GPU memory allocations protected with lock to prevent shared memory access issues. GPU usage limited for the best performance.
This commit is contained in:
mlemanczyk committed
1 parent
02d16bf05c
commit
5cac51479e
12 files changed
+236
-148
No files matched your search
@@ -117,7 +117,11 @@ public sealed class PrimeOrderCalculatorAccelerator
|
||||
|
||||
public readonly Kernel CheckFactorsKernel;
|
||||
public readonly Kernel ConvertToStandardKernel;
|
||||
|
||||
public readonly Action<AcceleratorStream, Index1D, ArrayView1D<ulong, Stride1D.Dense>, MontgomeryDivisorData, ArrayView1D<ulong, Stride1D.Dense>> ConvertToStandardKernelLauncher;
|
||||
|
||||
public readonly Kernel HeuristicCombinedTrialDivisionKernel;
|
||||
public readonly Action<AcceleratorStream, Index1D, byte, ArrayView<int>, ulong, ulong, HeuristicCombinedGpuViews> HeuristicCombinedTrialDivisionKernelLauncher;
|
||||
public readonly Kernel KeepMontgomeryKernel;
|
||||
public readonly Kernel PollardRhoKernel;
|
||||
public readonly Kernel Pow2ModWideKernel;
|
||||
@@ -133,29 +137,41 @@ public sealed class PrimeOrderCalculatorAccelerator
|
||||
MemoryBuffer1D<KeyValuePair<ulong, int>, Stride1D.Dense> pow2ModEntriesToTestOnDevice = Pow2ModEntriesToTestOnDevice;
|
||||
if (pow2ModEntriesToTestOnDevice.Length < factorsCount)
|
||||
{
|
||||
pow2ModEntriesToTestOnDevice.Dispose();
|
||||
Pow2ModEntriesToTestOnDevice = accelerator.Allocate1D<KeyValuePair<ulong, int>>(factorsCount);
|
||||
lock (accelerator)
|
||||
{
|
||||
pow2ModEntriesToTestOnDevice.Dispose();
|
||||
Pow2ModEntriesToTestOnDevice = accelerator.Allocate1D<KeyValuePair<ulong, int>>(factorsCount);
|
||||
}
|
||||
}
|
||||
|
||||
MemoryBuffer1D<ulong, Stride1D.Dense> input = Input;
|
||||
if (input.Length < primeTesterCapacity)
|
||||
{
|
||||
input.Dispose();
|
||||
Input = accelerator.Allocate1D<ulong>(primeTesterCapacity);
|
||||
lock (accelerator)
|
||||
{
|
||||
input.Dispose();
|
||||
Input = accelerator.Allocate1D<ulong>(primeTesterCapacity);
|
||||
}
|
||||
}
|
||||
|
||||
MemoryBuffer1D<byte, Stride1D.Dense> outputByte = OutputByte;
|
||||
if (outputByte.Length < primeTesterCapacity)
|
||||
{
|
||||
outputByte.Dispose();
|
||||
OutputByte = accelerator.Allocate1D<byte>(primeTesterCapacity);
|
||||
lock (accelerator)
|
||||
{
|
||||
outputByte.Dispose();
|
||||
OutputByte = accelerator.Allocate1D<byte>(primeTesterCapacity);
|
||||
}
|
||||
}
|
||||
|
||||
MemoryBuffer1D<int, Stride1D.Dense> outputInt = OutputInt;
|
||||
if (outputInt.Length < primeTesterCapacity)
|
||||
{
|
||||
outputInt.Dispose();
|
||||
OutputInt = accelerator.Allocate1D<int>(primeTesterCapacity);
|
||||
lock (accelerator)
|
||||
{
|
||||
outputInt.Dispose();
|
||||
OutputInt = accelerator.Allocate1D<int>(primeTesterCapacity);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
