Aldrin M
Also, I see now that `util/hashing.h` wraps xxHash, vendored from Cyan4973's github repo. Hashing32 has the following comment above it: // Implementations are based on xxh3 32-bit algorithm description from:...
> Biggest feedback is to not allocate a large temporary `std::vector` and instead write directly into the buffer you'll be returning. Also you made your TempVectorStack way too big. thanks!...
never mind, you actually mentioned how to do it already. the mobile interface hid it
@ursabot please benchmark lang=C++
before figuring out the conbench stuff, I put some of the local benchmark results in a google slide: https://docs.google.com/presentation/d/1cUU_F3jB6LsOLbClhl34YdQiodtbz7l76l3juHTsC5k/edit#slide=id.g13e9d117f47_0_63 Just wanted to put this in as a sneak peek even...
@ursabot please benchmark lang=C++
sorry, renaming to match the PR back to ARROW-8991. Since I am already trying to accommodate 64-bit, it didn't make sense to keep the smaller scope that the sub-task was...
This PR is ready for review. based on how long this has been opened, I tried to get important coverage through `util/hashing_benchmark.cc` and handle the main data types. This PR...
@ursabot please benchmark lang=C++
I think the absolute fastest way is to break up the columns into different files, in which case you'll have far fewer inefficiencies. I think structurally, IPC and parquet are...