This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: auto-vectorization


> Before working on a proper bug report, I wanted to find out whether
> the code is supposed to be functional, has any know issues, etc.

Here is the list of known issues (issues that I'm currently aware of):

1. PR18172 (fix will be sent shortly)
2. PR18173 (alignment issue)
3. PR17589 (possibly related alignment issue)
4. PR18009 (patch submitted, pending review -
            http://gcc.gnu.org/ml/gcc-patches/2004-10/msg01421.html)
5. PR18179 (patch is in the works)
6. PR18181 (patch is in the works)

(There are also PRs for missed optimization; I know of PR17165 and PR16362)

The vectorizer is known to pass SPEC on ppc-darwin, vectorizing over 200
loops (mostly simple initialization loops).

> Ie whether such a report is desired.

yes please.

dorit



                                                                       
                      James Cloos                                      
                      <cloos@jhcloos.co        To:       gcc@gcc.gnu.org
                      m>                       cc:                     
                      Sent by:                 Subject:  auto-vectorization
                      gcc-owner@gcc.gnu                                
                      .org                                             
                                                                       
                                                                       
                      26/10/2004 11:13                                 
                                                                       




What is the status of the auto-vectorization support in 4.0?

I grabbed the latest snapshot yesterday and tried compiling asterisk¹
with it.  (Incidently, before I found -ftree-vectorize in the docs,
the 4.0-compiled codecs were slower as shown by asterisk's show
translations command than the 3.4.2-compiled codecs.)  However, with
the -ftree-vectorize option the compiled crapped out when compiling
certain of the codecs with an internal compiler error.

Before working on a proper bug report, I wanted to find out whether
the code is supposed to be functional, has any know issues, etc.
Ie whether such a report is desired.

-JimC

¹ cf asterisk.org




Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]