@@ -166,15 +182,21 @@ public sealed class PrimeOrderCalculatorAccelerator
|
||||
MemoryBuffer1D<ulong, Stride1D.Dense> outputUlong = OutputUlong;
|
||||
if (outputUlong.Length < newSize)
|
||||
{
|
||||
outputUlong.Dispose();
|
||||
OutputUlong = accelerator.Allocate1D<ulong>(newSize);
|
||||
lock (accelerator)
|
||||
{
|
||||
outputUlong.Dispose();
|
||||
OutputUlong = accelerator.Allocate1D<ulong>(newSize);
|
||||
}
|
||||
}
|
||||
|
||||
MemoryBuffer1D<int, Stride1D.Dense> outputInt = OutputInt;
|
||||
if (outputInt.Length < newSize)
|
||||
{
|
||||
outputInt.Dispose();
|
||||
OutputInt = accelerator.Allocate1D<int>(newSize);
|
||||
lock (accelerator)
|
||||
{
|
||||
outputInt.Dispose();
|
||||
OutputInt = accelerator.Allocate1D<int>(newSize);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
@@ -184,15 +206,21 @@ public sealed class PrimeOrderCalculatorAccelerator
|
||||
MemoryBuffer1D<ulong, Stride1D.Dense> input = Input;
|
||||
if (input.Length < newSize)
|
||||
{
|
||||
input.Dispose();
|
||||
Input = accelerator.Allocate1D<ulong>(newSize);
|
||||
lock (accelerator)
|
||||
{
|
||||
input.Dispose();
|
||||
Input = accelerator.Allocate1D<ulong>(newSize);
|
||||
}
|
||||
}
|
||||
|
||||
MemoryBuffer1D<ulong, Stride1D.Dense> outputUlong = OutputUlong;
|
||||
if (outputUlong.Length < newSize)
|
||||
{
|
||||
outputUlong.Dispose();
|
||||
OutputUlong = accelerator.Allocate1D<ulong>(newSize);
|
||||
lock (accelerator)
|
||||
{
|
||||
outputUlong.Dispose();
|
||||
OutputUlong = accelerator.Allocate1D<ulong>(newSize);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
@@ -201,57 +229,65 @@ public sealed class PrimeOrderCalculatorAccelerator
|
||||
var accelerator = _accelerators[acceleratorIndex];
|
||||
Accelerator = accelerator;
|
||||
AcceleratorIndex = acceleratorIndex;
|
||||
Input = accelerator.Allocate1D<ulong>(Math.Max(specialMaxFactorCapacity, primeTesterCapacity));
|
||||
OutputByte = Accelerator.Allocate1D<byte>(primeTesterCapacity);
|
||||
OutputInt = Accelerator.Allocate1D<int>(Math.Max(primeTesterCapacity, smallPrimeFactorSlotCount));
|
||||
OutputUlong = accelerator.Allocate1D<ulong>(Math.Max(specialMaxFactorCapacity, smallPrimeFactorSlotCount));
|
||||
Pow2ModEntriesToTestOnDevice = accelerator.Allocate1D<KeyValuePair<ulong, int>>(factorsCount);
|
||||
SmallPrimeFactorCountSlot = accelerator.Allocate1D<int>(1);
|
||||
SpecialMaxResult = accelerator.Allocate1D<ulong>(1);
|
||||
|
||||
LastDigitGpuTables lastDigitSharedTables = LastDigitGpuTables.GetStaticTables(acceleratorIndex);
|
||||
|
||||
DevicePrimesLastOne = lastDigitSharedTables.DevicePrimesLastOne.View;
|
||||
DevicePrimesLastSeven = lastDigitSharedTables.DevicePrimesLastSeven.View;
|
||||
DevicePrimesLastThree = lastDigitSharedTables.DevicePrimesLastThree.View;
|
||||
DevicePrimesLastNine = lastDigitSharedTables.DevicePrimesLastNine.View;
