Struct HalfMaxKernel
Half max. Contiguous chunks (stride == 2) run a 4-accumulator unroll that breaks the per-element dependency chain (~1.3× the scalar fold); other strides take the scalar branch. NaN propagates: the kernel aborts the moment a NaN is seen and only flags it (SawNaN).
public readonly struct HalfMaxKernel : INDReducingInnerLoop<HalfMinMaxAccumulator>
- Implements
- Inherited Members
Remarks
VALUE-only: it decides neither WHICH NaN nor which ±0 a NumPy reduction returns, so the engine's float16
np.max / np.min no longer use it — they run NumPyHalfMinMaxReduce, NumPy's sequential HALF loop bit for bit
(Default.Reduction.MinMax.Half.cs). Kept for callers of this public kernel.
Methods
Execute(void**, long*, long, ref HalfMinMaxAccumulator)
public bool Execute(void** dataptrs, long* strides, long count, ref HalfMinMaxAccumulator a)
Parameters
dataptrsvoid**strideslong*countlongaHalfMinMaxAccumulator