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

MethodImplOptions.AggressiveOptimization & MethodImplOptions.AggressiveInlining optimized. Small optimizations.

Browse Source
This commit is contained in:
mlemanczyk committed 2025-12-10 19:59:51 +01:00
1 parent 7a242b3bfa
commit 05d3061c47
23 files changed
+628 -559
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

PerfectNumbers.Core/Gpu/Accelerators/HeuristicCombinedGpuTables.cs
+6
View File
@@ -1,3 +1,4 @@
using System.Runtime.CompilerServices;
using ILGPU;
using ILGPU.Runtime;
@@ -9,10 +10,13 @@ internal readonly struct HeuristicCombinedGpuTables
private static readonly HeuristicCombinedGpuTables[] _sharedTables = new HeuristicCombinedGpuTables[PerfectNumberConstants.RollingAccelerators];
[MethodImpl(MethodImplOptions.AggressiveOptimization)]
internal static void WarmUp(int acceleratorIndex, AcceleratorStream stream) => _sharedTables[acceleratorIndex] = new(_accelerators[acceleratorIndex], stream);
[MethodImpl(MethodImplOptions.AggressiveInlining | MethodImplOptions.AggressiveOptimization)]
internal static HeuristicCombinedGpuTables GetStaticTables(int acceleratorIndex) => _sharedTables[acceleratorIndex];
[MethodImpl(MethodImplOptions.AggressiveOptimization)]
internal HeuristicCombinedGpuTables(Accelerator accelerator, AcceleratorStream stream)
{
var combinedEnding1 = HeuristicCombinedPrimeTester.CombinedDivisorsEnding1;
@@ -45,6 +49,7 @@ internal readonly struct HeuristicCombinedGpuTables
internal readonly MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicCombinedDivisorsEnding9;
internal readonly MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicCombinedDivisorSquaresEnding9;
[MethodImpl(MethodImplOptions.AggressiveInlining | MethodImplOptions.AggressiveOptimization)]
internal HeuristicCombinedGpuViews CreateViews() => new(
HeuristicCombinedDivisorsEnding1.View,
HeuristicCombinedDivisorsEnding3.View,
@@ -55,6 +60,7 @@ internal readonly struct HeuristicCombinedGpuTables
HeuristicCombinedDivisorSquaresEnding7.View,
HeuristicCombinedDivisorSquaresEnding9.View);
[MethodImpl(MethodImplOptions.AggressiveOptimization)]
internal void Dispose()
{
HeuristicCombinedDivisorsEnding1.Dispose();
PerfectNumbers.Core/Gpu/Accelerators/PrimeOrderCalculatorAccelerator.cs
+3
View File
@@ -195,6 +195,7 @@ public sealed class PrimeOrderCalculatorAccelerator
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 Action<AcceleratorStream, Index1D, ulong, ArrayView<ulong>, int, ulong, ArrayView<ulong>> SpecialMaxKernelLauncher;
[MethodImpl(MethodImplOptions.AggressiveOptimization)]
public void EnsureCalculateOrderWideCapacity(int newSize)
{
if (newSize > Input.Length)
@@ -213,6 +214,7 @@ public sealed class PrimeOrderCalculatorAccelerator
}
}
[MethodImpl(MethodImplOptions.AggressiveOptimization)]
public void EnsureCapacity(int factorsCount, int primeTesterCapacity)
{
Accelerator accelerator = Accelerator;
@@ -325,6 +327,7 @@ public sealed class PrimeOrderCalculatorAccelerator
}
}
[MethodImpl(MethodImplOptions.AggressiveOptimization)]
public void EnsureUlongInputOutputCapacity(int newSize)
{
var accelerator = Accelerator;
PerfectNumbers.Core/Gpu/GpuKernelPool.cs
+2
View File
@@ -1,3 +1,4 @@
using System.Runtime.CompilerServices;
using ILGPU;
using ILGPU.Runtime;
using PerfectNumbers.Core.Gpu.Accelerators;
@@ -74,6 +75,7 @@ public class GpuKernelPool
});
}
[MethodImpl(MethodImplOptions.AggressiveOptimization)]
static void InitPow2ModKernelScan(Accelerator accelerator, KernelContainer kernels)
{
InitOnce(ref kernels.Pow2Mod, () =>
PerfectNumbers.Core/Gpu/Kernels/Pow2ModKernels.cs
+3
View File
@@ -1,3 +1,4 @@
using System.Runtime.CompilerServices;
using ILGPU;
using ILGPU.Runtime;
@@ -6,6 +7,7 @@ namespace PerfectNumbers.Core.Gpu;
internal static class Pow2ModKernels
{
/// This kernel always sets the result of the corresponding element. Callers don't need to clear the output buffers.
