Explore low level optimizations

[ErikPelli]

With the arrival of Go 1.27, an experimental SIMD package has been introduced.
I found this Zig library (https://github.com/zigcc/zig-msgpack) for MessagePack that uses SIMD, along with other Assembly optimizations, and according to their benchmarks the improvement of performance is between 10% and 90%.

Here is a list of features they added, according to the README:

Optimization Features

    CPU Cache Prefetching
        Platform-specific prefetch instructions (x86: PREFETCH*, ARM: PRFM)
        Intelligently prefetches data before it's needed for containers ≥256 bytes
        Multi-level cache hints (L1/L2/L3) for optimal cache utilization
        Streaming prefetch for non-temporal data access

    SIMD Operations
        Automatic detection of available SIMD features (AVX-512, AVX2, SSE2, NEON)
        Vectorized string comparison (16-64 byte chunks)
        Vectorized memory copying with alignment optimization
        Vectorized byte order conversion (big-endian ↔ little-endian)
        Batch integer array conversions (u32/u64)

    Memory Alignment Optimization
        Automatic alignment detection and fast path selection
        Aligned memory reads/writes for supported types
        Memory alignment preprocessing for better SIMD performance
        Large data copy optimization (≥64 bytes)

    Branch Prediction Optimization
        Hot path annotations for common cases
        Lookup tables for O(1) marker byte conversion (256-entry precomputed table)
        Switch expressions instead of if-else chains (jump table optimization)
        Optimized error handling paths

    HashMap-Based Maps
        O(1) average-case key lookups (vs O(n) linear search)
        Efficient hash function with depth limiting
        Support for any Payload type as keys
        getOrPut optimization (single hash computation)

Maybe should someone explore the possibility to add them to this Go library too?