Test Utf8JsonReader code shape for ARM64 regression

[jozkee]

Tests whether the compound pattern rewrites in the single-segment Utf8JsonReader affect the Linux ARM64 Perf_Get.GetUInt64 regression tracked by #131600.

This draft is stacked on #132399 only to retain the preserved #130976 source commits and the benchmark-image libunwind prerequisite. It does not modify #132399.

The benchmark should compare:

• 66b30d95: last measured good source stage
• 43560bc7: first measured bad source stage
• e92960b3: Utf8JsonReader.cs compound patterns restored to their previous code shape

Validation:

• dotnet build src/libraries/System.Text.Json/src/System.Text.Json.csproj
• dotnet build /t:test src/libraries/System.Text.Json/tests/System.Text.Json.Tests/System.Text.Json.Tests.csproj
  • net11.0: 53,694 passed
  • net481: 53,382 passed

Note

This pull request was prepared with GitHub Copilot.

[azure-pipelines]

Azure Pipelines:
Successfully started running 4 pipeline(s).
12 pipeline(s) were filtered out due to trigger conditions.
There may be pipelines that require an authorized user to comment /azp run to run.

[dotnet-policy-service]

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

[Copilot]

Pull request overview

This PR adjusts a few hot-path code shapes (notably in Utf8JsonReader) to help determine whether recent compound-pattern rewrites are implicated in the Linux/ARM64 System.Text.Json.Tests.Perf_Get.GetUInt64 regression tracked by #131600.

Changes:

• Replaces several C# pattern-based comparisons in Utf8JsonReader with equivalent ==/!= and &&/|| forms to restore an earlier code shape.
• Restores a pre-modernization ternary shape for JsonEncodedText.GetHashCode().
• Updates NativeAOT’s libunwind symbol-privatization step to avoid invoking llvm-link as the relocatable linker by preferring ld.lld/ld when necessary.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 1 comment.

File	Description
src/libraries/System.Text.Json/src/System/Text/Json/Reader/Utf8JsonReader.cs	Rewrites several compound/pattern comparisons in number parsing and token classification to different boolean forms for code-shape testing.
src/libraries/System.Text.Json/src/System/Text/Json/JsonEncodedText.cs	Changes GetHashCode() to a ternary form to match a prior code shape.
src/coreclr/nativeaot/Runtime/Full/CMakeLists.txt	Adds linker selection logic for the libunwind privatization custom command to avoid using llvm-link.

Suppressed comments (4)

src/libraries/System.Text.Json/src/System/Text/Json/Reader/Utf8JsonReader.cs:1515

• Use byte literals for these UTF-8 byte checks to avoid char-to-int promotions and to match the rest of the reader's byte-oriented parsing logic.

                if (nextByte != '.' && nextByte != 'E' && nextByte != 'e')
                {
                    _bytePositionInLine += i;
                    ThrowHelper.ThrowJsonReaderException(ref this, ExceptionResource.ExpectedEndOfDigitNotFound, nextByte);
                }
            }

            Debug.Assert(nextByte == '.' || nextByte == 'E' || nextByte == 'e');

src/libraries/System.Text.Json/src/System/Text/Json/Reader/Utf8JsonReader.cs:1539

• These are byte values from the UTF-8 payload; comparing to char literals promotes to int. Prefer (byte)'E'/(byte)'e' here (and in the assert) to keep the parsing logic purely byte-based.

                if (nextByte != 'E' && nextByte != 'e')
                {
                    _bytePositionInLine += i;
                    ThrowHelper.ThrowJsonReaderException(ref this, ExceptionResource.ExpectedNextDigitEValueNotFound, nextByte);
                }
            }

            Debug.Assert(nextByte == 'E' || nextByte == 'e');

src/libraries/System.Text.Json/src/System/Text/Json/Reader/Utf8JsonReader.cs:1628

• Use byte literals for these UTF-8 byte comparisons to avoid implicit promotions and keep parsing code consistent with other (byte)'0' usage in this file.