[MethodImpl(MethodImplOptions.AggressiveOptimization)]
public static void Pow2ModKernelScan(
Index1D index,
ulong exponent,
@@ -107,6 +109,7 @@ internal static class Pow2ModKernels
/// This kernel doesn't always set the result of the found array. It sets the value only when it needs to. The callers
/// must clean the output buffer before the call to get a deterministic result.
[MethodImpl(MethodImplOptions.AggressiveOptimization)]
public static void Pow2ModOrderKernelScan(
Index1D index,
ulong exponent,
PerfectNumbers.Core/Gpu/Kernels/Pow2MontgomeryKernels.cs
+4 -4
View File
@@ -117,7 +117,7 @@ public static class Pow2MontgomeryKernels
return sumHigh;
}
[MethodImpl(MethodImplOptions.AggressiveInlining)]
[MethodImpl(MethodImplOptions.AggressiveOptimization | MethodImplOptions.AggressiveInlining)]
private static void MultiplyPartsGpu(ulong left, ulong right, out ulong high, out ulong low)
{
ulong leftLow = (uint)left;
@@ -135,13 +135,13 @@ public static class Pow2MontgomeryKernels
high = unchecked(highProduct + (cross1 >> 32) + (cross2 >> 32) + (carry >> 32));
}
[MethodImpl(MethodImplOptions.AggressiveInlining)]
[MethodImpl(MethodImplOptions.AggressiveInlining | MethodImplOptions.AggressiveOptimization)]
internal static ulong Pow2MontgomeryModWindowedGpuKeepMontgomery(in MontgomeryDivisorDataGpu divisor, ulong exponent)
{
return Pow2MontgomeryModWindowedGpuMontgomeryResult(divisor, exponent);
}
[MethodImpl(MethodImplOptions.AggressiveInlining)]
[MethodImpl(MethodImplOptions.AggressiveInlining | MethodImplOptions.AggressiveOptimization)]
public static ulong Pow2MontgomeryModWindowedGpuConvertToStandard(this MontgomeryDivisorDataGpu divisor, ulong exponent)
{
ulong modulus = divisor.Modulus;
@@ -150,7 +150,7 @@ public static class Pow2MontgomeryKernels
return MontgomeryMultiplyGpu(montgomeryResult, 1UL, modulus, nPrime);
}
[MethodImpl(MethodImplOptions.AggressiveInlining)]
[MethodImpl(MethodImplOptions.AggressiveOptimization)]
private static ulong Pow2MontgomeryModWindowedGpuMontgomeryResult(in MontgomeryDivisorDataGpu divisor, ulong exponent)
{
ulong modulus = divisor.Modulus;
PerfectNumbers.Core/Gpu/Kernels/PrimeOrderKernels.cs
+5
View File
@@ -1182,6 +1182,7 @@ internal static partial class PrimeOrderKernels
return divisor.Pow2MontgomeryModWindowedGpuConvertToStandard(exponent) == 1UL;
}
[MethodImpl(MethodImplOptions.AggressiveOptimization)]
private static ulong CalculateByDoublingKernel(ulong prime)
{
ulong order = 1UL;
@@ -1203,6 +1204,7 @@ internal static partial class PrimeOrderKernels
/// This kernel always sets the result of the corresponding element. Callers don't need to clear the output buffers.
[MethodImpl(MethodImplOptions.AggressiveOptimization)]
internal static void Pow2ModKernel(Index1D index, ArrayView1D<ulong, Stride1D.Dense> exponents, ulong divisorModulus, ulong divisorNPrime, ulong divisorMontgomeryOne, ulong divisorMontgomeryTwo, ulong divisorMontgomeryTwoSquared, ArrayView1D<ulong, Stride1D.Dense> remainders)
{
ulong exponent = exponents[index];
@@ -1211,12 +1213,14 @@ internal static partial class PrimeOrderKernels
}
/// This kernel always sets the result of the corresponding element. Callers don't need to clear the output buffers.
[MethodImpl(MethodImplOptions.AggressiveOptimization)]
internal static void Pow2ModKernelWide(Index1D index, ArrayView1D<GpuUInt128, Stride1D.Dense> exponents, GpuUInt128 modulus, ArrayView1D<GpuUInt128, Stride1D.Dense> remainders)
{
GpuUInt128 exponent = exponents[index];
remainders[index] = Pow2ModKernelCore(exponent, modulus);
}
[MethodImpl(MethodImplOptions.AggressiveOptimization)]
private static GpuUInt128[] InitializeOddPowersTable(GpuUInt128 baseValue, GpuUInt128 modulus, int oddPowerCount)
{
GpuUInt128[] result = new GpuUInt128[PerfectNumberConstants.MaxOddPowersCount];
@@ -1242,6 +1246,7 @@ internal static partial class PrimeOrderKernels
return result;
}
[MethodImpl(MethodImplOptions.AggressiveOptimization)]
internal static GpuUInt128 Pow2ModKernelCore(GpuUInt128 exponent, GpuUInt128 modulus)
{
// This should never happen in production code.
