Vectorize IA5 and Visible strings in ASN.1 encoding and decoding. by vcsjones · Pull Request #131109 · dotnet/runtime

vcsjones · 2026-07-20T21:58:38Z

This vectorizes the IA5Encoding and VisibleStringEncoding for ASN.1 encoding and decoding for the .NET implementations. Downlevel remains unchanged. Validation, encoding, and decoding will now operate on length of Vector<T> at a time instead of processing each input character-by-character or byte-by-byte.

Overall this shows good performance improvements. Allocations remain unchanged, and through improves for typical inputs. This only uses high-level System.NumericsVector<T> - using intrinsics was not a goal of this change. This ensure we don't need to account for minor platform behavior differences.

For a 32 element input, decoding goes from 51.6ns to 17.7ns. Writing goes from 52.76ns to 21.06ns. Full benchmarks are available below the fold. Benchmark code is available at https://gist.github.com/vcsjones/73dbd0ae386b62a68b4d03cf71ed933d.

Error paths may be slightly slower. This is because we process data one vector of length at a time, including validation. However in the event of invalid input, we have to back up and process by scalar to find the exact position where the invalid element is to maintain exception compatibility. This feels like a worthy trade off, both because it's an error path and the exception throw and unwinding is still going to dominate, and for simplicity.

Some test cases were added to ensure all validation paths are correctly hit.

Performance


BenchmarkDotNet v0.15.8, macOS Tahoe 26.5.2 (25F84) [Darwin 25.5.0]
Apple M1 Ultra, 1 CPU, 20 logical and 20 physical cores
[Host] : .NET 11.0.0 (11.0.0-dev, 42.42.42.42424), Arm64 RyuJIT armv8.0-a
Main : .NET 11.0.0 (11.0.0-dev, 42.42.42.42424), Arm64 RyuJIT armv8.0-a
Vector : .NET 11.0.0 (11.0.0-dev, 42.42.42.42424), Arm64 RyuJIT armv8.0-a
Runtime=.NET 10.0 IterationCount=15 WarmupCount=5

Type	Method	Job	Toolchain	Input	Mean	Error	StdDev	Ratio	RatioSD	Gen0	Gen1	Allocated	Alloc Ratio
LongIa5Benchmarks	Decode	Main	Main	?	1,400.90 ns	2.113 ns	1.873 ns	1.00	0.00	0.3300	-	2072 B	1.00
LongIa5Benchmarks	Decode	Vector	Vector	?	226.35 ns	0.827 ns	0.691 ns	0.16	0.00	0.3302	-	2072 B	1.00

ShortIa5Benchmarks	Decode	Main	Main	?	20.32 ns	0.027 ns	0.024 ns	1.00	0.00	0.0063	-	40 B	1.00
ShortIa5Benchmarks	Decode	Vector	Vector	?	17.46 ns	0.044 ns	0.039 ns	0.86	0.00	0.0063	-	40 B	1.00

LongIa5Benchmarks	Write	Main	Main	?	1,370.51 ns	6.418 ns	5.359 ns	1.00	0.01	0.1755	-	1104 B	1.00
LongIa5Benchmarks	Write	Vector	Vector	?	265.40 ns	0.770 ns	0.720 ns	0.19	0.00	0.1760	-	1104 B	1.00

ShortIa5Benchmarks	Write	Main	Main	?	23.36 ns	0.039 ns	0.037 ns	1.00	0.00	0.0140	-	88 B	1.00
ShortIa5Benchmarks	Write	Vector	Vector	?	19.86 ns	0.046 ns	0.043 ns	0.85	0.00	0.0140	-	88 B	1.00

ValidIa5Benchmarks	Decode	Main	Main	DnsName	36.29 ns	0.081 ns	0.076 ns	1.00	0.00	0.0102	-	64 B	1.00
ValidIa5Benchmarks	Decode	Vector	Vector	DnsName	18.07 ns	0.040 ns	0.035 ns	0.50	0.00	0.0102	-	64 B	1.00

