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

Stream handling simplified to avoid dictionary lookups in favor of array lookups with known indexes.

Browse Source
This commit is contained in:
mlemanczyk committed 2025-11-16 15:01:51 +01:00
1 parent cd7636ddbe
commit a84445ef2a
15 files changed
+67 -73
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

PerfectNumbers.Core/Gpu/Accelerators/AcceleratorStreamPool.cs
+10 -10
View File
@@ -7,8 +7,7 @@ public static class AcceleratorStreamPool
{
private static readonly Accelerator[] _accelerators = AcceleratorPool.Shared.Accelerators;
private static readonly SemaphoreSlim[] _locks = new SemaphoreSlim[PerfectNumberConstants.RollingAccelerators];
private static readonly ConcurrentQueue<AcceleratorStream>[] _streams = new ConcurrentQueue<AcceleratorStream> [PerfectNumberConstants.RollingAccelerators];
private static readonly AcceleratorStream[] _streams = new AcceleratorStream[PerfectNumberConstants.RollingAccelerators];
private static SemaphoreSlim CreateLock() => new(PerfectNumberConstants.ThreadsByAccelerator);
@@ -17,26 +16,27 @@ public static class AcceleratorStreamPool
public static AcceleratorStream Rent(int acceleratorIndex)
{
var streamLock = _locks[acceleratorIndex];
var queue = _streams[acceleratorIndex];
streamLock.Wait();
var queue = _streams[acceleratorIndex];
if (queue != null)
{
return queue;
}
return queue.TryDequeue(out var stream)
? stream
: _accelerators[acceleratorIndex].CreateStream();
queue = _accelerators[acceleratorIndex].CreateStream();
_streams[acceleratorIndex] = queue;
return queue;
}
public static void Return(int acceleratorIndex, AcceleratorStream stream)
public static void Return(int acceleratorIndex)
{
var streamLock = _locks[acceleratorIndex];
var queue = _streams[acceleratorIndex];
queue.Enqueue(stream);
streamLock.Release();
}
public static void WarmUp(int acceleratorIndex)
{
_streams[acceleratorIndex] = new();
_locks[acceleratorIndex] = CreateLock();
}
}
PerfectNumbers.Core/Gpu/Accelerators/HeuristicGpuTables.cs
+16 -16
View File
@@ -45,25 +45,25 @@ internal sealed class HeuristicGpuTables
HeuristicGroupBDivisorsEnding9 = CopyUintSpanToDevice(accelerator, stream, DivisorGenerator.SmallPrimesLastNineWithoutLastThree);
HeuristicGroupBDivisorSquaresEnding9 = CopySpanToDevice(accelerator, stream, DivisorGenerator.SmallPrimesPow2LastNineWithoutLastThree);
var combinedEnding1 = HeuristicCombinedPrimeTester.CombinedDivisorsEnding1;
HeuristicCombinedDivisorsEnding1 = CopySpanToDevice(accelerator, stream, combinedEnding1);
var combinedSquaresEnding1 = HeuristicCombinedPrimeTester.CombinedDivisorsEnding1Squares;
HeuristicCombinedDivisorSquaresEnding1 = CopySpanToDevice(accelerator, stream, combinedSquaresEnding1);
// var combinedEnding1 = HeuristicCombinedPrimeTester.CombinedDivisorsEnding1;
// HeuristicCombinedDivisorsEnding1 = CopySpanToDevice(accelerator, stream, combinedEnding1);
// var combinedSquaresEnding1 = HeuristicCombinedPrimeTester.CombinedDivisorsEnding1Squares;
// HeuristicCombinedDivisorSquaresEnding1 = CopySpanToDevice(accelerator, stream, combinedSquaresEnding1);
var combinedEnding3 = HeuristicCombinedPrimeTester.CombinedDivisorsEnding3;
HeuristicCombinedDivisorsEnding3 = CopySpanToDevice(accelerator, stream, combinedEnding3);
var combinedSquaresEnding3 = HeuristicCombinedPrimeTester.CombinedDivisorsEnding3Squares;
HeuristicCombinedDivisorSquaresEnding3 = CopySpanToDevice(accelerator, stream, combinedSquaresEnding3);
