On 2017-03-25 20:59:27 -0700, Andres Freund wrote:
> On 2017-03-25 23:51:45 -0400, Tom Lane wrote:
> > Andres Freund <andres@anarazel.de> writes:
> > > On March 25, 2017 4:56:11 PM PDT, Ants Aasma <ants.aasma@eesti.ee> wrote:
> > >> I haven't had the time to research this properly, but initial tests
> > >> show that with GCC 6.2 adding
> > >>
> > >> #pragma GCC optimize ("no-crossjumping")
> > >>
> > >> fixes merging of the op tail jumps.
> > >>
> > >> Some quick and dirty benchmarking suggests that the benefit for the
> > >> interpreter is about 15% (5% speedup on a workload that spends 1/3 in
> > >> ExecInterpExpr). My idea of prefetching op->resnull/resvalue to local
> > >> vars before the indirect jump is somewhere between a tiny benefit and
> > >> no effect, certainly not worth introducing extra complexity. Clang 3.8
> > >> does the correct thing out of the box and is a couple of percent
> > >> faster than GCC with the pragma.
> >
> > > That's large enough to be worth doing (although I recall you seeing all jumps commonalized). We should probably
dothis on a per function basis however (either using pragma push option, or function attributes).
> >
> > Seems like it would be fine to do it on a per-file basis.
>
> I personally find per-function annotation ala
> __attribute__((optimize("no-crossjumping")))
> cleaner anyway. I tested that, and it seems to work.
>
> Obviously we'd have to hide that behind a configure test. Could also do
> tests based on __GNUC__ / __GNUC_MINOR__, but that seems uglier.
Checking for this isn't entirely pretty - see my attached attempt at
doing so. I considered hiding
__attribute__((optimize("no-crossjumping"))) in execInterpExpr.c behind
a macro (like PG_DISABLE_CROSSJUMPING), but I don't really think that
makes things better.
Comments?
Greetings,
Andres Freund