[Bug tree-optimization/116610] New: wrong-code with SLP induction vectorization when using partial vectors and alignment peeling via masking

[Bug tree-optimization/116610] New: wrong-code with SLP induction vectorization when using partial vectors and alignment peeling via masking

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116610

            Bug ID: 116610
           Summary: wrong-code with SLP induction vectorization when using
                    partial vectors and alignment peeling via masking
           Product: gcc
           Version: 15.0
            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: ---

gcc.target/aarch64/sve/peel_ind_1.c when forcing single-lane SLP vectorizes
using VNx4SI and

peel_ind_1.c:16:34: note:  misalignment for fully-masked loop: 1

but the initial value for the induction isn't adjusted:

  # vect_vec_iv_.5_2 = PHI <_11(5), { 5, 10, 15, 20, 25, 30, 35, 40 }(2)>

[Bug tree-optimization/116610] wrong-code with SLP induction vectorization when using partial vectors and alignment peeling via masking

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116610

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
     Ever confirmed|0                           |1
             Status|UNCONFIRMED                 |ASSIGNED
           Assignee|unassigned at gcc dot gnu.org      |rguenth at gcc dot gnu.org
   Last reconfirmed|                            |2024-09-05

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Mine.

[Bug tree-optimization/116610] wrong-code with SLP induction vectorization when using partial vectors and alignment peeling via masking

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116610

--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
The issue is quite old btw.

[Bug tree-optimization/116610] wrong-code with SLP induction vectorization when using partial vectors and alignment peeling via masking

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116610

--- Comment #3 from GCC 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:6a1a856ba78589f7f5285b00ecd40ba2bbeef8b0

commit r15-3507-g6a1a856ba78589f7f5285b00ecd40ba2bbeef8b0
Author: Richard Biener <rguenther@suse.de>
Date:   Thu Sep 5 11:18:57 2024 +0200

    tree-optimization/116610 - wrong SLP induction bias for mask peeling

    The following fixes a mistake when applying the bias for peeling via
    masking to the inital value of SLP inductions.

    This resolves gcc.target/aarch64/sve/peel_ind_1.c (a scan-assembler
    only unfortunately) when forcing single-lane SLP for it.

            PR tree-optimization/116610
            * tree-vect-loop.cc (vectorizable_induction): Use MINUS_EXPR
            to apply a mask peeling adjustment.

[Bug tree-optimization/116610] wrong-code with SLP induction vectorization when using partial vectors and alignment peeling via masking

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116610

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
      Known to work|                            |15.0

--- Comment #4 from Richard Biener <rguenth at gcc dot gnu.org> ---
Fixed on trunk sofar, latent on branches - queued for backporting.

[Bug tree-optimization/116610] wrong-code with SLP induction vectorization when using partial vectors and alignment peeling via masking

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116610

--- Comment #5 from GCC Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-14 branch has been updated by Richard Biener
<rguenth@gcc.gnu.org>:

https://gcc.gnu.org/g:cacc976a71027e7da8e3438b60da76ecdf990d38

commit r14-10684-gcacc976a71027e7da8e3438b60da76ecdf990d38
Author: Richard Biener <rguenther@suse.de>
Date:   Thu Sep 5 11:18:57 2024 +0200

    tree-optimization/116610 - wrong SLP induction bias for mask peeling

    The following fixes a mistake when applying the bias for peeling via
    masking to the inital value of SLP inductions.

    This resolves gcc.target/aarch64/sve/peel_ind_1.c (a scan-assembler
    only unfortunately) when forcing single-lane SLP for it.

            PR tree-optimization/116610
            * tree-vect-loop.cc (vectorizable_induction): Use MINUS_EXPR
            to apply a mask peeling adjustment.

    (cherry picked from commit 6a1a856ba78589f7f5285b00ecd40ba2bbeef8b0)

[Bug tree-optimization/116610] wrong-code with SLP induction vectorization when using partial vectors and alignment peeling via masking

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116610

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
      Known to work|                            |14.2.1
         Resolution|---                         |FIXED
             Status|ASSIGNED                    |RESOLVED

--- Comment #6 from Richard Biener <rguenth at gcc dot gnu.org> ---
Fixed, not backporting further - I think the code is basically unreachable w/o
single-lane SLP.