-
-
Notifications
You must be signed in to change notification settings - Fork 5.8k
Extrema is slower than maximum + minimum #31442
Copy link
Copy link
Closed
Labels
compiler:simdinstruction-level vectorizationinstruction-level vectorizationfoldsum, maximum, reduce, foldl, etc.sum, maximum, reduce, foldl, etc.good first issueIndicates a good issue for first-time contributors to JuliaIndicates a good issue for first-time contributors to Juliahelp wantedIndicates that a maintainer wants help on an issue or pull requestIndicates that a maintainer wants help on an issue or pull requestperformanceMust go fasterMust go faster
Description
Metadata
Metadata
Assignees
Labels
compiler:simdinstruction-level vectorizationinstruction-level vectorizationfoldsum, maximum, reduce, foldl, etc.sum, maximum, reduce, foldl, etc.good first issueIndicates a good issue for first-time contributors to JuliaIndicates a good issue for first-time contributors to Juliahelp wantedIndicates that a maintainer wants help on an issue or pull requestIndicates that a maintainer wants help on an issue or pull requestperformanceMust go fasterMust go faster
On AVX-512 systems, calling
extrema(x)is slower than callingmaximumand thenminimum. Looks like we need to incorporate the same techniques from #30320 over to extrema in order to get it to vectorize in a similar manner.