[lworld+fp16] RFR: 8345053: Add support for FP16 valueOf routines

[Bhavana Kilambi]

On Fri, 17 Jan 2025 17:47:20 GMT, Bhavana Kilambi <bkilambi at openjdk.org> wrote:

> This patch adds intrinsic support for Float16.valueOf() routines - conversions from int/long/double to half float.

Hi Jatin, I have uploaded a new commit with benchmark tests for valueOf() routines. Here are the results on an N1 (128-bit) machine - 


Benchmark                           (vectorDim)  Mode    Cnt	Gain
Float16OpsBenchmark.DoubleToFP16       1024	 thrpt   25     5.47
Float16OpsBenchmark.IntToFP16          1024	 thrpt   25     10.77
Float16OpsBenchmark.LongToFP16         1024	 thrpt   25     2.36


I have observed similar/larger gains on other Arm based machines as well.

Please review the patch. Thanks

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

PR Comment: https://git.openjdk.org/valhalla/pull/1332#issuecomment-2612560341