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

Kernels replaced with pre-compiled actions for better performance

Browse Source
This commit is contained in:
mlemanczyk committed 2025-11-20 03:00:11 +01:00
1 parent df04ba601c
commit 497b1fcf89
7 files changed
+39 -66
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

PerfectNumbers.Core/Gpu/Accelerators/PrimeOrderCalculatorAccelerator.cs
+18 -26
View File
@@ -115,20 +115,18 @@ public sealed class PrimeOrderCalculatorAccelerator
public Dictionary<ulong, int> Pow2ModEntriesToTestOnHost = []; public Dictionary<ulong, int> Pow2ModEntriesToTestOnHost = [];
public readonly Kernel CheckFactorsKernel; public readonly Action<AcceleratorStream, int, ulong, ArrayView1D<KeyValuePair<ulong, int>, Stride1D.Dense>, MontgomeryDivisorData, ArrayView1D<ulong, Stride1D.Dense>> CheckFactorsKernelLauncher;
public readonly Kernel ConvertToStandardKernel;
public readonly Action<AcceleratorStream, Index1D, ArrayView1D<ulong, Stride1D.Dense>, MontgomeryDivisorData, ArrayView1D<ulong, Stride1D.Dense>> ConvertToStandardKernelLauncher; public readonly Action<AcceleratorStream, Index1D, ArrayView1D<ulong, Stride1D.Dense>, MontgomeryDivisorData, ArrayView1D<ulong, Stride1D.Dense>> ConvertToStandardKernelLauncher;
public readonly Kernel HeuristicCombinedTrialDivisionKernel;
public readonly Action<AcceleratorStream, Index1D, byte, ArrayView<int>, ulong, ulong, HeuristicCombinedGpuViews> HeuristicCombinedTrialDivisionKernelLauncher; public readonly Action<AcceleratorStream, Index1D, byte, ArrayView<int>, ulong, ulong, HeuristicCombinedGpuViews> HeuristicCombinedTrialDivisionKernelLauncher;
public readonly Kernel KeepMontgomeryKernel; public readonly Action<AcceleratorStream, Index1D, ArrayView1D<ulong, Stride1D.Dense>, MontgomeryDivisorData, ArrayView1D<ulong, Stride1D.Dense>> KeepMontgomeryKernelLauncher;
public readonly Kernel PollardRhoKernel; public readonly Action<AcceleratorStream, ulong, int, ArrayView1D<ulong, Stride1D.Dense>, ArrayView1D<byte, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>> PollardRhoKernelLauncher;
public readonly Kernel Pow2ModWideKernel; public readonly Action<AcceleratorStream, Index1D, ArrayView1D<GpuUInt128, Stride1D.Dense>, GpuUInt128, ArrayView1D<GpuUInt128, Stride1D.Dense>> Pow2ModWideKernelLauncher;
public readonly Kernel SharesFactorKernel; public readonly Action<AcceleratorStream, Index1D, ArrayView<ulong>, ArrayView<byte>> SharesFactorKernelLauncher;
public readonly Kernel SmallPrimeFactorKernel; public readonly 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>> SmallPrimeFactorKernelLauncher;
public readonly Kernel SmallPrimeSieveKernel; public readonly Action<AcceleratorStream, Index1D, ArrayView<ulong>, ArrayView<uint>, ArrayView<uint>, ArrayView<uint>, ArrayView<uint>, ArrayView<ulong>, ArrayView<ulong>, ArrayView<ulong>, ArrayView<ulong>, ArrayView<byte>> SmallPrimeSieveKernelLauncher;
public readonly Kernel SpecialMaxKernel; public readonly Action<AcceleratorStream, Index1D, ulong, ArrayView1D<ulong, Stride1D.Dense>, int, MontgomeryDivisorData, ArrayView1D<ulong, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>> SpecialMaxKernelLauncher;