|
||||
DevicePrimesPow2LastOne = lastDigitSharedTables.DevicePrimesPow2LastOne.View;
|
||||
DevicePrimesPow2LastSeven = lastDigitSharedTables.DevicePrimesPow2LastSeven.View;
|
||||
DevicePrimesPow2LastThree = lastDigitSharedTables.DevicePrimesPow2LastThree.View;
|
||||
DevicePrimesPow2LastNine = lastDigitSharedTables.DevicePrimesPow2LastNine.View;
|
||||
|
||||
DivisorTables = HeuristicCombinedGpuTables
|
||||
.GetStaticTables(acceleratorIndex)
|
||||
.CreateViews();
|
||||
|
||||
var smallPrimeFactorTables = SmallPrimeFactorGpuTables.GetStaticTables(acceleratorIndex);
|
||||
|
||||
if (smallPrimeFactorTables.HasValue)
|
||||
lock (accelerator)
|
||||
{
|
||||
SmallPrimeFactorGpuTables tables = smallPrimeFactorTables.Value;
|
||||
SmallPrimeFactorPrimes = tables.Primes.View;
|
||||
SmallPrimeFactorSquares = tables.Squares.View;
|
||||
Input = accelerator.Allocate1D<ulong>(Math.Max(specialMaxFactorCapacity, primeTesterCapacity));
|
||||
OutputByte = Accelerator.Allocate1D<byte>(primeTesterCapacity);
|
||||
OutputInt = Accelerator.Allocate1D<int>(Math.Max(primeTesterCapacity, smallPrimeFactorSlotCount));
|
||||
OutputUlong = accelerator.Allocate1D<ulong>(Math.Max(specialMaxFactorCapacity, smallPrimeFactorSlotCount));
|
||||
Pow2ModEntriesToTestOnDevice = accelerator.Allocate1D<KeyValuePair<ulong, int>>(factorsCount);
|
||||
SmallPrimeFactorCountSlot = accelerator.Allocate1D<int>(1);
|
||||
SpecialMaxResult = accelerator.Allocate1D<ulong>(1);
|
||||
|
||||
LastDigitGpuTables lastDigitSharedTables = LastDigitGpuTables.GetStaticTables(acceleratorIndex);
|
||||
DevicePrimesLastOne = lastDigitSharedTables.DevicePrimesLastOne.View;
|
||||
DevicePrimesLastSeven = lastDigitSharedTables.DevicePrimesLastSeven.View;
|
||||
DevicePrimesLastThree = lastDigitSharedTables.DevicePrimesLastThree.View;
|
||||
DevicePrimesLastNine = lastDigitSharedTables.DevicePrimesLastNine.View;
|
||||
DevicePrimesPow2LastOne = lastDigitSharedTables.DevicePrimesPow2LastOne.View;
|
||||
DevicePrimesPow2LastSeven = lastDigitSharedTables.DevicePrimesPow2LastSeven.View;
|
||||
DevicePrimesPow2LastThree = lastDigitSharedTables.DevicePrimesPow2LastThree.View;
|
||||
DevicePrimesPow2LastNine = lastDigitSharedTables.DevicePrimesPow2LastNine.View;
|
||||
|
||||
DivisorTables = HeuristicCombinedGpuTables
|
||||
.GetStaticTables(acceleratorIndex)
|
||||
.CreateViews();
|
||||
|
||||
var smallPrimeFactorTables = SmallPrimeFactorGpuTables.GetStaticTables(acceleratorIndex);
|
||||
|
||||
if (smallPrimeFactorTables.HasValue)
|
||||
{
|
||||
SmallPrimeFactorGpuTables tables = smallPrimeFactorTables.Value;
|
||||
SmallPrimeFactorPrimes = tables.Primes.View;
|
||||
SmallPrimeFactorSquares = tables.Squares.View;
|
||||
}
|
||||
|
||||