// var combinedEnding3 = HeuristicCombinedPrimeTester.CombinedDivisorsEnding3;
// HeuristicCombinedDivisorsEnding3 = CopySpanToDevice(accelerator, stream, combinedEnding3);
// var combinedSquaresEnding3 = HeuristicCombinedPrimeTester.CombinedDivisorsEnding3Squares;
// HeuristicCombinedDivisorSquaresEnding3 = CopySpanToDevice(accelerator, stream, combinedSquaresEnding3);
var combinedEnding7 = HeuristicCombinedPrimeTester.CombinedDivisorsEnding7;
HeuristicCombinedDivisorsEnding7 = CopySpanToDevice(accelerator, stream, combinedEnding7);
var combinedSquaresEnding7 = HeuristicCombinedPrimeTester.CombinedDivisorsEnding7Squares;
HeuristicCombinedDivisorSquaresEnding7 = CopySpanToDevice(accelerator, stream, combinedSquaresEnding7);
// var combinedEnding7 = HeuristicCombinedPrimeTester.CombinedDivisorsEnding7;
// HeuristicCombinedDivisorsEnding7 = CopySpanToDevice(accelerator, stream, combinedEnding7);
// var combinedSquaresEnding7 = HeuristicCombinedPrimeTester.CombinedDivisorsEnding7Squares;
// HeuristicCombinedDivisorSquaresEnding7 = CopySpanToDevice(accelerator, stream, combinedSquaresEnding7);
var combinedEnding9 = HeuristicCombinedPrimeTester.CombinedDivisorsEnding9;
HeuristicCombinedDivisorsEnding9 = CopySpanToDevice(accelerator, stream, combinedEnding9);
var combinedSquaresEnding9 = HeuristicCombinedPrimeTester.CombinedDivisorsEnding9Squares;
HeuristicCombinedDivisorSquaresEnding9 = CopySpanToDevice(accelerator, stream, combinedSquaresEnding9);
// var combinedEnding9 = HeuristicCombinedPrimeTester.CombinedDivisorsEnding9;
// HeuristicCombinedDivisorsEnding9 = CopySpanToDevice(accelerator, stream, combinedEnding9);
// var combinedSquaresEnding9 = HeuristicCombinedPrimeTester.CombinedDivisorsEnding9Squares;
// HeuristicCombinedDivisorSquaresEnding9 = CopySpanToDevice(accelerator, stream, combinedSquaresEnding9);
}
internal readonly MemoryBuffer1D<ulong, Stride1D.Dense> HeuristicGroupADivisors;
PerfectNumbers.Core/Gpu/Accelerators/HeuristicPrimeTesterAccelerator.cs
+2 -2
View File
@@ -136,7 +136,7 @@ internal sealed class HeuristicPrimeTesterAccelerator
Accelerator = accelerator;
EnsureCapacity(minBufferCapacity);
HeuristicTrialDivisionKernel = KernelUtil.GetKernel(accelerator.LoadAutoGroupedStreamKernel<Index1D, ArrayView<int>, ulong, ulong, HeuristicGpuDivisorTableKind, HeuristicGpuDivisorTables>(PrimeTesterKernels.HeuristicTrialDivisionKernel));
HeuristicTrialDivisionKernel = KernelUtil.GetKernel(accelerator.LoadAutoGroupedStreamKernel<Index1D, ArrayView<int>, ulong, ulong, ArrayView1D<ulong, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>>(PrimeTesterKernels.HeuristicTrialDivisionKernel));
AcceleratorStream stream = AcceleratorStreamPool.Rent(acceleratorIndex);
@@ -146,7 +146,7 @@ internal sealed class HeuristicPrimeTesterAccelerator
_heuristicDivisorTables = _sharedTables.CreateHeuristicDivisorTables();
stream.Synchronize();
AcceleratorStreamPool.Return(acceleratorIndex, stream);
AcceleratorStreamPool.Return(acceleratorIndex);
}
public void EnsureCapacity(int minCapacity)
PerfectNumbers.Core/Gpu/GpuKernelPool.cs
+1 -1
View File
@@ -25,7 +25,7 @@ public class GpuKernelPool
internal static KernelContainer GetOrAddKernels(int acceleratorIndex, AcceleratorStream stream, KernelType kernelType)
{
var pool = _pool ??= [];
var pool = _pool ??= new KernelContainer[PerfectNumberConstants.RollingAccelerators];
if (pool[acceleratorIndex] is not {} kernels)
{
kernels = new();
PerfectNumbers.Core/Gpu/Kernels/PrimeTesterKernels.cs
+2 -12
View File
