[perf] System.Collections.** and System.Buffers.** performance regressions in Mono AOT x64 microbenchmarks from Preview 5 to Preview 7

[kotlarmilos]

Description

The following performance regressions occurred between the Preview 4 snapshot (May 2nd, 2023) and Preview 7 snapshot (July 19th, 2023).

The configuration used for the benchmarks is Mono AOT with LLVM on x64 ubuntu 18.04.

Name	Baseline Value	Compare Value	% Difference
System.Collections.Tests.Perf_PriorityQueue<Int32, Int32>.Enumerate(Size: 10)	48.24	55.58	15.22
System.Collections.Tests.Perf_BitArray.BitArraySet(Size: 4)	9.86	10.38	5.27
System.Collections.Tests.Perf_BitArray.BitArrayNot(Size: 512)	156.05	387.13	148.08
System.Collections.Tests.Perf_BitArray.BitArrayNot(Size: 4)	3.83	4.11	7.31
System.Collections.Tests.Perf_BitArray.BitArrayCopyToBoolArray(Size: 512)	7519.8	8942.84	18.92
System.Collections.Tests.Perf_BitArray.BitArrayByteArrayCtor(Size: 512)	1418.08	1509.38	6.44
System.Collections.Tests.Perf_BitArray.BitArrayBoolArrayCtor(Size: 512)	825.32	875.1	6.03
System.Collections.Tests.Perf_BitArray.BitArrayAnd(Size: 4)	4.94	5.39	9.11
System.Collections.Tests.Add_Remove_SteadyState<String>.Queue(Count: 512)	21.68	24.49	12.96
System.Collections.Sort<Int32>.Array_Comparison(Size: 512)	28171.27	29595.3	5.05
System.Collections.IterateForEach<String>.IEnumerable(Size: 512)	2962.18	3405.16	14.95
System.Collections.IterateForEach<Int32>.Dictionary(Size: 512)	2918.23	3256.52	11.59
System.Collections.IterateFor<String>.ReadOnlySpan(Size: 512)	461.08	683.44	48.23
System.Collections.IterateFor<String>.Array(Size: 512)	347.92	460.09	32.24
System.Collections.IndexerSet<String>.List(Size: 512)	2918.13	3256.95	11.61
System.Collections.IndexerSet<Int32>.List(Size: 512)	1321.05	1468.21	11.14
System.Collections.CtorFromCollection<Int32>.ConcurrentStack(Size: 512)	6904.3	7311.26	5.89
System.Collections.ContainsTrue<String>.HashSet(Size: 512)	17877.2	19374.79	8.38
System.Collections.ContainsTrue<Int32>.ImmutableArray(Size: 512)	106299.93	313108.63	194.55
System.Collections.ContainsTrue<Int32>.ICollection(Size: 512)	102156.81	309156.44	202.63
System.Collections.ContainsFalse<Int32>.ICollection(Size: 512)	194446.04	577782.73	197.14
System.Collections.Concurrent.Count<String>.Queue(Size: 512)	7.93	8.43	6.31
System.Buffers.Text.Tests.Utf8ParserTests.TryParseUInt64(value: 18446744073709551615)	41.01	57.29	39.7
System.Buffers.Text.Tests.Utf8ParserTests.TryParseBool(value: true)	11.85	12.5	5.49
System.Buffers.Tests.ReadOnlySequenceTests<Byte>.FirstSpanMemory	20.75	22.07	6.36

[kotlarmilos]

@MihaZupan Some of them may have regressed in #87621.

/cc: @vargaz

[MihaZupan]

That change should be cosmetic only

-bool someCondition = span.IndexOfAnyExcept(foo) >= 0;
+bool someCondition = span.ContainsAnyExcept(foo);

static bool ContainsAnyExcept(Span, value) => span.IndexOfAnyExcept(value) >= 0;

I suppose it's possible you could get worse codegen for some of these, but the differences here are way too excessive.

Besides, none of these benchmarks look like they'd be hitting the changed code.

[steveisok]

Since this is about .net 8, there's really not much action to take here. Closing.