ValidIa5Benchmarks	Write	Main	Main	DnsName	37.34 ns	0.247 ns	0.219 ns	1.00	0.01	0.0153	-	96 B	1.00
ValidIa5Benchmarks	Write	Vector	Vector	DnsName	20.51 ns	0.230 ns	0.179 ns	0.55	0.01	0.0153	-	96 B	1.00

InvalidIa5Benchmarks	DecodeInvalid	Main	Main	EmailEarly	2,880.45 ns	4.989 ns	4.666 ns	1.00	0.00	0.1984	-	1264 B	1.00
InvalidIa5Benchmarks	DecodeInvalid	Vector	Vector	EmailEarly	2,885.77 ns	3.808 ns	3.562 ns	1.00	0.00	0.1984	-	1264 B	1.00

InvalidIa5Benchmarks	WriteInvalid	Main	Main	EmailEarly	1,182.06 ns	1.420 ns	1.259 ns	1.00	0.00	0.1163	-	736 B	1.00
InvalidIa5Benchmarks	WriteInvalid	Vector	Vector	EmailEarly	1,168.29 ns	1.759 ns	1.560 ns	0.99	0.00	0.1163	-	736 B	1.00

InvalidIa5Benchmarks	DecodeInvalid	Main	Main	EmailMiddle	2,886.74 ns	4.166 ns	3.693 ns	1.00	0.00	0.1984	-	1264 B	1.00
InvalidIa5Benchmarks	DecodeInvalid	Vector	Vector	EmailMiddle	2,876.05 ns	5.428 ns	5.078 ns	1.00	0.00	0.1984	-	1264 B	1.00

InvalidIa5Benchmarks	WriteInvalid	Main	Main	EmailMiddle	1,184.04 ns	1.944 ns	1.819 ns	1.00	0.00	0.1163	-	736 B	1.00
InvalidIa5Benchmarks	WriteInvalid	Vector	Vector	EmailMiddle	1,183.79 ns	1.825 ns	1.708 ns	1.00	0.00	0.1163	-	736 B	1.00

InvalidIa5Benchmarks	DecodeInvalid	Main	Main	EmailLate	2,888.25 ns	3.983 ns	3.531 ns	1.00	0.00	0.1984	-	1264 B	1.00
InvalidIa5Benchmarks	DecodeInvalid	Vector	Vector	EmailLate	2,879.34 ns	3.686 ns	3.268 ns	1.00	0.00	0.1984	-	1264 B	1.00

InvalidIa5Benchmarks	WriteInvalid	Main	Main	EmailLate	1,181.78 ns	1.037 ns	0.970 ns	1.00	0.00	0.1163	-	736 B	1.00
InvalidIa5Benchmarks	WriteInvalid	Vector	Vector	EmailLate	1,176.87 ns	2.646 ns	2.345 ns	1.00	0.00	0.1163	-	736 B	1.00

InvalidIa5Benchmarks	DecodeInvalid	Main	Main	Length256Late	3,060.51 ns	2.382 ns	2.111 ns	1.00	0.00	0.1984	-	1264 B	1.00
InvalidIa5Benchmarks	DecodeInvalid	Vector	Vector	Length256Late	2,928.94 ns	3.417 ns	3.197 ns	0.96	0.00	0.1984	-	1264 B	1.00

InvalidIa5Benchmarks	WriteInvalid	Main	Main	Length256Late	1,362.22 ns	1.288 ns	1.205 ns	1.00	0.00	0.1564	-	984 B	1.00
InvalidIa5Benchmarks	WriteInvalid	Vector	Vector	Length256Late	1,223.28 ns	1.960 ns	1.833 ns	0.90	0.00	0.1564	-	984 B	1.00

InvalidIa5Benchmarks	DecodeInvalid	Main	Main	Length1024Late	3,558.98 ns	2.779 ns	2.600 ns	1.00	0.00	0.1984	-	1264 B	1.00
InvalidIa5Benchmarks	DecodeInvalid	Vector	Vector	Length1024Late	2,957.81 ns	2.656 ns	2.484 ns	0.83	0.00	0.1984	-	1264 B	1.00

