Benchmarks to compare Interlocked with Atomic added. Atomic replaced with Interlocked on CPU path as better performer.

PerfectNumbers.Core/Gpu/AcceleratorPool.cs

@@ -1,3 +1,4 @@
+using System.Runtime.CompilerServices;
 using ILGPU;
 using ILGPU.Runtime;
 
@@ -11,10 +12,11 @@ namespace PerfectNumbers.Core.Gpu
 
 		private int _index;
 
+		[MethodImpl(MethodImplOptions.AggressiveInlining | MethodImplOptions.AggressiveOptimization)]
 		public int Rent()
 		{
-			Atomic.CompareExchange(ref _index, Capacity, 0);
-			var index = Atomic.Add(ref _index, 1);
+			_ = Interlocked.CompareExchange(ref _index, Capacity, 0);
+			var index = Interlocked.Increment(ref _index);
 			if (index >= Capacity)
 			{
 				index -= Capacity;
@@ -23,6 +25,7 @@ namespace PerfectNumbers.Core.Gpu
 			return index;
 		}
 
+		[MethodImpl(MethodImplOptions.AggressiveInlining | MethodImplOptions.AggressiveOptimization)]
 		public void Return(int acceleratorIndex)
 		{
 			// Intentionally left empty - there's nothing to do here. We're not really renting anything.

PerfectNumbers.Core/PrimeOrderCalculator.Hybrid.cs

@@ -377,12 +377,12 @@ internal static partial class PrimeOrderCalculator
 		{
 			while (true)
 			{
-				if (Atomic.CompareExchange(ref _cpuCount, 0, 1) == 0)
+				if (Interlocked.CompareExchange(ref _cpuCount, 0, 1) == 0)
 				{
 					return true;
 				}
 
-				if (Atomic.CompareExchange(ref _cpuCount, 1, 0) == 1)
+				if (Interlocked.CompareExchange(ref _cpuCount, 1, 0) == 1)
 				{
 					return false;
 				}
@@ -391,10 +391,10 @@ internal static partial class PrimeOrderCalculator
 			}
 		}
 
-		int cpuCount = Atomic.Add(ref _cpuCount, 1);
+		int cpuCount = Interlocked.Increment(ref _cpuCount);
 		if (cpuCount == PerfectNumberConstants.GpuRatio)
 		{
-			Atomic.Add(ref _cpuCount, -PerfectNumberConstants.GpuRatio);
+			Interlocked.Add(ref _cpuCount, -PerfectNumberConstants.GpuRatio);
 		}
 		else if (cpuCount > PerfectNumberConstants.GpuRatio)
 		{