[Bug tree-optimization/58520] New: GCC fails to vectorize loop

[Bug tree-optimization/58520] New: GCC fails to vectorize loop

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58520

            Bug ID: 58520
           Summary: GCC fails to vectorize loop
           Product: gcc
           Version: 4.9.0
            Status: UNCONFIRMED
          Keywords: missed-optimization
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: burnus at gcc dot gnu.org

Created attachment 30886
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=30886&action=edit
C test case

On Intel(R) Xeon(R) CPU E5-2630, the following code doesn't vectorize with GCC
4.9 - but it does with ICC:

  for (yi = 0; yi < ycount - 1; ++yi)
    yheight[yi] = ybins[yi+1] - ybins[yi];

[Bug tree-optimization/58520] GCC fails to vectorize loop

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58520

--- Comment #1 from Tobias Burnus <burnus at gcc dot gnu.org> ---
Created attachment 30887
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=30887&action=edit
Assembler generated by ICC

[Bug tree-optimization/58520] GCC fails to vectorize loop

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58520

Tobias Burnus <burnus at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|---                         |INVALID

--- Comment #2 from Tobias Burnus <burnus at gcc dot gnu.org> ---
INVALID.

Pilot error - I had -fsanitize=address in use. Still, I have another example,
which still fails.

[Bug tree-optimization/58520] GCC fails to vectorize loop

[paolo.carlini at oracle dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58520

--- Comment #3 from Paolo Carlini <paolo.carlini at oracle dot com> ---
By the way, I think that saying which specific command line options you are
using is *much* more important than providing CPU brand and model, because the
optimizers do not magically divine what you hardware exactly supports (like,
eg, availability of vector instructions, etc) just assume by default a very
outdated Pentium-class CPU. Sorry, I pointed out something obvious, but
somebody may find it useful to know.