CheckFactorsKernel = KernelUtil.GetKernel(accelerator.LoadStreamKernel<int, ulong, ArrayView1D<KeyValuePair<ulong, int>, Stride1D.Dense>, MontgomeryDivisorData, ArrayView1D<ulong, Stride1D.Dense>>(PrimeOrderGpuHeuristics.CheckFactorsKernel));
|
||||
|
||||
Kernel convertToStandardKernel = KernelUtil.GetKernel(accelerator.LoadAutoGroupedStreamKernel<Index1D, ArrayView1D<ulong, Stride1D.Dense>, MontgomeryDivisorData, ArrayView1D<ulong, Stride1D.Dense>>(Pow2MontgomeryKernels.Pow2MontgomeryKernelConvertToStandard));
|
||||
|
||||
ConvertToStandardKernel = convertToStandardKernel;
|
||||
ConvertToStandardKernelLauncher = convertToStandardKernel.CreateLauncherDelegate<Action<AcceleratorStream, Index1D, ArrayView1D<ulong, Stride1D.Dense>, MontgomeryDivisorData, ArrayView1D<ulong, Stride1D.Dense>>>();
|
||||
|
||||
var heuristicCombinedTrialDivisionKernel = KernelUtil.GetKernel(accelerator.LoadAutoGroupedStreamKernel<Index1D, byte, ArrayView<int>, ulong, ulong, HeuristicCombinedGpuViews>(PrimeTesterKernels.HeuristicTrialCombinedDivisionKernel));
|
||||
|
||||
HeuristicCombinedTrialDivisionKernel = heuristicCombinedTrialDivisionKernel;
|
||||
HeuristicCombinedTrialDivisionKernelLauncher = heuristicCombinedTrialDivisionKernel.CreateLauncherDelegate<Action<AcceleratorStream, Index1D, byte, ArrayView<int>, ulong, ulong, HeuristicCombinedGpuViews>>();
|
||||
|
||||
KeepMontgomeryKernel = KernelUtil.GetKernel(accelerator.LoadAutoGroupedStreamKernel<Index1D, ArrayView1D<ulong, Stride1D.Dense>, MontgomeryDivisorData, ArrayView1D<ulong, Stride1D.Dense>>(Pow2MontgomeryKernels.Pow2MontgomeryKernelKeepMontgomery));
|
||||
|
||||
PollardRhoKernel = KernelUtil.GetKernel(accelerator.LoadStreamKernel<ulong, int, ArrayView1D<ulong, Stride1D.Dense>, ArrayView1D<byte, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>>(Pow2MontgomeryKernels.TryPollardRhoKernel));
|
||||
|
||||
Pow2ModWideKernel = KernelUtil.GetKernel(accelerator.LoadAutoGroupedStreamKernel<Index1D, ArrayView1D<GpuUInt128, Stride1D.Dense>, GpuUInt128, ArrayView1D<GpuUInt128, Stride1D.Dense>>(PrimeOrderGpuHeuristics.Pow2ModKernelWide));
|
||||
|
||||
SharesFactorKernel = KernelUtil.GetKernel(accelerator.LoadAutoGroupedStreamKernel<Index1D, ArrayView<ulong>, ArrayView<byte>>(PrimeTesterKernels.SharesFactorKernel));
|
||||
|
||||
SmallPrimeFactorKernel = KernelUtil.GetKernel(accelerator.LoadAutoGroupedStreamKernel<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>>(SmallPrimeFactorKernels.SmallPrimeFactorKernelScan));
|
||||
|
||||
SmallPrimeSieveKernel = KernelUtil.GetKernel(accelerator.LoadAutoGroupedStreamKernel<Index1D, ArrayView<ulong>, ArrayView<uint>, ArrayView<uint>, ArrayView<uint>, ArrayView<uint>, ArrayView<ulong>, ArrayView<ulong>, ArrayView<ulong>, ArrayView<ulong>, ArrayView<byte>>(PrimeTesterKernels.SmallPrimeSieveKernel));
|
||||
|
||||
