Benchmark Regressions from Profile Maintenance and Block Layout Changes

[amanasifkhalid]

Part of #107749. I've collated all the regression reports assigned to me since #103450 was merged. For now, I'm excluding reports on Tiger machine configs to skip JCC errata-related regressions.

[amanasifkhalid]

Notes	Recent Score	Orig Score	arm64 ubuntu	Surface Windows	Viper Ubuntu	Viper Windows	Benchmark
#111736	2.65	2.67	1.76 1.79		3.99 3.98		System.Memory.Span(Byte).Reverse(Size: 512)
Unrelated	2.00	2.00			2.00 2.00		PerfLabTests.CastingPerf2.CastingPerf.ScalarValueTypeObj
#111736	1.97	1.95	1.43 1.42		2.72 2.68		System.Memory.Span(Char).Reverse(Size: 512)
#111736	1.84	1.85	1.68 1.68		2.02 2.04		System.Memory.Span(Int32).Reverse(Size: 512)
Unrelated	1.81	1.81				1.81 1.81	System.Linq.Tests.Perf_Enumerable.SingleWithPredicate_FirstElementMatches(input: Array)
Noisy	1.70	2.03	1.70 2.03				System.Text.Json.Document.Tests.Perf_DocumentParse.Parse(IsDataIndented: True, TestRandomAccess: True, TestCase: Json400KB)
#111736	1.56	1.06		1.56 1.06			System.Text.Json.Tests.Perf_Reader.ReadMultiSpanSequenceEmptyLoop(IsDataCompact: False, TestCase: DeepTree)
Unrelated	1.48	1.46	0.98 1.20			3.14 2.37	System.Collections.Tests.Perf_BitArray.BitArrayNot(Size: 512)
Noisy	1.45	1.12				1.45 1.12	System.Text.Json.Node.Tests.Perf_ParseThenWrite.ParseThenWrite(IsDataIndented: True, TestCase: Json400KB)
#111915	1.43	1.47				1.43 1.47	System.Reflection.Invoke.StaticMethod5_arrayNotCached_int_string_struct_class_bool
#111736	1.39	1.39			1.39 1.39		System.Memory.Span(Byte).IndexOfValue(Size: 512)
#111736	1.38	1.39				1.38 1.39	System.Numerics.Tensors.Tests.Perf_NumberTensorPrimitives(Int32).BitwiseAnd_Scalar(BufferLength: 128)
#111736	1.35	1.31			1.35 1.31		System.Numerics.Tests.Perf_BigInteger.Equals(arguments: 259 bytes, Same)
#112041	1.33	1.33				1.33 1.33	System.Numerics.Tensors.Tests.Perf_FloatingPointTensorPrimitives(Single).Pow_Vectors(BufferLength: 3079)
#112041	1.33	1.33				1.33 1.33	System.Numerics.Tensors.Tests.Perf_FloatingPointTensorPrimitives(Single).Pow_ScalarExponent(BufferLength: 3079)
#112041	1.33	1.33				1.33 1.33	System.Numerics.Tensors.Tests.Perf_FloatingPointTensorPrimitives(Single).Pow_ScalarBase(BufferLength: 3079)
#112041	1.33	1.32				1.33 1.32	System.Numerics.Tensors.Tests.Perf_FloatingPointTensorPrimitives(Single).Pow_Vectors(BufferLength: 128)
#111736	1.33	1.33	1.33 1.33				Benchstone.BenchI.BubbleSort.Test
#112041	1.32	1.32				1.32 1.32	System.Numerics.Tensors.Tests.Perf_FloatingPointTensorPrimitives(Single).Pow_ScalarExponent(BufferLength: 128)
#112041	1.32	1.32				1.32 1.32	System.Numerics.Tensors.Tests.Perf_FloatingPointTensorPrimitives(Single).Pow_ScalarBase(BufferLength: 128)
#111736	1.32	1.33			1.32 1.33		System.Memory.Span(Int32).LastIndexOfValue(Size: 512)
	1.32	1.26				1.32 1.26	System.Memory.Span(Int32).SequenceEqual(Size: 512)
	1.31	1.30			1.31 1.30		SciMark2.kernel.benchmarkLU
#103450	1.30	1.29	1.30 1.29				System.Memory.Span(Int32).IndexOfAnyFourValues(Size: 512)
#111736	1.27	1.27			1.27 1.27		System.Memory.Span(Byte).StartsWith(Size: 512)
#112041	1.26	1.26				1.26 1.26	System.MathBenchmarks.Single.AtanPi
#111736	1.25	1.28				1.25 1.28	System.Numerics.Tensors.Tests.Perf_FloatingPointTensorPrimitives(Single).Truncate(BufferLength: 128)
#111736	1.25	1.18	1.25 1.18				Microsoft.Extensions.Primitives.StringSegmentBenchmark.Equals_Object_Valid
#111736	1.25	1.24	1.41 1.42		1.11 1.09		System.Buffers.Text.Tests.Base64Tests.ConvertToBase64CharArray(NumberOfBytes: 1000)
Unrelated	1.25	1.60				1.25 1.60	System.Threading.Tasks.ValueTaskPerfTest.CreateAndAwait_FromResult
#110277	1.25	1.25	1.24 1.25	1.25 1.25			System.Collections.IterateForNonGeneric(String).ArrayList(Size: 512)
#111736	1.25	1.56			1.25 1.56		System.Numerics.Tensors.Tests.Perf_NumberTensorPrimitives(Int32).Max_Scalar(BufferLength: 128)