InvalidIa5Benchmarks	WriteInvalid	Main	Main	Length1024Late	1,888.57 ns	8.916 ns	7.904 ns	1.00	0.01	0.2785	-	1752 B	1.00
InvalidIa5Benchmarks	WriteInvalid	Vector	Vector	Length1024Late	1,326.31 ns	2.821 ns	2.639 ns	0.70	0.00	0.2785	-	1752 B	1.00

InvalidIa5Benchmarks	DecodeInvalid	Main	Main	Length4096Late	5,496.13 ns	6.284 ns	5.570 ns	1.00	0.00	0.1984	-	1264 B	1.00
InvalidIa5Benchmarks	DecodeInvalid	Vector	Vector	Length4096Late	3,119.58 ns	4.459 ns	3.952 ns	0.57	0.00	0.1984	-	1264 B	1.00

InvalidIa5Benchmarks	WriteInvalid	Main	Main	Length4096Late	3,961.88 ns	7.917 ns	6.611 ns	1.00	0.00	0.7629	0.0076	4824 B	1.00
InvalidIa5Benchmarks	WriteInvalid	Vector	Vector	Length4096Late	1,740.70 ns	2.692 ns	2.519 ns	0.44	0.00	0.7668	0.0114	4824 B	1.00

VisibleStringBenchmarks	Decode	Main	Main	Length64	96.40 ns	0.116 ns	0.109 ns	1.00	0.00	0.0242	-	152 B	1.00
VisibleStringBenchmarks	Decode	Vector	Vector	Length64	24.41 ns	0.081 ns	0.068 ns	0.25	0.00	0.0242	-	152 B	1.00

VisibleStringBenchmarks	Write	Main	Main	Length64	101.00 ns	2.010 ns	1.880 ns	1.00	0.03	0.0229	-	144 B	1.00
VisibleStringBenchmarks	Write	Vector	Vector	Length64	28.27 ns	0.059 ns	0.055 ns	0.28	0.01	0.0229	-	144 B	1.00

VisibleStringBenchmarks	Decode	Main	Main	Length1024	1,386.68 ns	1.321 ns	1.171 ns	1.00	0.00	0.3300	-	2072 B	1.00
VisibleStringBenchmarks	Decode	Vector	Vector	Length1024	221.22 ns	0.353 ns	0.330 ns	0.16	0.00	0.3302	-	2072 B	1.00

VisibleStringBenchmarks	Write	Main	Main	Length1024	1,348.01 ns	1.441 ns	1.348 ns	1.00	0.00	0.1755	-	1104 B	1.00
VisibleStringBenchmarks	Write	Vector	Vector	Length1024	258.67 ns	0.799 ns	0.747 ns	0.19	0.00	0.1760	-	1104 B	1.00

InvalidIa5Benchmarks	DecodeInvalid	Main	Main	Length64Late	2,966.07 ns	5.094 ns	3.977 ns	1.00	0.00	0.1984	-	1264 B	1.00
InvalidIa5Benchmarks	DecodeInvalid	Vector	Vector	Length64Late	2,901.35 ns	4.689 ns	4.386 ns	0.98	0.00	0.1984	-	1264 B	1.00

InvalidIa5Benchmarks	WriteInvalid	Main	Main	Length64Late	1,230.04 ns	2.293 ns	2.145 ns	1.00	0.00	0.1240	-	784 B	1.00
InvalidIa5Benchmarks	WriteInvalid	Vector	Vector	Length64Late	1,193.77 ns	1.603 ns	1.500 ns	0.97	0.00	0.1240	-	784 B	1.00

ValidIa5Benchmarks	Decode	Main	Main	ShortEmail	32.26 ns	0.060 ns	0.056 ns	1.00	0.00	0.0089	-	56 B	1.00
ValidIa5Benchmarks	Decode	Vector	Vector	ShortEmail	15.58 ns	0.055 ns	0.052 ns	0.48	0.00	0.0089	-	56 B	1.00

