[API] Support np.where via ILKernelGenerator#606

Merged

Nucs merged 19 commits intomasterfrom

Apr 20, 2026

Commits on Apr 20, 2026

feat(where): Add IL-generated SIMD optimization for np.where(condition, x, y)
Nucs
committed
perf(where): AVX2/SSE4.1 optimize mask expansion in np.where kernel
Nucs
committed
perf(where): inline mask creation in IL - 5.4x faster kernel
Nucs
committed
fix(where): implement NumPy 2.x NEP50 type promotion for np.where
Nucs
committed
feat(asanyarray): support all built-in C# collection types
Nucs
committed
feat(asanyarray): add non-generic IEnumerable/IEnumerator fallback
Nucs
committed
refactor(asanyarray): consolidate duplicate code
Nucs
committed
fix(asanyarray): add Tuple/ValueTuple support, fix empty collection handling
Nucs
committed
fix(asanyarray): add NumPy-like type promotion for mixed-type collections
Nucs
committed
perf(asanyarray): optimize non-generic collection conversion ~4x faster
Nucs
committed
perf(asanyarray): add ToArrayFast with CollectionsMarshal/CopyTo optimization
Nucs
committed
perf(asanyarray): use GC.AllocateUninitializedArray to skip zeroing
Nucs
committed
perf(asanyarray): optimize with CollectionsMarshal.AsSpan and early exit
Nucs
committed
refactor(tests): migrate np.where and np.asanyarray tests from TUnit to MSTest v3
Nucs
committed
fix(asanyarray): handle object[] via type-promotion path
Nucs
committed
cleanup(where, asanyarray): remove dead code and trim noisy comments
Nucs
committed
fix(asanyarray,where): pure-float object[] promotion + hot-path short-circuits
Nucs
committed
refactor(where): consolidate reflection cache into partial CachedMethods
Nucs
committed
fix(where): gate x86-specific SIMD path on Sse41/Avx2 for ARM64 compatibility
Nucs
committed