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

Optimizations. Unnecessary variables removed.

Browse Source
This commit is contained in:
Marek Lemańczyk committed 2025-09-24 01:41:54 +02:00
1 parent 15bc2686cc
commit 01bb731220
1 file changed
+43 -52
Download Patch File Download Diff File Expand all files Collapse all files
PerfectNumbers.Core/Gpu/MersenneNumberDivisorByDivisorGpuTester.cs
+43 -52
View File
@@ -98,6 +98,7 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester
private bool CheckDivisors(ulong prime, ulong allowedMax, out ulong lastProcessed, out bool coveredRange)
{
// TODO: We should be able to calculate allowedMax outside in the parent method, unless it depends from some changing value.
allowedMax = Math.Min(allowedMax, _divisorLimit);
lastProcessed = 0UL;
@@ -107,53 +108,58 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester
return false;
}
// TODO: Rent these in the parent method before the loops - batch size doesn't change
var gpuLease = GpuContextPool.RentPreferred(preferCpu: false);
var accelerator = gpuLease.Accelerator;
var kernel = GetKernel(accelerator);
int batchCapacity = _gpuBatchSize;
// TODO: Rent buffers in the parent method before the loops - batch size doesn't change
var divisorsBuffer = accelerator.Allocate1D<MontgomeryDivisorData>(batchCapacity);
var hitsBuffer = accelerator.Allocate1D<byte>(batchCapacity);
ulong[] divisors = ArrayPool<ulong>.Shared.Rent(batchCapacity);
byte[] hits = ArrayPool<byte>.Shared.Rent(batchCapacity);
MontgomeryDivisorData[] divisorData = ArrayPool<MontgomeryDivisorData>.Shared.Rent(batchCapacity);
bool composite = false;
bool processedAll = false;
ulong divisor = 3UL;
ulong cycle, divisor = 3UL, next;
int batchSize, i;
bool reachedEndInBatch, useCycles = _useDivisorCycles;
Span<MontgomeryDivisorData> divisorDataSpan;
Span<ulong> divisorSpan;
Span<byte> hitsSpan;
MersenneDivisorCycles divisorCycles = MersenneDivisorCycles.Shared;
ArrayView1D<MontgomeryDivisorData, Stride1D.Dense> divisorView;
ArrayView1D<byte, Stride1D.Dense> hitsView;
try
{
bool useCycles = _useDivisorCycles;
MersenneDivisorCycles? divisorCycles = useCycles ? MersenneDivisorCycles.Shared : null;
while (divisor <= allowedMax)
{
int batchSize = 0;
bool reachedEndInBatch = false;
batchSize = 0;
reachedEndInBatch = false;
while (batchSize < batchCapacity && divisor <= allowedMax)
{
ulong current = divisor;
ulong next = current + 2UL;
bool include = true;
if (useCycles)
{
ulong cycle = divisorCycles!.GetCycle(current);
cycle = divisorCycles!.GetCycle(divisor);
if (cycle == 0UL || prime % cycle != 0UL)
{
include = false;
lastProcessed = current;
ReportStatus(lastProcessed);
ReportStatus(divisor);
}
}
if (include)
else
{
divisors[batchSize++] = current;
divisors[batchSize++] = divisor;
}
}
else
{
divisors[batchSize++] = divisor;
}
if (next <= current)
next = divisor + 2UL;
if (next <= divisor)
{
reachedEndInBatch = true;
break;
@@ -174,14 +180,14 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester
break;
}
Span<MontgomeryDivisorData> divisorDataSpan = divisorData.AsSpan(0, batchSize);
var divisorView = divisorsBuffer.View.SubView(0, batchSize);
var hitsView = hitsBuffer.View.SubView(0, batchSize);
divisorDataSpan = divisorData.AsSpan(0, batchSize);
divisorView = divisorsBuffer.View.SubView(0, batchSize);
hitsView = hitsBuffer.View.SubView(0, batchSize);
Span<ulong> divisorSpan = divisors.AsSpan(0, batchSize);
Span<byte> hitsSpan = hits.AsSpan(0, batchSize);
divisorSpan = divisors.AsSpan(0, batchSize);
hitsSpan = hits.AsSpan(0, batchSize);
for (int i = 0; i < batchSize; i++)
for (i = 0; i < batchSize; i++)
{
divisorDataSpan[i] = CreateMontgomeryDivisorData(divisorSpan[i]);
}
