On Thu, 1 Oct 2020 01:01:23 GMT, Paul Sandoz <psan...@openjdk.org> wrote:
>> Hi @PaulSandoz , >> >> I think it would be better to integrate it [1] in this MR. >> >> I have tested this MR on our AVX512 machines and it still crashes. >> Also, for the sake of maintenance, it seems NOT a good idea to push a >> problematic commit into the jdk main-line repo. >> >> As for the review process, I don't think it's a problem since the fix [1] is >> clear and small enough. >> >> What do you think? >> >> Thanks. >> >> [1] >> https://github.com/openjdk/panama-vector/commit/1af35c357066743935bd3f48ce3610a41761f89a > > @DamonFool I appreciate your efforts on this but i want to hold back on that > issue and follow up very quickly after > integration of this PR. This change has been through an extremely long and > arduous review process, and i want to stick > to what was reviewed and not ask reviewers to go through further cycles on > what overall is a very large change. > Unfortunately this change is in a holding pattern waiting for the CSR to be > approved thereby increasing the window > where we might find further issues (that if we had already integrated may > have been dealt with separately perhaps in a > less timely fashion with respect to that integration). Unless an issue is > extremely severe I think we should queue them > up in `panama-vector/vectorIntrinsics` (there is at least one more for ARM > SVE that is queued up). Since the issue you > describe effects one instruction, for one type, on AVX512, its impact is > limited and will be mitigated by a quick > follow up. Okay. I can understand it. Vector API is very valuable to us. Hope the follow-ups can be integrated as soon as possible. And thank you all for your great work. Best regards, Jie ------------- PR: https://git.openjdk.java.net/jdk/pull/367