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

ConcurrentDictionary removed. Kernel made dependent on the shared GPU context.

Browse Source
This commit is contained in:
mlemanczyk committed 2025-11-07 10:16:33 +01:00
1 parent 9537addf38
commit 119690086a
11 files changed
+343 -315
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

PerfectNumbers.Core/Gpu/GpuKernelPool.cs
+7 -16
View File
@@ -83,7 +83,7 @@ public readonly struct ResiduePrimeViews(
public readonly ArrayView1D<ulong, Stride1D.Dense> LastSevenPow2 = lastSevenPow2;
}
public readonly struct SmallPrimeFactorTables(
public readonly struct SmallPrimeFactorViews(
MemoryBuffer1D<uint, Stride1D.Dense> primes,
MemoryBuffer1D<ulong, Stride1D.Dense> squares,
int count)
@@ -184,18 +184,13 @@ public class GpuKernelPool
return new ResiduePrimeViews(deviceLastOne.View, deviceLastSeven.View, deviceLastOnePow2.View, deviceLastSevenPow2.View);
}
public static SmallPrimeFactorTables EnsureSmallPrimeFactorTables(KernelContainer kernels, Accelerator accelerator, AcceleratorStream stream)
public static SmallPrimeFactorViews EnsureSmallPrimeFactorTables(KernelContainer kernels, Accelerator accelerator, AcceleratorStream stream)
{
if (kernels.SmallPrimeFactorsPrimes is { } primeBuffer && kernels.SmallPrimeFactorsSquares is { } squareBuffer)
{
return new SmallPrimeFactorTables(primeBuffer, squareBuffer, (int)primeBuffer.Length);
return new SmallPrimeFactorViews(primeBuffer, squareBuffer, (int)primeBuffer.Length);
}
// if (kernels.SmallPrimeFactorsPrimes is { } existingPrimes && kernels.SmallPrimeFactorsSquares is { } existingSquares)
// {
// return new SmallPrimeFactorTables(existingPrimes, existingSquares, (int)existingPrimes.Length);
// }
var hostPrimes = PrimesGenerator.SmallPrimes;
var hostSquares = PrimesGenerator.SmallPrimesPow2;
@@ -207,17 +202,13 @@ public class GpuKernelPool
kernels.SmallPrimeFactorsPrimes = devicePrimes;
kernels.SmallPrimeFactorsSquares = deviceSquares;
return new SmallPrimeFactorTables(devicePrimes, deviceSquares, hostPrimes.Length);
return new SmallPrimeFactorViews(devicePrimes, deviceSquares, hostPrimes.Length);
}
public static GpuKernelLease GetKernel()
public static GpuKernelLease Rent()
{
GpuPrimeWorkLimiter.Acquire();
var gpu = Rent();
var gpu = GpuContextPool.Rent();
var stream = gpu.Accelerator.CreateStream();
var kernels = GetKernels(gpu.Accelerator, stream);
return GpuKernelLease.Rent(gpu, stream, kernels);
@@ -229,7 +220,7 @@ public class GpuKernelPool
/// <param name="action">Action to run with (Accelerator, Stream).</param>
public static void Run(Action<Accelerator, AcceleratorStream> action)
{
var lease = GetKernel();
var lease = Rent();
var accelerator = lease.Accelerator;
var stream = lease.Stream;
action(accelerator, stream);
PerfectNumbers.Core/Gpu/MersenneNumberDivisorByDivisorGpuTester.cs
+2 -2
View File
@@ -95,7 +95,7 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
ulong processedCount;
ulong lastProcessed;
var gpuLease = GpuKernelPool.GetKernel();
var gpuLease = GpuKernelPool.Rent();
var accelerator = gpuLease.Accelerator;
var stream = gpuLease.Stream;
@@ -501,7 +501,7 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
internal DivisorScanSession(MersenneNumberDivisorByDivisorGpuTester owner)
{
_owner = owner;
_lease = GpuKernelPool.GetKernel();
_lease = GpuKernelPool.Rent();
_accelerator = _lease.Accelerator;
_stream = _lease.Stream;
_capacity = Math.Max(1, owner._gpuBatchSize);
PerfectNumbers.Core/Gpu/MersenneNumberIncrementalGpuTester.cs
+1 -1
View File
@@ -16,7 +16,7 @@ public class MersenneNumberIncrementalGpuTester(GpuKernelType kernelType, bool u
{
throw new NotImplementedException($"GPU incremental scanning requires the device cycle heuristics implementation (kernel: {_kernelType}, GPU order: {_useGpuOrder}).");
var gpuLease = GpuKernelPool.GetKernel();
var gpuLease = GpuKernelPool.Rent();
var accelerator = gpuLease.Accelerator;
var stream = gpuLease.Stream;
int batchSize = GpuConstants.ScanBatchSize; // large batch improves GPU occupancy
PerfectNumbers.Core/Gpu/MersenneNumberOrderGpuTester.cs
+1 -1
View File
@@ -9,7 +9,7 @@ public class MersenneNumberOrderGpuTester(GpuKernelType kernelType, bool useGpuO
public void Scan(ulong exponent, UInt128 twoP, LastDigit lastDigit, UInt128 maxK, ref bool isPrime)
{
var gpuLease = GpuKernelPool.GetKernel();
var gpuLease = GpuKernelPool.Rent();
var accelerator = gpuLease.Accelerator;
var stream = gpuLease.Stream;
int batchSize = GpuConstants.ScanBatchSize; // large batch improves GPU occupancy and avoids TDR
PerfectNumbers.Core/Gpu/PrimeOrderGpuHeuristics.cs
+4 -4
View File
@@ -198,7 +198,7 @@ internal static partial class PrimeOrderGpuHeuristics
try
{
var lease = GpuKernelPool.GetKernel();
var lease = GpuKernelPool.Rent();
Accelerator accelerator = lease.Accelerator;
AcceleratorStream stream = lease.Stream;
@@ -365,7 +365,7 @@ internal static partial class PrimeOrderGpuHeuristics
{
order = 0UL;
var lease = GpuKernelPool.GetKernel();
var lease = GpuKernelPool.Rent();
Accelerator accelerator = lease.Accelerator;
AcceleratorStream stream = lease.Stream;
@@ -526,7 +526,7 @@ internal static partial class PrimeOrderGpuHeuristics
private static bool TryComputeOnGpu(ReadOnlySpan<ulong> exponents, ulong prime, in MontgomeryDivisorData divisorData, Span<ulong> results)
{
var lease = GpuKernelPool.GetKernel();
var lease = GpuKernelPool.Rent();
Accelerator accelerator = lease.Accelerator;
AcceleratorStream stream = lease.Stream;
var kernel = GetPow2ModKernel(accelerator);
@@ -581,7 +581,7 @@ internal static partial class PrimeOrderGpuHeuristics
GpuUInt128[]? rentedExponents = null;
GpuUInt128[]? rentedResults = null;
var lease = GpuKernelPool.GetKernel();
var lease = GpuKernelPool.Rent();
ArrayPool<GpuUInt128> gpuUInt128Pool = ThreadStaticPools.GpuUInt128Pool;
try