GpuScratchBuffer / -Pool naively integrated with PriemOrderCalculatorAccelerator (formerly Pow2MontgomeryAccelerator). FactorEntry replaced with 2 arrays - unused field removed. Small cleanup.
This commit is contained in:
mlemanczyk committed
1 parent
43863d03fb
commit
c53aa517f7
14 files changed
+295
-420
No files matched your search
@@ -1,10 +1,6 @@
|
||||
using System.Diagnostics;
|
||||
using System.Runtime.CompilerServices;
|
||||
using System.Runtime.InteropServices;
|
||||
using ILGPU;
|
||||
using ILGPU.Runtime;
|
||||
using ILGPU.Util;
|
||||
using Open.Collections;
|
||||
using PerfectNumbers.Core.Gpu;
|
||||
using PerfectNumbers.Core.Gpu.Accelerators;
|
||||
|
||||
@@ -16,7 +12,7 @@ internal static partial class PrimeOrderCalculator
|
||||
|
||||
private const int GpuSmallPrimeFactorSlots = 64;
|
||||
|
||||
private static ulong CalculateByFactorizationGpu(Pow2MontgomeryAccelerator gpu, ulong prime, in MontgomeryDivisorData divisorData)
|
||||
private static ulong CalculateByFactorizationGpu(PrimeOrderCalculatorAccelerator gpu, ulong prime, in MontgomeryDivisorData divisorData)
|
||||
{
|
||||
ulong phi = prime - 1UL;
|
||||
Dictionary<ulong, int> counts = gpu.Pow2ModEntriesToTestOnHost;
|
||||
@@ -59,7 +55,7 @@ internal static partial class PrimeOrderCalculator
|
||||
return order;
|
||||
}
|
||||
|
||||
private static bool TryPopulateSmallPrimeFactorsGpu(ulong value, uint limit, Dictionary<ulong, int> counts, out int factorCount, out ulong remaining)
|
||||
private static bool TryPopulateSmallPrimeFactorsGpu(PrimeOrderCalculatorAccelerator gpu, ulong value, uint limit, Dictionary<ulong, int> counts, out int factorCount, out ulong remaining)
|
||||
{
|
||||
var primeBufferArray = ThreadStaticPools.UlongPool.Rent(GpuSmallPrimeFactorSlots);
|
||||
var exponentBufferArray = ThreadStaticPools.IntPool.Rent(GpuSmallPrimeFactorSlots);
|
||||
@@ -67,8 +63,8 @@ internal static partial class PrimeOrderCalculator
|
||||
Span<int> exponentBuffer = exponentBufferArray.AsSpan(0, GpuSmallPrimeFactorSlots);
|
||||
remaining = value;
|
||||
|
||||
Accelerator accelerator = AcceleratorPool.Shared.Rent();
|
||||
GpuScratchBuffer scratch = GpuScratchBufferPool.Rent(accelerator, GpuSmallPrimeFactorSlots, 0);
|
||||
gpu.EnsureSmallPrimeFactorSlotsCapacity(GpuSmallPrimeFactorSlots);
|
||||
var accelerator = gpu.Accelerator;
|
||||
|
||||
GpuPrimeWorkLimiter.Acquire();
|
||||
var stream = AcceleratorStreamPool.Rent(accelerator);
|
||||
@@ -78,10 +74,10 @@ internal static partial class PrimeOrderCalculator
|
||||
var kernel = kernels.SmallPrimeFactor!;
|
||||
var kernelLauncher = kernel.CreateLauncherDelegate<Action<AcceleratorStream, Index1D, ulong, uint, ArrayView1D<uint, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>, int, ArrayView1D<ulong, Stride1D.Dense>, ArrayView1D<int, Stride1D.Dense>, ArrayView1D<int, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>>>();
|
||||
|
||||
ArrayView1D<int, Stride1D.Dense> smallPrimeFactorCountSlotView = scratch.SmallPrimeFactorCountSlot.View;
|
||||
ArrayView1D<ulong, Stride1D.Dense> smallPrimeFactorRemainingSlotView = scratch.SmallPrimeFactorRemainingSlot.View;
|
||||
ArrayView1D<ulong, Stride1D.Dense> smallPrimeFactorPrimeSlotsView = scratch.SmallPrimeFactorPrimeSlots.View;
|
||||
ArrayView1D<int, Stride1D.Dense> smallPrimeFactorExponentSlotsView = scratch.SmallPrimeFactorExponentSlots.View;
|
||||
ArrayView1D<int, Stride1D.Dense> smallPrimeFactorCountSlotView = gpu.SmallPrimeFactorCountSlot.View;
|
||||
ArrayView1D<ulong, Stride1D.Dense> smallPrimeFactorRemainingSlotView = gpu.SmallPrimeFactorRemainingSlot.View;
|
||||
ArrayView1D<ulong, Stride1D.Dense> smallPrimeFactorPrimeSlotsView = gpu.SmallPrimeFactorPrimeSlots.View;
|
||||
