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

Merge pull request #241 from mlemanczyk/codex/remove-executionscope-structure

Browse Source 
Remove GPU context creation lock
This commit is contained in:
Marek Lemańczyk authored and GitHub committed 2025-10-28 13:39:43 +01:00
2 parents 139899fa5e 86990ae0f7
commit a89ed7576e
10 files changed
+41 -119
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

PerfectNumbers.Core/DivisorCycleCache.cs
-2
View File
@@ -227,7 +227,6 @@ public sealed class DivisorCycleCache
{ {
int length = divisors.Length; int length = divisors.Length;
var gpuLease = GpuKernelPool.GetKernel(useGpuOrder: true); var gpuLease = GpuKernelPool.GetKernel(useGpuOrder: true);
var execution = gpuLease.EnterExecutionScope();
Accelerator accelerator = gpuLease.Accelerator; Accelerator accelerator = gpuLease.Accelerator;
var kernel = _gpuKernelCache.GetOrAdd(accelerator, LoadKernel); var kernel = _gpuKernelCache.GetOrAdd(accelerator, LoadKernel);
@@ -309,7 +308,6 @@ public sealed class DivisorCycleCache
ArrayPool<byte>.Shared.Return(rentedStatus!, clearArray: false); ArrayPool<byte>.Shared.Return(rentedStatus!, clearArray: false);
} }
execution.Dispose();
gpuLease.Dispose(); gpuLease.Dispose();
} }
PerfectNumbers.Core/Gpu/GpuContextPool.cs
+6 -16
View File
@@ -47,7 +47,6 @@ public static class GpuContextPool
private static readonly ConcurrentQueue<PooledContext> CpuPool = new(); private static readonly ConcurrentQueue<PooledContext> CpuPool = new();
private static readonly ConcurrentQueue<PooledContext> GpuPool = new(); private static readonly ConcurrentQueue<PooledContext> GpuPool = new();
private static readonly object CreationLock = new();
public static GpuContextLease Rent() public static GpuContextLease Rent()
{ {
@@ -61,20 +60,17 @@ public static class GpuContextPool
{ {
if (preferCpu && CpuPool.TryDequeue(out var cpu)) if (preferCpu && CpuPool.TryDequeue(out var cpu))
{ {
return new GpuContextLease(cpu, CreationLock); return new GpuContextLease(cpu);
} }
if (!preferCpu && GpuPool.TryDequeue(out var gpu)) if (!preferCpu && GpuPool.TryDequeue(out var gpu))
{ {
return new GpuContextLease(gpu, CreationLock); return new GpuContextLease(gpu);
} }
} }
// Serialize accelerator creation to avoid concurrent CL device/stream setup across threads. // Create a new accelerator when the pool does not have one available.
lock (CreationLock) return new GpuContextLease(new PooledContext(preferCpu));
{
return new GpuContextLease(new PooledContext(preferCpu), CreationLock);
}
} }
public static void DisposeAll() public static void DisposeAll()
@@ -114,10 +110,7 @@ public static class GpuContextPool
else else
{ {
lock (CreationLock) ctx.Dispose();
{
ctx.Dispose();
}
} }
} }
@@ -125,18 +118,15 @@ public static class GpuContextPool
{ {
private readonly PooledContext _ctx; private readonly PooledContext _ctx;
internal GpuContextLease(PooledContext ctx, object kernelInitLock) internal GpuContextLease(PooledContext ctx)
{ {
_ctx = ctx; _ctx = ctx;
KernelInitLock = kernelInitLock;
} }
public Context Context => _ctx.Context; public Context Context => _ctx.Context;
public Accelerator Accelerator => _ctx.Accelerator; public Accelerator Accelerator => _ctx.Accelerator;
public object KernelInitLock { get; }
public object ExecutionLock => _ctx.ExecutionLock; public object ExecutionLock => _ctx.ExecutionLock;
PerfectNumbers.Core/Gpu/GpuKernelLease.cs
+30 -75
View File
@@ -14,7 +14,6 @@ public sealed class GpuKernelLease
private GpuContextLease? _gpu; private GpuContextLease? _gpu;
private KernelContainer? _kernels; private KernelContainer? _kernels;
private AcceleratorStream? _stream; private AcceleratorStream? _stream;
private object? _executionLock;
private GpuKernelLease() private GpuKernelLease()
{ {
@@ -31,7 +30,6 @@ public sealed class GpuKernelLease
