Pipe block operators proposal

Block-template operator

• Excellent parallel
  • map
  • filter
• Good parallel (local accumulators)
  • reduce (makes sure that size >= 1)
  • reduce0 (also takes zero element. Can accept zero-sized stream)
  • group_by (in fact it's reduce by google whitepaper)
  • partition_by (??? должен разделять по условию... формат???)
• Bad parallel
  • sort
  • unique (alias disctinct)
• Shrinkable
  • skip
  • limit
  • any
  • all
• Debug
  • peek(dst)
  • print (==peek(stdout))

Frame operators

• Pack N frames into 1.
• Unpack 1 frame into N. (in fact combos with pack-blocking)

Block operators

• Pack each block to make it operational over N-subblock-frame
  • unroll
  • vectorize

Alignment

• Array of objects
• Object of arrays
  • Object of mixed arrays (fusing same type into single array, partial fusing)

Black magic ???

• Loop Stream Detector size/format tuning
  • some info
  • intel info
• CPU time quant granularity rounding/tuning (enough work for thread)
• CPU cache line tuning