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

GpuPrimeWorkLimiter.Acquire() / .Release() removed to save time on the hot-path with billions of calls. We'll rely on AcceleratorStreamPool limits and synchronization.

Browse Source
This commit is contained in:
mlemanczyk committed 2025-11-13 17:52:13 +01:00
1 parent 1552e10ff0
commit 0d2a88c45f
17 files changed
+108 -124
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

PerfectNumbers.Core/Gpu/Pow2MontgomeryGpuCalculator.cs
+19 -20
View File
@@ -10,28 +10,29 @@ public static class Pow2MontgomeryGpuCalculator
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public static ulong CalculateKeep(ulong exponent, in MontgomeryDivisorData divisor)
{
GpuPrimeWorkLimiter.Acquire();
// GpuPrimeWorkLimiter.Acquire();
ulong result = 0UL;
var gpu = PrimeOrderCalculatorAccelerator.Rent(1);
Accelerator? accelerator = gpu.Accelerator;
AcceleratorStream stream = AcceleratorStreamPool.Rent(accelerator);
MemoryBuffer1D<ulong, Stride1D.Dense> exponentBuffer = gpu.Input;
MemoryBuffer1D<ulong, Stride1D.Dense> resultBuffer = gpu.OutputUlong;
var kernel = gpu.KeepMontgomeryKernel;
AcceleratorStream stream = AcceleratorStreamPool.Rent(accelerator);
exponentBuffer.View.CopyFromCPU(stream, ref exponent, 1);
// We don't need to worry about any left-overs here.
// resultBuffer.MemSetToZero(stream);
MemoryBuffer1D<ulong, Stride1D.Dense> resultBuffer = gpu.OutputUlong;
var kernel = gpu.KeepMontgomeryKernel;
var kernelLauncher = kernel.CreateLauncherDelegate<Action<AcceleratorStream, Index1D, ArrayView1D<ulong, Stride1D.Dense>, MontgomeryDivisorData, ArrayView1D<ulong, Stride1D.Dense>>>();
kernelLauncher(stream, 1, exponentBuffer.View, divisor, resultBuffer.View);
ulong result = 0UL;
resultBuffer.View.CopyToCPU(stream, ref result, 1);
stream.Synchronize();
AcceleratorStreamPool.Return(stream);
GpuPrimeWorkLimiter.Release();
// GpuPrimeWorkLimiter.Release();
PrimeOrderCalculatorAccelerator.Return(gpu);
return result;
@@ -40,50 +41,48 @@ public static class Pow2MontgomeryGpuCalculator
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public static ulong CalculateConvert(ulong exponent, in MontgomeryDivisorData divisor)
{
GpuPrimeWorkLimiter.Acquire();
// GpuPrimeWorkLimiter.Acquire();
ulong result = 0UL;
var gpu = PrimeOrderCalculatorAccelerator.Rent(1);
AcceleratorStream stream = AcceleratorStreamPool.Rent(gpu.Accelerator);
MemoryBuffer1D<ulong, Stride1D.Dense> exponentBuffer = gpu.Input;
MemoryBuffer1D<ulong, Stride1D.Dense> resultBuffer = gpu.OutputUlong;
AcceleratorStream stream = AcceleratorStreamPool.Rent(gpu.Accelerator);
exponentBuffer.View.CopyFromCPU(stream, ref exponent, 1);
// We don't need to worry about any left-overs here.
// resultBuffer.MemSetToZero(stream);
MemoryBuffer1D<ulong, Stride1D.Dense> resultBuffer = gpu.OutputUlong;
var kernel = gpu.ConvertToStandardKernel;
kernel.Launch(stream, 1, 1, exponentBuffer.View, divisor, resultBuffer.View);
ulong result = 0UL;
resultBuffer.View.CopyToCPU(stream, ref result, 1);
stream.Synchronize();
AcceleratorStreamPool.Return(stream);
GpuPrimeWorkLimiter.Release();
AcceleratorStreamPool.Return(stream);
// GpuPrimeWorkLimiter.Release();
PrimeOrderCalculatorAccelerator.Return(gpu);
return result;
}
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public static ulong CalculateConvert(PrimeOrderCalculatorAccelerator gpu, ulong exponent, in MontgomeryDivisorData divisor)
{
AcceleratorStream stream = AcceleratorStreamPool.Rent(gpu.Accelerator);
ulong result = 0UL;
MemoryBuffer1D<ulong, Stride1D.Dense> exponentBuffer = gpu.Input;
MemoryBuffer1D<ulong, Stride1D.Dense> resultBuffer = gpu.OutputUlong;
var kernel = gpu.ConvertToStandardKernel;
AcceleratorStream stream = AcceleratorStreamPool.Rent(gpu.Accelerator);
exponentBuffer.View.CopyFromCPU(stream, ref exponent, 1);
// We don't need to worry about any left-overs here.
// resultBuffer.MemSetToZero(stream);
MemoryBuffer1D<ulong, Stride1D.Dense> resultBuffer = gpu.OutputUlong;
var kernel = gpu.ConvertToStandardKernel;
var kernelLauncher = kernel.CreateLauncherDelegate<Action<AcceleratorStream, Index1D, ArrayView1D<ulong, Stride1D.Dense>, MontgomeryDivisorData, ArrayView1D<ulong, Stride1D.Dense>>>();
kernelLauncher(stream, 1,exponentBuffer.View, divisor, resultBuffer.View);
ulong result = 0UL;
resultBuffer.View.CopyToCPU(stream, ref result, 1);
stream.Synchronize();
AcceleratorStreamPool.Return(stream);
return result;