[vectorIntrinsics+mask] RFR: 8270349: Initial X86 backend support for optimizing masking operations on AVX512 targets. [v6]

[Xiaohong Gong]

On Thu, 12 Aug 2021 12:20:23 GMT, Jatin Bhateja <jbhateja at openjdk.org> wrote:

>> Ok, can the opcodes like `AndVMask` be blocked for non-predicated targets in `Matcher::match_rule_supported` ?
>
> Matcher::match_rule_supported only check the existence of pattern based on the opcode and default return value is true. 
> AddVI v1 v2
> AddVI v1 v2 mask
> 
> Where as match_rule_supported_vector_masked initially calls match_rule_supported, it then return false in default case and true only for operations for which masked patterns exist.

But I think `AndVMask/OrVMask/XorVMask` are different from the masked vector opcodes like `AddVI`. I was intending to make nodes like `AndVMask/OrVMask/XorVMask` also work for the non-mask supported platforms. Other platforms can also support them by adding the backend implementations if they want. We can just treat them as a normal vector operations for vector mask.

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

PR: https://git.openjdk.java.net/panama-vector/pull/99