ArrayView1D<int, Stride1D.Dense> smallPrimeFactorExponentSlotsView = gpu.SmallPrimeFactorExponentSlots.View;
|
||||
ArrayView1D<uint, Stride1D.Dense> smallPrimeFactorsPrimesView = kernels.SmallPrimeFactorsPrimes!.View;
|
||||
|
||||
kernelLauncher(
|
||||
@@ -117,7 +113,6 @@ internal static partial class PrimeOrderCalculator
|
||||
stream.Synchronize();
|
||||
|
||||
AcceleratorStreamPool.Return(stream);
|
||||
GpuScratchBufferPool.Return(scratch);
|
||||
GpuPrimeWorkLimiter.Release();
|
||||
|
||||
for (int i = 0; i < factorCount; i++)
|
||||
@@ -131,38 +126,26 @@ internal static partial class PrimeOrderCalculator
|
||||
}
|
||||
|
||||
private static bool EvaluateSpecialMaxCandidatesGpu(
|
||||
Pow2MontgomeryAccelerator gpu,
|
||||
Span<ulong> buffer,
|
||||
ReadOnlySpan<FactorEntry> factors,
|
||||
PrimeOrderCalculatorAccelerator gpu,
|
||||
ReadOnlySpan<ulong> factors,
|
||||
ulong phi,
|
||||
ulong prime,
|
||||
in MontgomeryDivisorData divisorData)
|
||||
{
|
||||
if (factors.Length == 0)
|
||||
{
|
||||
return true;
|
||||
}
|
||||
|
||||
int factorCount = factors.Length;
|
||||
Span<ulong> factorSpan = buffer[..factorCount];
|
||||
for (int i = 0; i < factorCount; i++)
|
||||
{
|
||||
factorSpan[i] = factors[i].Value;
|
||||
}
|
||||
gpu.EnsureSpecialMaxFactorsCapacity(factorCount);
|
||||
|
||||
Accelerator accelerator = gpu.Accelerator;
|
||||
GpuPrimeWorkLimiter.Acquire();
|
||||
Accelerator accelerator = gpu.Accelerator; //AcceleratorPool.Shared.Rent();
|
||||
var stream = AcceleratorStreamPool.Rent(accelerator);// accelerator.CreateStream();
|
||||
var stream = AcceleratorStreamPool.Rent(accelerator);
|
||||
var kernels = GpuKernelPool.GetOrAddKernels(accelerator, stream, KernelType.EvaluateSpecialMaxCandidatesKernel);
|
||||
|
||||
GpuScratchBuffer scratch = GpuScratchBufferPool.Rent(accelerator, 0, factorCount);
|
||||
|
||||
ArrayView1D<ulong, Stride1D.Dense> specialMaxFactorsView = scratch.SpecialMaxFactors.View;
|
||||
specialMaxFactorsView.SubView(0, factorCount).CopyFromCPU(stream, ref MemoryMarshal.GetReference(factorSpan), factorCount);
|
||||
ArrayView1D<ulong, Stride1D.Dense> specialMaxFactorsView = gpu.SpecialMaxFactors.View;
|
||||
specialMaxFactorsView.SubView(0, factorCount).CopyFromCPU(stream, ref MemoryMarshal.GetReference(factors), factorCount);
|
||||
|
||||
var kernel = kernels.SpecialMax!;
|
||||
|
||||
ArrayView1D<ulong, Stride1D.Dense> specialMaxResultView = scratch.SpecialMaxResult.View;
|
||||
ArrayView1D<ulong, Stride1D.Dense> specialMaxResultView = gpu.SpecialMaxResult.View;
|
||||
|
||||
kernel(
|
||||
stream,
|
||||
@@ -171,28 +154,20 @@ internal static partial class PrimeOrderCalculator
|
||||
specialMaxFactorsView,
|
||||
factorCount,
|
||||
divisorData,
|
||||
scratch.SpecialMaxCandidates.View,
|
||||
gpu.SpecialMaxCandidates.View,
|
||||
specialMaxResultView);
|
||||
|
||||
// Span<ulong> result = stackalloc ulong[1];
|
||||
ulong result = 0UL;
|
||||
// specialMaxResultView.CopyToCPU(stream, ref result, 1);
|
||||
specialMaxResultView.CopyToCPU(stream, ref result, 1);
|
||||
stream.Synchronize();
|
||||
|
||||
AcceleratorStreamPool.Return(stream);
|
||||
// stream.Dispose();
|
||||
|
||||
// stream.Synchronize();
|
||||
// return result != 0;
|
||||
GpuScratchBufferPool.Return(scratch);
|
||||
GpuPrimeWorkLimiter.Release();
|
||||
return result != 0;
|
||||
}
|
||||
|
||||
private static bool TryPollardRhoGpu(Pow2MontgomeryAccelerator gpu, ulong n, out ulong factor)
|
||||
private static bool TryPollardRhoGpu(PrimeOrderCalculatorAccelerator gpu, ulong n, out ulong factor)
|
||||
{
|
||||
// var stream = gpu.Stream!;
|
||||
Span<ulong> randomStateSpan = stackalloc ulong[1];
|
||||
randomStateSpan[0] = ThreadStaticDeterministicRandomGpu.Exclusive.State;
|
||||
|
||||
|
||||
Reference in new issue
Block a user