Isolate allocation measurement.

Remove `GC.Collect()` from allocation measurement.

Author: timcassell

src/BenchmarkDotNet/Engines/Engine.cs

@@ -212,8 +212,6 @@ private ClockSpan Measure(Action<long> action, long invokeCount)
             return clock.GetElapsed();
         }
 
-        // Make sure tier0 jit doesn't cause any unexpected allocations in this method.
-        [MethodImpl(CodeGenHelper.AggressiveOptimizationOption)]
         private (GcStats, ThreadingStats, double) GetExtraStats(IterationData data)
         {
             // Warm up the GetAllocatedBytes function before starting the actual measurement.
@@ -224,23 +222,33 @@ private ClockSpan Measure(Action<long> action, long invokeCount)
             var initialThreadingStats = ThreadingStats.ReadInitial(); // this method might allocate
             var exceptionsStats = new ExceptionsStats(); // allocates
             exceptionsStats.StartListening(); // this method might allocate
-            var initialGcStats = GcStats.ReadInitial();
 
-            WorkloadAction(data.InvokeCount / data.UnrollFactor);
+            // GC collect before measuring allocations, as we do not collect during the measurement.
+            ForceGcCollect();
+            var gcStats = MeasureWithGc(data.InvokeCount / data.UnrollFactor);
 
-            var finalGcStats = GcStats.ReadFinal();
             exceptionsStats.Stop(); // this method might (de)allocate
             var finalThreadingStats = ThreadingStats.ReadFinal();
 
             IterationCleanupAction(); // we run iteration cleanup after collecting GC stats
 
             var totalOperationsCount = data.InvokeCount * OperationsPerInvoke;
-            GcStats gcStats = (finalGcStats - initialGcStats).WithTotalOperations(totalOperationsCount);
+            gcStats = gcStats.WithTotalOperations(totalOperationsCount);
             ThreadingStats threadingStats = (finalThreadingStats - initialThreadingStats).WithTotalOperations(data.InvokeCount * OperationsPerInvoke);
 
             return (gcStats, threadingStats, exceptionsStats.ExceptionsCount / (double)totalOperationsCount);
         }
 
+        // Isolate the allocation measurement and skip tier0 jit to make sure we don't get any unexpected allocations.
+        [MethodImpl(MethodImplOptions.NoInlining | CodeGenHelper.AggressiveOptimizationOption)]
+        private GcStats MeasureWithGc(long invokeCount)
+        {
+            var initialGcStats = GcStats.ReadInitial();
+            WorkloadAction(invokeCount);
+            var finalGcStats = GcStats.ReadFinal();
+            return finalGcStats - initialGcStats;
+        }
+
         private void RandomizeManagedHeapMemory()
         {
             // invoke global cleanup before global setup
@@ -267,8 +275,6 @@ private void GcCollect()
             ForceGcCollect();
         }
 
-        // Make sure tier0 jit doesn't cause any unexpected allocations in this method.
-        [MethodImpl(CodeGenHelper.AggressiveOptimizationOption)]
         internal static void ForceGcCollect()
         {
             GC.Collect();

src/BenchmarkDotNet/Engines/GcStats.cs

@@ -111,7 +111,6 @@ public int GetCollectionsCount(int generation)
         [MethodImpl(CodeGenHelper.AggressiveOptimizationOption)]
         public static GcStats ReadInitial()
         {
-            // this will force GC.Collect, so we want to do this before collecting collections counts
             long? allocatedBytes = GetAllocatedBytes();
 
             return new GcStats(
@@ -130,9 +129,6 @@ public static GcStats ReadFinal()
                 GC.CollectionCount(0),
                 GC.CollectionCount(1),
                 GC.CollectionCount(2),
-
-                // this will force GC.Collect, so we want to do this after collecting collections counts
-                // to exclude this single full forced collection from results
                 GetAllocatedBytes(),
                 0);
         }
@@ -149,11 +145,8 @@ public static GcStats FromForced(int forcedFullGarbageCollections)
             if (RuntimeInformation.IsWasm)
                 return null;
 
-            // "This instance Int64 property returns the number of bytes that have been allocated by a specific
-            // AppDomain. The number is accurate as of the last garbage collection." - CLR via C#
-            // so we enforce GC.Collect here just to make sure we get accurate results
-            Engine.ForceGcCollect();
-
+            // Calling GC.Collect() before calling GC.GetTotalAllocatedBytes appears to interfere with the results for some reason,
+            // so we just call the API without forcing a collection.
 #if NET6_0_OR_GREATER
             return GC.GetTotalAllocatedBytes(precise: true);
 #else