ValidIa5Benchmarks	Write	Main	Main	ShortEmail	34.38 ns	0.046 ns	0.043 ns	1.00	0.00	0.0153	-	96 B	1.00
ValidIa5Benchmarks	Write	Vector	Vector	ShortEmail	18.39 ns	0.120 ns	0.100 ns	0.53	0.00	0.0153	-	96 B	1.00

ValidIa5Benchmarks	Decode	Main	Main	LongEmail	85.55 ns	0.106 ns	0.099 ns	1.00	0.00	0.0216	-	136 B	1.00
ValidIa5Benchmarks	Decode	Vector	Vector	LongEmail	29.65 ns	0.068 ns	0.064 ns	0.35	0.00	0.0216	-	136 B	1.00

ValidIa5Benchmarks	Write	Main	Main	LongEmail	85.93 ns	0.120 ns	0.112 ns	1.00	0.00	0.0216	-	136 B	1.00
ValidIa5Benchmarks	Write	Vector	Vector	LongEmail	32.17 ns	0.109 ns	0.102 ns	0.37	0.00	0.0216	-	136 B	1.00

ValidIa5Benchmarks	Decode	Main	Main	Uri	92.23 ns	0.885 ns	0.784 ns	1.00	0.01	0.0242	-	152 B	1.00
ValidIa5Benchmarks	Decode	Vector	Vector	Uri	33.22 ns	0.060 ns	0.056 ns	0.36	0.00	0.0242	-	152 B	1.00

ValidIa5Benchmarks	Write	Main	Main	Uri	94.65 ns	0.060 ns	0.050 ns	1.00	0.00	0.0216	-	136 B	1.00
ValidIa5Benchmarks	Write	Vector	Vector	Uri	35.81 ns	0.071 ns	0.067 ns	0.38	0.00	0.0216	-	136 B	1.00

ValidIa5Benchmarks	Decode	Main	Main	Length8	20.08 ns	0.025 ns	0.024 ns	1.00	0.00	0.0063	-	40 B	1.00
ValidIa5Benchmarks	Decode	Vector	Vector	Length8	17.22 ns	0.029 ns	0.027 ns	0.86	0.00	0.0063	-	40 B	1.00

ValidIa5Benchmarks	Write	Main	Main	Length8	23.01 ns	0.048 ns	0.044 ns	1.00	0.00	0.0140	-	88 B	1.00
ValidIa5Benchmarks	Write	Vector	Vector	Length8	19.68 ns	0.038 ns	0.036 ns	0.86	0.00	0.0140	-	88 B	1.00

ValidIa5Benchmarks	Decode	Main	Main	Length16	30.80 ns	0.047 ns	0.044 ns	1.00	0.00	0.0089	-	56 B	1.00
ValidIa5Benchmarks	Decode	Vector	Vector	Length16	15.08 ns	0.135 ns	0.126 ns	0.49	0.00	0.0089	-	56 B	1.00

ValidIa5Benchmarks	Write	Main	Main	Length16	32.43 ns	0.054 ns	0.050 ns	1.00	0.00	0.0153	-	96 B	1.00
ValidIa5Benchmarks	Write	Vector	Vector	Length16	17.36 ns	0.045 ns	0.042 ns	0.54	0.00	0.0153	-	96 B	1.00

ValidIa5Benchmarks	Decode	Main	Main	Length32	51.66 ns	0.082 ns	0.077 ns	1.00	0.00	0.0140	-	88 B	1.00
ValidIa5Benchmarks	Decode	Vector	Vector	Length32	17.70 ns	0.068 ns	0.061 ns	0.34	0.00	0.0140	-	88 B	1.00

ValidIa5Benchmarks	Write	Main	Main	Length32	52.76 ns	0.074 ns	0.069 ns	1.00	0.00	0.0178	-	112 B	1.00
ValidIa5Benchmarks	Write	Vector	Vector	Length32	21.06 ns	0.079 ns	0.062 ns	0.40	0.00	0.0179	-	112 B	1.00