#111915	1.25	1.22				1.25 1.22	System.Collections.Tests.Add_Remove_SteadyState(String).Queue(Count: 512)
#111736	1.24	1.24			1.24 1.24		System.Numerics.Tensors.Tests.Perf_NumberTensorPrimitives(Single).Add_Scalar(BufferLength: 128)
#103450	1.24	1.25	1.24 1.25				System.Memory.Span(Int32).IndexOfAnyFiveValues(Size: 512)
#111915	1.24	1.23	1.24 1.24		1.23 1.23		System.Collections.IterateForNonGeneric(Int32).ArrayList(Size: 512)
#111736	1.24	1.23			1.24 1.23		System.Memory.Span(Byte).SequenceEqual(Size: 512)
#111736	1.23	1.28			1.23 1.28		System.Memory.Span(Char).SequenceEqual(Size: 512)
#111047	1.23	1.22				1.23 1.22	System.Reflection.Invoke.Field_SetStatic_int
#112041	1.22	1.22				1.22 1.22	System.MathBenchmarks.Single.AcosPi
Regressed by 3-opt and profile changes	1.21	1.31	1.13 1.13			1.29 1.52	Benchstone.BenchI.IniArray.Test
#110277	1.20	1.20		1.20 1.20			System.Tests.Perf_Char.Char_ToUpperInvariant(input: "Hello World!")
#110277	1.19	1.20			1.18 1.16	1.21 1.25	Span.Sorting.QuickSortArray(Size: 512)
#112041	1.19	1.19				1.19 1.19	System.MathBenchmarks.Double.AtanPi
#103450	1.18	1.33	1.19 1.11		1.17 1.60		Span.Sorting.BubbleSortSpan(Size: 512)
#111915	1.18	1.15			1.27 1.21	1.09 1.09	System.Collections.IterateForEachNonGeneric(String).ArrayList(Size: 512)
Unrelated	1.17	1.17				1.17 1.17	System.Numerics.Tests.Perf_VectorConvert.Convert_float_int
#110277	1.17	1.19				1.17 1.19	System.Collections.IterateForEach(String).ImmutableSortedSet(Size: 512)
#110277	1.17	1.16	1.15 1.14			1.19 1.18	System.Collections.IterateFor(Int32).IList(Size: 512)
#112041	1.16	1.16				1.16 1.16	System.Numerics.Tensors.Tests.Perf_FloatingPointTensorPrimitives(Single).AtanPi(BufferLength: 128)
	1.16	1.87			1.16 1.87		System.Buffers.Tests.RentReturnArrayPoolTests(Byte).ProducerConsumer(RentalSize: 4096, ManipulateArray: False, Async: True, UseSharedPool: True)
#112041	1.16	1.16				1.16 1.16	System.Numerics.Tensors.Tests.Perf_FloatingPointTensorPrimitives(Single).AtanPi(BufferLength: 3079)
#110277	1.16	1.18	1.06 1.08		1.20 1.26	1.23 1.21	Span.Sorting.QuickSortSpan(Size: 512)
#111736	1.16	1.16				1.16 1.16	System.Numerics.Tensors.Tests.Perf_NumberTensorPrimitives(Single).Divide_Vector(BufferLength: 128)
#111736	1.15	1.15	1.15 1.15				System.Text.Perf_Ascii.ToUpper_Bytes_Chars(Size: 128)
#111736	1.15	1.14		1.15 1.14			System.Tests.Perf_Int64.TryFormat(value: 12345)
	1.15	1.14				1.15 1.14	System.Collections.CreateAddAndClear(String).ConcurrentQueue(Size: 512)
#111736	1.15	1.30	1.15 1.30				System.Memory.Span(Byte).IndexOfAnyThreeValues(Size: 33)
#111915	1.14	1.07				1.14 1.07	System.Collections.Tests.Perf_PriorityQueue(Int32, Int32).HeapSort(Size: 1000)
#111915	1.14	1.11				1.14 1.11	System.Text.RegularExpressions.Tests.Perf_Regex_Industry_BoostDocs_Simple.IsMatch(Id: 6, Options: NonBacktracking)
	1.14	1.14		1.14 1.14			System.Text.Json.Document.Tests.Perf_EnumerateObject.EnumerateProperties(TestCase: StringProperties)
	1.14	1.14	1.14 1.14				Span.IndexerBench.CoveredIndex2(length: 1024)
	1.14	1.14			1.07 1.07	1.21 1.21	System.MathBenchmarks.Single.AsinPi
	1.14	1.14			1.11 1.11	1.17 1.17	System.MathBenchmarks.Double.AsinPi
	1.14	1.13				1.14 1.13	System.Numerics.Tensors.Tests.Perf_FloatingPointTensorPrimitives(Double).AtanPi(BufferLength: 128)
	1.14	1.11				1.14 1.11	System.Text.RegularExpressions.Tests.Perf_Regex_Industry_BoostDocs_Simple.IsMatch(Id: 10, Options: NonBacktracking)
	1.13	1.14			1.13 1.14		Benchstone.BenchF.Trap.Test
	1.13	1.12			1.12 1.11	1.14 1.14	System.MathBenchmarks.Double.AcosPi
	1.13	1.20	1.13 1.20				XmlDocumentTests.XmlNodeListTests.Perf_XmlNodeList.GetCount
	1.12	1.12				1.12 1.12	System.Numerics.Tensors.Tests.Perf_FloatingPointTensorPrimitives(Double).AtanPi(BufferLength: 3079)
	1.12	1.13			1.09 1.06	1.16 1.21	System.Numerics.Tensors.Tests.Perf_FloatingPointTensorPrimitives(Double).Pow_Vectors(BufferLength: 3079)
