On Tue, 14 Oct 2025 06:43:05 GMT, Alan Bateman <[email protected]> wrote:

> Are you planning to add a micro benchmark to go with this?

No; this code path is not performance sensitive, but the new code should run 
faster.

Hex floating-point output, while informative for testing and debugging, is not 
commonly used. Subnormal values, the subset of numbers affected here, are rarer 
still, so it is a corner-case squared situation.

>From a recent refactoring in Float16, one of the reviewers questioned the use 
>of regular expressions for this kind of low-level code and it seemed 
>reasonable to replace the idiom with plain string operations.

(If performance were a concern, we would replicate the tricky logic present to 
handle double subnormals.)

-------------

PR Comment: https://git.openjdk.org/jdk/pull/27780#issuecomment-3402564984

Reply via email to