ValidIa5Benchmarks	Decode	Main	Main	Length64	96.62 ns	0.143 ns	0.134 ns	1.00	0.00	0.0242	-	152 B	1.00
ValidIa5Benchmarks	Decode	Vector	Vector	Length64	24.60 ns	0.115 ns	0.096 ns	0.25	0.00	0.0242	-	152 B	1.00

ValidIa5Benchmarks	Write	Main	Main	Length64	106.81 ns	0.158 ns	0.140 ns	1.00	0.00	0.0229	-	144 B	1.00
ValidIa5Benchmarks	Write	Vector	Vector	Length64	28.05 ns	0.050 ns	0.047 ns	0.26	0.00	0.0229	-	144 B	1.00

ValidIa5Benchmarks	Decode	Main	Main	Length128	195.11 ns	0.174 ns	0.155 ns	1.00	0.00	0.0446	-	280 B	1.00
ValidIa5Benchmarks	Decode	Vector	Vector	Length128	38.80 ns	0.120 ns	0.112 ns	0.20	0.00	0.0446	-	280 B	1.00

ValidIa5Benchmarks	Write	Main	Main	Length128	196.89 ns	0.083 ns	0.073 ns	1.00	0.00	0.0331	-	208 B	1.00
ValidIa5Benchmarks	Write	Vector	Vector	Length128	43.85 ns	0.104 ns	0.087 ns	0.22	0.00	0.0331	-	208 B	1.00

ValidIa5Benchmarks	Decode	Main	Main	Length256	367.03 ns	0.271 ns	0.240 ns	1.00	0.00	0.0854	-	536 B	1.00
ValidIa5Benchmarks	Decode	Vector	Vector	Length256	64.32 ns	0.209 ns	0.195 ns	0.18	0.00	0.0854	-	536 B	1.00

ValidIa5Benchmarks	Write	Main	Main	Length256	359.23 ns	0.373 ns	0.348 ns	1.00	0.00	0.0534	-	336 B	1.00
ValidIa5Benchmarks	Write	Vector	Vector	Length256	73.64 ns	0.117 ns	0.104 ns	0.20	0.00	0.0535	-	336 B	1.00

ValidIa5Benchmarks	Decode	Main	Main	Length1024	1,389.09 ns	12.097 ns	10.724 ns	1.00	0.01	0.3300	-	2072 B	1.00
ValidIa5Benchmarks	Decode	Vector	Vector	Length1024	222.03 ns	0.713 ns	0.557 ns	0.16	0.00	0.3302	-	2072 B	1.00

ValidIa5Benchmarks	Write	Main	Main	Length1024	1,350.11 ns	1.211 ns	1.073 ns	1.00	0.00	0.1755	-	1104 B	1.00
ValidIa5Benchmarks	Write	Vector	Vector	Length1024	253.19 ns	0.493 ns	0.461 ns	0.19	0.00	0.1760	-	1104 B	1.00

ValidIa5Benchmarks	Decode	Main	Main	Length4096	5,436.07 ns	6.509 ns	6.088 ns	1.00	0.00	1.3046	-	8216 B	1.00
ValidIa5Benchmarks	Decode	Vector	Vector	Length4096	863.95 ns	1.696 ns	1.324 ns	0.16	0.00	1.3084	-	8216 B	1.00

ValidIa5Benchmarks	Write	Main	Main	Length4096	5,328.17 ns	5.326 ns	4.982 ns	1.00	0.00	0.6638	0.0076	4176 B	1.00
ValidIa5Benchmarks	Write	Vector	Vector	Length4096	1,000.79 ns	1.553 ns	1.453 ns	0.19	0.00	0.6657	0.0095	4176 B	1.00

InvalidIa5Benchmarks	DecodeInvalid	Main	Main	LongEmailLate	2,933.39 ns	13.165 ns	11.671 ns	1.00	0.01	0.1984	-	1264 B	1.00
InvalidIa5Benchmarks	DecodeInvalid	Vector	Vector	LongEmailLate	2,918.78 ns	5.534 ns	4.321 ns	1.00	0.00	0.1984	-	1264 B	1.00

