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

PerfectNumbers.Core/Gpu/GpuKernelPool.cs

@@ -280,13 +280,13 @@ public class GpuKernelPool
 	/// <param name="action">Action to run with (Accelerator, Stream).</param>
 	public static void Run(Action<Accelerator, AcceleratorStream> action)
 	{
-		GpuPrimeWorkLimiter.Acquire();
+		// GpuPrimeWorkLimiter.Acquire();
 		var accelerator = SharedGpuContext.CreateAccelerator();
 		var stream = accelerator.CreateStream();
 		action(accelerator, stream);
 		stream.Synchronize();
 		stream.Dispose();
 		accelerator.Dispose();
-		GpuPrimeWorkLimiter.Release();
+		// GpuPrimeWorkLimiter.Release();
 	}
 }

PerfectNumbers.Core/Gpu/MersenneNumberDivisorByDivisorGpuTester.cs

@@ -103,7 +103,7 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
         ulong processedCount;
         ulong lastProcessed;
 
-        GpuPrimeWorkLimiter.Acquire();
+        // GpuPrimeWorkLimiter();
         var accelerator = AcceleratorPool.Shared.Rent();
         var stream = accelerator.CreateStream();
 
@@ -137,7 +137,7 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
 		stream.Dispose();
         ReturnBatchResources(resources);
 		// Monitor.Exit(gpuLease.ExecutionLock);
-		GpuPrimeWorkLimiter.Release();
+		// GpuPrimeWorkLimiter.Release();
 		
         if (composite)
         {
@@ -509,7 +509,7 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
         internal DivisorScanSession(MersenneNumberDivisorByDivisorGpuTester owner)
         {
             _owner = owner;
-            GpuPrimeWorkLimiter.Acquire();
+            // GpuPrimeWorkLimiter.Acquire();
 			Accelerator accelerator = AcceleratorPool.Shared.Rent();
             _accelerator = accelerator;
             _stream = accelerator.CreateStream();

PerfectNumbers.Core/Gpu/MersenneNumberIncrementalGpuTester.cs

@@ -16,7 +16,7 @@ public class MersenneNumberIncrementalGpuTester(GpuKernelType kernelType, bool u
     {
         throw new NotImplementedException($"GPU incremental scanning requires the device cycle heuristics implementation (kernel: {_kernelType}, GPU order: {_useGpuOrder}).");
 
-        GpuPrimeWorkLimiter.Acquire();
+        // GpuPrimeWorkLimiter.Acquire();
         var accelerator = AcceleratorPool.Shared.Rent();
         var stream = accelerator.CreateStream();
         int batchSize = GpuConstants.ScanBatchSize; // large batch improves GPU occupancy
@@ -125,6 +125,6 @@ public class MersenneNumberIncrementalGpuTester(GpuKernelType kernelType, bool u
 		stream.Dispose();
         ulongPool.Return(orderArray);
 		orderBuffer.Dispose();
-        GpuPrimeWorkLimiter.Release();
+        // GpuPrimeWorkLimiter.Release();
     }
 }

PerfectNumbers.Core/Gpu/MersenneNumberLucasLehmerGpuTester.cs

@@ -159,7 +159,7 @@ public class MersenneNumberLucasLehmerGpuTester
         }
         else
         {
-        	GpuPrimeWorkLimiter.Acquire();
+        	// GpuPrimeWorkLimiter.Acquire();
             var accelerator = AcceleratorPool.Shared.Rent();
             var stream = accelerator.CreateStream();
             var modulus = new GpuUInt128(((UInt128)1 << (int)exponent) - 1UL); // TODO: Cache these Mersenne moduli per exponent so LL GPU runs skip rebuilding them every launch.
@@ -173,7 +173,7 @@ public class MersenneNumberLucasLehmerGpuTester
             stream.Dispose();
         }
 
-		GpuPrimeWorkLimiter.Release();
+		// GpuPrimeWorkLimiter.Release();
         return result;
     }
 

PerfectNumbers.Core/Gpu/MersenneNumberOrderGpuTester.cs

@@ -9,7 +9,7 @@ public class MersenneNumberOrderGpuTester(GpuKernelType kernelType, bool useGpuO
 
 	public void Scan(ulong exponent, UInt128 twoP, LastDigit lastDigit, UInt128 maxK, ref bool isPrime)
 	{
-		GpuPrimeWorkLimiter.Acquire();
+		// GpuPrimeWorkLimiter.Acquire();
 		// var accelerator = SharedGpuContext.Accelerator;
 		var accelerator = AcceleratorPool.Shared.Rent();
 		var stream = accelerator.CreateStream();
@@ -71,6 +71,6 @@ public class MersenneNumberOrderGpuTester(GpuKernelType kernelType, bool useGpuO
 
 		stream.Dispose();
 		foundBuffer.Dispose();
-		GpuPrimeWorkLimiter.Release();
+		// GpuPrimeWorkLimiter.Release();
 	}
 }