lease._gpu = gpu; lease._gpu = gpu;
lease._kernels = kernels; lease._kernels = kernels;
lease._stream = null; lease._stream = null;
lease._executionLock = gpu.ExecutionLock;
return lease; return lease;
} }
@@ -53,22 +51,17 @@ public sealed class GpuKernelLease
} }
} }
public ExecutionScope EnterExecutionScope() => new ExecutionScope(_executionLock);
public Action<AcceleratorStream, Index1D, ulong, ulong, ArrayView<GpuUInt128>, ArrayView<ulong>> OrderKernel public Action<AcceleratorStream, Index1D, ulong, ulong, ArrayView<GpuUInt128>, ArrayView<ulong>> OrderKernel
{ {
get get
{ {
var accel = Accelerator; // avoid capturing 'this' in lambda var accel = Accelerator; // avoid capturing 'this' in lambda
lock (_gpu!.Value.KernelInitLock) return KernelContainer.InitOnce(ref _kernels!.Order, () =>
{ {
return KernelContainer.InitOnce(ref _kernels!.Order, () => var loaded = accel.LoadAutoGroupedStreamKernel<Index1D, ulong, ulong, ArrayView<GpuUInt128>, ArrayView<ulong>>(OrderKernels.OrderKernelScan);
{ var kernel = KernelUtil.GetKernel(loaded);
var loaded = accel.LoadAutoGroupedStreamKernel<Index1D, ulong, ulong, ArrayView<GpuUInt128>, ArrayView<ulong>>(OrderKernels.OrderKernelScan); return kernel.CreateLauncherDelegate<Action<AcceleratorStream, Index1D, ulong, ulong, ArrayView<GpuUInt128>, ArrayView<ulong>>>();
var kernel = KernelUtil.GetKernel(loaded); });
return kernel.CreateLauncherDelegate<Action<AcceleratorStream, Index1D, ulong, ulong, ArrayView<GpuUInt128>, ArrayView<ulong>>>();
});
}
} }
} }
@@ -77,15 +70,12 @@ public sealed class GpuKernelLease
get get
{ {
var accel = Accelerator; var accel = Accelerator;
lock (_gpu!.Value.KernelInitLock) return KernelContainer.InitOnce(ref _kernels!.Pow2Mod, () =>
{ {
return KernelContainer.InitOnce(ref _kernels!.Pow2Mod, () => var loaded = accel.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);
var loaded = accel.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); 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>>>();
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>>>();
});
}
} }
} }
@@ -94,15 +84,12 @@ public sealed class GpuKernelLease
get get
{ {
var accel = Accelerator; var accel = Accelerator;
lock (_gpu!.Value.KernelInitLock) return KernelContainer.InitOnce(ref _kernels!.Incremental, () =>
{ {
return KernelContainer.InitOnce(ref _kernels!.Incremental, () => var loaded = accel.LoadAutoGroupedStreamKernel<Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ulong, ulong, ulong, ulong, ArrayView<ulong>, ArrayView1D<ulong, Stride1D.Dense>>(IncrementalKernels.IncrementalKernelScan);
{ var kernel = KernelUtil.GetKernel(loaded);
var loaded = accel.LoadAutoGroupedStreamKernel<Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ulong, ulong, ulong, ulong, ArrayView<ulong>, ArrayView1D<ulong, Stride1D.Dense>>(IncrementalKernels.IncrementalKernelScan); return kernel.CreateLauncherDelegate<Action<AcceleratorStream, Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ulong, ulong, ulong, ulong, ArrayView<ulong>, ArrayView1D<ulong, Stride1D.Dense>>>();
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>>>();
});
}
} }
} }
@@ -111,15 +98,12 @@ public sealed class GpuKernelLease
get get
{ {
var accel = Accelerator; var accel = Accelerator;
lock (_gpu!.Value.KernelInitLock) return KernelContainer.InitOnce(ref _kernels!.IncrementalOrder, () =>
{ {
return KernelContainer.InitOnce(ref _kernels!.IncrementalOrder, () => var loaded = accel.LoadAutoGroupedStreamKernel<Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ResidueAutomatonArgs, ArrayView<int>, ArrayView1D<ulong, Stride1D.Dense>>(IncrementalKernels.IncrementalOrderKernelScan);
{ var kernel = KernelUtil.GetKernel(loaded);
