توجيه حلقة باستخدام المتجهات

A plain loop touches one element per step. It is clear but leaves your CPU's vector lanes mostly unused.

Vectorizing means each loop step handles a whole pack of elements instead of one. You do more work per iteration.

Instead of one value, you load several at once into a SIMD vector. That single load grabs a slice of your data.

With both chunks in SIMD form, add them in one element-wise step. The math covers every lane at once.

Write the whole SIMD result back in one move with a vector store. One write covers all the lanes you computed.

A vectorized loop advances by the width, not by one. Each pass covers a full pack of elements.

Mojo's vectorize helper runs a closure across the range in vector steps and handles the bookkeeping for you.

You write a small parameterized function that processes one chunk. Mojo calls it with the right width as it sweeps the range.

Call vectorize with your chunk function, the width, and the total size. It does the looping and the tail for you.

The vectorized loop gives identical output to the scalar one but moves through data in big steps, so it finishes sooner.

Always time both versions. Vectorization usually wins big, but a benchmark proves the gain on your real data.

You replace a one-element-per-step loop with SIMD loads and stores. How should the loop advance?

Vectorizing loads and stores whole chunks, steps by the width, and lets vectorize handle the sweep and tail, giving the same result far faster. 🚀