RFR: 8351016: RA support for EVEX to REX/REX2 demotion to optimize NDD instructions [v2]

[Jatin Bhateja]

On Tue, 26 Aug 2025 23:37:01 GMT, Sandhya Viswanathan <sviswanathan at openjdk.org> wrote:

>> Jatin Bhateja has updated the pull request with a new target base due to a merge or a rebase. The pull request now contains four commits:
>> 
>>  - Updating as per reivew suggestions
>>  - Merge branch 'master' of http://github.com/openjdk/jdk into JDK-8351016
>>  - Some refactoring
>>  - 8351016: RA support for EVEX to REX/REX2 demotion to optimize NDD instructions
>
> src/hotspot/share/opto/chaitin.cpp line 1655:
> 
>> 1653:     };
>> 1654: 
>> 1655:     if (X86_ONLY(UseAPX) NOT_X86(false)) {
> 
> The change looks to be generically applicable and not APX or X86 specific.

Hi @sviswa7, I have generalized the fix by lifting X86/APX checks as per the suggestion. Though, our intent here is to facilitate the demotion of NDD instructions having 4 byte EEVEX prefix, in other scenarios of 3-operand instructions, we may not see any benefit from biasing. If a use's live range (LRG) surpasses its user's LRG then, RA automatically prevents sharing of register,  in other case **it may** assign the same register to definition as per first allocation policy. Thus, biasing is only favorable to APX NDD use case where assembler layer is equipped to perform demotion.

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

PR Review Comment: https://git.openjdk.org/jdk/pull/26283#discussion_r2362786986