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

Merge remote-tracking branch 'github/codex/optimize-evenperfectbitscanner-for-static-arrays-khd9w8' into work

Browse Source 
# Conflicts:
#	PerfectNumbers.Core/Gpu/GpuContextPool.cs
#	PerfectNumbers.Core/Gpu/GpuKernelLease.cs
#	PerfectNumbers.Core/Gpu/GpuKernelPool.cs
#	PerfectNumbers.Core/Gpu/GpuStaticTableInitializer.cs
#	PerfectNumbers.Core/Gpu/Kernels/PrimeOrderSpecialMaxKernels.cs
#	PerfectNumbers.Core/PrimeOrderCalculator.Cpu.cs
#	PerfectNumbers.Core/PrimeOrderCalculator.Gpu.cs
#	PerfectNumbers.Core/PrimeTester.cs
This commit is contained in:
Marek Lemańczyk committed 2025-11-03 23:59:12 +01:00
1 parent 507daaddb9
commit 4c4d4243fb
23 files changed
+858 -633
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

PerfectNumbers.Core/HeuristicPrimeTester.cs
+62 -62
View File
@@ -213,7 +213,7 @@ public sealed class HeuristicPrimeTester
int processed = 0;
while (processed < interleaveBatchSize)
{
if (!TrySelectNextGroupBDivisor(endingOrder, indices, maxDivisorSquare, out ulong divisor))
if (!TrySelectNextGroupBDivisor(endingOrder, indices, maxDivisorSquare, out ulong divisor))
{
groupBHasMore = false;
break;
@@ -238,77 +238,77 @@ public sealed class HeuristicPrimeTester
}
private bool HeuristicTrialDivisionGpuDetectsDivisor(ulong n, ulong maxDivisorSquare, byte nMod10)
{
var limiter = GpuPrimeWorkLimiter.Acquire();
var gpu = PrimeTester.PrimeTesterGpuContextPool.Rent(1);
var accelerator = gpu.Accelerator;
var kernel = gpu.HeuristicTrialDivisionKernel;
var flagView1D = gpu.HeuristicFlag.View;
var flagView = flagView1D.AsContiguous();
private bool HeuristicTrialDivisionGpuDetectsDivisor(ulong n, ulong maxDivisorSquare, byte nMod10)
{
GpuPrimeWorkLimiter.Acquire();
var gpu = PrimeTester.PrimeTesterGpuContextPool.Rent(1);
var accelerator = gpu.Accelerator;
var kernel = PrimeTester.PrimeTesterGpuContextPool.PrimeTesterGpuContextLease.HeuristicTrialDivisionKernel;
var flagView1D = gpu.HeuristicFlag.View;
var flagView = flagView1D.AsContiguous();
bool compositeDetected = false;
int compositeFlag = 0;
bool compositeDetected = false;
int compositeFlag = 0;
int groupALength = HeuristicPrimeSieves.GroupADivisors.Length;
int groupALength = HeuristicPrimeSieves.GroupADivisors.Length;
flagView1D.CopyFromCPU(ref compositeFlag, 1);
kernel(
groupALength,
flagView,
n,
maxDivisorSquare,
HeuristicGpuDivisorTableKind.GroupA,
gpu.HeuristicGpuTables);
accelerator.Synchronize();
flagView1D.CopyToCPU(ref compositeFlag, 1);
compositeDetected = compositeFlag != 0;
flagView1D.CopyFromCPU(ref compositeFlag, 1);
kernel(
groupALength,
flagView,
n,
maxDivisorSquare,
HeuristicGpuDivisorTableKind.GroupA,
gpu.HeuristicGpuTables);
accelerator.Synchronize();
flagView1D.CopyToCPU(ref compositeFlag, 1);
compositeDetected = compositeFlag != 0;
if (!compositeDetected)
{
ReadOnlySpan<byte> endingOrder = GetGroupBEndingOrder(nMod10);
for (int i = 0; i < endingOrder.Length && !compositeDetected; i++)
{
byte ending = endingOrder[i];
HeuristicGpuDivisorTableKind tableKind = ending switch
{
1 => HeuristicGpuDivisorTableKind.GroupBEnding1,
7 => HeuristicGpuDivisorTableKind.GroupBEnding7,
9 => HeuristicGpuDivisorTableKind.GroupBEnding9,
_ => HeuristicGpuDivisorTableKind.GroupA,
};
if (!compositeDetected)
{
ReadOnlySpan<byte> endingOrder = GetGroupBEndingOrder(nMod10);
for (int i = 0; i < endingOrder.Length && !compositeDetected; i++)
{
byte ending = endingOrder[i];
HeuristicGpuDivisorTableKind tableKind = ending switch
{
1 => HeuristicGpuDivisorTableKind.GroupBEnding1,
7 => HeuristicGpuDivisorTableKind.GroupBEnding7,
9 => HeuristicGpuDivisorTableKind.GroupBEnding9,
_ => HeuristicGpuDivisorTableKind.GroupA,
};
if (tableKind == HeuristicGpuDivisorTableKind.GroupA)
{
continue;
}
if (tableKind == HeuristicGpuDivisorTableKind.GroupA)
{
continue;
}
int divisorLength = GetGroupBDivisors(ending).Length;
int divisorLength = GetGroupBDivisors(ending).Length;
compositeFlag = 0;
flagView1D.CopyFromCPU(ref compositeFlag, 1);
kernel(
divisorLength,
flagView,
n,
maxDivisorSquare,
tableKind,
gpu.HeuristicGpuTables);
accelerator.Synchronize();
flagView1D.CopyToCPU(ref compositeFlag, 1);
compositeDetected = compositeFlag != 0;
}
}
compositeFlag = 0;
flagView1D.CopyFromCPU(ref compositeFlag, 1);
kernel(
divisorLength,
flagView,
n,
maxDivisorSquare,
tableKind,
gpu.HeuristicGpuTables);
accelerator.Synchronize();
flagView1D.CopyToCPU(ref compositeFlag, 1);
compositeDetected = compositeFlag != 0;
}
}
gpu.Dispose();
limiter.Dispose();
return compositeDetected;
}
gpu.Dispose();
GpuPrimeWorkLimiter.Release();
return compositeDetected;
}
[MethodImpl(MethodImplOptions.AggressiveInlining)]
private static bool EvaluateWithOpenNumericFallback(ulong n)
[MethodImpl(MethodImplOptions.AggressiveInlining)]
private static bool EvaluateWithOpenNumericFallback(ulong n)
{
return Prime.Numbers.IsPrime(n);
}
@@ -327,7 +327,7 @@ public sealed class HeuristicPrimeTester
};
[MethodImpl(MethodImplOptions.AggressiveInlining)]
private static bool TrySelectNextGroupBDivisor(ReadOnlySpan<byte> endingOrder, Span<int> indices, ulong maxDivisorSquare, out ulong divisor)
private static bool TrySelectNextGroupBDivisor(ReadOnlySpan<byte> endingOrder, Span<int> indices, ulong maxDivisorSquare, out ulong divisor)
{
ulong bestCandidate = ulong.MaxValue;
int bestEndingIndex = -1;