InvalidIa5Benchmarks	WriteInvalid	Main	Main	LongEmailLate	1,232.75 ns	8.312 ns	7.369 ns	1.00	0.01	0.1221	-	776 B	1.00
InvalidIa5Benchmarks	WriteInvalid	Vector	Vector	LongEmailLate	1,194.64 ns	5.756 ns	5.103 ns	0.97	0.01	0.1221	-	776 B	1.00

VisibleStringBenchmarks	Decode	Main	Main	ShortText	24.09 ns	0.029 ns	0.026 ns	1.00	0.00	0.0076	-	48 B	1.00
VisibleStringBenchmarks	Decode	Vector	Vector	ShortText	19.82 ns	0.027 ns	0.025 ns	0.82	0.00	0.0076	-	48 B	1.00

VisibleStringBenchmarks	Write	Main	Main	ShortText	26.24 ns	0.038 ns	0.033 ns	1.00	0.00	0.0140	-	88 B	1.00
VisibleStringBenchmarks	Write	Vector	Vector	ShortText	21.92 ns	0.047 ns	0.042 ns	0.84	0.00	0.0140	-	88 B	1.00

VisibleStringBenchmarks	Decode	Main	Main	Uri	95.28 ns	0.210 ns	0.197 ns	1.00	0.00	0.0242	-	152 B	1.00
VisibleStringBenchmarks	Decode	Vector	Vector	Uri	33.43 ns	0.084 ns	0.079 ns	0.35	0.00	0.0242	-	152 B	1.00

VisibleStringBenchmarks	Write	Main	Main	Uri	94.32 ns	0.072 ns	0.068 ns	1.00	0.00	0.0216	-	136 B	1.00
VisibleStringBenchmarks	Write	Vector	Vector	Uri	35.50 ns	0.070 ns	0.065 ns	0.38	0.00	0.0216	-	136 B	1.00

This vectorizes the IA5Encoding and VisibleStringEncoding for ASN.1 encoding and decoding.

azure-pipelines · 2026-07-20T21:59:17Z

Azure Pipelines: Successfully started running 3 pipeline(s). 13 pipeline(s) were filtered out due to trigger conditions. There may be pipelines that require an authorized user to comment /azp run to run.

vcsjones · 2026-07-20T21:59:18Z

/cc @tannergooding since this is adjacent to your wheelhouse.

vcsjones · 2026-07-20T22:00:40Z

cc @dotnet/area-system-formats-asn1

dotnet-policy-service · 2026-07-20T22:00:47Z

Tagging subscribers to this area: @dotnet/area-system-formats-asn1, @bartonjs, @vcsjones
See info in area-owners.md if you want to be subscribed.

Copilot

Pull request overview

This PR improves System.Formats.Asn1 character-string handling by introducing a vectorized fast-path for IA5String and VisibleString encoding/decoding on .NET (while keeping downlevel behavior unchanged), and extends tests to cover vector-width boundaries and invalid-data index reporting.

Changes:

Introduces RestrictedAsciiRangeEncoding for .NETCoreApp builds, using System.Numerics.Vector<T> to validate/convert in vector-sized chunks with scalar fallback to preserve exception-index behavior.
Switches IA5Encoding and VisibleStringEncoding to derive from RestrictedAsciiRangeEncoding (range-based validation) instead of the existing allowed-table implementation.
Adds targeted reader/writer tests for vector-sized payloads and “invalid after vector prefix” scenarios.

