Initial PollardRho implementation for GPU. Added pool for accelerator streams shared across the threads. GPU processing performance and stabilization improvements. GpuUInt128.BinaryGcd switched to the version which outperformed other in benchmarks.

PerfectNumbers.Core/HeuristicCombinedPrimeTester.cs

@@ -224,6 +224,7 @@ public sealed class HeuristicCombinedPrimeTester
 
 	private static bool HeuristicTrialDivisionGpuDetectsDivisor(ulong n, ulong maxDivisorSquare, byte nMod10)
 	{
+		GpuPrimeWorkLimiter.Acquire();
 		var gpu = HeuristicPrimeTesterAccelerator.Rent(1);
 		var accelerator = gpu.Accelerator;
 		var stream = gpu.Stream;
@@ -238,7 +239,6 @@ public sealed class HeuristicCombinedPrimeTester
 		flagView1D.CopyFromCPU(stream, ref compositeFlag, 1);
 		stream!.Synchronize();
 
-		GpuPrimeWorkLimiter.Acquire();
 
 		kernel(
 				stream,