PerfectNumbers.Core/Gpu/MersenneNumberResidueGpuTester.cs
+8 -1
View File
@@ -23,6 +23,7 @@ public class MersenneNumberResidueGpuTester(bool useGpuOrder)
// requested by the caller, skip queuing additional block generation, and keep the snapshot cache untouched.
// GPU residue variant: check 2^p % q == 1 for q = 2*p*k + 1.
[MethodImpl(MethodImplOptions.AggressiveOptimization)]
public void Scan(PrimeOrderCalculatorAccelerator gpu, ulong exponent, UInt128 twoP, LastDigit lastDigit, UInt128 maxK, ref bool isPrime)
{
// GpuPrimeWorkLimiter.Acquire();
@@ -114,6 +115,7 @@ public class MersenneNumberResidueGpuTester(bool useGpuOrder)
// GpuPrimeWorkLimiter.Release();
}
[MethodImpl(MethodImplOptions.AggressiveOptimization)]
private static 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>> GetPow2ModKernel(Accelerator accelerator)
{
@@ -128,6 +130,7 @@ public class MersenneNumberResidueGpuTester(bool useGpuOrder)
return cached;
}
[MethodImpl(MethodImplOptions.AggressiveOptimization)]
private static 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>> CreatePow2ModKernel(Accelerator accelerator)
{
var loaded = 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);
@@ -140,6 +143,7 @@ public class MersenneNumberResidueGpuTester(bool useGpuOrder)
}
[MethodImpl(MethodImplOptions.AggressiveOptimization)]
private static ResidueResources RentResources(Accelerator accelerator, int capacity)
{
var resourcePools = _resourcePools ??= new();
@@ -152,6 +156,7 @@ public class MersenneNumberResidueGpuTester(bool useGpuOrder)
return new ResidueResources(accelerator, capacity);
}
[MethodImpl(MethodImplOptions.AggressiveOptimization)]
private static void ReturnResources(ResidueResources resources)
{
var resourcePools = _resourcePools ??= new();
@@ -160,6 +165,7 @@ public class MersenneNumberResidueGpuTester(bool useGpuOrder)
private sealed class ResidueResources
{
[MethodImpl(MethodImplOptions.AggressiveOptimization)]
internal ResidueResources(Accelerator accelerator, int capacity)
{
Capacity = Math.Max(1, capacity);
@@ -171,6 +177,7 @@ public class MersenneNumberResidueGpuTester(bool useGpuOrder)
OrderArray = ThreadStaticPools.UlongPool.Rent(Capacity);
}
[MethodImpl(MethodImplOptions.AggressiveOptimization)]
internal void EnsureCapacity(Accelerator accelerator, int capacity)
{
if (OrderBuffer.Length >= capacity)
@@ -213,4 +220,4 @@ public class MersenneNumberResidueGpuTester(bool useGpuOrder)
}
}
}
PerfectNumbers.Core/Gpu/Pow2MontgomeryGpuCalculator.cs
+2 -2
View File
@@ -7,7 +7,7 @@ namespace PerfectNumbers.Core.Gpu;
public static class Pow2MontgomeryGpuCalculator
{
[MethodImpl(MethodImplOptions.AggressiveInlining)]
[MethodImpl(MethodImplOptions.AggressiveInlining | MethodImplOptions.AggressiveOptimization)]
public static ulong CalculateKeep(PrimeOrderCalculatorAccelerator gpu, ulong exponent, in MontgomeryDivisorData divisor)
{
// GpuPrimeWorkLimiter.Acquire();
@@ -34,7 +34,7 @@ public static class Pow2MontgomeryGpuCalculator
return result;
}
[MethodImpl(MethodImplOptions.AggressiveInlining)]
[MethodImpl(MethodImplOptions.AggressiveInlining | MethodImplOptions.AggressiveOptimization)]
public static ulong CalculateConvert(PrimeOrderCalculatorAccelerator gpu, ulong exponent, in MontgomeryDivisorData divisor)
{
ulong result = 0UL;
PerfectNumbers.Core/Gpu/PrimeOrderGpuHeuristics.cs
+12
View File
@@ -1,5 +1,6 @@
using System.Buffers;
using System.Collections.Concurrent;
using System.Runtime.CompilerServices;
using System.Runtime.InteropServices;
using ILGPU.Runtime;
