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

Various try-out changes

Browse Source
This commit is contained in:
mlemanczyk committed 2025-10-17 06:53:54 +02:00
1 parent 301e715179
commit fddec17b3f
6 files changed
+813 -821
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

PerfectNumbers.Core/PrimeOrderCalculator.Gpu.cs
+71 -85
View File
@@ -9,100 +9,86 @@ namespace PerfectNumbers.Core;
internal static partial class PrimeOrderCalculator
{
private static bool IsGpuPow2Allowed => s_pow2ModeInitialized && s_allowGpuPow2;
private static bool IsGpuPow2Allowed => s_pow2ModeInitialized && s_allowGpuPow2;
private const int GpuSmallPrimeFactorSlots = 64;
private const int GpuSmallPrimeFactorSlots = 64;
private static bool TryPopulateSmallPrimeFactorsGpu(ulong value, uint limit, Dictionary<ulong, int> counts, out ulong remaining)
{
ulong[] primeBufferArray = ThreadStaticPools.UlongPool.Rent(GpuSmallPrimeFactorSlots);
int[] exponentBufferArray = ThreadStaticPools.IntPool.Rent(GpuSmallPrimeFactorSlots);
Span<ulong> primeBuffer = primeBufferArray.AsSpan(0, GpuSmallPrimeFactorSlots);
Span<int> exponentBuffer = exponentBufferArray.AsSpan(0, GpuSmallPrimeFactorSlots);
primeBuffer.Clear();
exponentBuffer.Clear();
remaining = value;
private static bool TryPopulateSmallPrimeFactorsGpu(ulong value, uint limit, Span<ulong> primeBuffer, Span<int> exponentBuffer, out int factorCount, out ulong remaining)
{
// primeBufferArray = ThreadStaticPools.UlongPool.Rent(GpuSmallPrimeFactorSlots);
// exponentBufferArray = ThreadStaticPools.IntPool.Rent(GpuSmallPrimeFactorSlots);
// Span<ulong> primeBuffer = primeBufferArray.AsSpan(0, GpuSmallPrimeFactorSlots);
// Span<int> exponentBuffer = exponentBufferArray.AsSpan(0, GpuSmallPrimeFactorSlots);
// primeBuffer.Clear();
// exponentBuffer.Clear();
remaining = value;
try
{
var lease = GpuKernelPool.GetKernel(useGpuOrder: true);
try
{
var execution = lease.EnterExecutionScope();
try
{
Accelerator accelerator = lease.Accelerator;
AcceleratorStream stream = lease.Stream;
// try
// {
var lease = GpuKernelPool.GetKernel(useGpuOrder: true);
var execution = lease.EnterExecutionScope();
try
{
Accelerator accelerator = lease.Accelerator;
AcceleratorStream stream = lease.Stream;
SmallPrimeFactorTables tables = GpuKernelPool.EnsureSmallPrimeFactorTables(accelerator);
SmallPrimeFactorScratch scratch = GpuKernelPool.EnsureSmallPrimeFactorScratch(accelerator, GpuSmallPrimeFactorSlots);
scratch.Clear();
SmallPrimeFactorTables tables = GpuKernelPool.EnsureSmallPrimeFactorTables(accelerator);
SmallPrimeFactorScratch scratch = GpuKernelPool.EnsureSmallPrimeFactorScratch(accelerator, GpuSmallPrimeFactorSlots);
// scratch.Clear();
var kernel = lease.SmallPrimeFactorKernel;
kernel(
stream,
1,
value,
limit,
tables.PrimesView,
tables.SquaresView,
tables.Count,
scratch.PrimeSlots.View,
scratch.ExponentSlots.View,
scratch.CountSlot.View,
scratch.RemainingSlot.View);
var kernel = lease.SmallPrimeFactorKernel;
kernel(
stream,
1,
value,
limit,
tables.PrimesView,
tables.SquaresView,
tables.Count,
scratch.PrimeSlots.View,
scratch.ExponentSlots.View,
scratch.CountSlot.View,
scratch.RemainingSlot.View);
stream.Synchronize();
stream.Synchronize();
int factorCount = 0;
scratch.CountSlot.View.CopyToCPU(ref factorCount, 1);
factorCount = Math.Min(factorCount, GpuSmallPrimeFactorSlots);
factorCount = 0;
scratch.CountSlot.View.CopyToCPU(ref factorCount, 1);
factorCount = Math.Min(factorCount, GpuSmallPrimeFactorSlots);
if (factorCount > 0)
{
scratch.PrimeSlots.View.CopyToCPU(ref MemoryMarshal.GetReference(primeBuffer), factorCount);
scratch.ExponentSlots.View.CopyToCPU(ref MemoryMarshal.GetReference(exponentBuffer), factorCount);
}
scratch.RemainingSlot.View.CopyToCPU(ref remaining, 1);
if (factorCount > 0)
{
scratch.PrimeSlots.View.CopyToCPU(ref MemoryMarshal.GetReference(primeBuffer), factorCount);
scratch.ExponentSlots.View.CopyToCPU(ref MemoryMarshal.GetReference(exponentBuffer), factorCount);
}
for (int i = 0; i < factorCount; i++)
{
ulong primeValue = primeBuffer[i];
int exponent = exponentBuffer[i];
if (primeValue == 0UL || exponent == 0)
{
continue;
}
scratch.RemainingSlot.View.CopyToCPU(ref remaining, 1);
counts[primeValue] = exponent;
}
// for (int i = 0; i < factorCount; i++)
// {
// ulong primeValue = primeBuffer[i];
// int exponent = exponentBuffer[i];
// // This will never happen in production code
// // if (primeValue == 0UL || exponent == 0)
// // {
// // continue;
// // }
return true;
}
finally
{
execution.Dispose();
}
}
finally
{
lease.Dispose();
}
}
catch (CLException)
{
remaining = value;
return false;
}
catch (Exception ex) when (ex is AcceleratorException or InternalCompilerException or NotSupportedException or InvalidOperationException or AggregateException)
{
remaining = value;
return false;
}
finally
{
ThreadStaticPools.UlongPool.Return(primeBufferArray);
ThreadStaticPools.IntPool.Return(exponentBufferArray);
}
}
// counts[primeValue] = exponent;
// }
return true;
}
finally
{
execution.Dispose();
lease.Dispose();
}
// }
// finally
// {
// ThreadStaticPools.UlongPool.Return(primeBufferArray);
// ThreadStaticPools.IntPool.Return(exponentBufferArray);
// }
}
}