RFR: 8268229: Aarch64: Use Neon in intrinsics for String.equals

[Dong Bo]

On Wed, 9 Jun 2021 08:28:37 GMT, Andrew Haley <aph at openjdk.org> wrote:

>> src/hotspot/cpu/aarch64/macroAssembler_aarch64.cpp line 4807:
>> 
>>> 4805:   mov(tmp2, v0, T2D, 1);
>>> 4806:   cbnz(tmp2, DONE);
>>> 4807:   b(SAME);
>> 
>> Shouldn't this be
>> 
>> mov(tmp1, v0, T2D, 0);
>> mov(tmp2, v0, T2D, 1);
>> orr(tmp1, tmp1, tmp2);
>> cbnz(tmp1, DONE);
>> 
>> 
>> ... which would use up fewer branch prediction resources.
>
> ... or maybe do the OR in the vector unit?

I guess it can be done with:

umaxv(v1, T4S, v0);
mov(tmp1, v1, T4S, 0);
cbnz(tmp1, DONE0);

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

PR: https://git.openjdk.java.net/jdk/pull/4423