	1.12	1.12			1.12 1.12		System.Numerics.Tensors.Tests.Perf_FloatingPointTensorPrimitives(Single).Distance(BufferLength: 3079)
	1.11	1.11	1.11 1.11				System.Numerics.Tensors.Tests.Perf_NumberTensorPrimitives(Single).Max(BufferLength: 3079)
	1.11	1.11			1.11 1.11		System.Collections.Perf_LengthBucketsFrozenDictionary.TryGetValue_False_FrozenDictionary(Count: 10000, ItemsPerBucket: 5)
	1.11	1.10	1.11 1.10				Benchmark.GetChildKeysTests.AddChainedConfigurationEmpty
	1.11	1.20	1.05 1.06		1.17 1.36		Benchstone.BenchI.BubbleSort2.Test
	1.11	1.11	1.11 1.11				System.Text.Perf_Ascii.ToLower_Bytes_Chars(Size: 128)
	1.11	1.08				1.11 1.08	LinqBenchmarks.Count00LinqMethodX
	1.11	1.09			1.11 1.09		PerfLabTests.DelegatePerf.MulticastDelegateInvoke(length: 1000)
	1.10	1.09	1.10 1.09				System.Numerics.Tensors.Tests.Perf_FloatingPointTensorPrimitives(Single).FusedMultiplyAdd_ScalarMultiplier(BufferLength: 128)
	1.10	1.52	1.06 1.71		1.15 1.35		System.Collections.IterateFor(String).ImmutableList(Size: 512)
	1.10	1.10		1.10 1.10			System.Text.Json.Tests.Perf_Reader.ReadReturnBytes(IsDataCompact: False, TestCase: DeepTree)
	1.10	1.20		1.10 1.20			System.Collections.TryGetValueTrue(SmallClass, SmallClass).ImmutableDictionary(Size: 512)
	1.10	1.07	1.10 1.07				System.Collections.Perf_LengthBucketsFrozenDictionary.TryGetValue_True_FrozenDictionary(Count: 1000, ItemsPerBucket: 5)
	1.10	1.14	1.01 1.13		1.15 1.16	1.14 1.14	SeekUnroll.Test(boxedIndex: 3)
	1.10	1.07				1.10 1.07	Benchstone.MDBenchI.MDNDhrystone.Test
	1.10	1.07			1.10 1.07		System.Tests.Perf_Version.Parse4
	1.09	1.15			1.09 1.15		System.Tests.Perf_Enum.ToString_Format_NonFlags(value: Thursday, format: "f")
	1.09	1.07		1.07 1.07	1.12 1.07		System.Tests.Perf_Enum.ToString_Flags(value: 32)
	1.09	1.09		1.09 1.09			System.Globalization.Tests.StringSearch.IsPrefix_FirstHalf(Options: (, IgnoreCase, True))
	1.09	1.09				1.09 1.09	System.Collections.IterateForEachNonGeneric(Int32).ArrayList(Size: 512)
	1.09	1.10	1.09 1.10				Benchstone.BenchI.Array2.Test
	1.09	1.05			1.09 1.05		System.IO.Tests.Perf_Path.GetDirectoryName
	1.09	1.08			1.09 1.08		LinqBenchmarks.Where00ForX
	1.09	1.19	1.20 1.20		0.99 1.17		Span.IndexerBench.CoveredIndex3(length: 1024)
	1.09	1.11		1.09 1.11			System.Globalization.Tests.StringSearch.IsPrefix_FirstHalf(Options: (en-US, IgnoreCase, True))
	1.09	1.09				1.09 1.09	System.Numerics.Tensors.Tests.Perf_NumberTensorPrimitives(Single).BitwiseAnd_Vector(BufferLength: 128)
	1.09	1.11			1.07 1.07	1.11 1.16	System.Numerics.Tensors.Tests.Perf_FloatingPointTensorPrimitives(Double).Pow_ScalarExponent(BufferLength: 3079)
	1.09	1.09			1.09 1.09		System.Linq.Tests.Perf_Enumerable.SelectToList(input: IList)
	1.09	1.07		1.09 1.07			System.Collections.IndexerSet(Int32).SortedDictionary(Size: 512)
	1.09	1.11			1.06 1.07	1.11 1.16	System.Numerics.Tensors.Tests.Perf_FloatingPointTensorPrimitives(Double).Pow_ScalarBase(BufferLength: 128)
	1.08	1.11	1.08 1.11				System.Numerics.Tensors.Tests.Perf_FloatingPointTensorPrimitives(Single).FusedMultiplyAdd_Vectors(BufferLength: 128)
	1.08	1.09				1.08 1.09	LinqBenchmarks.Where00LinqQueryX
	1.08	1.08		1.08 1.08			System.Memory.Span(Byte).LastIndexOfAnyValues(Size: 512)
	1.08	1.09				1.08 1.09	LinqBenchmarks.Where00LinqMethodX
	1.08	1.14				1.08 1.14	System.Buffers.Tests.ReadOnlySequenceTests(Byte).IterateGetPositionArray
	1.08	1.13				1.08 1.13	System.Buffers.Tests.RentReturnArrayPoolTests(Byte).MultipleSerial(RentalSize: 4096, ManipulateArray: False, Async: False, UseSharedPool: True)
	1.08	1.12			1.08 1.12		System.Collections.Sort(IntStruct).List(Size: 512)
	1.08	1.09				1.08 1.09	System.Collections.IterateForEach(Int32).FrozenDictionary(Size: 512)
	1.08	1.08	1.08 1.08				Burgers.Test3
	1.08	1.12	1.00 1.13		1.16 1.17	1.07 1.07	SeekUnroll.Test(boxedIndex: 1)
	1.07	1.17				1.07 1.17	System.Threading.Tasks.ValueTaskPerfTest.CreateAndAwait_FromCompletedTask