public void EnsureCapacity(int factorsCount, int primeTesterCapacity) public void EnsureCapacity(int factorsCount, int primeTesterCapacity)
{ {
@@ -262,31 +260,25 @@ public sealed class PrimeOrderCalculatorAccelerator
SmallPrimeFactorSquares = tables.Squares.View; SmallPrimeFactorSquares = tables.Squares.View;
} }
CheckFactorsKernel = KernelUtil.GetKernel(accelerator.LoadStreamKernel<int, ulong, ArrayView1D<KeyValuePair<ulong, int>, Stride1D.Dense>, MontgomeryDivisorData, ArrayView1D<ulong, Stride1D.Dense>>(PrimeOrderGpuHeuristics.CheckFactorsKernel)); CheckFactorsKernelLauncher = KernelUtil.GetKernel(accelerator.LoadStreamKernel<int, ulong, ArrayView1D<KeyValuePair<ulong, int>, Stride1D.Dense>, MontgomeryDivisorData, ArrayView1D<ulong, Stride1D.Dense>>(PrimeOrderGpuHeuristics.CheckFactorsKernel)).CreateLauncherDelegate<Action<AcceleratorStream, int, ulong, ArrayView1D<KeyValuePair<ulong, int>, Stride1D.Dense>, MontgomeryDivisorData, ArrayView1D<ulong, Stride1D.Dense>>>();
Kernel convertToStandardKernel = KernelUtil.GetKernel(accelerator.LoadAutoGroupedStreamKernel<Index1D, ArrayView1D<ulong, Stride1D.Dense>, MontgomeryDivisorData, ArrayView1D<ulong, Stride1D.Dense>>(Pow2MontgomeryKernels.Pow2MontgomeryKernelConvertToStandard)); ConvertToStandardKernelLauncher = KernelUtil.GetKernel(accelerator.LoadAutoGroupedStreamKernel<Index1D, ArrayView1D<ulong, Stride1D.Dense>, MontgomeryDivisorData, ArrayView1D<ulong, Stride1D.Dense>>(Pow2MontgomeryKernels.Pow2MontgomeryKernelConvertToStandard)).CreateLauncherDelegate<Action<AcceleratorStream, Index1D, ArrayView1D<ulong, Stride1D.Dense>, MontgomeryDivisorData, ArrayView1D<ulong, Stride1D.Dense>>>();
ConvertToStandardKernel = convertToStandardKernel; HeuristicCombinedTrialDivisionKernelLauncher = KernelUtil.GetKernel(accelerator.LoadAutoGroupedStreamKernel<Index1D, byte, ArrayView<int>, ulong, ulong, HeuristicCombinedGpuViews>(PrimeTesterKernels.HeuristicTrialCombinedDivisionKernel)).CreateLauncherDelegate<Action<AcceleratorStream, Index1D, byte, ArrayView<int>, ulong, ulong, HeuristicCombinedGpuViews>>();
ConvertToStandardKernelLauncher = convertToStandardKernel.CreateLauncherDelegate<Action<AcceleratorStream, Index1D, ArrayView1D<ulong, Stride1D.Dense>, MontgomeryDivisorData, ArrayView1D<ulong, Stride1D.Dense>>>();
var heuristicCombinedTrialDivisionKernel = KernelUtil.GetKernel(accelerator.LoadAutoGroupedStreamKernel<Index1D, byte, ArrayView<int>, ulong, ulong, HeuristicCombinedGpuViews>(PrimeTesterKernels.HeuristicTrialCombinedDivisionKernel)); KeepMontgomeryKernelLauncher = KernelUtil.GetKernel(accelerator.LoadAutoGroupedStreamKernel<Index1D, ArrayView1D<ulong, Stride1D.Dense>, MontgomeryDivisorData, ArrayView1D<ulong, Stride1D.Dense>>(Pow2MontgomeryKernels.Pow2MontgomeryKernelKeepMontgomery)).CreateLauncherDelegate<Action<AcceleratorStream, Index1D, ArrayView1D<ulong, Stride1D.Dense>, MontgomeryDivisorData, ArrayView1D<ulong, Stride1D.Dense>>>();
HeuristicCombinedTrialDivisionKernel = heuristicCombinedTrialDivisionKernel; PollardRhoKernelLauncher = KernelUtil.GetKernel(accelerator.LoadStreamKernel<ulong, int, ArrayView1D<ulong, Stride1D.Dense>, ArrayView1D<byte, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>>(Pow2MontgomeryKernels.TryPollardRhoKernel)).CreateLauncherDelegate<Action<AcceleratorStream, ulong, int, ArrayView1D<ulong, Stride1D.Dense>, ArrayView1D<byte, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>>>();
