[Perf] Regressions in TryParseInt64

[DrewScoggins]

Run Information

Architecture	x64
OS	Windows 10.0.18362
Baseline	f280419a07a9445e1c6724e5717b3da7bdc5be7d
Compare	7b19ccefccb4d116a64bf09c9bb1db3dd1df35e8
Diff	Diff

Regressions in System.Buffers.Text.Tests.Utf8ParserTests

Benchmark	Baseline	Test	Test/Base	Test Quality	Edge Detector	Baseline IR	Compare IR	IR Ratio	Baseline ETL	Compare ETL
TryParseInt64 - Duration of single invocation	23.23 ns	27.85 ns	1.20	0.02	False
TryParseInt64 - Duration of single invocation	23.37 ns	27.23 ns	1.17	0.02	False
TryParseUInt64 - Duration of single invocation	16.33 ns	29.67 ns	1.82	0.06	True

Historical Data in Reporting System

Repro

git clone https://github.com/dotnet/performance.git
py .\performance\scripts\benchmarks_ci.py -f netcoreapp5.0 --filter 'System.Buffers.Text.Tests.Utf8ParserTests*'

Details

Payloads

Baseline
Compare

Histogram

System.Buffers.Text.Tests.Utf8ParserTests.TryParseInt64(value: -9223372036854775808)

System.Buffers.Text.Tests.Utf8ParserTests.TryParseInt64(value: 9223372036854775807)

System.Buffers.Text.Tests.Utf8ParserTests.TryParseUInt64(value: 18446744073709551615)

Docs

Profiling workflow for dotnet/runtime repository
Benchmarking workflow for dotnet/runtime repository

I couldn't figure out the best area label to add to this issue. If you have write-permissions please help me learn by adding exactly one area label.

[DrewScoggins]

Seems related to #52423

[stephentoub]

cc: @GrabYourPitchforks

[GrabYourPitchforks]

@DrewScoggins Do we know what processor this test was running on? The benchmarks I gave in #52423 were from my personal box (Ryzen 3950X), and I saw similar results on my work desktop (6th gen Intel).

If these optimizations ended up being too sensitive to the particular processors I tested against then we can easily revert, no harm done.

[GrabYourPitchforks]

Another interesting consideration here is that the regression only seems to be for the extremes, where things like overflow checks start to kick in. According to DrewScoggins/performance-2#7394 and DrewScoggins/performance-2#7395 the original PR actually improved performance for more typical inputs.

Still need to understand why the extremes saw a regression here, especially since my own benchmarking doesn't match what our labs are showing. But it makes me feel a bit better about the original PR not being bad across the board.

Tagging subscribers to this area: @dotnet/area-system-runtime
See info in area-owners.md if you want to be subscribed.

Issue Details

---

Run Information

Architecture	x64
OS	Windows 10.0.18362
Baseline	f280419a07a9445e1c6724e5717b3da7bdc5be7d
Compare	7b19ccefccb4d116a64bf09c9bb1db3dd1df35e8
Diff	Diff

Regressions in System.Buffers.Text.Tests.Utf8ParserTests

Benchmark	Baseline	Test	Test/Base	Test Quality	Edge Detector	Baseline IR	Compare IR	IR Ratio	Baseline ETL	Compare ETL
TryParseInt64 - Duration of single invocation	23.23 ns	27.85 ns	1.20	0.02	False
TryParseInt64 - Duration of single invocation	23.37 ns	27.23 ns	1.17	0.02	False
TryParseUInt64 - Duration of single invocation	16.33 ns	29.67 ns	1.82	0.06	True

Historical Data in Reporting System

Repro

git clone https://github.com/dotnet/performance.git
py .\performance\scripts\benchmarks_ci.py -f netcoreapp5.0 --filter 'System.Buffers.Text.Tests.Utf8ParserTests*'

Details

Payloads

Baseline
Compare

Histogram

System.Buffers.Text.Tests.Utf8ParserTests.TryParseInt64(value: -9223372036854775808)

System.Buffers.Text.Tests.Utf8ParserTests.TryParseInt64(value: 9223372036854775807)

System.Buffers.Text.Tests.Utf8ParserTests.TryParseUInt64(value: 18446744073709551615)

Docs

Profiling workflow for dotnet/runtime repository
Benchmarking workflow for dotnet/runtime repository

Author:	DrewScoggins
Assignees:	-
Labels:	arch-x64, area-System.Runtime, os-windows, tenet-performance, tenet-performance-benchmarks, untriaged
Milestone:	6.0.0

[kunalspathak]

Arm64 regression - DrewScoggins/performance-2#7611