SIMDe is a header-only library that provides portable implementations of SIMD intrinsics on hardware which does not natively support them.
The library solves the problem of porting SIMD code across different processor architectures. When hardware natively supports an instruction set, SIMDe uses the native implementation with no performance penalty. When native support is unavailable, the library provides fallback implementations. This approach allows developers to write code using a single SIMD intrinsic API that works across x86, ARM, PowerPC, and other architectures, with the library automatically selecting the best available implementation at compile time.
Developers should choose SIMDe when porting performance-critical code to multiple architectures without rewriting for each target. It suits projects that need to support both x86 and ARM, or that require testing SIMD code for instruction sets not available on the development machine. The library's comprehensive coverage of instruction set functionality distinguishes it from abstraction layers that limit themselves to common denominators. Developers can keep SSE, AVX, and NEON implementations side-by-side in the same codebase, gradually optimizing performance-critical sections for each architecture while maintaining portability through SIMDe's fallbacks.
The project maintains active development with regular additions to its instruction set coverage. Implementation strategies include accelerated paths using cross-ISA intrinsics, compiler-specific vector extensions and built-ins, and compiler auto-vectorization hints via OpenMP, Cilk Plus, and GCC or Clang pragmas. The project provides an online testing environment through Compiler Explorer integration.
