On Thu, Aug 22, 2019 at 2:16 AM Kyrill Tkachov
<kyrylo.tkac...@foss.arm.com> wrote:
>
> Hi all,
>
> The optimisation to optimise:
> typedef unsigned long long u64;
>
> void bar(u64 *x)
> {
> *x = 0xabcdef10abcdef10;
> }
>
> from:
> mov x1, 61200
> movk x1, 0xabcd, lsl 16
> movk x1, 0xef10, lsl 32
> movk x1, 0xabcd, lsl 48
> str x1, [x0]
>
> into:
> mov w1, 61200
> movk w1, 0xabcd, lsl 16
> stp w1, w1, [x0]
>
> ends up producing two distinct stores if the destination is volatile:
> void bar(u64 *x)
> {
> *(volatile u64 *)x = 0xabcdef10abcdef10;
> }
> mov w1, 61200
> movk w1, 0xabcd, lsl 16
> str w1, [x0]
> str w1, [x0, 4]
>
> because we end up not merging the strs into an stp. It's questionable
> whether the use of STP is valid for volatile in the first place.
It is not valid as the architecture does not require stp to be atomic
so then the order of the stores can change
Thanks,
Andrew Pinski
> To avoid unnecessary pain in a context where it's unlikely to be
> performance critical [1] (use of volatile), this patch avoids this
> transformation for volatile destinations, so we produce the original
> single STR-X.
>
> Bootstrapped and tested on aarch64-none-linux-gnu.
>
> Ok for trunk (and eventual backports)?
>
> Thanks,
> Kyrill
>
> [1]
> https://lore.kernel.org/lkml/20190821103200.kpufwtviqhpbuv2n@willie-the-truck/
>
>
> gcc/
> 2019-08-22 Kyrylo Tkachov <kyrylo.tkac...@arm.com>
>
> * config/aarch64/aarch64.md (mov<mode>): Don't call
> aarch64_split_dimode_const_store on volatile MEM.
>
> gcc/testsuite/
> 2019-08-22 Kyrylo Tkachov <kyrylo.tkac...@arm.com>
>
> * gcc.target/aarch64/nosplit-di-const-volatile_1.c: New test.
>
