Bug 12446 – std.parallelism.amap prefer iteration to indexing

Currently std.parallelism.amap uses the random access range primitives instead of the input range primitives to traverse the per-thread slices. This matters when random access is more expensive than iteration. Iteration, if properly implemented, should always at least as fast as indexing.

(In reply to comment #0) > This matters when random access is more expensive than iteration. > > Iteration, if properly implemented, should always at least as fast as indexing. It occurred to me that the cost of slicing a range could be disproportionate to indexing. Are there any use cases where this is true?

I think the differences should be negligible among the two approaches.

... But there's always a way to test by measuring :o). Try it on a few typical ranges. ("Typical" == cost of random access, if offered, shouldn't be onerous.)

(In reply to comment #4) > ... But there's always a way to test by measuring :o). Try it on a few typical > ranges. ("Typical" == cost of random access, if offered, shouldn't be onerous.) It's not typical array-like random access ranges that will benefit, it's atypical ones where iteration is simple but indexing is non-trivial. For example, my "triangular" random access range which returns the tuples: [0, 0], [1, 0], [1, 1], [2, 0], [2, 1], [2, 2], [3, 0], [3, 1], [3, 2], [3, 3] Iteration is simple, but indexing involves solving a quadratic + some other operations. Even though one might not call this "onerous," there is still enough relative difference for it to be noticeable.

Commits pushed to master at https://github.com/D-Programming-Language/phobos https://github.com/D-Programming-Language/phobos/commit/fc3cda0beb95d5164af5767c9f8134359968d294 Fix issue 12446 - std.parallelism.amap: prefer iteration to indexing https://github.com/D-Programming-Language/phobos/commit/35617d87b30a3bf17d3dc50bf41bac4c4627c12f Merge pull request #2042 from Safety0ff/fix12446 Fix issue 12446 - std.parallelism.amap: prefer iteration to indexing