Enable SymbolicSgdLogisticRegression (SymSgdNative) on arm64

[vladimir-aubrecht]

Summary

Enables the SymbolicSgdLogisticRegression trainer (backed by SymSgdNative) on arm/arm64, where Intel MKL is unavailable. Fixes #5798.

SymSGD uses only four CBLAS routines (sdot, saxpy, sdoti, saxpyi). This PR provides them via a tiny, self-contained libMklImports shim built from portable C — with no external BLAS dependency — so it configures and links in the CI cross-compilation sysroots (which ship no OpenBLAS/BLAS).

Changes

• src/Native/MklImportsArm/ (new): implements the four CBLAS routines as plain C loops (-O3 autovectorizes the dense paths to NEON) plus MKL DFTI stubs. No find_package(BLAS) — zero external deps. Symbols are exported explicitly because the native build uses -fvisibility=hidden.
• src/Native/CMakeLists.txt: build MklImportsArm + SymSgdNative on arm.
• src/Native/SymSgdNative/CMakeLists.txt: link SymSgdNative against the shim on arm.
• src/Native/SymSgdNative/SparseBLAS.h: make the CBLAS calling convention portable (__cdecl only on _WIN32).
• Directory.Build.targets: ship libMklImports and libSymSgdNative next to the managed assemblies on arm.
• SymSgdClassificationTrainer.cs: marshal LearnAll's native bool parameters as UnmanagedType.I1. The default 4-byte bool marshalling corrupts later stack arguments and segfaults on arm64 (works on x64 by luck).

Testing

Built native (Release + Debug) and ran the SymSGD trainer tests on Apple Silicon (arm64):

• TestEstimatorSymSgdClassificationTrainer
• TestEstimatorSymSgdInitPredictor
• SimpleTrainAndPredictSymSGD

3 passed, 0 failed. Without the bool marshalling fix, these crash the test host with a SIGSEGV inside LearnAll.

Notes

• The initial arm enablement scaffolding (MklImportsArm skeleton, CMake gating, SparseBLAS.h calling-convention guard) is based on prior work by @anicka-net (credited as co-author).
• Broader MKL consumers in Microsoft.ML.Mkl.Components (OLS, VectorWhitening) and Microsoft.ML.TimeSeries still require a real BLAS/LAPACK and are out of scope here.

[vladimir-aubrecht]

@dotnet-policy-service agree company="Microsoft"

[Copilot]

Pull request overview

Enables SymbolicSgdLogisticRegression / SymSgdNative on arm/arm64 by adding an ARM-specific MklImports shim (implementing the small subset of CBLAS APIs SymSGD needs), wiring it into the native build, and fixing managed P/Invoke marshalling that can crash on arm64.

Changes:

• Add src/Native/MklImportsArm/ to build a self-contained MklImports shim on arm/arm64 and link SymSgdNative against it.
• Update native build gating so SymSgdNative is built on arm/arm64 and shipped next to managed assemblies.
• Fix LearnAll P/Invoke bool marshalling to avoid stack corruption / SIGSEGV on arm64.

Reviewed changes

Copilot reviewed 7 out of 7 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
src/Native/SymSgdNative/SparseBLAS.h	Makes CBLAS calling convention portable across Windows vs non-Windows builds.
src/Native/SymSgdNative/CMakeLists.txt	Adjusts how MklImports is resolved/linked on arm platforms.
src/Native/MklImportsArm/MklImportsArm.c	Implements the minimal CBLAS subset + DFTI stubs for arm/arm64 as a self-contained shim.
src/Native/MklImportsArm/CMakeLists.txt	Adds CMake target to build/install the ARM MklImports shim.
src/Native/CMakeLists.txt	Enables building SymSgdNative on arm/arm64 and adds MklImportsArm to the build graph.
src/Microsoft.ML.Mkl.Components/SymSgdClassificationTrainer.cs	Fixes P/Invoke bool marshalling for LearnAll to prevent arm64 crashes.
Directory.Build.targets	Ensures MklImports and SymSgdNative are copied for arm/arm64 outputs (no longer removed).

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.