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

Remove temporary by-divisor review documents

Browse Source
This commit is contained in:
Marek Lemańczyk committed 2025-10-28 02:01:28 +01:00
1 parent bd0727131f
commit 6dd5cea79f
27 files changed
+498 -600
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

PerfectNumbers.Core/MersenneNumberTester.cs
+41 -46
View File
@@ -152,56 +152,51 @@ public sealed class MersenneNumberTester(
var stream = gpuLease.Stream;
var orderKernel = gpuLease.OrderKernel!;
try
// Guard long-running kernels by chunking the warm-up across batches.
// Reuse the same ScanBatchSize knob used for scanning.
int batchSize = GpuConstants.ScanBatchSize;
ulong divMul = (ulong)((((UInt128)1 << 64) - 1UL) / exponent) + 1UL;
int offset = 0;
while (offset < idx)
{
// Guard long-running kernels by chunking the warm-up across batches.
// Reuse the same ScanBatchSize knob used for scanning.
int batchSize = GpuConstants.ScanBatchSize;
ulong divMul = (ulong)((((UInt128)1 << 64) - 1UL) / exponent) + 1UL;
int offset = 0;
while (offset < idx)
int count = Math.Min(batchSize, idx - offset);
var qBuffer = accelerator.Allocate1D<PerfectNumbers.Core.Gpu.GpuUInt128>(count);
var orderBuffer = accelerator.Allocate1D<ulong>(count);
// Convert to device-friendly GpuUInt128 on the fly
var tmp = System.Buffers.ArrayPool<PerfectNumbers.Core.Gpu.GpuUInt128>.Shared.Rent(count);
for (int i = 0; i < count; i++)
{
int count = Math.Min(batchSize, idx - offset);
var qBuffer = accelerator.Allocate1D<PerfectNumbers.Core.Gpu.GpuUInt128>(count);
var orderBuffer = accelerator.Allocate1D<ulong>(count);
// Convert to device-friendly GpuUInt128 on the fly
var tmp = System.Buffers.ArrayPool<PerfectNumbers.Core.Gpu.GpuUInt128>.Shared.Rent(count);
for (int i = 0; i < count; i++)
{
tmp[i] = (PerfectNumbers.Core.Gpu.GpuUInt128)qs[offset + i];
}
// Copy only the populated elements to the device buffer to avoid overrun on pooled arrays.
qBuffer.View.CopyFromCPU(ref tmp[0], count);
System.Buffers.ArrayPool<PerfectNumbers.Core.Gpu.GpuUInt128>.Shared.Return(tmp);
orderKernel(stream, count, exponent, divMul, qBuffer.View, orderBuffer.View);
stream.Synchronize();
ulong[] orders = orderBuffer.GetAsArray1D();
for (int i = 0; i < count; i++)
{
ulong order = orders[i];
if (order == 0UL)
{
order = qs[offset + i].CalculateOrder();
}
if (useOrderCache)
{
OrderCache[qs[offset + i]] = order;
}
}
offset += count;
orderBuffer.Dispose();
qBuffer.Dispose();
tmp[i] = (PerfectNumbers.Core.Gpu.GpuUInt128)qs[offset + i];
}
// Copy only the populated elements to the device buffer to avoid overrun on pooled arrays.
qBuffer.View.CopyFromCPU(ref tmp[0], count);
System.Buffers.ArrayPool<PerfectNumbers.Core.Gpu.GpuUInt128>.Shared.Return(tmp);
orderKernel(stream, count, exponent, divMul, qBuffer.View, orderBuffer.View);
stream.Synchronize();
ulong[] orders = orderBuffer.GetAsArray1D();
for (int i = 0; i < count; i++)
{
ulong order = orders[i];
if (order == 0UL)
{
order = qs[offset + i].CalculateOrder();
}
if (useOrderCache)
{
OrderCache[qs[offset + i]] = order;
}
}
offset += count;
orderBuffer.Dispose();
qBuffer.Dispose();
}
finally
{
execution.Dispose();
gpuLease.Dispose();
ArrayPool<UInt128>.Shared.Return(qs);
}
execution.Dispose();
gpuLease.Dispose();
ArrayPool<UInt128>.Shared.Return(qs);
}
public bool IsMersennePrime(ulong exponent)