Memory & SIMD

OaMemcpy — AVX2 Memory Operations

Always use OaMemcpy instead of std::memcpy. 5.1x faster on large buffers (64MB+). Non-temporal streaming for buffers >2MB.

• OaMemcpy(dst, src, size) — general purpose
• OaMemcpyStream(dst, src, size) — non-temporal for >2MB
• OaMemzero(dst, size) — fast zero-fill
• OaMemEqual(a, b, size) — AVX2 comparison

OaSimd — Google Highway

• OaSimdDotF32(a, b, n) — vectorized dot product
• OaSimdScaleF32(data, factor, n) — vectorized scale
• OaSimdMemzero(dst, size) — SIMD zero-fill