HeuristicCombinedTrialDivisionKernelLauncher = heuristicCombinedTrialDivisionKernel.CreateLauncherDelegate<Action<AcceleratorStream, Index1D, byte, ArrayView<int>, ulong, ulong, HeuristicCombinedGpuViews>>();
KeepMontgomeryKernel = KernelUtil.GetKernel(accelerator.LoadAutoGroupedStreamKernel<Index1D, ArrayView1D<ulong, Stride1D.Dense>, MontgomeryDivisorData, ArrayView1D<ulong, Stride1D.Dense>>(Pow2MontgomeryKernels.Pow2MontgomeryKernelKeepMontgomery)); Pow2ModWideKernelLauncher = KernelUtil.GetKernel(accelerator.LoadAutoGroupedStreamKernel<Index1D, ArrayView1D<GpuUInt128, Stride1D.Dense>, GpuUInt128, ArrayView1D<GpuUInt128, Stride1D.Dense>>(PrimeOrderGpuHeuristics.Pow2ModKernelWide)).CreateLauncherDelegate<Action<AcceleratorStream, Index1D, ArrayView1D<GpuUInt128, Stride1D.Dense>, GpuUInt128, ArrayView1D<GpuUInt128, Stride1D.Dense>>>();
PollardRhoKernel = KernelUtil.GetKernel(accelerator.LoadStreamKernel<ulong, int, ArrayView1D<ulong, Stride1D.Dense>, ArrayView1D<byte, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>>(Pow2MontgomeryKernels.TryPollardRhoKernel)); SharesFactorKernelLauncher = KernelUtil.GetKernel(accelerator.LoadAutoGroupedStreamKernel<Index1D, ArrayView<ulong>, ArrayView<byte>>(PrimeTesterKernels.SharesFactorKernel)).CreateLauncherDelegate<Action<AcceleratorStream, Index1D, ArrayView<ulong>, ArrayView<byte>>>();
Pow2ModWideKernel = KernelUtil.GetKernel(accelerator.LoadAutoGroupedStreamKernel<Index1D, ArrayView1D<GpuUInt128, Stride1D.Dense>, GpuUInt128, ArrayView1D<GpuUInt128, Stride1D.Dense>>(PrimeOrderGpuHeuristics.Pow2ModKernelWide)); SmallPrimeFactorKernelLauncher = KernelUtil.GetKernel(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)).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>>>();
SharesFactorKernel = KernelUtil.GetKernel(accelerator.LoadAutoGroupedStreamKernel<Index1D, ArrayView<ulong>, ArrayView<byte>>(PrimeTesterKernels.SharesFactorKernel)); SmallPrimeSieveKernelLauncher = KernelUtil.GetKernel(accelerator.LoadAutoGroupedStreamKernel<Index1D, ArrayView<ulong>, ArrayView<uint>, ArrayView<uint>, ArrayView<uint>, ArrayView<uint>, ArrayView<ulong>, ArrayView<ulong>, ArrayView<ulong>, ArrayView<ulong>, ArrayView<byte>>(PrimeTesterKernels.SmallPrimeSieveKernel)).CreateLauncherDelegate<Action<AcceleratorStream, Index1D, ArrayView<ulong>, ArrayView<uint>, ArrayView<uint>, ArrayView<uint>, ArrayView<uint>, ArrayView<ulong>, ArrayView<ulong>, ArrayView<ulong>, ArrayView<ulong>, ArrayView<byte>>>();
SmallPrimeFactorKernel = KernelUtil.GetKernel(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)); SpecialMaxKernelLauncher = KernelUtil.GetKernel(accelerator.LoadAutoGroupedStreamKernel<Index1D, ulong, ArrayView1D<ulong, Stride1D.Dense>, int, MontgomeryDivisorData, ArrayView1D<ulong, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>>(PrimeOrderGpuHeuristics.EvaluateSpecialMaxCandidatesKernel)).CreateLauncherDelegate<Action<AcceleratorStream, Index1D, ulong, ArrayView1D<ulong, Stride1D.Dense>, int, MontgomeryDivisorData, ArrayView1D<ulong, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>>>();
