[Bug tree-optimization/108752] New: word_mode vectorization is pessimized by hard limit on nunits

[Bug tree-optimization/108752] New: word_mode vectorization is pessimized by hard limit on nunits

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108752

            Bug ID: 108752
           Summary: word_mode vectorization is pessimized by hard limit on
                    nunits
           Product: gcc
           Version: 13.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: rguenth at gcc dot gnu.org
  Target Milestone: ---

r13-5771-gdc87e1391c55c6 re-introduced a hard nunits limit to the vectorizer
when using emulated vectors (aka word_mode vectorization).  That's because
this feature relies on vector lowering to implement plus, minus and negate
with bit operations and that has such limit in place for when dealing with
user written code that didn't have any cost modeling applied.

The fix is to emit supported operations from the vectorizer.

[Bug tree-optimization/108752] word_mode vectorization is pessimized by hard limit on nunits

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108752

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Assignee|unassigned at gcc dot gnu.org      |rguenth at gcc dot gnu.org
           Keywords|                            |missed-optimization
             Status|UNCONFIRMED                 |ASSIGNED
   Last reconfirmed|                            |2023-02-10
     Ever confirmed|0                           |1

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Mine, hopefully for GCC 14.

[Bug tree-optimization/108752] word_mode vectorization is pessimized by hard limit on nunits

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108752

--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
Created attachment 54447
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=54447&action=edit
prototype

Prototype patch.  Would benefit from a vect_finish_stmt_generation with
a gimple_seq overload and using gimple_build and some interleaved comments.

[Bug tree-optimization/108752] word_mode vectorization is pessimized by hard limit on nunits

[hp at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108752

Hans-Peter Nilsson <hp at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |hp at gcc dot gnu.org

--- Comment #3 from Hans-Peter Nilsson <hp at gcc dot gnu.org> ---
(In reply to Richard Biener from comment #0)
> emulated vectors (aka word_mode vectorization).

Ackchyually, more commonly known as SWAR: https://en.wikipedia.org/wiki/SWAR.
(IWBN if options and identifiers were keyed off that acronym.)

[Bug tree-optimization/108752] word_mode vectorization is pessimized by hard limit on nunits

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108752

--- Comment #4 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Richard Biener <rguenth@gcc.gnu.org>:

https://gcc.gnu.org/g:821ef93976e750c118d42a2ad33b96dbd1b9f3a5

commit r14-322-g821ef93976e750c118d42a2ad33b96dbd1b9f3a5
Author: Richard Biener <rguenther@suse.de>
Date:   Fri Feb 10 13:09:10 2023 +0100

    tree-optimization/108752 - vectorize emulated vectors in lowered form

    The following makes sure to emit operations lowered to bit operations
    when vectorizing using emulated vectors.  This avoids relying on
    the vector lowering pass adhering to the exact same cost considerations
    as the vectorizer.

            PR tree-optimization/108752
            * tree-vect-generic.cc (build_replicated_const): Rename
            to build_replicated_int_cst and move to tree.{h,cc}.
            (do_plus_minus): Adjust.
            (do_negate): Likewise.
            * tree-vect-stmts.cc (vectorizable_operation): Emit emulated
            arithmetic vector operations in lowered form.
            * tree.h (build_replicated_int_cst): Declare.
            * tree.cc (build_replicated_int_cst): Moved from
            tree-vect-generic.cc build_replicated_const.

[Bug tree-optimization/108752] word_mode vectorization is pessimized by hard limit on nunits

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108752

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
         Resolution|---                         |FIXED
   Target Milestone|---                         |14.0

--- Comment #5 from Richard Biener <rguenth at gcc dot gnu.org> ---
Fixed for GCC 14.

[Bug tree-optimization/108752] word_mode vectorization is pessimized by hard limit on nunits

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108752

--- Comment #6 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Richard Biener <rguenth@gcc.gnu.org>:

https://gcc.gnu.org/g:9d5034aad9868bed45472ef5bfac22dd9ac0e0cb

commit r14-1136-g9d5034aad9868bed45472ef5bfac22dd9ac0e0cb
Author: Richard Biener <rguenther@suse.de>
Date:   Tue May 23 15:58:52 2023 +0200

    Generic vector op costing adjustment

    This is a small adjustment to the work done for PR108752 and
    better reflects the cost of the generated sequence.

            PR tree-optimization/108752
            * tree-vect-stmts.cc (vectorizable_operation): For bit
            operations with generic word_mode vectors do not cost
            an extra stmt.  For plus, minus and negate also cost the
            constant materialization.