How to write C/C++ loops so that GCC, Clang, and MSVC automatically vectorize them with 256-bit AVX2 instructions:

1. Use #pragma omp simd or #pragma clang loop vectorize(enable).
2. Avoid loop-carried data dependencies: Each iteration must be independent of prior iterations.
3. Use __restrict__ pointers: Informs the compiler that output and input memory buffers never overlap in memory.
4. Keep trip counts aligned: Iterating in multiples of 8 floats avoids scalar cleanup loop epilogues.

Compiles to native vaddps / vmulps SIMD instructions with an instant 4x to 8x throughput boost!

Adding __restrict__ is the number one fix for compiler auto-vectorization failures. Compilers are paranoid about pointer aliasing.

Checking generated disassembly in Godbolt Compiler Explorer is the best way to verify if your loops successfully vectorized.