using PerfectNumbers.Core.Gpu.Accelerators;
@@ -16,6 +17,7 @@ internal enum GpuPow2ModStatus
internal static partial class PrimeOrderGpuHeuristics
{
[MethodImpl(MethodImplOptions.AggressiveInlining | MethodImplOptions.AggressiveOptimization)]
internal static void PreloadStaticTables(AcceleratorStream stream)
{
// _ = GetSmallPrimeDeviceCache(stream);
@@ -31,16 +33,19 @@ internal static partial class PrimeOrderGpuHeuristics
internal static ConcurrentDictionary<ulong, byte> OverflowRegistry => OverflowedPrimes;
internal static ConcurrentDictionary<UInt128, byte> OverflowRegistryWide => OverflowedPrimesWide;
[MethodImpl(MethodImplOptions.AggressiveInlining | MethodImplOptions.AggressiveOptimization)]
internal static void OverrideCapabilitiesForTesting(PrimeOrderGpuCapability capability)
{
s_capability = capability;
}
[MethodImpl(MethodImplOptions.AggressiveInlining | MethodImplOptions.AggressiveOptimization)]
internal static void ResetCapabilitiesForTesting()
{
s_capability = PrimeOrderGpuCapability.Default;
}
[MethodImpl(MethodImplOptions.AggressiveOptimization)]
public static bool TryPartialFactor(
PrimeOrderCalculatorAccelerator gpu,
ulong value,
@@ -88,6 +93,7 @@ internal static partial class PrimeOrderGpuHeuristics
private static readonly Accelerator[] _accelerators = AcceleratorPool.Shared.Accelerators;
[MethodImpl(MethodImplOptions.AggressiveOptimization)]
private static bool TryLaunchPartialFactorKernel(
PrimeOrderCalculatorAccelerator gpu,
ulong value,
@@ -156,6 +162,7 @@ internal static partial class PrimeOrderGpuHeuristics
return true;
}
[MethodImpl(MethodImplOptions.AggressiveOptimization)]
public static GpuPow2ModStatus TryPow2Mod(PrimeOrderCalculatorAccelerator gpu, ulong exponent, ulong prime, out ulong remainder, in MontgomeryDivisorData? divisorData)
{
Span<ulong> exponents = stackalloc ulong[1];
@@ -167,6 +174,7 @@ internal static partial class PrimeOrderGpuHeuristics
return status;
}
[MethodImpl(MethodImplOptions.AggressiveOptimization)]
public static GpuPow2ModStatus TryPow2ModBatch(PrimeOrderCalculatorAccelerator gpu, ReadOnlySpan<ulong> exponents, ulong prime, Span<ulong> remainders, in MontgomeryDivisorData? divisorData)
{
// This will never occur in production code
@@ -222,6 +230,7 @@ internal static partial class PrimeOrderGpuHeuristics
return computed ? GpuPow2ModStatus.Success : GpuPow2ModStatus.Unavailable;
}
[MethodImpl(MethodImplOptions.AggressiveOptimization)]
public static GpuPow2ModStatus TryPow2Mod(PrimeOrderCalculatorAccelerator gpu, in UInt128 exponent, in UInt128 prime, out UInt128 remainder)
{
Span<UInt128> exponents = stackalloc UInt128[1];
@@ -233,11 +242,13 @@ internal static partial class PrimeOrderGpuHeuristics
return status;
}
[MethodImpl(MethodImplOptions.AggressiveOptimization)]
public static GpuPow2ModStatus TryPow2ModBatch(PrimeOrderCalculatorAccelerator gpu, ReadOnlySpan<UInt128> exponents, UInt128 prime, Span<UInt128> remainders)
{
return TryPow2ModBatchInternal(gpu, exponents, prime, remainders);
}
[MethodImpl(MethodImplOptions.AggressiveOptimization)]
internal static bool TryCalculateOrder(
PrimeOrderCalculatorAccelerator gpu,
ulong prime,
@@ -310,6 +321,7 @@ internal static partial class PrimeOrderGpuHeuristics
return order != 0UL;
}
[MethodImpl(MethodImplOptions.AggressiveOptimization)]
private static GpuPow2ModStatus TryPow2ModBatchInternal(PrimeOrderCalculatorAccelerator gpu, ReadOnlySpan<UInt128> exponents, UInt128 prime, Span<UInt128> remainders)
{
if (exponents.Length == 0)