On Fri, 2020-05-01 at 12:52 +0100, Richard Earnshaw wrote: > On 01/05/2020 12:01, Kyrylo Tkachov wrote: > > Hi JangNing (please reply inline in the future as that is the preferred > > style > > on this mailing list) > > > > > -----Original Message----- > > > From: JiangNing OS <jiangn...@os.amperecomputing.com> > > > Sent: 01 May 2020 11:49 > > > To: Richard Earnshaw <richard.earns...@foss.arm.com>; Kyrylo Tkachov > > > <kyrylo.tkac...@arm.com>; Andrew Pinski <pins...@gmail.com>; Florian > > > Weimer <fwei...@redhat.com> > > > Cc: gcc-patches@gcc.gnu.org; nmeye...@amzn.com > > > Subject: RE: Should ARMv8-A generic tuning default to -moutline-atomics > > > > > > In reality, a lot of users are still using old gcc versions running on new > > > hardware. OpenJDK is a typical example, I think. > > > > Although this option is not an ABI change and thus I don't expect it to > > break > > anything by design, it is definitely good to be cautious as it is a > > significant change in code generation. Let's wait for a few weeks to make > > sure the default change in GCC 10.1 works for everyone and then revisit the > > backport story. In any case, GCC 9 and 8 are not extremely close to release > > currently IIUC. > > Frankly, I don't think there is anything to consider. We should not > change codegen for an existing release series unless there is a bug. It > upsets users who expect stability in dot releases. > > There's an option that can be used to change the behaviour, and that's > enough. I tend to agree. ISTM that the best course of action for the older releases and distros is to have the options available and the distros can determine on their own if they want to flip the default by configuring GCC differently or just build packages with the magic option.
Jeff
