RFR: 8293198: [vectorapi] Improve the implementation of VectorMask.indexInRange()

[Xiaohong Gong]

On Thu, 19 Jan 2023 03:51:28 GMT, Quan Anh Mai <qamai at openjdk.org> wrote:

>> This "offset < 0" path can be optimized out by compiler if the "offset >= 0" in the common cases (i.e. normal loop with no tail loop).
>
> Yes but I don't think it is common enough to deserve special treatment. A fast path should be common for this branch otherwise we are regressing the common path for the corner cases.

The special handling here is necessary for me, because we do not have the instruction supports in compiler intrinsification. Besides, the original API implementation also has the special handle for "offset < 0", which means this API itself accepts the negative offset as the valid argument.  We have to cover it here as well to make the implementation comprehensive.

 Maybe I misunderstood what you mean. Could you please give an alternative solution besides  the current implementation for such corner cases? Thanks in advance!

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

PR: https://git.openjdk.org/jdk/pull/12064