the aim of this PR was firstly to have a correct implementation of the spec. Speed improvement was a secondary goal, which was intentionally pursued with conventional Java arithmetic and just a few calls to simple methods in the standard library.

Whether the Vector API (which is still in incubation phase) might help in this specific case remains to be explored.

As a first impression, there seem to be no good spots to parallelize using SIMD instructions. But engineers with more SIMD experience might find creative ways to make good use of them.


On 2022-06-07 18:35, LifeIsStrange wrote:
On Mon, 9 May 2022 12:50:31 GMT, Raffaello Giulietti <d...@openjdk.java.net> 

Marked as reviewed by limck...@github.com (no known OpenJDK username).

While we at OpenJDK appreciate constructive reviews from GitHub users not 
registered in the [census](https://openjdk.java.net/census), only officially 
nominated reviewers have the authority to approve this PR.

@rgiulietti friendly ping, noob question: Do you believe some parts of the 
algorithm could exploit explicit SIMD instructions (via the high level [Vector 
API](https://openjdk.java.net/jeps/417))  for even better performance (such as 
https://github.com/wrandelshofer/FastDoubleParser ) ?


PR: https://git.openjdk.java.net/jdk/pull/3402

Reply via email to