Multi-output broadcast / map

[willtebbutt]

Currently, when broadcast and map are used to apply a function f which accepts multiple inputs to the appropriate number of arrays of data, each input is treated individually as Julia does not appear to have built-in functionality for performing a multiple-output broadcast / map. This strategy will incur an increased number of memops if it is not the case that all of the input and output data can be cached (L2?), specifically, if N is the number of inputs and P the length of the arrays being broadcasted / mapped over, we incur an extra O(PN) memops due to the need to re-cache the output for each input. Furthermore, we incur an extra O(PN) integer arithmetic operations due to the extra for-loops required. Having Julia implement multi-output map and broadcast operations would therefore yield significant computational savings in certain situations.

[willtebbutt]

See this Issue in the Julia repo: JuliaLang/julia#22129