RFR: 8370013: Refactor Double.toHexString to eliminate regex and StringBuilder
Raffaello Giulietti
rgiulietti at openjdk.org
Mon Oct 20 15:06:08 UTC 2025
On Wed, 15 Oct 2025 00:02:58 GMT, Shaojin Wen <swen at openjdk.org> wrote:
>> This PR refactors the Double.toHexString method to improve its performance by eliminating the use of StringBuilder and regex operations. The new implementation uses a direct byte array approach to construct the hex string representation, which avoids the overhead of StringBuilder operations and regex pattern matching.
>>
>> Existing tests in `java/lang/Double/ToHexString.java`.
>
> The test results run under Mac Book M1 Pro are as follows
>
> # before 3f11c4307ee1ca918926c8c522d2fc496947980e
> -Benchmark Mode Cnt Score Error Units
> -Doubles.toHexString avgt 15 228.383 ± 17.172 ns/op
>
> # after c18f7632f2f2cf0752b3f4651d4cadf95bb859c6
> +Benchmark Mode Cnt Score Error Units
> +Doubles.toHexString avgt 15 15.138 ± 1.070 ns/op
>
> More than 10 times performance improvement
@wenshao What's wrong with my other suggestions?
In particular, the `abs()` line can be moved after the check for `0.0`, and I cannot see the need for the `0x1000_0000_0000_0000L` bit in `signifBits`.
-------------
PR Comment: https://git.openjdk.org/jdk/pull/27811#issuecomment-3422475171
More information about the core-libs-dev
mailing list