	1.07	1.18			1.07 1.18		System.Tests.Perf_Enum.ToString_Format_Flags_Large(value: All, format: "g")
	1.07	1.08			1.07 1.08		System.Numerics.Tensors.Tests.Perf_NumberTensorPrimitives(Single).AddMultiply_Vectors(BufferLength: 128)
	1.07	1.09		1.07 1.09			System.Memory.Span(Int32).IndexOfAnyThreeValues(Size: 33)
	1.07	1.09				1.07 1.09	System.Text.RegularExpressions.Tests.Perf_Regex_Industry_BoostDocs_Simple.IsMatch(Id: 3, Options: NonBacktracking)
	1.07	1.10	1.08 1.11	1.07 1.09			System.Globalization.Tests.StringSearch.IndexOf_Word_NotFound(Options: (en-US, IgnoreCase, False))
	1.07	1.07	1.07 1.06	1.08 1.08			System.Text.Json.Document.Tests.Perf_ParseThenWrite.ParseThenWrite(IsDataIndented: True, TestCase: DeepTree)
	1.07	1.08	1.07 1.08				System.Net.Tests.Perf_WebUtility.Decode_DecodingRequired
	1.07	1.07			1.07 1.07		System.Numerics.Tensors.Tests.Perf_FloatingPointTensorPrimitives(Double).Pow_ScalarBase(BufferLength: 3079)
	1.07	1.07	1.07 1.07				System.Buffers.Tests.SearchValuesByteTests.ContainsAny(Values: "abcdefABCDEF0123456789")
	1.07	1.11				1.07 1.11	System.Collections.ContainsTrueComparer(String).HashSet(Size: 512)
	1.07	1.08	1.07 1.08				System.Text.Encodings.Web.Tests.Perf_Encoders.EncodeUtf8(arguments: UnsafeRelaxed,no (escaping /) required,16)
	1.07	1.06				1.07 1.06	System.Collections.TryGetValueTrue(BigStruct, BigStruct).ConcurrentDictionary(Size: 512)
	1.07	1.07	1.07 1.07				Benchstone.BenchI.CSieve.Test
	1.07	1.09			1.07 1.06	1.07 1.12	System.Numerics.Tensors.Tests.Perf_FloatingPointTensorPrimitives(Double).Pow_Vectors(BufferLength: 128)
	1.07	1.07	1.07 1.07				System.Text.Encodings.Web.Tests.Perf_Encoders.EncodeUtf8(arguments: UnsafeRelaxed,no (escaping /) required,512)
	1.07	1.09	1.07 1.10	1.07 1.09			System.Globalization.Tests.StringSearch.LastIndexOf_Word_NotFound(Options: (, IgnoreCase, False))
	1.07	1.11	1.07 1.11				System.Globalization.Tests.StringSearch.IndexOf_Word_NotFound(Options: (, IgnoreCase, False))
	1.07	1.06			1.07 1.06		System.Numerics.Tensors.Tests.Perf_FloatingPointTensorPrimitives(Double).Pow_ScalarExponent(BufferLength: 128)
	1.07	1.07			1.07 1.07		System.Tests.Perf_Version.TryParse4
	1.07	1.27	1.24 1.29		0.92 1.26		System.Tests.Perf_Char.Char_IsLower(input: "Good afternoon, Constable!")
	1.07	1.11	1.07 1.11				System.Globalization.Tests.StringSearch.LastIndexOf_Word_NotFound(Options: (en-US, IgnoreCase, False))
	1.07	1.07				1.07 1.07	System.Collections.Tests.Perf_PriorityQueue(Int32, Int32).Dequeue_And_Enqueue(Size: 100)
	1.06	1.06			1.06 1.06		System.Tests.Perf_Int128.TryParse(value: "-170141183460469231731687303715884105728")
	1.06	1.06	1.06 1.06				System.Text.Perf_Ascii.Equals_Bytes_Chars(Size: 128)
	1.06	1.06				1.06 1.06	System.Numerics.Tensors.Tests.Perf_FloatingPointTensorPrimitives(Double).Sqrt(BufferLength: 128)
	1.06	1.06			1.07 1.06	1.06 1.06	System.Tests.Perf_Int128.ParseSpan(value: "-170141183460469231731687303715884105728")
	1.06	1.08		1.06 1.08			System.Memory.Span(Byte).IndexOfAnyTwoValues(Size: 512)
	1.06	1.06			1.05 1.06	1.07 1.07	System.Linq.Tests.Perf_Enumerable.Average(input: IEnumerable)
	1.06	1.05				1.06 1.05	System.Reflection.Invoke.Property_Set_int
	1.06	1.15	1.00 1.08	1.00 1.08		1.20 1.32	System.Globalization.Tests.StringSearch.LastIndexOf_Word_NotFound(Options: (en-US, OrdinalIgnoreCase, False))
	1.06	1.15				1.06 1.15	System.Text.Json.Tests.Perf_Reader.ReadReturnBytes(IsDataCompact: True, TestCase: LotsOfStrings)
	1.06	1.06			1.06 1.07	1.05 1.06	System.Tests.Perf_Int128.Parse(value: "-170141183460469231731687303715884105728")
	1.06	1.06				1.06 1.06	System.MathBenchmarks.Single.ExpM1
	1.06	1.06			1.06 1.06	1.05 1.06	System.Tests.Perf_Int128.TryParseSpan(value: "-170141183460469231731687303715884105728")
	1.06	1.14			1.06 1.14		System.Tests.Perf_Int32.TryParse(value: "2147483647")
	1.06	1.06			1.06 1.06		System.IO.Tests.Perf_StreamWriter.WritePartialCharArray(writeLength: 2)