SmallPrimeSieveKernel = KernelUtil.GetKernel(accelerator.LoadAutoGroupedStreamKernel<Index1D, ArrayView<ulong>, ArrayView<uint>, ArrayView<uint>, ArrayView<uint>, ArrayView<uint>, ArrayView<ulong>, ArrayView<ulong>, ArrayView<ulong>, ArrayView<ulong>, ArrayView<byte>>(PrimeTesterKernels.SmallPrimeSieveKernel));
SpecialMaxKernel = KernelUtil.GetKernel(accelerator.LoadAutoGroupedStreamKernel<Index1D, ulong, ArrayView1D<ulong, Stride1D.Dense>, int, MontgomeryDivisorData, ArrayView1D<ulong, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>>(PrimeOrderGpuHeuristics.EvaluateSpecialMaxCandidatesKernel));
} }
} }
PerfectNumbers.Core/Gpu/Pow2MontgomeryGpuCalculator.cs
+1 -4
View File
@@ -18,16 +18,13 @@ public static class Pow2MontgomeryGpuCalculator
Accelerator? accelerator = gpu.Accelerator; Accelerator? accelerator = gpu.Accelerator;
MemoryBuffer1D<ulong, Stride1D.Dense> exponentBuffer = gpu.Input; MemoryBuffer1D<ulong, Stride1D.Dense> exponentBuffer = gpu.Input;
MemoryBuffer1D<ulong, Stride1D.Dense> resultBuffer = gpu.OutputUlong; MemoryBuffer1D<ulong, Stride1D.Dense> resultBuffer = gpu.OutputUlong;
var kernel = gpu.KeepMontgomeryKernel;
AcceleratorStream stream = AcceleratorStreamPool.Rent(acceleratorIndex); AcceleratorStream stream = AcceleratorStreamPool.Rent(acceleratorIndex);
exponentBuffer.View.CopyFromCPU(stream, ref exponent, 1); exponentBuffer.View.CopyFromCPU(stream, ref exponent, 1);
// We don't need to worry about any left-overs here. // We don't need to worry about any left-overs here.
// resultBuffer.MemSetToZero(stream); // resultBuffer.MemSetToZero(stream);
var kernelLauncher = kernel.CreateLauncherDelegate<Action<AcceleratorStream, Index1D, ArrayView1D<ulong, Stride1D.Dense>, MontgomeryDivisorData, ArrayView1D<ulong, Stride1D.Dense>>>(); var kernelLauncher = gpu.KeepMontgomeryKernelLauncher;
kernelLauncher(stream, 1, exponentBuffer.View, divisor, resultBuffer.View); kernelLauncher(stream, 1, exponentBuffer.View, divisor, resultBuffer.View);
resultBuffer.View.CopyToCPU(stream, ref result, 1); resultBuffer.View.CopyToCPU(stream, ref result, 1);
PerfectNumbers.Core/Gpu/PrimeOrderGpuHeuristics.cs
+1 -1
View File
@@ -604,7 +604,7 @@ internal static partial class PrimeOrderGpuHeuristics
} }
exponentBuffer.View.CopyFromCPU(stream, exponentSpan); exponentBuffer.View.CopyFromCPU(stream, exponentSpan);
var kernelLauncher = gpu.Pow2ModWideKernel.CreateLauncherDelegate<Action<AcceleratorStream, Index1D, ArrayView1D<GpuUInt128, Stride1D.Dense>, GpuUInt128, ArrayView1D<GpuUInt128, Stride1D.Dense>>>(); var kernelLauncher = gpu.Pow2ModWideKernelLauncher;
kernelLauncher(stream, length, exponentBuffer.View, modulus, remainderBuffer.View); kernelLauncher(stream, length, exponentBuffer.View, modulus, remainderBuffer.View);
remainderBuffer.View.CopyToCPU(stream, ref MemoryMarshal.GetReference(resultSpan), length); remainderBuffer.View.CopyToCPU(stream, ref MemoryMarshal.GetReference(resultSpan), length);
PerfectNumbers.Core/HeuristicCombinedPrimeTester.cs
+3 -4
View File
@@ -235,13 +235,12 @@ public sealed class HeuristicCombinedPrimeTester
