[Bug target/81635] New: nvptx SLP test cases regressions

[tschwinge at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=81635

            Bug ID: 81635
           Summary: nvptx SLP test cases regressions
           Product: gcc
           Version: 8.0
            Status: UNCONFIRMED
          Severity: minor
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: tschwinge at gcc dot gnu.org
                CC: rsandifo at gcc dot gnu.org, vries at gcc dot gnu.org
  Target Milestone: ---
            Target: nvptx

I noticed that trunk r249896 "Tweak BB analysis for dr_analyze_innermost"
breaks some of the nvptx SLP test cases recently contributed by Tom.  (Tom
committed the nvptx SLP patches later, but had tested these on an earlier GCC
sources revision, apparently.)

    PASS: gcc.target/nvptx/slp.c (test for excess errors)
    FAIL: gcc.target/nvptx/slp.c scan-assembler ld.v2.u32
    FAIL: gcc.target/nvptx/slp.c scan-assembler st.v2.u32

    PASS: gcc.target/nvptx/slp-2.c (test for excess errors)
    FAIL: gcc.target/nvptx/slp-2.c scan-assembler ld.v2.u64
    FAIL: gcc.target/nvptx/slp-2.c scan-assembler st.v2.u64

I have not analyzed what the actual problem is, whether it is something latent
that r249896 just uncovered, for example.

Testing on top of last week's trunk r250635, I can make these SLP test cases
PASS again (with no other regressions/changes) if reverting trunk r249896, plus
a handful of later patches touching the same code: r249911, r249913, r249914,
r249915, r249916, r249917.