PerfectNumbers.Core/Gpu/MersenneNumberResidueGpuTester.cs

@@ -25,7 +25,7 @@ public class MersenneNumberResidueGpuTester(bool useGpuOrder)
 	// GPU residue variant: check 2^p % q == 1 for q = 2*p*k + 1.
 	public void Scan(ulong exponent, UInt128 twoP, LastDigit lastDigit, UInt128 maxK, ref bool isPrime)
 	{
-		GpuPrimeWorkLimiter.Acquire();
+		// GpuPrimeWorkLimiter.Acquire();
 		var accelerator = AcceleratorPool.Shared.Rent();
 
 		// Monitor.Enter(gpuLease.ExecutionLock);
@@ -110,7 +110,7 @@ public class MersenneNumberResidueGpuTester(bool useGpuOrder)
 		stream.Dispose();
 		// Monitor.Exit(gpuLease.ExecutionLock);
 		gpuKernels.Dispose();
-		GpuPrimeWorkLimiter.Release();
+		// GpuPrimeWorkLimiter.Release();
 	}
 
 	private static Action<AcceleratorStream, Index1D, ulong, GpuUInt128, GpuUInt128, byte, ulong, ResidueAutomatonArgs, ArrayView<ulong>, ArrayView1D<ulong, Stride1D.Dense>, ArrayView1D<uint, Stride1D.Dense>, ArrayView1D<uint, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>, ArrayView1D<ulong, Stride1D.Dense>> GetPow2ModKernel(Accelerator accelerator)

PerfectNumbers.Core/Gpu/Pow2MontgomeryGpuCalculator.cs

@@ -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;

PerfectNumbers.Core/Gpu/PrimeOrderGpuHeuristics.cs

@@ -225,7 +225,7 @@ internal static partial class PrimeOrderGpuHeuristics
 
 		try
 		{
-			GpuPrimeWorkLimiter.Acquire();
+			// GpuPrimeWorkLimiter.Acquire();
 			Accelerator accelerator = AcceleratorPool.Shared.Rent();
 			AcceleratorStream stream = accelerator.CreateStream();
 
@@ -290,7 +290,7 @@ internal static partial class PrimeOrderGpuHeuristics
 			countBuffer.Dispose();
 			remainingBuffer.Dispose();
 			fullyFactoredBuffer.Dispose();
-			GpuPrimeWorkLimiter.Release();
+			// GpuPrimeWorkLimiter.Release();
 			return true;
 		}
 		catch (CLException ex)
@@ -402,7 +402,7 @@ internal static partial class PrimeOrderGpuHeuristics
 	{
 		order = 0UL;
 
-		GpuPrimeWorkLimiter.Acquire();
+		// GpuPrimeWorkLimiter.Acquire();
 		Accelerator accelerator = AcceleratorPool.Shared.Rent();
 		AcceleratorStream stream = accelerator.CreateStream();
 
@@ -494,7 +494,7 @@ internal static partial class PrimeOrderGpuHeuristics
 		stackProductBuffer.Dispose();
 		resultBuffer.Dispose();
 		statusBuffer.Dispose();
-		GpuPrimeWorkLimiter.Release();
+		// GpuPrimeWorkLimiter.Release();
 
 		PrimeOrderKernelStatus kernelStatus = (PrimeOrderKernelStatus)status;
 		if (kernelStatus == PrimeOrderKernelStatus.Fallback)
@@ -574,7 +574,7 @@ internal static partial class PrimeOrderGpuHeuristics
 
 	private static bool TryComputeOnGpu(ReadOnlySpan<ulong> exponents, ulong prime, in MontgomeryDivisorData divisorData, Span<ulong> results)
 	{
-		GpuPrimeWorkLimiter.Acquire();
+		// GpuPrimeWorkLimiter.Acquire();
 		Accelerator accelerator = AcceleratorPool.Shared.Rent();
 		AcceleratorStream? stream = accelerator.CreateStream();
 		// TODO: These buffers should be allocated once per accelerator and only reallocated when the new length exceeds capacity.
@@ -604,7 +604,7 @@ internal static partial class PrimeOrderGpuHeuristics
 			stream = null;
 			exponentBuffer.Dispose();
 			remainderBuffer.Dispose();
-			GpuPrimeWorkLimiter.Release();
+			// GpuPrimeWorkLimiter.Release();
 		}
 		catch (Exception)
 		{
@@ -612,7 +612,7 @@ internal static partial class PrimeOrderGpuHeuristics
 			stream?.Dispose();
 			exponentBuffer.Dispose();
 			remainderBuffer.Dispose();
-			GpuPrimeWorkLimiter.Release();
+			// GpuPrimeWorkLimiter.Release();
 			throw;
 		}
 
@@ -638,7 +638,7 @@ internal static partial class PrimeOrderGpuHeuristics
 
 		GpuUInt128[]? rentedExponents = null;
 		GpuUInt128[]? rentedResults = null;
-		GpuPrimeWorkLimiter.Acquire();
+		// GpuPrimeWorkLimiter.Acquire();
 		ArrayPool<GpuUInt128> gpuUInt128Pool = ThreadStaticPools.GpuUInt128Pool;
 
 		try
@@ -704,7 +704,7 @@ internal static partial class PrimeOrderGpuHeuristics
 				gpuUInt128Pool.Return(rentedResults!, clearArray: false);
 			}
 
-			GpuPrimeWorkLimiter.Release();
+			// GpuPrimeWorkLimiter.Release();
 		}
 	}