The ARM NEON qdmlal (saturating doubling multiply‑add long) instructions are missing saturation behavior…
The ARM NEON qdmlal (saturating doubling multiply‑add long) instructions are missing saturation behavior…: a task in LegoFlow-SWE (Harbor dataset). Observed behavior When vqdmlal s16 is called with accumulator values near INT32 MAX and the doubling multiply produces a product that would overflow…
The task
**Observed behavior** When `vqdmlal_s16` is called with accumulator values near `INT32_MAX` and the doubling multiply produces a product that would overflow `int32_t`, the current implementation wraps around (e.g., returns a negative number) instead of clamping to `INT32_MAX`. Values that should saturate to…
Part of Lego-X/LegoFlow-SWE.