[Bug target/63848] [5 Regression] FAIL: c-c++-common/torture/builtin-arith-overflow-17.c -O0 execution test

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63848

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |ASSIGNED
   Last reconfirmed|                            |2014-11-21
           Assignee|unassigned at gcc dot gnu.org      |jakub at gcc dot gnu.org
     Ever confirmed|0                           |1

--- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Created attachment 34069
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=34069&action=edit
gcc5-pr63848.patch

Untested fix.

Apparently on some targets for some modes emit_cmp_and_jump_insns silently
generates wrong code, so one needs to use do_compare_rtx_and_jump which wraps
that and handles the wider modes.  Unfortunately the two functions have
different order of arguments (and do_compare_rtx_and_jump one additional one).
built-arith-overflow* still passes on x86_64-linux and i686-linux with this,
haven't bootstrapped/regtested it yet (but, on these targets it doesn't make
much difference, as there are even TImode comparison patterns for -m64 and
DImode for -m32).