[lworld+fp16] RFR: 8338102: x86 backend support for newly added Float16 intrinsics.

[Bhavana Kilambi]

On Mon, 12 Aug 2024 10:40:31 GMT, Jatin Bhateja <jbhateja at openjdk.org> wrote:

>>> FTR, other missing intrinsics are conversion operations (Float16.longValue/intValue/doubleValue etc), x86 FP16 ISA has direct instructions for conversions from / to Float16 values..
>> 
>> I can add implementation for these conversions (if it's not already taken by anyone else or you :)
>
>> > FTR, other missing intrinsics are conversion operations (Float16.longValue/intValue/doubleValue etc), x86 FP16 ISA has direct instructions for conversions from / to Float16 values..
>> 
>> I can add implementation for these conversions (if it's not already taken by anyone else or you :)
> 
> Hi @Bhavana-Kilambi , this will be really helpful. 
> 
> Also, your comments have been incorporated, kindly approve if looks ok.

@jatin-bhateja Thanks for your responses for my questions above. I am still not sure how useful this approach would be for aarch64 as we still would have to generate separate match rules anyway for every operation (unlike x86). But this idea does sound interesting and  I am looking forward to a discussion with other members of OpenJDK community as well on this.

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

PR Comment: https://git.openjdk.org/valhalla/pull/1196#issuecomment-2286269209