RFR: 8346664: C2: Optimize mask check with constant offset [v3]

[Emanuel Peter]

On Wed, 29 Jan 2025 10:49:21 GMT, Matthias Ernst <duke at openjdk.org> wrote:

>> @mernst-github Thanks for the test update!
>> 
>> I see there is still discussion around this line:
>> `  return zeros > 0 && ((((jlong)1) << zeros) > mask_t->hi_as_long() && mask_t->lo_as_long() >= 0);`
>> 
>> It is indeed difficult to read. Maybe even just some clever names of temporary variables could help. But I leave that up to you ;)
>> 
>> Ping me when you want me to run testing and review again :)
>
> @eme64 After 5+ weeks I'm getting very tired of running after reviewer engagement here. I've been in this business long enough to know that leaving a drive-by comment every few days and kicking the can down the road a little further is a very effective way to signal you're not interested in this contribution. You could have "run testing" a long time ago. If you're overworked, pass it on to a different reviewer but don't string along the contributor. If you care about a contrasting experience, https://github.com/openjdk/jdk/pull/23142 with a much larger design surface took under 2 weeks because of engaged and constructive review. Maybe it might be wise for your group and you to define or revisit goals of engagement.
> 
> This is not a pet peeve feature but a really basic performance fix for foreign memory access; I want Panama to be great, but it hurts credibility if getting segment[i+1] right doesn't appear to be a priority.
> 
> Feel free to integrate this or hit the close button, I think I've done my due diligence and do not intend to engage on it any further. Thanks.

@mernst-github Can you please merge with master? My test launching script has issues with automatic merging.
It is good practice to merge closer to integration anyway, it lowers the risk of undetected during testing bugs.

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

PR Comment: https://git.openjdk.org/jdk/pull/22856#issuecomment-2621795639