RFR: 8241502: Migrate x86_64.ad to MacroAssembler [v6]
Vladimir Ivanov
vlivanov at openjdk.java.net
Thu Mar 25 11:30:39 UTC 2021
On Thu, 25 Mar 2021 11:24:21 GMT, Vladimir Ivanov <vlivanov at openjdk.org> wrote:
>> Looks good.
>
>> I didn't remove the, now unused, enc_classes and related code; I think it would be better to keep them for a few more days to make easier any eventual debugging. I'll be happy to remove them if you think that's a better approach.
>
> Can you elaborate, please, what you mean here? I don't see "enc_classes and related code" in the latest patch.
> I guess you counted an instruction validated if it was generated at least once (so at least 1 possible encoding)? On RISC architectures that would probably be enough, but I know on x86/x64, register choice can change the instruction encoding and length, thanks to prefixes, etc.
One possibility is to write a native test (using gtest) and exercise some encoding methods by enumerating all the possible input value locations. Probably, too much for a regularly executed test (there's a lot of stuff missing right now to support such scenario), but something which can be quickly hacked for a one-time run.
-------------
PR: https://git.openjdk.java.net/jdk/pull/2420
More information about the hotspot-compiler-dev
mailing list