GPU stabilization improvements to prevent calculation crashes

PerfectNumbers.Core.Tests/Gpu/GpuKernelCompilationTests.cs

@@ -348,32 +348,40 @@ public class GpuKernelCompilationTests
 
     private static void CompileGpuKernelPoolKernels(Accelerator accelerator)
     {
-        _ = accelerator;
-        GpuPrimeWorkLimiter.Acquire();
+
+		GpuPrimeWorkLimiter.Acquire();
+		AcceleratorStream stream = accelerator.CreateStream();
         try
-        {
-            _ = GpuKernelPool.Kernels.Order;
-            _ = GpuKernelPool.Kernels.Incremental;
-            _ = GpuKernelPool.Kernels.Pow2Mod;
-            _ = GpuKernelPool.Kernels.IncrementalOrder;
-            _ = GpuKernelPool.Kernels.Pow2ModOrder;
+		{
+			var kernels = GpuKernelPool.GetOrAddKernels(accelerator, stream);
+            _ = kernels.Order;
+            _ = kernels.Incremental;
+            _ = kernels.Pow2Mod;
+            _ = kernels.IncrementalOrder;
+			_ = kernels.Pow2ModOrder;
         }
         finally
-        {
+		{
+			stream.Synchronize();
+			stream.Dispose();
             GpuPrimeWorkLimiter.Release();
         }
 
-        GpuPrimeWorkLimiter.Acquire();
+		stream = accelerator.CreateStream();
+		GpuPrimeWorkLimiter.Acquire();
         try
         {
-            _ = GpuKernelPool.Kernels.Order;
-            _ = GpuKernelPool.Kernels.Incremental;
-            _ = GpuKernelPool.Kernels.Pow2Mod;
-            _ = GpuKernelPool.Kernels.IncrementalOrder;
-            _ = GpuKernelPool.Kernels.Pow2ModOrder;
+			var kernels = GpuKernelPool.GetOrAddKernels(accelerator, stream);
+            _ = kernels.Order;
+            _ = kernels.Incremental;
+            _ = kernels.Pow2Mod;
+            _ = kernels.IncrementalOrder;
+            _ = kernels.Pow2ModOrder;
         }
         finally
         {
+			stream.Synchronize();
+			stream.Dispose();
             GpuPrimeWorkLimiter.Release();
         }
     }

PerfectNumbers.Core.Tests/Gpu/NttGpuMathTests.cs

@@ -150,7 +150,7 @@ public class NttGpuMathTests
 		var modulus = new GpuUInt128(0UL, 17UL);
 		var primitiveRoot = new GpuUInt128(0UL, 3UL);
 
-		var accelerator = SharedGpuContext.Accelerator;
+		var accelerator = SharedGpuContext.CreateAccelerator();
 		var stream = accelerator.CreateStream();
 
 		var method = typeof(NttGpuMath).GetMethod("GetSquareCache", BindingFlags.NonPublic | BindingFlags.Static);
@@ -170,6 +170,7 @@ public class NttGpuMathTests
 		twiddlesInvMont.CopyToCPU(stream, inverse);
 		stream.Synchronize();
 		stream.Dispose();
+		accelerator.Dispose();
 
 		for (int i = 0; i < count; i++)
 		{
@@ -340,7 +341,7 @@ public class NttGpuMathTests
 		values[1] = new GpuUInt128(0UL, 2UL);
 		var modulus = new GpuUInt128(0UL, 17UL);
 		var primitiveRoot = new GpuUInt128(0UL, 3UL);
-		Accelerator accelerator = SharedGpuContext.Accelerator;
+		Accelerator accelerator = SharedGpuContext.CreateAccelerator();
 		var stream = accelerator.CreateStream();
 		var buffer = accelerator.Allocate1D<GpuUInt128>(values.Length);
 		buffer.View.CopyFromCPU(stream, ref values[0], values.Length);
@@ -348,6 +349,7 @@ public class NttGpuMathTests
 		stream.Synchronize();
 		stream.Dispose();
 		buffer.Dispose();
+		accelerator.Dispose();
 
 		// Verify that the square cache now holds entries for the accelerator.
 		var field = typeof(NttGpuMath).GetField("SquareCache", BindingFlags.NonPublic | BindingFlags.Static)!;

PerfectNumbers.Core.Tests/Gpu/NttMontgomeryTests.cs

@@ -33,20 +33,21 @@ public class NttMontgomeryTests
 		NttGpuMath.Square(cpu, modulus, primitiveRoot);
 
 		// Device execution
-		var accelerator = SharedGpuContext.Accelerator;
+		var accelerator = SharedGpuContext.CreateAccelerator();
 		var stream = accelerator.CreateStream();
 		using var buffer = accelerator.Allocate1D<GpuUInt128>(gpu.Length);
 		buffer.View.CopyFromCPU(stream, ref gpu[0], gpu.Length);
 		NttGpuMath.SquareDevice(accelerator, stream, buffer.View, modulus, primitiveRoot);
 		buffer.View.CopyToCPU(stream, ref gpu[0], gpu.Length);
 		stream.Synchronize();
+		stream.Dispose();
+		accelerator.Dispose();
 
 		for (int i = 0; i < cpu.Length; i++)
 		{
 			gpu[i].Should().Be(cpu[i]);
 		}
 
-		stream.Dispose();
 	}
 
 	[Fact]