Yes, this was spotted and fixed already. Annoyingly the test I added didn't detect this so GHA was green, but it failed some tier2 tests on aarch64. I added extra safeguards by predicating matching the encode_iso_array instructions on the node being !ascii, which will cause C2 to report an error rather than silently using the iso variant for ascii-only nodes.
Hämta Outlook för Android<https://aka.ms/AAb9ysg> ________________________________ From: core-libs-dev <core-libs-dev-r...@openjdk.java.net> on behalf of Volker Simonis <simo...@openjdk.java.net> Sent: Tuesday, September 28, 2021 1:53:22 PM To: core-libs-dev@openjdk.java.net <core-libs-dev@openjdk.java.net>; hotspot-compiler-...@openjdk.java.net <hotspot-compiler-...@openjdk.java.net>; nio-...@openjdk.java.net <nio-...@openjdk.java.net> Subject: Re: RFR: 8274242: Implement fast-path for ASCII-compatible CharsetEncoders on x86 [v5] On Tue, 28 Sep 2021 10:01:43 GMT, Claes Redestad <redes...@openjdk.org> wrote: >> Not too much work. I recently introduced platform-specific `matcher_*.hpp` >> files, so since then adding a boolean constant is easy (no need to muck with >> the .ad files). > > Does the changes in 9800a99 resolve your concerns? In principle yes, but shouldn't the condition read: if (!Matcher::match_rule_supported(Op_EncodeISOArray) || !Matcher::supports_encode_ascii_array) return false; I.e. the intrinisc is supported if both conditions are true and not supported if either one of them is false? ------------- PR: https://git.openjdk.java.net/jdk/pull/5621