private static bool HeuristicTrialDivisionGpuDetectsDivisor(PrimeOrderCalculatorAccelerator gpu, ulong n, ulong maxDivisorSquare, byte nMod10) private static bool HeuristicTrialDivisionGpuDetectsDivisor(PrimeOrderCalculatorAccelerator gpu, ulong n, ulong maxDivisorSquare, byte nMod10)
{ {
var acceleratorIndex = gpu.AcceleratorIndex; var acceleratorIndex = gpu.AcceleratorIndex;
var stream = AcceleratorStreamPool.Rent(acceleratorIndex);
var kernel = gpu.HeuristicCombinedTrialDivisionKernel;
var flagView1D = gpu.OutputInt!.View; var flagView1D = gpu.OutputInt!.View;
bool compositeDetected; bool compositeDetected;
// int compositeFlag = 0;
Span<int> compositeFlag = stackalloc int[1]; Span<int> compositeFlag = stackalloc int[1];
// int compositeFlag = 0;
var stream = AcceleratorStreamPool.Rent(acceleratorIndex);
flagView1D.MemSetToZero(stream); flagView1D.MemSetToZero(stream);
var kernelLauncher = gpu.HeuristicCombinedTrialDivisionKernelLauncher; var kernelLauncher = gpu.HeuristicCombinedTrialDivisionKernelLauncher;
PerfectNumbers.Core/HeuristicPrimeTester.cs
+1 -4
View File
@@ -3,7 +3,6 @@ using System.Runtime.CompilerServices;
using ILGPU.Runtime; using ILGPU.Runtime;
using PerfectNumbers.Core.Gpu; using PerfectNumbers.Core.Gpu;
using PerfectNumbers.Core.Gpu.Accelerators; using PerfectNumbers.Core.Gpu.Accelerators;
using ILGPU;
namespace PerfectNumbers.Core; namespace PerfectNumbers.Core;
@@ -246,7 +245,6 @@ public sealed class HeuristicPrimeTester
var gpu = PrimeOrderCalculatorAccelerator.Rent(1); var gpu = PrimeOrderCalculatorAccelerator.Rent(1);
int acceleratorIndex = gpu.AcceleratorIndex; int acceleratorIndex = gpu.AcceleratorIndex;
var stream = AcceleratorStreamPool.Rent(acceleratorIndex); var stream = AcceleratorStreamPool.Rent(acceleratorIndex);
var kernel = gpu.HeuristicCombinedTrialDivisionKernel;
var flagView1D = gpu.OutputInt!.View; var flagView1D = gpu.OutputInt!.View;
bool compositeDetected; bool compositeDetected;
@@ -261,7 +259,7 @@ public sealed class HeuristicPrimeTester
// HeuristicGpuDivisorTableKind tableKind, // HeuristicGpuDivisorTableKind tableKind,
// HeuristicGpuDivisorTables tables) // HeuristicGpuDivisorTables tables)
var kernelLauncher = kernel.CreateLauncherDelegate<Action<AcceleratorStream, Index1D, byte, ArrayView<int>, ulong, ulong, HeuristicGpuDivisorTableKind, HeuristicCombinedGpuViews>>(); var kernelLauncher = gpu.HeuristicCombinedTrialDivisionKernelLauncher;
kernelLauncher( kernelLauncher(
stream, stream,
1, 1,
@@ -269,7 +267,6 @@ public sealed class HeuristicPrimeTester
flagView1D, flagView1D,
n, n,
maxDivisorSquare, maxDivisorSquare,
HeuristicGpuDivisorTableKind.GroupA,
gpu.DivisorTables); gpu.DivisorTables);
flagView1D.CopyToCPU(stream, ref compositeFlag, 1); flagView1D.CopyToCPU(stream, ref compositeFlag, 1);
PerfectNumbers.Core/PrimeOrderCalculator.Gpu.cs
+9 -17
View File
@@ -36,15 +36,13 @@ internal static partial class PrimeOrderCalculator
ulong order = phi; ulong order = phi;
var acceleratorIndex = gpu.AcceleratorIndex; var acceleratorIndex = gpu.AcceleratorIndex;
var kernel = gpu.CheckFactorsKernel;
gpu.EnsureCapacity(entryCount, 1); gpu.EnsureCapacity(entryCount, 1);
