> On Mon, Nov 25, 2013 at 06:15:50PM +0100, Jan Hubicka wrote:
> > > > What's the reason you cannot defer SIMD cloning to LTRANS stage
> > > > as simple IPA pass next to IPA-PTA?
> > >
> > > Ok, deferring till after IPA-PTA was easy, just small ipa-cp.c changes
> > > (look at the attribute rather than simd*clone* fields), passes.def and
> > > had to tweak ipa_add_new_function which assumed that all new functions
> > > must be definitions with gimple body.
> >
> > Note that any small IPA pass at ltrans will increase peak memory use of
> > ltrans copmilation by loading all function bodies into memory (since
> > IPA transformations needs to be applied first).
> >
> > It would be nice to avoid these enabled by default unless we have really
> > good reason for it.
>
> bool gate () { return flag_openmp || flag_openmp_simd
> || flag_enable_cilkplus; }
>
> isn't exactly enabled by default ;)
OK :))
> Anyway, all the pass needs is bodies of functions with "omp declare simd"
> attribute which will be defined in the current partition, for functions
> defined in other partitions all it wants to do is just clone the
> DECL_EXTERNAL FUNCTION_DECL. But it needs to be called before any
> caller's of that function (whether extern/in another partition, or local
> (defined in the current partition) will run through vectorization.
Yep, we will need to add an interface for late passes that needs to look
only into specific bodies. (in fact, I already added cgraph_get_body
and perhaps I can just integrate IPA transformation into that and make late IPA
passes
to use them)
Honza
>
> Jakub
