[Perf] Regressions in System.Numerics.Tests.Perf_BigInteger

[performanceautofiler]

Run Information

Architecture	arm64
OS	ubuntu 18.04
Baseline	749fa1381f122f7a0bb23374a9e220f834ccbbb3
Compare	f83a9d9689433ce522b91e74a9631c83847e3b64
Diff	Diff

Regressions in System.Numerics.Tests.Perf_BigInteger

Benchmark	Baseline	Test	Test/Base	Test Quality	Edge Detector	Baseline IR	Compare IR	IR Ratio	Baseline ETL	Compare ETL
Ctor_ByteArray - Duration of single invocation	24.62 ns	32.53 ns	1.32	0.43	False
Ctor_ByteArray - Duration of single invocation	12.83 ns	26.17 ns	2.04	0.41	False

Historical Data in Reporting System

Repro

git clone https://github.com/dotnet/performance.git
python3 .\performance\scripts\benchmarks_ci.py -f netcoreapp5.0 --filter 'System.Numerics.Tests.Perf_BigInteger*'

Payloads

Baseline
Compare

Histogram

System.Numerics.Tests.Perf_BigInteger.Ctor_ByteArray(numberString: -2147483648)

System.Numerics.Tests.Perf_BigInteger.Ctor_ByteArray(numberString: 123)

Docs

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

[ghost]

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

Issue Details

---

Run Information

Architecture	arm64
OS	ubuntu 18.04
Baseline	749fa1381f122f7a0bb23374a9e220f834ccbbb3
Compare	f83a9d9689433ce522b91e74a9631c83847e3b64
Diff	Diff

Regressions in System.Numerics.Tests.Perf_BigInteger

Benchmark	Baseline	Test	Test/Base	Test Quality	Edge Detector	Baseline IR	Compare IR	IR Ratio	Baseline ETL	Compare ETL
Ctor_ByteArray - Duration of single invocation	24.62 ns	32.53 ns	1.32	0.43	False
Ctor_ByteArray - Duration of single invocation	12.83 ns	26.17 ns	2.04	0.41	False

Historical Data in Reporting System

Repro

git clone https://github.com/dotnet/performance.git
python3 .\performance\scripts\benchmarks_ci.py -f netcoreapp5.0 --filter 'System.Numerics.Tests.Perf_BigInteger*'

Payloads

Baseline
Compare

Histogram

System.Numerics.Tests.Perf_BigInteger.Ctor_ByteArray(numberString: -2147483648)

System.Numerics.Tests.Perf_BigInteger.Ctor_ByteArray(numberString: 123)

Docs

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

Author:	performanceautofiler[bot]
Assignees:	-
Labels:	area-System.Numerics, untriaged
Milestone:	-

[DrewScoggins]

Seems related to https://github.com/dotnet/runtime/pull/53984/files

[stephentoub]

Assuming #53984, as something intended to be a minor cleanup but regressing some benchmarks by 30%-100%, at this point for .NET 6 it should likely be reverted and revisted for .NET 7.
cc: @tannergooding, @huoyaoyuan

[tannergooding]

Agreed. Additionally, we should likely look into why the regression is so bad as a follow up.

I don't believe its expected that using indexers should be this impactful.

[huoyaoyuan]

https://sharplab.io/#v2:EYLgxg9gTgpgtADwGwBYA0AXEBDAzgWwB8ABAJgEYBYAKGIGYACMhgYQYG8aHuGBLAOwwBtALoNgvDLgDcXHvSbkkfQQwCy5ABRsAbtgA2AVxgBKOd07Ue1pgHYGeozAB0EqUIB65EbKs8AvuZMjMRKKhjqpNoOBsZmfhZB1sT2jsaukrhCaS5uuM4AMjD8AOYYABYMcAzevtaB1P5AA

JIT looks happy with index expression and generates slightly better code for it.

[stephentoub]

https://sharplab.io/#v2:EYLgxg9gTgpgtADwGwBYA0AXEBDAzgWwB8ABAJgEYBYAKGIGYACMhgYQYG8aHuGBLAOwwBtALoNgvDLgDcXHvSbkkfQQwCy5ABRsAbtgA2AVxgBKOd07Ue1pgHYGeozAB0EqUIB65EbKs8AvuZMjMRKKhjqpNoOBsZmfhZB1sT2jsaukrhCaS5uuM4AMjD8AOYYABYMcAzevtaB1P5AA
JIT looks happy with index expression and generates slightly better code for it.

FYI, that's on .NET 5 and amd64, rather than .NET 6 and arm64.

[tannergooding]

Notably, that's a relatively simple example. Looking through other scenarios and directly copying the full expression/context they are used in shows worse codegen in several cases.

I'm investigating more so I can log some issues on the runtime and/or compiler.

[AndyAyersMS]

Possibly relevant: #11870.