[Bug target/115254] New: [15 Regression] GCN regressions from "Avoid splitting store dataref groups during SLP discovery"

[tschwinge at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115254

            Bug ID: 115254
           Summary: [15 Regression] GCN regressions from "Avoid splitting
                    store dataref groups during SLP discovery"
           Product: gcc
           Version: 14.0
            Status: UNCONFIRMED
          Keywords: testsuite-fail
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: tschwinge at gcc dot gnu.org
                CC: ams at gcc dot gnu.org, rguenth at gcc dot gnu.org
  Target Milestone: ---
            Target: GCN

If my tracking is to be believed, the recent commit
r15-812-gc71886f2ca2e46ce1449c7064d6f1b447d02fcba "Avoid splitting store
dataref groups during SLP discovery" is causing the following regressions for
GCN target testing (tested '-march=gfx908'):

    PASS: gcc.dg/vect/slp-cond-2-big-array.c (test for excess errors)
    PASS: gcc.dg/vect/slp-cond-2-big-array.c execution test
    [-PASS:-]{+FAIL:+} gcc.dg/vect/slp-cond-2-big-array.c scan-tree-dump-times
vect "vectorizing stmts using SLP" 3

    gcc.dg/vect/slp-cond-2-big-array.c: pattern found 4 times


    PASS: gcc.dg/vect/slp-cond-2.c (test for excess errors)
    PASS: gcc.dg/vect/slp-cond-2.c execution test
    [-PASS:-]{+FAIL:+} gcc.dg/vect/slp-cond-2.c scan-tree-dump-times vect
"vectorizing stmts using SLP" 3

    gcc.dg/vect/slp-cond-2.c: pattern found 4 times


    PASS: gcc.dg/vect/vect-gather-4.c (test for excess errors)
    [-PASS:-]{+FAIL:+} gcc.dg/vect/vect-gather-4.c scan-tree-dump-not vect
"Loop contains only SLP stmts"

Per commit 85e2ce10f76aee93e43aab6558cf8e39cec911e4 "Fix
gcc.dg/vect/vect-gather-4.c for cascadelake", that one later additionally
changed:

    FAIL: gcc.dg/vect/vect-gather-4.c scan-tree-dump-not vect [-"Loop contains
only SLP stmts"-]{+"vectorizing stmts using SLP"+}

..., but that's not relevant for the original regression.


In other words, if I locally revert that patch, these all PASS again.