@@ -83,20 +83,10 @@ internal static class PrimeTesterKernels
ArrayView<int> resultFlag,
ulong n,
ulong maxDivisorSquare,
HeuristicGpuDivisorTableKind tableKind,
HeuristicGpuDivisorTables tables)
ArrayView1D<ulong, Stride1D.Dense> divisors,
ArrayView1D<ulong, Stride1D.Dense> divisorSquares)
{
byte nMod10 = (byte)(n % 10UL);
ArrayView1D<ulong, Stride1D.Dense> divisors = tables.SelectDivisors(tableKind, nMod10);
ArrayView1D<ulong, Stride1D.Dense> divisorSquares = tables.SelectDivisorSquares(tableKind, nMod10);
int divisorLength = (int)divisorSquares.Length;
int threadIndex = index;
if (threadIndex >= divisorLength)
{
return;
}
ulong divisorSquare = divisorSquares[threadIndex];
if (divisorSquare > maxDivisorSquare)
{
PerfectNumbers.Core/Gpu/Pow2MontgomeryGpuCalculator.cs
+3 -3
View File
@@ -32,7 +32,7 @@ public static class Pow2MontgomeryGpuCalculator
resultBuffer.View.CopyToCPU(stream, ref result, 1);
stream.Synchronize();
AcceleratorStreamPool.Return(acceleratorIndex, stream);
AcceleratorStreamPool.Return(acceleratorIndex);
// GpuPrimeWorkLimiter.Release();
PrimeOrderCalculatorAccelerator.Return(gpu);
@@ -60,7 +60,7 @@ public static class Pow2MontgomeryGpuCalculator
resultBuffer.View.CopyToCPU(stream, ref result, 1);
stream.Synchronize();
AcceleratorStreamPool.Return(acceleratorIndex, stream);
AcceleratorStreamPool.Return(acceleratorIndex);
// GpuPrimeWorkLimiter.Release();
PrimeOrderCalculatorAccelerator.Return(gpu);
@@ -86,7 +86,7 @@ public static class Pow2MontgomeryGpuCalculator
resultBuffer.View.CopyToCPU(stream, ref result, 1);
stream.Synchronize();
AcceleratorStreamPool.Return(acceleratorIndex, stream);
AcceleratorStreamPool.Return(acceleratorIndex);
return result;
}
PerfectNumbers.Core/Gpu/PrimeOrderGpuHeuristics.cs
+5 -5
View File
@@ -102,7 +102,7 @@ internal static partial class PrimeOrderGpuHeuristics
private static SmallPrimeDeviceCache GetSmallPrimeDeviceCache(int acceleratorIndex, AcceleratorStream stream)
{
var pool = _smallPrimeDeviceCache ??= [];
var pool = _smallPrimeDeviceCache ??= new SmallPrimeDeviceCache[PerfectNumberConstants.RollingAccelerators];
if (pool[acceleratorIndex] is {} cached)
{
return cached;
@@ -132,7 +132,7 @@ internal static partial class PrimeOrderGpuHeuristics
private static Kernel GetPartialFactorKernel(int acceleratorIndex)
{
var pool = _partialFactorKernel ??= [];
var pool = _partialFactorKernel ??= new Kernel[PerfectNumberConstants.RollingAccelerators];
if (pool[acceleratorIndex] is {} cached)
{
return cached;
@@ -487,7 +487,7 @@ internal static partial class PrimeOrderGpuHeuristics
private static Kernel GetOrderKernel(int acceleratorIndex)
{
var pool = _orderKernel ??= [];
var pool = _orderKernel ??= new Kernel[PerfectNumberConstants.RollingAccelerators];
if (pool[acceleratorIndex] is {} cached)
{
return cached;
@@ -637,7 +637,7 @@ internal static partial class PrimeOrderGpuHeuristics
remainderBuffer.View.CopyToCPU(stream, ref MemoryMarshal.GetReference(resultSpan), length);
stream.Synchronize();
AcceleratorStreamPool.Return(acceleratorIndex, stream);
AcceleratorStreamPool.Return(acceleratorIndex);
for (int i = 0; i < length; i++)
{
results[i] = (UInt128)resultSpan[i];
@@ -675,7 +675,7 @@ internal static partial class PrimeOrderGpuHeuristics
private static Kernel GetPow2ModKernel(int acceleratorIndex)
{
var pool = _pow2ModKernel ??= [];
var pool = _pow2ModKernel ??= new Kernel[PerfectNumberConstants.RollingAccelerators];
if (pool[acceleratorIndex] is {} cached)
{
return cached;