	1.06	1.06		1.06 1.06			System.Text.Json.Tests.Perf_Guids.WriteGuids(Formatted: False, SkipValidation: False)
	1.06	1.13	1.06 1.09	1.05 1.17			MicroBenchmarks.Serializers.Xml_ToStream(IndexViewModel).XmlSerializer_
	1.06	1.06	1.06 1.06				System.Numerics.Tensors.Tests.Perf_NumberTensorPrimitives(Single).IndexOfMax(BufferLength: 128)
	1.05	1.05	1.05 1.05				Layout.SearchLoops.LoopReturn
	1.05	1.05	1.05 1.05				System.Numerics.Tensors.Tests.Perf_NumberTensorPrimitives(Int32).MaxMagnitude_Vector(BufferLength: 128)
	1.05	1.05				1.05 1.05	System.MathBenchmarks.Double.ExpM1
	1.05	1.10				1.05 1.10	System.Collections.CreateAddAndClear(String).ImmutableArray(Size: 512)
	1.05	1.21	1.08 1.08	1.11 1.09	0.97 1.51		System.Memory.Span(Int32).IndexOfAnyFiveValues(Size: 33)
	1.04	1.10			1.04 1.10		System.Collections.TryGetValueTrue(SmallClass, SmallClass).SortedList(Size: 512)
	1.04	1.08		1.07 1.09		1.02 1.07	System.Text.RegularExpressions.Tests.Perf_Regex_Industry_BoostDocs_Simple.IsMatch(Id: 4, Options: NonBacktracking)
	1.04	1.20		1.04 1.20			System.Collections.TryGetValueTrue(SmallClass, SmallClass).ImmutableSortedDictionary(Size: 512)
	1.04	1.37	0.98 1.63		1.12 1.32	1.02 1.19	System.Collections.IterateFor(Int32).ImmutableList(Size: 512)
#111736	1.03	1.05	1.03 1.05				System.Memory.Span(Char).Clear(Size: 512)
	1.03	1.08				1.03 1.08	System.Linq.Tests.Perf_Enumerable.AppendPrepend(input: IEnumerable)
	1.03	1.80				1.03 1.80	System.Linq.Tests.Perf_Enumerable.SingleWithPredicate_LastElementMatches(input: Array)
	1.03	1.07				1.03 1.07	System.Diagnostics.Perf_Process.GetCurrentProcess
	1.03	1.09				1.03 1.09	System.Text.Json.Serialization.Tests.WriteJson(ClassRecord).SerializeToUtf8Bytes(Mode: Reflection)
	1.03	1.11				1.03 1.11	System.Collections.IndexerSet(String).SortedList(Size: 512)
	1.03	1.09				1.03 1.09	System.Collections.CreateAddAndRemove(String).List(Size: 512)
	1.02	2.08		1.02 2.08			System.Threading.Tests.Perf_Interlocked.CompareExchange_int
	1.02	1.10			1.02 1.10		System.Numerics.Tensors.Tests.Perf_NumberTensorPrimitives(Int32).AddMultiply_Vectors(BufferLength: 128)
	1.02	1.09			1.02 1.09		Span.Sorting.BubbleSortArray(Size: 512)
	1.02	1.22			1.02 1.22		System.Security.Cryptography.Primitives.Tests.Performance.Perf_FixedTimeEquals.FixedTimeEquals_256Bit_FirstBitDifferent
	1.02	1.08	1.02 1.08				MicroBenchmarks.Serializers.Json_ToStream(Location).DataContractJsonSerializer_
	1.02	1.20			1.02 1.20		System.Linq.Tests.Perf_Enumerable.SelectToList(input: Array)
	1.02	1.17	1.02 1.17				System.Buffers.Tests.RentReturnArrayPoolTests(Object).ProducerConsumer(RentalSize: 4096, ManipulateArray: False, Async: True, UseSharedPool: False)
	1.02	1.22			1.02 1.22		System.Security.Cryptography.Primitives.Tests.Performance.Perf_FixedTimeEquals.FixedTimeEquals_256Bit_Equal
	1.02	1.22			1.02 1.22		System.Security.Cryptography.Primitives.Tests.Performance.Perf_FixedTimeEquals.FixedTimeEquals_256Bit_VersusZero
	1.02	1.22			1.02 1.22		System.Security.Cryptography.Primitives.Tests.Performance.Perf_FixedTimeEquals.FixedTimeEquals_256Bit_LastBitDifferent
	1.02	1.22			1.02 1.22		System.Security.Cryptography.Primitives.Tests.Performance.Perf_FixedTimeEquals.FixedTimeEquals_256Bit_AllBitsDifferent
	1.02	1.22			1.02 1.22		System.Security.Cryptography.Primitives.Tests.Performance.Perf_FixedTimeEquals.FixedTimeEquals_256Bit_CascadingErrors
	1.01	1.29	1.04 1.60		1.00 1.18	1.01 1.14	System.Collections.IterateFor(Int32).ImmutableSortedSet(Size: 512)
	1.01	1.08			1.01 1.08		SciMark2.kernel.benchMonteCarlo
	1.01	1.10				1.01 1.10	System.Collections.IterateForEach(String).SortedSet(Size: 512)
	1.01	1.09			1.01 1.09		System.Collections.Perf_LengthBucketsFrozenDictionary.TryGetValue_True_FrozenDictionary(Count: 10, ItemsPerBucket: 5)
	1.01	1.13	1.00 1.19		1.02 1.07		PerfLabTests.LowLevelPerf.ForeachOverList100Elements
	1.01	1.54	1.02 1.48		1.00 1.87	1.00 1.33	Struct.SpanWrapper.WrapperSum

