[vectorIntrinsics] RFR: 8265817: Add LoadVectorGather nodes implementation for Arm SVE

[Wang Huang]

On Sun, 25 Apr 2021 08:29:44 GMT, Xiaohong Gong <xgong at openjdk.org> wrote:

>> * Add LoadVectorGather nodes implementation for Arm SVE
>> * add new instruction `sve_ld1w_gather`[1] & `sve_ld1d_gather`[2]
>> * move a small code into `op_sve_supported` as @nsjian suggested[3]
>> * some code styles fix 
>> [1] https://developer.arm.com/documentation/ddi0596/2020-12/SVE-Instructions/LD1W--scalar-plus-vector---Gather-load-unsigned-words-to-vector--vector-index--
>> [2] https://developer.arm.com/documentation/ddi0596/2020-12/SVE-Instructions/LD1D--scalar-plus-vector---Gather-load-doublewords-to-vector--vector-index--
>> [3]https://github.com/openjdk/panama-vector/pull/70#discussion_r618240204
>
> src/hotspot/cpu/aarch64/aarch64_sve.ad line 4115:
> 
>> 4113:   %}
>> 4114:   ins_pipe(pipe_slow);
>> 4115: %}
> 
> Thanks for your contribution! This patch looks good to me. I just have one question: since currently the SVE supports the vector operations for vector length smaller than the max vector size, does it need to  consider the partial implementation for the `LoadVectorGather `op? Thanks!

Thank you for your review. In my opinion, partial version of `LoadVectorGather` is useful. However, I don't know the plan of `mask`  work of SVE, so I have no plan to implement that now.

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

PR: https://git.openjdk.java.net/panama-vector/pull/72