            nextByte = data[i];
            if (nextByte != '.' && nextByte != 'E' && nextByte != 'e')
            {

src/libraries/System.Text.Json/src/System/Text/Json/Reader/Utf8JsonReader.cs:1707

• Since nextByte is a byte from the UTF-8 data, prefer comparing against byte literals to avoid implicit numeric promotion (and to keep the parsing logic byte-oriented).

            byte nextByte = data[i];
            if (nextByte == '+' || nextByte == '-')
            {

[jozkee]

Comparing the pattern-rewrite commit with the current PR head, which selectively restores the Utf8JsonReader expressions. Both full SHAs were verified against PR metadata and local Git.

@EgorBot -ubuntu24_azure_ampere -pr 132430 -commits 43560bc --filter "System.Text.Json.Tests.Perf_Get.GetUInt64"

Note

This benchmark request was prepared with GitHub Copilot.

[jozkee]

First-pass cumulative bisection of the confirmed Utf8JsonReader code-shape regression. The stages split the ten one-line restores into candidate groups of 1, 4, 3, and 2 changes. Every full SHA was verified against the current PR history and local Git.

@EgorBot -ubuntu24_azure_ampere -pr 132430 -commits d354506,c514b6e24c36c921ccaf7ea6c95f528c52694b63,9d37db381a99a26be0e757f6aa56f96b9adf429c --filter "System.Text.Json.Tests.Perf_Get.GetUInt64"

Note

This benchmark request was prepared with GitHub Copilot.

[Copilot]

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated no new comments.

Suppressed comments (1)

src/coreclr/nativeaot/Runtime/Full/CMakeLists.txt:44

• ld.lld is advertised as the preferred linker when CMAKE_LINKER resolves to llvm-link, but the current find_program(... NO_DEFAULT_PATH) only searches the llvm-link directory. If ld.lld is available on PATH (or via CMake defaults) but not colocated with llvm-link, the build will silently fall back to ld (or even fail if ld isn’t present) despite ld.lld being available. Consider falling back to a default-path find_program for ld.lld before trying ld.

    find_program(NATIVEAOT_PRIVATE_LIBUNWIND_LD_LLD
      NAMES "ld.lld${NATIVEAOT_PRIVATE_LIBUNWIND_LINKER_VERSION}${NATIVEAOT_PRIVATE_LIBUNWIND_LINKER_EXTENSION}" ld.lld
      PATHS "${NATIVEAOT_PRIVATE_LIBUNWIND_TOOL_DIR}"
      NO_DEFAULT_PATH)

[jozkee]

Final split for the recovery boundary from issue #509. fa936cfa restores IsTokenTypeString, which is used by ValueTextEquals; 69b2a7cb restores the Release-effective number-terminator condition in TryGetNumber. Every full SHA was verified against the current PR history and local Git.

@EgorBot -ubuntu24_azure_ampere -pr 132430 -commits d354506,fa936cfa0ec1e7d1f634fa8885fd3a5519b2f432,69b2a7cbfc7dfc8031deacace4bd770255522706 --filter "System.Text.Json.Tests.Perf_Get.GetUInt64"

Note

This benchmark request was prepared with GitHub Copilot.

[jozkee]

ARM64 optimized disassembly at the confirmed regression boundary. This limits BenchmarkDotNet to one warmup and one measured invocation while requesting disassembly for TryGetNumber and GetUInt64. Both full SHAs were verified against the published PR history and local Git.

@EgorBot -ubuntu24_azure_ampere -pr 132430 -commits fa936cf,69b2a7cbfc7dfc8031deacace4bd770255522706 --filter "System.Text.Json.Tests.Perf_Get.GetUInt64" --warmupCount 1 --iterationCount 1 --minIterationCount 1 --maxIterationCount 1 --invocationCount 1 --unrollFactor 1 --envvars "DOTNET_JitDisasm:TryGetNumber GetUInt64" DOTNET_JitDisasmDiffable:1 DOTNET_ReadyToRun:0 DOTNET_TieredCompilation:0

Note

This benchmark request was prepared with GitHub Copilot.