[dotnet-policy-service]

Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch
See info in area-owners.md if you want to be subscribed.

[amanasifkhalid]

Starting from the top, the System.Memory.Span.Reverse regressions are quite large, and were likely introduced by #111736. The only interesting change in that PR is we no longer mark blocks unreachable via normal flow as rarely run in optSetBlockWeights. As far as I can tell, this only makes a difference for handler regions.

Currently, profile synthesis will set the weight of handler region entries to something close to zero, but it won't mark them as rarely run. We probably don't want to change this, because marking them as rarely run would block inlining within handlers, which matters for finally regions. If we keep handler regions from being marked as rarely run after optSetBlockWeights, these blocks will be considered by a handful of later phases, in particular CSE. Considering CSEs in handler blocks is probably a poor use of our CSE budget, and we already treat handler regions as cold elsewhere, but #111736 had a decent number of improvements too, so I'm hesitant to reintroduce the previous behavior.

I wasn't able to repro the regression with EgorBot, but I'll take a look locally.

[amanasifkhalid]

I wasn't able to repro the regression with EgorBot, but I'll take a look locally.

I'm able to repro the regression on my Windows AMD machine, but the difference is quite inconsistent. From run to run, the difference is either just a nanosecond, or over 2x. For example:

| Method  | Job        | Toolchain                                                                         | Size | Mean     | Error     | StdDev    | Median   | Min      | Max      | Ratio | RatioSD | Allocated | Alloc Ratio |
|-------- |----------- |---------------------------------------------------------------------------------- |----- |---------:|----------:|----------:|---------:|---------:|---------:|------:|--------:|----------:|------------:|
| Reverse | Job-IGPZVU | \runtime2\artifacts\tests\coreclr\windows.x64.Checked\Tests\Core_Root\corerun.exe | 512  | 8.496 ns | 0.2086 ns | 0.1742 ns | 8.452 ns | 8.308 ns | 8.898 ns |  1.00 |    0.03 |         - |          NA |
| Reverse | Job-DKUPJI | \runtime\artifacts\tests\coreclr\windows.x64.Checked\Tests\Core_Root\corerun.exe  | 512  | 7.777 ns | 0.1223 ns | 0.1085 ns | 7.746 ns | 7.630 ns | 8.008 ns |  0.92 |    0.02 |         - |          NA |

| Method  | Job        | Toolchain                                                                         | Size | Mean      | Error     | StdDev    | Median    | Min       | Max       | Ratio | RatioSD | Allocated | Alloc Ratio |
|-------- |----------- |---------------------------------------------------------------------------------- |----- |----------:|----------:|----------:|----------:|----------:|----------:|------:|--------:|----------:|------------:|
| Reverse | Job-SIJWDS | \runtime2\artifacts\tests\coreclr\windows.x64.Checked\Tests\Core_Root\corerun.exe | 512  | 21.037 ns | 0.2629 ns | 0.2459 ns | 20.927 ns | 20.858 ns | 21.567 ns |  1.00 |    0.02 |         - |          NA |
| Reverse | Job-MXGQIK | \runtime\artifacts\tests\coreclr\windows.x64.Checked\Tests\Core_Root\corerun.exe  | 512  |  8.090 ns | 0.1682 ns | 0.1405 ns |  8.051 ns |  7.973 ns |  8.485 ns |  0.38 |    0.01 |         - |          NA |

I'm failing to find any codegen diffs along the call stack of this benchmark, and the potential fix would likely have its own slew of regressions, so the return on pursuing these regressions likely aren't as high as others that are directly related to layout churn. I'll keep this one in mind for later...

