[Bug testsuite/91626] [9/10 Regression] FAIL: gcc.dg/lto/pr48622 c_lto_pr48622_0.o-c_lto_pr48622_0.o link, -O -flto -finline-small-functions -fno-early-inlining

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91626

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Assignee|hubicka at gcc dot gnu.org         |matz at gcc dot gnu.org
          Component|lto                         |testsuite
           Priority|P3                          |P2
                 CC|                            |hubicka at gcc dot gnu.org

--- Comment #6 from Richard Biener <rguenth at gcc dot gnu.org> ---
Micha you added the testcase - it probably fails on hppa64-hpux since
introudction(?) and I don't see how it can link since nothing provides
ashift_qi_1.  On x86 we somehow optimize the program to

main:
.LFB0:  
        .cfi_startproc
        subq    $8, %rsp
        .cfi_def_cfa_offset 16
        call    abort

but only during RTL opts.  Likely because .optimized:

  <bb 2> [local count: 1073741824]:
  if (_28(D) != 255)
    goto <bb 3>; [0.00%]
  else
    goto <bb 4>; [100.00%]

  <bb 3> [count: 0]:
  abort ();

so somehow the UNDEF is simplified (possibly becomes zero through init-regs
or so) but not on hppa.

I think the issue lies in the testcase which either should be changed
to build a shared object or provide a definition of ashift_qi_1.
Verifying it still tests what it was supposed to.