@@ -189,12 +195,10 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester
divisorView.CopyFromCPU(ref MemoryMarshal.GetReference(divisorDataSpan), batchSize);
kernel(batchSize, prime, divisorView, hitsView);
accelerator.Synchronize();
hitsView.CopyToCPU(ref MemoryMarshal.GetReference(hitsSpan), batchSize);
for (int i = 0; i < batchSize; i++)
for (i = 0; i < batchSize; i++)
{
lastProcessed = divisorSpan[i];
if (hitsSpan[i] != 0)
{
composite = true;
@@ -202,7 +206,7 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester
}
}
ReportStatus(lastProcessed);
ReportStatus(divisorSpan[i < batchSize ? i : i - 1]);
if (composite)
{
break;
@@ -216,16 +220,13 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester
}
coveredRange = composite || processedAll || divisor > allowedMax;
}
finally
{
ArrayPool<byte>.Shared.Return(hits, clearArray: true);
ArrayPool<ulong>.Shared.Return(divisors, clearArray: true);
ArrayPool<MontgomeryDivisorData>.Shared.Return(divisorData, clearArray: true);
hitsBuffer.Dispose();
divisorsBuffer.Dispose();
gpuLease.Dispose();
}
return composite;
}
@@ -317,7 +318,6 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester
}
int offset = 0;
int consoleStatus = 0;
MersenneNumberDivisorByDivisorGpuTester owner = _owner;
int gpuBatchSize = owner._gpuBatchSize;
bool useCycles = owner._useDivisorCycles;
@@ -335,23 +335,17 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester
MersenneDivisorCycles? divisorCycles = useCycles ? MersenneDivisorCycles.Shared : null;
ulong cycle = useCycles ? divisorCycles!.GetCycle(divisor) : 0UL, primeValue;
MontgomeryDivisorData divisorData = CreateMontgomeryDivisorData(divisor);
Span<ulong> primeSpan = Span<ulong>.Empty;
Span<int> indexSpan = Span<int>.Empty;
Span<byte> hitSpan = Span<byte>.Empty;
Span<ulong> primeSpan = [];
ReadOnlySpan<ulong> primesSlice;
Span<int> indexSpan = [];
Span<byte> hitSpan = [];
Span<byte> hitsSlice;
int batchSize, currentSpanSize = -1, gpuCount, i, primesLength = primes.Length;
while (offset < primesLength)
{
batchSize = Math.Min(gpuBatchSize, primesLength - offset);
ReadOnlySpan<ulong> primesSlice = primes.Slice(offset, batchSize);
if (++consoleStatus == PerfectNumberConstants.ConsoleInterval)
{
Console.WriteLine($"...processed by-divisor candidate = {primesSlice[0]}");
consoleStatus = 0;
}
primesSlice = primes.Slice(offset, batchSize);
hitsSlice = hits.Slice(offset, batchSize);
if (useCycles)
@@ -364,8 +358,6 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester
currentSpanSize = batchSize;
}
ref ulong primesSpanRef = ref MemoryMarshal.GetReference(primeSpan);
ref byte hitSpanRef = ref MemoryMarshal.GetReference(hitSpan);
gpuCount = 0;
for (i = 0; i < batchSize; i++)
@@ -387,11 +379,11 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester
primesView = primesBufferView.SubView(0, gpuCount);
hitsView = hitsBufferView.SubView(0, gpuCount);
primesView.CopyFromCPU(ref primesSpanRef, gpuCount);
primesView.CopyFromCPU(ref MemoryMarshal.GetReference(primeSpan), gpuCount);
kernel(gpuCount, divisorData, primesView, hitsView);
accelerator.Synchronize();
hitsView.CopyToCPU(ref hitSpanRef, gpuCount);
hitsView.CopyToCPU(ref MemoryMarshal.GetReference(hitSpan), gpuCount);
for (i = 0; i < gpuCount; i++)
{
hitsSlice[indexSpan[i]] = hitSpan[i];
@@ -403,8 +395,7 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester
primesView = primesBufferView.SubView(0, batchSize);
hitsView = hitsBufferView.SubView(0, batchSize);
ref ulong primesSliceRef = ref MemoryMarshal.GetReference(primesSlice);
primesView.CopyFromCPU(ref primesSliceRef, batchSize);
primesView.CopyFromCPU(ref MemoryMarshal.GetReference(primesSlice), batchSize);
kernel(batchSize, divisorData, primesView, hitsView);
accelerator.Synchronize();