[amanasifkhalid]

PerfLabTests.CastingPerf2.CastingPerf.ScalarValueTypeObj regressed and then later improved due to flowgraph/profile churn. Its current regressed status is from #112060, and is tracked in #112657. This is also true for System.Linq.Tests.Perf_Enumerable.SingleWithPredicate_FirstElementMatches(input: Array). I'll remove these from the above chart.

[amanasifkhalid]

System.Collections.Tests.Perf_BitArray.BitArrayNot(Size: 512) improved on both platforms once late profile repair was enabled. On Viper Windows, it regressed again around 2/27/2025. I can't find an issue tracking this regression, but nothing in the commit range is flowgraph-related.

[amanasifkhalid]

I've annotated the most-regressed benchmarks with culprit PRs based on when the last notable regression occurred.

• JIT: Fix profile maintenance in optSetBlockWeights, funclet creation #111736 comes up much more often than I expected it to; I suspect it's pessimizing CSE quite a bit.
• There are a handful of PRs that regressed from the initial 3-opt or 4-opt PRs, and didn't budge all that much since. These likely expose the limitations of our cost model's ability to reflect actual performance.
• Unsurprisingly, several regressions are attributed to changes in profile data. These could be regressions in block layout as well, and/or regressions elsewhere.

I looked at a few of the regressions from #103450, and for these cases, 3-opt is keen to make bottom-tested loops top-tested. For example, here's the initial layout for System.SpanHelpers:IndexOfAny, which is the hot call in System.Memory.Span.IndexOfAnyFourValues and System.Memory.Span.IndexOfAnyFiveValues:

---------------------------------------------------------------------------------------------------------------------------------------------------------------------
BBnum BBid ref try hnd preds           weight    IBC [IL range]   [jump]                            [EH region]        [flags]
---------------------------------------------------------------------------------------------------------------------------------------------------------------------
BB01 [0000]  1                             1      18 [000..011)-> BB03(0.2),BB02(0.8)     ( cond )                     i LIR IBC
BB02 [0032]  1       BB01                  0.80   14 [000..001)-> BB03(1)                 (always)                     i LIR IBC hascall gcsafe
BB03 [0033]  2       BB01,BB02             1      18 [000..022)-> BB05(0.2),BB04(0.8)     ( cond )                     i LIR IBC
BB04 [0037]  1       BB03                  0.80   14 [011..012)-> BB05(1)                 (always)                     i LIR IBC hascall gcsafe
BB05 [0038]  2       BB03,BB04             1      18 [011..025)-> BB06(0.99),BB20(0.01)   ( cond )                     i LIR IBC
BB06 [0002]  1       BB05                  0.99   18 [027..03C)-> BB20(0.00595),BB17(0.994)   ( cond )                     i LIR IBC
BB17 [0043]  2       BB06,BB19            37.70  679 [03C..04D)-> BB19(0.195),BB18(0.805) ( cond )                     i LIR IBC bwd
BB18 [0044]  2       BB10,BB17           152.87 2752 [04D..062)-> BB10(0.995),BB09(0.005) ( cond )                     i LIR IBC bwd
BB10 [0007]  1       BB18                152.11 2738 [064..06C)-> BB18(0.805),BB19(0.195) ( cond )                     i LIR IBC bwd
BB19 [0045]  2       BB10,BB17            36.93  665 [06C..074)-> BB17(0.994),BB20(0.00595)   ( cond )                     i LIR IBC bwd
BB09 [0006]  1       BB18                  0.76   14 [062..064)                           (return)                     i LIR IBC
BB20 [0046]  3       BB06,BB05,BB19        0.24    4 [025..114)                           (return)                     i LIR IBC
---------------------------------------------------------------------------------------------------------------------------------------------------------------------

Both loops are bottom-tested, and we fall out of the inner loop into the outer loop. 3-opt transforms this into the following:

Initial layout cost: 3037.297884
Running greedy 3-opt pass.
Swapping partitions [BB17, BB10] and [BB19, BB19] (cost change = -106.408088)
Swapping partitions [BB19, BB10] and [BB20, BB09] (cost change = -0.105988)
Final layout cost: 2930.783808
...

