yongster commented on PR #11245: URL: https://github.com/apache/arrow-rs/pull/11245#issuecomment-6001837415
Closing this. #11219 already routes infallible numeric casts, including Float16 ↔ Float32, through `unary`, so the `unary_opt` and validity-bitmap cost described here is gone. What this PR still did was switch those two directions to `half`'s slice conversion. `arrow-cast` depends on `half` with default features off, so there is no runtime `fp16` / `f16c` detection. On a normal build that path is still a scalar loop, and the 3.4×–6.3× numbers were measured against the old `unary_opt` path. I don't think the special case is worth landing without a new measurement against current main. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
