Re: ping [PATCH] [4.3 projects] vectorization of induction

[Dorit Nuzman <DORIT at il dot ibm dot com>]

Zdenek Dvorak <rakdver@atrey.karlin.mff.cuni.cz> wrote on 15/01/2007
12:41:48:

> Hello,
>
> > http://gcc.gnu.org/ml/gcc-patches/2007-01/msg00005.html
>
> > *************** vect_is_simple_reduction (struct loop *l
> > *** 2048,2054 ****
> >      */
> >     def1 = SSA_NAME_DEF_STMT (op1);
> >     def2 = SSA_NAME_DEF_STMT (op2);
> > !   if (!def1 || !def2)
> >       {
> >         if (vect_print_dump_info (REPORT_DETAILS))
> >           {
> > --- 2041,2047 ----
> >      */
> >     def1 = SSA_NAME_DEF_STMT (op1);
> >     def2 = SSA_NAME_DEF_STMT (op2);
> > !   if (!def1 || !def2 || IS_EMPTY_STMT (def1) || IS_EMPTY_STMT (def2))
> >       {
> >         if (vect_print_dump_info (REPORT_DETAILS))
> >           {
>
> Can it really happen that def1 or def2 is NULL?
>

I'm note sure. I may have stumbled upon that once - I'll see if I can find
the testcase.

> >   (vectorizable_type_demotion): Use def-type of stmt argument rather
> >       than dummy def-type.
>
> What does this change?
>

I think this was more of a cleanup - even if functionality doesn't change -
just send the right def-type instead of a made-up one. I don't remember why
it was ever otherwise to begin with. I'll check, just in case, if it
actually has an effect on any testcase.

> > tested on the vectorizer testcases on powerpc-linux.
>
> Full regtesting is required, IMHO.
>

sure, and it was done:
http://gcc.gnu.org/ml/gcc-patches/2007-01/msg00075.html

> Otherwise, the patch looks OK to me (although obviously, it introduces a
new
> functionality, so I cannot approve it).
>

thanks for reviewing!

> It might help to submit the formatting-only changes separately the next
time,
> to make the patch easier to read (I think you can commit such changes
without
> approval).
>

sorry about that - will do in the future,

dorit

> Zdenek