---------------------------------------------------------------------------------------------------------------------------------------------------------------------
BBnum BBid ref try hnd preds           weight    IBC [IL range]   [jump]                            [EH region]        [flags]
---------------------------------------------------------------------------------------------------------------------------------------------------------------------
BB01 [0000]  1                             1      18 [000..011)-> BB03(0.2),BB02(0.8)     ( cond )                     i LIR IBC
BB02 [0032]  1       BB01                  0.80   14 [000..001)-> BB03(1)                 (always)                     i LIR IBC hascall gcsafe
BB03 [0033]  2       BB01,BB02             1      18 [000..022)-> BB05(0.2),BB04(0.8)     ( cond )                     i LIR IBC
BB04 [0037]  1       BB03                  0.80   14 [011..012)-> BB05(1)                 (always)                     i LIR IBC hascall gcsafe
BB05 [0038]  2       BB03,BB04             1      18 [011..025)-> BB06(0.99),BB20(0.01)   ( cond )                     i LIR IBC
BB06 [0002]  1       BB05                  0.99   18 [027..03C)-> BB20(0.00595),BB17(0.994)   ( cond )                     i LIR IBC
BB20 [0046]  3       BB06,BB05,BB19        0.24    4 [025..114)                           (return)                     i LIR IBC
BB09 [0006]  1       BB18                  0.76   14 [062..064)                           (return)                     i LIR IBC
BB19 [0045]  2       BB10,BB17            36.93  665 [06C..074)-> BB17(0.994),BB20(0.00595)   ( cond )                     i LIR IBC bwd
BB17 [0043]  2       BB06,BB19            37.70  679 [03C..04D)-> BB19(0.195),BB18(0.805) ( cond )                     i LIR IBC bwd
BB18 [0044]  2       BB10,BB17           152.87 2752 [04D..062)-> BB10(0.995),BB09(0.005) ( cond )                     i LIR IBC bwd
BB10 [0007]  1       BB18                152.11 2738 [064..06C)-> BB18(0.805),BB19(0.195) ( cond )                     i LIR IBC bwd
---------------------------------------------------------------------------------------------------------------------------------------------------------------------

It's the same case for Span.Sorting.BubbleSortSpan:

---------------------------------------------------------------------------------------------------------------------------------------------------------------------
BBnum BBid ref try hnd preds           weight       IBC [IL range]   [jump]                            [EH region]        [flags]
---------------------------------------------------------------------------------------------------------------------------------------------------------------------
BB01 [0000]  1                             1        100 [000..00A)-> BB02(1)                 (always)                     i LIR IBC
BB02 [0001]  2       BB01,BB11           168.07   16807 [00A..010)-> BB11(0.00597),BB10(0.994)   ( cond )                     i LIR IBC bwd bwd-target
BB10 [0009]  2       BB02,BB05           27969. 2796903 [010..026)-> BB05(0.0455),BB04(0.954)  ( cond )                     i LIR IBC bwd
BB04 [0003]  1       BB10                26696. 2669566 [026..052)-> BB05(1)                 (always)                     i LIR IBC bwd
BB05 [0004]  2       BB04,BB10           27969. 2796903 [052..05A)-> BB10(0.994),BB11(0.00597)   ( cond )                     i LIR IBC bwd
BB11 [0010]  2       BB02,BB05           168.07   16807 [05A..061)-> BB02(0.994),BB08(0.00595)   ( cond )                     i LIR IBC bwd
BB08 [0007]  1       BB11                  1.00     100 [061..062)                           (return)                     i LIR IBC
BB12 [0011]  0                             0            [???..???)                           (throw )                     i LIR rare keep internal
---------------------------------------------------------------------------------------------------------------------------------------------------------------------

Initial layout cost: 2924440.995680
Running greedy 3-opt pass.
Swapping partitions [BB10, BB04] and [BB05, BB05] (cost change = -77218.213563)
Swapping partitions [BB02, BB04] and [BB11, BB11] (cost change = -16506.722689)
Final layout cost: 2830716.059428

---------------------------------------------------------------------------------------------------------------------------------------------------------------------
BBnum BBid ref try hnd preds           weight       IBC [IL range]   [jump]                            [EH region]        [flags]
---------------------------------------------------------------------------------------------------------------------------------------------------------------------
BB01 [0000]  1                             1        100 [000..00A)-> BB02(1)                 (always)                     i LIR IBC
BB11 [0010]  2       BB02,BB05           168.07   16807 [05A..061)-> BB02(0.994),BB08(0.00595)   ( cond )                     i LIR IBC bwd
BB02 [0001]  2       BB01,BB11           168.07   16807 [00A..010)-> BB11(0.00597),BB10(0.994)   ( cond )                     i LIR IBC bwd bwd-target
BB05 [0004]  2       BB04,BB10           27969. 2796903 [052..05A)-> BB10(0.994),BB11(0.00597)   ( cond )                     i LIR IBC bwd
BB10 [0009]  2       BB02,BB05           27969. 2796903 [010..026)-> BB05(0.0455),BB04(0.954)  ( cond )                     i LIR IBC bwd
BB04 [0003]  1       BB10                26696. 2669566 [026..052)-> BB05(1)                 (always)                     i LIR IBC bwd
BB08 [0007]  1       BB11                  1.00     100 [061..062)                           (return)                     i LIR IBC
BB12 [0011]  0                             0            [???..???)                           (throw )                     i LIR rare keep internal
---------------------------------------------------------------------------------------------------------------------------------------------------------------------

For hot loops, 3-opt finds it massively profitable to create fallthrough along a loop backedge, since the entry edges are considerably lower in weight, but this gives a worse loop shape overall. I think we'd be justified in inhibiting the creation of fallthrough along loop backedges if it would make the loop top-tested. A targeted implementation would be to detect loop backedges in ThreeOptLayout::ConsiderEdge, and if the edge is a conditional jump, we don't consider it. I'd prefer ConsiderEdge continue to only check layout invariants and not introduce its own profitability checks, so we could also try tweaking the cost model to give loop backedges a cost of zero such that we don't try to create fallthrough along them, but I don't know all of the ramifications of that just yet.