Table of Contents

Delegate PolyCalcKernel

Namespace
NumSharp.Backends.Kernels
Assembly
NumSharp.dll

One numpy.polynomial calculus kernel call (see the file header): loads/scales and differentiates (orders fused orders) or integrates (one order) a C-contiguous working buffer in place.

public delegate void PolyCalcKernel(byte* src, long srcRow, long srcCol, byte* buf, long bufRow, long cols, long n, long orders, long block, byte* scl)

Parameters

src byte*

The source element (row 0, column 0) to LOAD from, or null when the series already sits in the buffer (then only c *= scl runs over it, when the kernel scales).

srcRow long

Byte stride of the source along the series axis (any sign, 0 for a broadcast).

srcCol long

Byte stride of the source between consecutive columns (C order of the other axes).

buf byte*

Buffer row 0: c[0]'s row for a derivative, the row tmp[0] goes to for an integral (c[0] then sits at row 1).

bufRow long

Byte stride between buffer rows (cols * itemsize).

cols long

Columns (the product of the other axes; 1 for a 1-D series, 0 does nothing).

n long

len(c) — coefficients in the series before the first order.

orders long

A derivative's order count, 1 ≤ orders < n (each order shortens the series by one); an integral kernel runs exactly one order and ignores it.

block long

Columns per block (≥ 1): how many columns stay cache-resident through the recurrence.

scl byte*

One element of the SCALE LOOP dtype (NumSharp.Backends.Kernels.PolyCalcKey.ScaleLoop — the coefficient dtype, or the wider loop a promoting strong scalar makes) holding NumPy's scl after its conversion into that loop (read only by a kernel compiled with NumSharp.Backends.Kernels.PolyCalcKey.Scale).