// GpuPrimeWorkLimiter.Acquire(); // GpuPrimeWorkLimiter.Acquire();
AcceleratorStream stream = AcceleratorStreamPool.Rent(acceleratorIndex); AcceleratorStream stream = AcceleratorStreamPool.Rent(acceleratorIndex);
gpu.Pow2ModEntriesToTestOnDevice.View.CopyFromCPU(stream, entries); gpu.Pow2ModEntriesToTestOnDevice.View.CopyFromCPU(stream, entries);
var kernelLauncher = kernel.CreateLauncherDelegate<Action<AcceleratorStream, int, ulong, ArrayView1D<KeyValuePair<ulong, int>, Stride1D.Dense>, MontgomeryDivisorData, ArrayView1D<ulong, Stride1D.Dense>>>(); var kernelLauncher = gpu.CheckFactorsKernelLauncher;
kernelLauncher(stream, entryCount, phi, gpu.Pow2ModEntriesToTestOnDevice.View, divisorData, gpu.OutputUlong.View); kernelLauncher(stream, entryCount, phi, gpu.Pow2ModEntriesToTestOnDevice.View, divisorData, gpu.OutputUlong.View);
gpu.OutputUlong.View.CopyToCPU(stream, ref order, 1); gpu.OutputUlong.View.CopyToCPU(stream, ref order, 1);
@@ -67,19 +65,16 @@ internal static partial class PrimeOrderCalculator
int acceleratorIndex = gpu.AcceleratorIndex; int acceleratorIndex = gpu.AcceleratorIndex;
var accelerator = gpu.Accelerator; var accelerator = gpu.Accelerator;
// GpuPrimeWorkLimiter.Acquire(); var kernelLauncher = gpu.SmallPrimeFactorKernelLauncher;
var stream = AcceleratorStreamPool.Rent(acceleratorIndex);
var kernel = gpu.SmallPrimeFactorKernel;
var kernelLauncher = 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>>>();
ArrayView1D<int, Stride1D.Dense> smallPrimeFactorCountSlotView = gpu.SmallPrimeFactorCountSlot.View; ArrayView1D<int, Stride1D.Dense> smallPrimeFactorCountSlotView = gpu.SmallPrimeFactorCountSlot.View;
ArrayView1D<ulong, Stride1D.Dense> smallPrimeFactorRemainingSlotView = gpu.Input.View; ArrayView1D<ulong, Stride1D.Dense> smallPrimeFactorRemainingSlotView = gpu.Input.View;
ArrayView1D<ulong, Stride1D.Dense> smallPrimeFactorPrimeSlotsView = gpu.OutputUlong.View; ArrayView1D<ulong, Stride1D.Dense> smallPrimeFactorPrimeSlotsView = gpu.OutputUlong.View;
ArrayView1D<int, Stride1D.Dense> smallPrimeFactorExponentSlotsView = gpu.OutputInt.View; ArrayView1D<int, Stride1D.Dense> smallPrimeFactorExponentSlotsView = gpu.OutputInt.View;
ArrayView1D<uint, Stride1D.Dense> smallPrimeFactorsPrimesView = gpu.SmallPrimeFactorPrimes;
ArrayView1D<ulong, Stride1D.Dense> smallPrimeFactorsSquaresView = gpu.SmallPrimeFactorSquares; ArrayView1D<ulong, Stride1D.Dense> smallPrimeFactorsSquaresView = gpu.SmallPrimeFactorSquares;
ArrayView1D<uint, Stride1D.Dense> smallPrimeFactorsPrimesView = gpu.SmallPrimeFactorPrimes;
// GpuPrimeWorkLimiter.Acquire();
var stream = AcceleratorStreamPool.Rent(acceleratorIndex);
kernelLauncher( kernelLauncher(
stream, stream,
1, 1,
@@ -136,17 +131,14 @@ internal static partial class PrimeOrderCalculator
int acceleratorIndex = gpu.AcceleratorIndex; int acceleratorIndex = gpu.AcceleratorIndex;
// GpuPrimeWorkLimiter.Acquire(); // GpuPrimeWorkLimiter.Acquire();
var stream = AcceleratorStreamPool.Rent(acceleratorIndex);
ArrayView1D<ulong, Stride1D.Dense> specialMaxFactorsView = gpu.Input.View; ArrayView1D<ulong, Stride1D.Dense> specialMaxFactorsView = gpu.Input.View;
specialMaxFactorsView.SubView(0, factorCount).CopyFromCPU(stream, factors);