var loaded = accel.LoadAutoGroupedStreamKernel<Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ResidueAutomatonArgs, ArrayView<int>, ArrayView1D<ulong, Stride1D.Dense>>(IncrementalKernels.IncrementalOrderKernelScan); return kernel.CreateLauncherDelegate<Action<AcceleratorStream, Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ResidueAutomatonArgs, ArrayView<int>, ArrayView1D<ulong, Stride1D.Dense>>>();
var kernel = KernelUtil.GetKernel(loaded); });
return kernel.CreateLauncherDelegate<Action<AcceleratorStream, Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ResidueAutomatonArgs, ArrayView<int>, ArrayView1D<ulong, Stride1D.Dense>>>();
});
}
} }
} }
@@ -128,15 +112,12 @@ public sealed class GpuKernelLease
get get
{ {
var accel = Accelerator; var accel = Accelerator;
lock (_gpu!.Value.KernelInitLock) return KernelContainer.InitOnce(ref _kernels!.Pow2ModOrder, () =>
{ {
return KernelContainer.InitOnce(ref _kernels!.Pow2ModOrder, () => var loaded = accel.LoadAutoGroupedStreamKernel<Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ResidueAutomatonArgs, ArrayView<int>, ArrayView1D<ulong, Stride1D.Dense>>(Pow2ModKernels.Pow2ModOrderKernelScan);
{ var kernel = KernelUtil.GetKernel(loaded);
var loaded = accel.LoadAutoGroupedStreamKernel<Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ResidueAutomatonArgs, ArrayView<int>, ArrayView1D<ulong, Stride1D.Dense>>(Pow2ModKernels.Pow2ModOrderKernelScan); return kernel.CreateLauncherDelegate<Action<AcceleratorStream, Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ResidueAutomatonArgs, ArrayView<int>, ArrayView1D<ulong, Stride1D.Dense>>>();
var kernel = KernelUtil.GetKernel(loaded); });
return kernel.CreateLauncherDelegate<Action<AcceleratorStream, Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ResidueAutomatonArgs, ArrayView<int>, ArrayView1D<ulong, Stride1D.Dense>>>();
});
}
} }
} }
@@ -145,15 +126,12 @@ public sealed class GpuKernelLease
get get
{ {
var accel = Accelerator; var accel = Accelerator;
lock (_gpu!.Value.KernelInitLock) return KernelContainer.InitOnce(ref _kernels!.SmallPrimeFactor, () =>
{ {
return KernelContainer.InitOnce(ref _kernels!.SmallPrimeFactor, () => var loaded = accel.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);
var loaded = accel.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); 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>>>();
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>>>();
});
}
} }
} }
@@ -165,8 +143,6 @@ public sealed class GpuKernelLease
_gpu?.Dispose(); _gpu?.Dispose();
_gpu = null; _gpu = null;
_executionLock = null;
_releaseLimiter?.Invoke(); _releaseLimiter?.Invoke();
_releaseLimiter = null; _releaseLimiter = null;
@@ -174,25 +150,4 @@ public sealed class GpuKernelLease
Pool.Enqueue(this); Pool.Enqueue(this);
} }
public readonly struct ExecutionScope
{
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);
}
}
}
} }
PerfectNumbers.Core/Gpu/GpuKernelPool.cs
-2
View File
@@ -293,12 +293,10 @@ public class GpuKernelPool
public static void Run(Action<Accelerator, AcceleratorStream> action) public static void Run(Action<Accelerator, AcceleratorStream> action)
{ {
var lease = GetKernel(useGpuOrder: true); var lease = GetKernel(useGpuOrder: true);
var execution = lease.EnterExecutionScope();
var accelerator = lease.Accelerator; var accelerator = lease.Accelerator;
var stream = accelerator.CreateStream(); var stream = accelerator.CreateStream();
action(accelerator, stream); action(accelerator, stream);
stream.Dispose(); stream.Dispose();
execution.Dispose();
lease.Dispose(); lease.Dispose();
} }
PerfectNumbers.Core/Gpu/MersenneNumberIncrementalGpuTester.cs
-2
View File
@@ -17,7 +17,6 @@ public class MersenneNumberIncrementalGpuTester(GpuKernelType kernelType, bool u
throw new NotImplementedException($"GPU incremental scanning requires the device cycle heuristics implementation (kernel: {_kernelType}, GPU order: {_useGpuOrder})."); throw new NotImplementedException($"GPU incremental scanning requires the device cycle heuristics implementation (kernel: {_kernelType}, GPU order: {_useGpuOrder}).");