Show a summary per file

File	Description
src/libraries/System.Formats.Asn1/src/System/Formats/Asn1/AsnCharacterStringEncodings.net.cs	Adds vectorized range-based encoding/decoding implementation for IA5/VisibleString on .NET.
src/libraries/System.Formats.Asn1/src/System/Formats/Asn1/AsnCharacterStringEncodings.downlevel.cs	Keeps downlevel behavior by mapping `RestrictedAsciiRangeEncoding` to the existing restricted ASCII encoder.
src/libraries/System.Formats.Asn1/src/System/Formats/Asn1/AsnCharacterStringEncodings.cs	Updates IA5/VisibleString encodings to use the new range-based base type.
src/libraries/System.Formats.Asn1/src/System.Formats.Asn1.csproj	Adds conditional compile includes for the new files and references `System.Numerics.Vectors` for .NET builds.
src/libraries/System.Formats.Asn1/tests/Writer/WriteIA5String.cs	Adds an invalid IA5 input with an invalid character after a vector-sized prefix.
src/libraries/System.Formats.Asn1/tests/Writer/SimpleWriterTests.cs	Adds VisibleString writer tests for vector-sized payloads and invalid character index reporting.
src/libraries/System.Formats.Asn1/tests/Reader/ReadIA5String.cs	Adds IA5 reader coverage for a longer valid input and a vector-prefix invalid input case.
src/libraries/System.Formats.Asn1/tests/Reader/ComprehensiveReadTests.cs	Adds VisibleString reader tests for vector-sized payloads and invalid byte index reporting.

Copilot's findings

Files reviewed: 8/8 changed files
Comments generated: 1

tannergooding · 2026-07-20T22:38:08Z

Should be able to get to this later tonight. At a high level glance it generally looks good/correct and following all the guidance.

I notably did just get a rewrite of the public SIMD (on docs.microsoft.com) done and have a PR up adding a vectorization skill for review/authoring, so if you're up for it then it might be a good first trial of it (not merged yet): #131108

bartonjs · 2026-07-20T23:38:20Z

Generally looks good to me, nothing stood out in review (though I didn't stop and think hard about the bounds). Have the benchmarks been run against strings at various offsets, just making sure alignment isn't a problem.

Since one of the tests clearly is operating at alignment+3, I'm going to guess it's all good, but it still feels like something worth checking.

vcsjones · 2026-07-21T02:58:00Z

Have the benchmarks been run against strings at various offsets, just making sure alignment isn't a problem.

Seems fine and still significant performance gains. Results (too big for a GH comment): https://gist.github.com/vcsjones/6500b884c3f145f0c3fb11749dbf566c

tannergooding

LGTM.

I also tested this locally against the new vectorization skill and depending on model the feedback was generally along the lines of:

You might be able to add a fast path to reuse Ascii.IsValid for the cases the range is "all ascii"
This is using Vector<T> rather than the fixed-width vectors and so may leave perf on the table
No explicit tests using BoundedMemory were added to help ensure safety

None of those I view as "blocking", particularly given the area and the way the code is setup.

vcsjones · 2026-07-21T15:51:26Z

You might be able to add a fast path to reuse Ascii.IsValid for the cases the range is "all ascii"

This is using Vector rather than the fixed-width vectors and so may leave perf on the table

I think the code reuse between IA5String and VisibleString is more valuable. Neither I don't think are quite used enough to warrant intense optimization to eke out every ounce of performance.

No explicit tests using BoundedMemory were added to help ensure safety

My impression was that Vector<T> is "high level" enough that it should on its own have bounds checking that does the right thing, but adding bounded memory tests is easy enough and I needed to fix a typo, so done.

tannergooding · 2026-07-21T15:56:19Z

My impression was that Vector is "high level" enough that it should on its own have bounds checking that does the right thing, but adding bounded memory tests is easy enough and I needed to fix a typo, so done.

It should be, especially if you use the Create(Span<T>) APIs instead of Load/LoadUnsafe. But it's also one of the cases that can often just integrate naturally into the edges you should already be testing (and which in your case you already were covering) so the guidance is kept as just general guidance.

Like I said, I didn't personally view any of that as blocking, more just "automatic callouts" that vectorization PRs may get from Copilot in the future (which will likely always be a bit stricter than a human reviewer).