var kernel = gpu.SpecialMaxKernel; var stream = AcceleratorStreamPool.Rent(acceleratorIndex);
specialMaxFactorsView.SubView(0, factorCount).CopyFromCPU(stream, factors);
ArrayView1D<ulong, Stride1D.Dense> specialMaxResultView = gpu.SpecialMaxResult.View; ArrayView1D<ulong, Stride1D.Dense> specialMaxResultView = gpu.SpecialMaxResult.View;
var kernelLauncher = kernel.CreateLauncherDelegate<Action<AcceleratorStream, Index1D, ulong, ArrayView1D<ulong, Stride1D.Dense>, int, MontgomeryDivisorData, ArrayView1D<ulong, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>>>(); var kernelLauncher = gpu.SpecialMaxKernelLauncher;
kernelLauncher( kernelLauncher(
stream, stream,
1, 1,
@@ -178,7 +170,7 @@ internal static partial class PrimeOrderCalculator
AcceleratorStream stream = AcceleratorStreamPool.Rent(acceleratorIndex); AcceleratorStream stream = AcceleratorStreamPool.Rent(acceleratorIndex);
gpu.Input.View.CopyFromCPU(stream, randomStateSpan); gpu.Input.View.CopyFromCPU(stream, randomStateSpan);
var kernelLauncher = gpu.PollardRhoKernel.CreateLauncherDelegate<Action<AcceleratorStream, ulong, int, ArrayView1D<ulong, Stride1D.Dense>, ArrayView1D<byte, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>>>(); var kernelLauncher = gpu.PollardRhoKernelLauncher;
kernelLauncher(stream, n, 1, gpu.Input.View, gpu.OutputByte.View, gpu.OutputUlong.View); kernelLauncher(stream, n, 1, gpu.Input.View, gpu.OutputByte.View, gpu.OutputUlong.View);
PerfectNumbers.Core/PrimeTester.cs
+6 -10
View File
@@ -90,19 +90,17 @@ public sealed class PrimeTester
[MethodImpl(MethodImplOptions.AggressiveInlining)] [MethodImpl(MethodImplOptions.AggressiveInlining)]
public static bool IsPrimeGpu(ulong n) public static bool IsPrimeGpu(ulong n)
{ {
byte flag = 0;
var gpu = PrimeOrderCalculatorAccelerator.Rent(1); var gpu = PrimeOrderCalculatorAccelerator.Rent(1);
int acceleratorIndex = gpu.AcceleratorIndex; int acceleratorIndex = gpu.AcceleratorIndex;
var inputView = gpu.Input.View; var inputView = gpu.Input.View;
var outputView = gpu.OutputByte.View; var outputView = gpu.OutputByte.View;
var kernel = gpu.SmallPrimeSieveKernel!;
// GpuPrimeWorkLimiter.Acquire(); // GpuPrimeWorkLimiter.Acquire();
AcceleratorStream stream = AcceleratorStreamPool.Rent(acceleratorIndex); AcceleratorStream stream = AcceleratorStreamPool.Rent(acceleratorIndex);
inputView.CopyFromCPU(stream, ref n, 1); inputView.CopyFromCPU(stream, ref n, 1);
var kernelLauncher = kernel.CreateLauncherDelegate<Action<AcceleratorStream, Index1D, ArrayView<ulong>, ArrayView<uint>, ArrayView<uint>, ArrayView<uint>, ArrayView<uint>, ArrayView<ulong>, ArrayView<ulong>, ArrayView<ulong>, ArrayView<ulong>, ArrayView<byte>>>(); var kernelLauncher = gpu.SmallPrimeSieveKernelLauncher;
kernelLauncher( kernelLauncher(
stream, stream,
@@ -118,6 +116,7 @@ public sealed class PrimeTester
gpu.DevicePrimesPow2LastNine, gpu.DevicePrimesPow2LastNine,
outputView); outputView);
byte flag = 0;
outputView.CopyToCPU(stream, ref flag, 1); outputView.CopyToCPU(stream, ref flag, 1);
stream.Synchronize(); stream.Synchronize();
@@ -136,13 +135,12 @@ public sealed class PrimeTester
var inputView = input.View; var inputView = input.View;