var gpuLease = GpuKernelPool.GetKernel(_useGpuOrder); var gpuLease = GpuKernelPool.GetKernel(_useGpuOrder);
var execution = gpuLease.EnterExecutionScope();
var accelerator = gpuLease.Accelerator; var accelerator = gpuLease.Accelerator;
var stream = gpuLease.Stream; var stream = gpuLease.Stream;
int batchSize = GpuConstants.ScanBatchSize; // large batch improves GPU occupancy int batchSize = GpuConstants.ScanBatchSize; // large batch improves GPU occupancy
@@ -123,7 +122,6 @@ public class MersenneNumberIncrementalGpuTester(GpuKernelType kernelType, bool u
ArrayPool<ulong>.Shared.Return(orderArray); ArrayPool<ulong>.Shared.Return(orderArray);
orderBuffer.Dispose(); orderBuffer.Dispose();
execution.Dispose();
gpuLease.Dispose(); gpuLease.Dispose();
} }
} }
PerfectNumbers.Core/Gpu/MersenneNumberOrderGpuTester.cs
-2
View File
@@ -10,7 +10,6 @@ public class MersenneNumberOrderGpuTester(GpuKernelType kernelType, bool useGpuO
public void Scan(ulong exponent, UInt128 twoP, LastDigit lastDigit, UInt128 maxK, ref bool isPrime) public void Scan(ulong exponent, UInt128 twoP, LastDigit lastDigit, UInt128 maxK, ref bool isPrime)
{ {
var gpuLease = GpuKernelPool.GetKernel(_useGpuOrder); var gpuLease = GpuKernelPool.GetKernel(_useGpuOrder);
var execution = gpuLease.EnterExecutionScope();
var accelerator = gpuLease.Accelerator; var accelerator = gpuLease.Accelerator;
var stream = gpuLease.Stream; var stream = gpuLease.Stream;
int batchSize = GpuConstants.ScanBatchSize; // large batch improves GPU occupancy and avoids TDR int batchSize = GpuConstants.ScanBatchSize; // large batch improves GPU occupancy and avoids TDR
@@ -61,7 +60,6 @@ public class MersenneNumberOrderGpuTester(GpuKernelType kernelType, bool useGpuO
} }
foundBuffer.Dispose(); foundBuffer.Dispose();
execution.Dispose();
gpuLease.Dispose(); gpuLease.Dispose();
} }
} }
PerfectNumbers.Core/Gpu/PrimeOrderGpuHeuristics.cs
+5 -14
View File
@@ -194,21 +194,20 @@ internal static partial class PrimeOrderGpuHeuristics
try try
{ {
var lease = GpuKernelPool.GetKernel(useGpuOrder: true); var lease = GpuKernelPool.GetKernel(useGpuOrder: true);
var execution = lease.EnterExecutionScope();
Accelerator accelerator = lease.Accelerator; Accelerator accelerator = lease.Accelerator;
AcceleratorStream stream = lease.Stream; AcceleratorStream stream = lease.Stream;
var kernel = GetPartialFactorKernel(accelerator); var kernel = GetPartialFactorKernel(accelerator);
SmallPrimeDeviceCache cache = GetSmallPrimeDeviceCache(accelerator); SmallPrimeDeviceCache cache = GetSmallPrimeDeviceCache(accelerator);
// TODO: We should create / reallocate these buffer only once or if the new required length is bigger than capacity. // TODO: We should create / reallocate these buffer only once or if the new required length is bigger than capacity.
var factorBuffer = accelerator.Allocate1D<ulong>(primeTargets.Length); var factorBuffer = accelerator.Allocate1D<ulong>(primeTargets.Length);
var exponentBuffer = accelerator.Allocate1D<int>(exponentTargets.Length); var exponentBuffer = accelerator.Allocate1D<int>(exponentTargets.Length);
var countBuffer = accelerator.Allocate1D<int>(1); var countBuffer = accelerator.Allocate1D<int>(1);
var remainingBuffer = accelerator.Allocate1D<ulong>(1); var remainingBuffer = accelerator.Allocate1D<ulong>(1);
var fullyFactoredBuffer = accelerator.Allocate1D<byte>(1); var fullyFactoredBuffer = accelerator.Allocate1D<byte>(1);
// TODO: There is no need to clear these buffers because the kernel will always assign values within the required bounds. // TODO: There is no need to clear these buffers because the kernel will always assign values within the required bounds.