Copilot

Copilot's findings

Comments suppressed due to low confidence (1)

src/libraries/System.Formats.Asn1/src/System/Formats/Asn1/AsnCharacterStringEncodings.net.cs:154

Same as above: the comment says it returns the position "before the current vector's width", but the code returns the current position. Please align the comment with the actual behavior.

 // If any element in the vector is not allowed, we break out and return the position before the
// current vector's width so that it goes down the scalar path. The scalar path will determine the
// precise location of the invalid element.

Files reviewed: 8/8 changed files
Comments generated: 1

As a follow up to #131109, this introduces vectorization for the `BMPEncoding`. BMPEncoding is big-endian UCS-2 encoding. It is a subset of UTF-16-BE, with the exception that it is truly fixed width and only supports plane 0 without the surrogate pairs. This also makes some changes to the existing vectorization logic for JIT friendliness.

Vectorize IA5 and Visible strings in ASN.1 encoding and decoding.
4411e14
This vectorizes the IA5Encoding and VisibleStringEncoding for ASN.1 encoding and decoding.

CopilotAI review requested due to automatic review settings July 20, 2026 21:58

github-actionsBot added the area-Infrastructure-libraries label Jul 20, 2026

Copilotstarted reviewing on behalf of vcsjones July 20, 2026 21:59 View session

dotnet-policy-serviceBot assigned vcsjones Jul 20, 2026

vcsjones added area-System.Formats.Asn1 and removed area-Infrastructure-libraries labels Jul 20, 2026

vcsjones added the tenet-performance Performance related issue label Jul 20, 2026

CopilotAI reviewed Jul 20, 2026
View reviewed changes

Comment threadsrc/libraries/System.Formats.Asn1/src/System/Formats/Asn1/AsnCharacterStringEncodings.net.cs Outdated

tannergooding self-requested a review July 20, 2026 22:35

Merge remote-tracking branch 'ms/main' into vector-asn1-string
c579a13

tannergooding approved these changes Jul 21, 2026
View reviewed changes

Code review feedback
e3d015f

CopilotAI review requested due to automatic review settings July 21, 2026 15:48

Copilotstarted reviewing on behalf of vcsjones July 21, 2026 15:48 View session

tannergooding approved these changes Jul 21, 2026
View reviewed changes

CopilotAI reviewed Jul 21, 2026
View reviewed changes

Comment threadsrc/libraries/System.Formats.Asn1/src/System/Formats/Asn1/AsnCharacterStringEncodings.net.cs

bartonjs approved these changes Jul 21, 2026
View reviewed changes

vcsjones enabled auto-merge (squash) July 21, 2026 16:31

vcsjones merged commit 2d21ab7 into dotnet:mainJul 21, 2026
83 checks passed

vcsjones deleted the vector-asn1-string branch July 21, 2026 19:17

vcsjones mentioned this pull request Jul 21, 2026
Vectorize BMP string #131170
Merged

dotnet-milestone-botBot modified the milestones: 11.0-rc1, 11.0-preview7 Jul 24, 2026

github-actionsBot locked and limited conversation to collaborators Aug 23, 2026

Uh oh!

Conversation

vcsjones commented Jul 20, 2026

Uh oh!

azure-pipelinesBot commented Jul 20, 2026

Uh oh!

vcsjones commented Jul 20, 2026

Uh oh!

vcsjones commented Jul 20, 2026

Uh oh!

dotnet-policy-serviceBot commented Jul 20, 2026

Uh oh!

CopilotAI left a comment

Choose a reason for hiding this comment

Pull request overview

Copilot's findings

Uh oh!

Uh oh!

tannergooding commented Jul 20, 2026• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

bartonjs commented Jul 20, 2026

Uh oh!

vcsjones commented Jul 21, 2026

Uh oh!

tannergooding left a comment

Choose a reason for hiding this comment

Uh oh!

vcsjones commented Jul 21, 2026

Uh oh!

tannergooding commented Jul 21, 2026

Uh oh!

CopilotAI left a comment

Choose a reason for hiding this comment

Copilot's findings

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

tannergooding commented Jul 20, 2026•
edited
Loading