, Stride1D.Dense>>(IncrementalKernels.IncrementalOrderKernelScan);
|
|
var kernel = KernelUtil.GetKernel(loaded);
|
|
return kernel.CreateLauncherDelegate<Action<AcceleratorStream, Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ResidueAutomatonArgs, ArrayView<int>, ArrayView1D<ulong, Stride1D.Dense>>>();
|
|
});
|
|
}
|
|
}
|
|
|
|
public Action<AcceleratorStream, Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ResidueAutomatonArgs, ArrayView<int>, ArrayView1D<ulong, Stride1D.Dense>> Pow2ModOrderKernel
|
|
{
|
|
get
|
|
{
|
|
return GpuKernelPool.InitOnce(ref Kernels!.Pow2ModOrder, () =>
|
|
{
|
|
var loaded = SharedGpuContext.Accelerator.LoadAutoGroupedStreamKernel<Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ResidueAutomatonArgs, ArrayView<int>, ArrayView1D<ulong, Stride1D.Dense>>(Pow2ModKernels.Pow2ModOrderKernelScan);
|
|
var kernel = KernelUtil.GetKernel(loaded);
|
|
return kernel.CreateLauncherDelegate<Action<AcceleratorStream, Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ResidueAutomatonArgs, ArrayView<int>, ArrayView1D<ulong, Stride1D.Dense>>>();
|
|
});
|
|
}
|
|
}
|
|
|
|
public Action<AcceleratorStream, Index1D, ulong, ArrayView1D<ulong, Stride1D.Dense>, int, MontgomeryDivisorData, ArrayView1D<ulong, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>> SpecialMaxKernel
|
|
{
|
|
get
|
|
{
|
|
return GpuKernelPool.InitOnce(ref Kernels!.SpecialMax, () =>
|
|
{
|
|
var loaded = SharedGpuContext.Accelerator.LoadAutoGroupedStreamKernel<Index1D, ulong, ArrayView1D<ulong, Stride1D.Dense>, int, MontgomeryDivisorData, ArrayView1D<ulong, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>>(PrimeOrderGpuHeuristics.EvaluateSpecialMaxCandidatesKernel);
|
|
var kernel = KernelUtil.GetKernel(loaded);
|
|
return kernel.CreateLauncherDelegate<Action<AcceleratorStream, Index1D, ulong, ArrayView1D<ulong, Stride1D.Dense>, int, MontgomeryDivisorData, ArrayView1D<ulong, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>>>();
|
|
});
|
|
}
|
|
}
|
|
|
|
public Action<AcceleratorStream, Index1D, ulong, uint, ArrayView1D<uint, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>, int, ArrayView1D<ulong, Stride1D.Dense>, ArrayView1D<int, Stride1D.Dense>, ArrayView1D<int, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>> SmallPrimeFactorKernel
|
|
{
|
|
get
|
|
{
|
|
return GpuKernelPool.InitOnce(ref Kernels!.SmallPrimeFactor, () =>
|
|
{
|
|
var loaded = SharedGpuContext.Accelerator.LoadAutoGroupedStreamKernel<Index1D, ulong, uint, ArrayView1D<uint, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>, int, ArrayView1D<ulong, Stride1D.Dense>, ArrayView1D<int, Stride1D.Dense>, ArrayView1D<int, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>>(SmallPrimeFactorKernels.SmallPrimeFactorKernelScan);
|
|
var kernel = KernelUtil.GetKernel(loaded);
|
|
return kernel.CreateLauncherDelegate<Action<AcceleratorStream, Index1D, ulong, uint, ArrayView1D<uint, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>, int, ArrayView1D<ulong, Stride1D.Dense>, ArrayView1D<int, Stride1D.Dense>, ArrayView1D<int, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>>>();
|
|
});
|
|
}
|
|
}
|
|
|
|
public void Dispose()
|
|
{
|
|
Stream.Dispose();
|
|
Pool.Enqueue(this);
|
|
GpuPrimeWorkLimiter.Release();
|
|
}
|
|
}
|