// factorBuffer.MemSetToZero(); // factorBuffer.MemSetToZero();
// exponentBuffer.MemSetToZero(); // exponentBuffer.MemSetToZero();
// countBuffer.MemSetToZero(); // countBuffer.MemSetToZero();
@@ -242,13 +241,12 @@ internal static partial class PrimeOrderGpuHeuristics
fullyFactoredBuffer.View.CopyToCPU(ref fullyFactoredFlag, 1); fullyFactoredBuffer.View.CopyToCPU(ref fullyFactoredFlag, 1);
fullyFactored = fullyFactoredFlag != 0; fullyFactored = fullyFactoredFlag != 0;
// TODO: These buffers shouldn't be disposed but rather reused accross call with the assigned accelerator. // TODO: These buffers shouldn't be disposed but rather reused accross call with the assigned accelerator.
factorBuffer.Dispose(); factorBuffer.Dispose();
exponentBuffer.Dispose(); exponentBuffer.Dispose();
countBuffer.Dispose(); countBuffer.Dispose();
remainingBuffer.Dispose(); remainingBuffer.Dispose();
fullyFactoredBuffer.Dispose(); fullyFactoredBuffer.Dispose();
execution.Dispose();
lease.Dispose(); lease.Dispose();
return true; return true;
} }
@@ -362,12 +360,11 @@ internal static partial class PrimeOrderGpuHeuristics
order = 0UL; order = 0UL;
var lease = GpuKernelPool.GetKernel(useGpuOrder: true); var lease = GpuKernelPool.GetKernel(useGpuOrder: true);
var execution = lease.EnterExecutionScope();
Accelerator accelerator = lease.Accelerator; Accelerator accelerator = lease.Accelerator;
AcceleratorStream stream = lease.Stream; AcceleratorStream stream = lease.Stream;
var kernel = GetOrderKernel(accelerator); var kernel = GetOrderKernel(accelerator);
SmallPrimeDeviceCache cache = GetSmallPrimeDeviceCache(accelerator); SmallPrimeDeviceCache cache = GetSmallPrimeDeviceCache(accelerator);
// TODO: These buffers should be created reallocated once and assigned to an accelerator. Callers should use their own // TODO: These buffers should be created reallocated once and assigned to an accelerator. Callers should use their own
// thread static cache to prevent other threads from using taking accelerators with pre-allocated buffers if they don't // thread static cache to prevent other threads from using taking accelerators with pre-allocated buffers if they don't
@@ -467,7 +464,6 @@ internal static partial class PrimeOrderGpuHeuristics
stackProductBuffer.Dispose(); stackProductBuffer.Dispose();
resultBuffer.Dispose(); resultBuffer.Dispose();
statusBuffer.Dispose(); statusBuffer.Dispose();
execution.Dispose();
} }
} }
@@ -526,7 +522,6 @@ internal static partial class PrimeOrderGpuHeuristics
private static bool TryComputeOnGpu(ReadOnlySpan<ulong> exponents, ulong prime, in MontgomeryDivisorData divisorData, Span<ulong> results) private static bool TryComputeOnGpu(ReadOnlySpan<ulong> exponents, ulong prime, in MontgomeryDivisorData divisorData, Span<ulong> results)
{ {
var lease = GpuKernelPool.GetKernel(useGpuOrder: true); var lease = GpuKernelPool.GetKernel(useGpuOrder: true);
var execution = lease.EnterExecutionScope();
Accelerator accelerator = lease.Accelerator; Accelerator accelerator = lease.Accelerator;
AcceleratorStream stream = lease.Stream; AcceleratorStream stream = lease.Stream;
var kernel = GetPow2ModKernel(accelerator); var kernel = GetPow2ModKernel(accelerator);
@@ -551,7 +546,6 @@ internal static partial class PrimeOrderGpuHeuristics
Console.WriteLine($"Exception for {prime} and exponents: {string.Join(",", exponents.ToArray())}."); Console.WriteLine($"Exception for {prime} and exponents: {string.Join(",", exponents.ToArray())}.");
exponentBuffer.Dispose(); exponentBuffer.Dispose();
remainderBuffer.Dispose(); remainderBuffer.Dispose();
execution.Dispose();
lease.Dispose(); lease.Dispose();
throw; throw;
} }
@@ -559,7 +553,6 @@ internal static partial class PrimeOrderGpuHeuristics