var output = gpu.OutputByte; var output = gpu.OutputByte;
var outputView = output.View; var outputView = output.View;
Kernel kernel = gpu.SmallPrimeSieveKernel!;
int acceleratorIndex = gpu.AcceleratorIndex; int acceleratorIndex = gpu.AcceleratorIndex;
AcceleratorStream stream = AcceleratorStreamPool.Rent(acceleratorIndex); AcceleratorStream stream = AcceleratorStreamPool.Rent(acceleratorIndex);
inputView.CopyFromCPU(stream, ref n, 1); inputView.CopyFromCPU(stream, ref n, 1);
var kernelLauncher = kernel.CreateLauncherDelegate<Action<AcceleratorStream, Index1D, ArrayView<ulong>, ArrayView<uint>, ArrayView<uint>, ArrayView<uint>, ArrayView<uint>, ArrayView<ulong>, ArrayView<ulong>, ArrayView<ulong>, ArrayView<ulong>, ArrayView<byte>>>(); var kernelLauncher = gpu.SmallPrimeSieveKernelLauncher;
kernelLauncher( kernelLauncher(
stream, stream,
@@ -196,7 +194,6 @@ public sealed class PrimeTester
// GpuPrimeWorkLimiter.Acquire(); // GpuPrimeWorkLimiter.Acquire();
var gpu = PrimeOrderCalculatorAccelerator.Rent(GpuBatchSize); var gpu = PrimeOrderCalculatorAccelerator.Rent(GpuBatchSize);
int acceleratorIndex = gpu.AcceleratorIndex; int acceleratorIndex = gpu.AcceleratorIndex;
var kernel = gpu.SmallPrimeSieveKernel!;
int totalLength = values.Length; int totalLength = values.Length;
int batchSize = GpuBatchSize; int batchSize = GpuBatchSize;
@@ -207,7 +204,7 @@ public sealed class PrimeTester
int pos = 0; int pos = 0;
AcceleratorStream stream = AcceleratorStreamPool.Rent(acceleratorIndex); AcceleratorStream stream = AcceleratorStreamPool.Rent(acceleratorIndex);
var kernelLauncher = kernel.CreateLauncherDelegate<Action<AcceleratorStream, Index1D, ArrayView<ulong>, ArrayView<uint>, ArrayView<uint>, ArrayView<uint>, ArrayView<uint>, ArrayView<ulong>, ArrayView<ulong>, ArrayView<ulong>, ArrayView<ulong>, ArrayView<byte>>>(); var kernelLauncher = gpu.SmallPrimeSieveKernelLauncher;
while (pos < totalLength) while (pos < totalLength)
{ {
@@ -280,15 +277,14 @@ public sealed class PrimeTester
var gpu = PrimeOrderCalculatorAccelerator.Rent(1); var gpu = PrimeOrderCalculatorAccelerator.Rent(1);
int acceleratorIndex = gpu.AcceleratorIndex; int acceleratorIndex = gpu.AcceleratorIndex;
var kernel = gpu.SharesFactorKernel;
gpu.EnsureCapacity(0, length); gpu.EnsureCapacity(0, length);
MemoryBuffer1D<ulong, Stride1D.Dense>? inputBuffer = gpu.Input; MemoryBuffer1D<ulong, Stride1D.Dense>? inputBuffer = gpu.Input;
MemoryBuffer1D<byte, Stride1D.Dense>? resultBuffer = gpu.OutputByte;
AcceleratorStream stream = AcceleratorStreamPool.Rent(acceleratorIndex); AcceleratorStream stream = AcceleratorStreamPool.Rent(acceleratorIndex);
inputBuffer.View.CopyFromCPU(stream, values); inputBuffer.View.CopyFromCPU(stream, values);
var kernelLauncher = kernel.CreateLauncherDelegate<Action<AcceleratorStream, Index1D, ArrayView<ulong>, ArrayView<byte>>>();
MemoryBuffer1D<byte, Stride1D.Dense>? resultBuffer = gpu.OutputByte;
var kernelLauncher = gpu.SharesFactorKernelLauncher;
kernelLauncher(stream, length, inputBuffer.View, resultBuffer.View); kernelLauncher(stream, length, inputBuffer.View, resultBuffer.View);
resultBuffer.View.CopyToCPU(stream, in results); resultBuffer.View.CopyToCPU(stream, in results);