### Current Behavior: See [this forum thread](https://discourse.mc-stan.org/t/improving-the-efficiency-of-dot-product-for-trinary-and-redundant-matrices/29940/10?u=wardbrian) by @mike-lawrence which contains an example that is ~5x _slower_ at `--O1` compared to the default. ### Expected Behavior: SoA optimization should not promote things being assigned to inside of a tight `for` loop