Delegate TakeLeadingKernel
The fancy-index gather along a non-leading axis, in NumPy's MapIter buffer order: for every index
j and outer row o, copies innerSize bytes from
src[(o * maxItem + indices[j]) * innerSize] to dst[(j * outerSize + o) * innerSize].
public delegate void TakeLeadingKernel(byte* src, long* indices, long indicesCount, long outerSize, long maxItem, long innerSize, long tileRows, byte* dst)
Parameters
srcbyte*The source, laid out (outerSize, maxItem, innerSize bytes) C-contiguous.
indiceslong*int64 indices, each already in
[0, maxItem)— not checked here.indicesCountlongThe number of indices (the leading dimension of dst).
outerSizelongThe product of the source dims before the indexed axis.
maxItemlongThe extent of the indexed axis in the source.
innerSizelongBytes per slab (element bytes times the dims after the indexed axis).
tileRowslongSource rows processed per tile (at least 1); sizes the cache-resident read window.
dstbyte*The destination, (indicesCount, outerSize, innerSize bytes) C-contiguous.