remainderBuffer.View.CopyToCPU(ref MemoryMarshal.GetReference(results), exponents.Length); remainderBuffer.View.CopyToCPU(ref MemoryMarshal.GetReference(results), exponents.Length);
exponentBuffer.Dispose(); exponentBuffer.Dispose();
remainderBuffer.Dispose(); remainderBuffer.Dispose();
execution.Dispose();
lease.Dispose(); lease.Dispose();
return true; return true;
} }
@@ -587,7 +580,6 @@ internal static partial class PrimeOrderGpuHeuristics
try try
{ {
var execution = lease.EnterExecutionScope();
Accelerator accelerator = lease.Accelerator; Accelerator accelerator = lease.Accelerator;
AcceleratorStream stream = lease.Stream; AcceleratorStream stream = lease.Stream;
var kernel = GetPow2ModWideKernel(accelerator); var kernel = GetPow2ModWideKernel(accelerator);
@@ -628,7 +620,6 @@ internal static partial class PrimeOrderGpuHeuristics
exponentBuffer.Dispose(); exponentBuffer.Dispose();
remainderBuffer.Dispose(); remainderBuffer.Dispose();
execution.Dispose();
return true; return true;
} }
catch (Exception) catch (Exception)
PerfectNumbers.Core/MersenneNumberTester.cs
-2
View File
@@ -147,7 +147,6 @@ public sealed class MersenneNumberTester(
UInt128[] qs = qsBuffer; UInt128[] qs = qsBuffer;
var gpuLease = GpuKernelPool.GetKernel(_useGpuOrder); var gpuLease = GpuKernelPool.GetKernel(_useGpuOrder);
var execution = gpuLease.EnterExecutionScope();
var accelerator = gpuLease.Accelerator; var accelerator = gpuLease.Accelerator;
var stream = gpuLease.Stream; var stream = gpuLease.Stream;
var orderKernel = gpuLease.OrderKernel!; var orderKernel = gpuLease.OrderKernel!;
@@ -194,7 +193,6 @@ public sealed class MersenneNumberTester(
qBuffer.Dispose(); qBuffer.Dispose();
} }
execution.Dispose();
gpuLease.Dispose(); gpuLease.Dispose();
ArrayPool<UInt128>.Shared.Return(qs); ArrayPool<UInt128>.Shared.Return(qs);
} }
PerfectNumbers.Core/PrimeOrderCalculator.Gpu.cs
-2
View File
@@ -22,7 +22,6 @@ internal static partial class PrimeOrderCalculator
remaining = value; remaining = value;
var lease = GpuKernelPool.GetKernel(useGpuOrder: true); var lease = GpuKernelPool.GetKernel(useGpuOrder: true);
var execution = lease.EnterExecutionScope();
Accelerator accelerator = lease.Accelerator; Accelerator accelerator = lease.Accelerator;
AcceleratorStream stream = lease.Stream; AcceleratorStream stream = lease.Stream;
@@ -64,7 +63,6 @@ internal static partial class PrimeOrderCalculator
counts.Add(primeValue, exponent); counts.Add(primeValue, exponent);
} }
execution.Dispose();
lease.Dispose(); lease.Dispose();
ThreadStaticPools.UlongPool.Return(primeBufferArray); ThreadStaticPools.UlongPool.Return(primeBufferArray);
ThreadStaticPools.IntPool.Return(exponentBufferArray); ThreadStaticPools.IntPool.Return(exponentBufferArray);
PerfectNumbers.Core/ULongExtensions.Gpu.cs
-2
View File
@@ -417,7 +417,6 @@ public static partial class ULongExtensions
ulong result = 0UL; ulong result = 0UL;
GpuKernelLease lease = GpuKernelPool.GetKernel(useGpuOrder: true); GpuKernelLease lease = GpuKernelPool.GetKernel(useGpuOrder: true);
var execution = lease.EnterExecutionScope();
Accelerator accelerator = lease.Accelerator; Accelerator accelerator = lease.Accelerator;
// Keep this commented out. It should never happen in production code. // Keep this commented out. It should never happen in production code.
@@ -444,7 +443,6 @@ public static partial class ULongExtensions
resultBuffer.Dispose(); resultBuffer.Dispose();
exponentBuffer.Dispose(); exponentBuffer.Dispose();
execution.Dispose();
lease.Dispose(); lease.Dispose();
// Keep this commented. We don't want to catch any exceptions. All should crash the scanner. // Keep this commented. We don't want to catch any exceptions. All should crash the scanner.
// catch (AcceleratorException) // catch (AcceleratorException)