On Tue, 6 Jan 2026 08:52:35 GMT, Eirik Bjørsnøs <[email protected]> wrote:
>> Joe Darcy has updated the pull request incrementally with one additional
>> commit since the last revision:
>>
>> Respond to review feedback and add conclusion.
>
> src/java.base/share/classes/java/lang/Math.java line 2381:
>
>> 2379: @IntrinsicCandidate
>> 2380: public static double fma(double a, double b, double c) {
>> 2381: // Implementation note: this method is intentionally coded in
>
> The note seems currently structured as:
>
> * Actual implementation design
> * Hypothetical/alternative implementation design
> * Performance context
> * Actual implementation design ("-ilities")
>
> The part suggesting how an alternative Java implementation could be made more
> performant seems counterproductive or at the least unnecessary.
>
> I think the structure can be simplified as:
>
> * Performance context
> * Actual implementation design/"-ilities"
>
> Here's a sketch intended to demonstrate the suggested stucture (needs work
> for correctness, grammar, etc.):
>
>
> // The Java implementation below will only be used when intrinsics
> // using hardware fma support is not available.
> // Therefore, simplicity, maintainability, and ease of testing
> // of the code is more important than direct performance.
> // The implementation is intentionally kept straightforward,
> // relying on BigDecimal for numerical computation.
The intention of the PR is to avert future attempts to re-writing Math.fma to
make it "faster"; see
https://mail.openjdk.org/pipermail/core-libs-dev/2025-December/156810.html
-------------
PR Review Comment: https://git.openjdk.org/jdk/pull/29044#discussion_r2665707030