SpecialMaxKernel = KernelUtil.GetKernel(accelerator.LoadAutoGroupedStreamKernel<Index1D, ulong, ArrayView1D<ulong, Stride1D.Dense>, int, MontgomeryDivisorData, ArrayView1D<ulong, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>>(PrimeOrderGpuHeuristics.EvaluateSpecialMaxCandidatesKernel));
|
||||
}
|
||||
|
||||
CheckFactorsKernel = KernelUtil.GetKernel(accelerator.LoadStreamKernel<int, ulong, ArrayView1D<KeyValuePair<ulong, int>, Stride1D.Dense>, MontgomeryDivisorData, ArrayView1D<ulong, Stride1D.Dense>>(PrimeOrderGpuHeuristics.CheckFactorsKernel));
|
||||
|
||||
ConvertToStandardKernel = KernelUtil.GetKernel(accelerator.LoadAutoGroupedStreamKernel<Index1D, ArrayView1D<ulong, Stride1D.Dense>, MontgomeryDivisorData, ArrayView1D<ulong, Stride1D.Dense>>(Pow2MontgomeryKernels.Pow2MontgomeryKernelConvertToStandard));
|
||||
|
||||
HeuristicCombinedTrialDivisionKernel = KernelUtil.GetKernel(accelerator.LoadAutoGroupedStreamKernel<Index1D, byte, ArrayView<int>, ulong, ulong, HeuristicCombinedGpuViews>(PrimeTesterKernels.HeuristicTrialCombinedDivisionKernel));
|
||||
|
||||
KeepMontgomeryKernel = KernelUtil.GetKernel(accelerator.LoadAutoGroupedStreamKernel<Index1D, ArrayView1D<ulong, Stride1D.Dense>, MontgomeryDivisorData, ArrayView1D<ulong, Stride1D.Dense>>(Pow2MontgomeryKernels.Pow2MontgomeryKernelKeepMontgomery));
|
||||
|
||||
PollardRhoKernel = KernelUtil.GetKernel(accelerator.LoadStreamKernel<ulong, int, ArrayView1D<ulong, Stride1D.Dense>, ArrayView1D<byte, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>>(Pow2MontgomeryKernels.TryPollardRhoKernel));
|
||||
|
||||
Pow2ModWideKernel = KernelUtil.GetKernel(accelerator.LoadAutoGroupedStreamKernel<Index1D, ArrayView1D<GpuUInt128, Stride1D.Dense>, GpuUInt128, ArrayView1D<GpuUInt128, Stride1D.Dense>>(PrimeOrderGpuHeuristics.Pow2ModKernelWide));
|
||||
|
||||
SharesFactorKernel = KernelUtil.GetKernel(accelerator.LoadAutoGroupedStreamKernel<Index1D, ArrayView<ulong>, ArrayView<byte>>(PrimeTesterKernels.SharesFactorKernel));
|
||||
|
||||
SmallPrimeFactorKernel = KernelUtil.GetKernel(accelerator.LoadAutoGroupedStreamKernel<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>>(SmallPrimeFactorKernels.SmallPrimeFactorKernelScan));
|
||||
|
||||
SmallPrimeSieveKernel = KernelUtil.GetKernel(accelerator.LoadAutoGroupedStreamKernel<Index1D, ArrayView<ulong>, ArrayView<uint>, ArrayView<uint>, ArrayView<uint>, ArrayView<uint>, ArrayView<ulong>, ArrayView<ulong>, ArrayView<ulong>, ArrayView<ulong>, ArrayView<byte>>(PrimeTesterKernels.SmallPrimeSieveKernel));
|
||||
|
||||
SpecialMaxKernel = KernelUtil.GetKernel(accelerator.LoadAutoGroupedStreamKernel<Index1D, ulong, ArrayView1D<ulong, Stride1D.Dense>, int, MontgomeryDivisorData, ArrayView1D<ulong, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>>(PrimeOrderGpuHeuristics.EvaluateSpecialMaxCandidatesKernel));
|
||||
}
|
||||
|
||||
public void Dispose()
|
||||
|
||||
Reference in new issue
Block a user