Avoid redundant compaction buffer clearing on GPU

PerfectNumbers.Core/Gpu/Kernels/DivisorByDivisorKernels.cs

@@ -232,8 +232,10 @@ internal static class DivisorByDivisorKernels
 
         int lastLaneMask = globalIndex == length - 1 ? 1 : 0;
         int countValue = shared[prefixBase + localIndex];
-        int storedCount = lastLaneMask != 0 ? countValue : 0;
-        Atomic.Max(ref compactedCount[0], storedCount);
+        if (lastLaneMask != 0)
+        {
+            compactedCount[0] = countValue;
+        }
     }
 
     public static void PublishFirstHitKernel(

PerfectNumbers.Core/Gpu/MersenneNumberDivisorByDivisorGpuTester.cs

@@ -600,7 +600,6 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
         //     return 0;
         // }
 
-        countView.MemSetToZero();
         SharedMemoryConfig sharedMemoryConfig = SharedMemoryConfig.RequestDynamic<int>(count * 3);
         Index1D gridDim = new Index1D(1);
         Index1D groupDim = new Index1D(count);

PerfectNumbers.Core/MersenneNumberDivisorByDivisorTester.cs

@@ -73,7 +73,7 @@ public static class MersenneNumberDivisorByDivisorTester
 				candidateCount = candidateSpan.Length;
 			}
 
-			pool.Return(recordedCandidates, clearArray: true);
+			pool.Return(recordedCandidates, clearArray: false);
 		}
 
 		if (skippedByPreviousResults > 0)
@@ -211,8 +211,8 @@ public static class MersenneNumberDivisorByDivisorTester
 			filteredPrimes.Add(prime);
 		}
 
-		pool.Return(allowedMaxBatch, clearArray: true);
-		pool.Return(primeBatch, clearArray: true);
+		pool.Return(allowedMaxBatch, clearArray: false);
+		pool.Return(primeBatch, clearArray: false);
 
 		// The filtered list mirrors primesToTest in the CPU flow, so the guard below never triggers after the earlier
 		// emptiness checks.