On Wed, Feb 24, 2021 at 6:39 PM Segher Boessenkool <seg...@kernel.crashing.org> wrote: > > Hi! > > On Tue, Feb 23, 2021 at 09:48:28AM +0100, Richard Biener wrote: > > On Tue, Feb 23, 2021 at 4:48 AM acsawdey--- via Gcc-patches > > <gcc-patches@gcc.gnu.org> wrote: > > > > > > From: Aaron Sawdey <acsaw...@linux.ibm.com> > > > > > > This patch implements a RTL pass that looks for pc-relative loads of the > > > address of an external variable using the PCREL_GOT relocation and a > > > single load or store that uses that external address. > > > > > > Produced by a cast of thousands: > > > * Michael Meissner > > > * Peter Bergner > > > * Bill Schmidt > > > * Alan Modra > > > * Segher Boessenkool > > > * Aaron Sawdey > > > > > > This incorporates the changes requested in Segher's review. A few things I > > > did not change were the insn-at-a-time scan that could be done with DF, > > > and > > > I did not change to using statistics.[ch] for the counters struct. I did > > > try > > > to improve the naming, and rewrote a number of comments to make them > > > consistent > > > with the code, and generally tried to make things more readable. > > > > > > OK for trunk if bootstrap/regtest passes? > > > > stage1 please? > > Versions of this patch were first submitted in August. Of 2019. > > We have quite a lot of experience with it so far. There are no known > wrong-code (or ICE) problems with it. > > All of this is only when targetting power10 (which isn't shipping > hardware yet), and is behind a flag anyway, so you can easily disable > the optimisation. > > So, there is virtually no risk.
So it's enabled by default when targeting power10? But power10 won't ship before the GCC 11 release? > On the other hand, we do want to get this into the hands of everyone > who should test it, and trunk works so much better for that than private > builds or patch sets. > > If it misses 11.1, we'll have to backport it to 11.2 . I see. > So, I'd really like it to get in now. Do any of these arguments change > your position on that? Well, in the end target maintainers have the say on target specific changes. I guess hitting problems with the patch in testing GCC 11 is unlikely to happen, and if it does we won't block shipping 11.1. I do note that GCC 10 seems to know -mpower10 so technically such "regression" would block a release. Richard. > > Segher