[Bug tree-optimization/79472] [7 Regression] x86-64: Switch table generation fails if default case has different code

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=79472

--- Comment #3 from Martin Liška <marxin at gcc dot gnu.org> ---
GCC 6:
void frobulate_for_gcc(uint32_t) (uint32_t v)
{
  const char * s;

  <bb 2>:
  switch (v_2(D)) <default: <L3>, case 0: <L7>, case 1: <L1>, case 2: <L2>>

<L1>:
  goto <bb 6> (<L7>);

<L2>:
  goto <bb 6> (<L7>);

<L3>:

  # s_1 = PHI <"foo"(2), "bar"(3), "spam"(4), 0B(5)>
<L7>:
  if (s_1 == 0B)
    goto <bb 7>;
  else
    goto <bb 8>;

  <bb 7>:
  abort ();

  <bb 8>:
  __builtin_puts (s_1);
  return;

}

GCC7:
Bailing out - no common successor to all case label target blocks found
--------------------------------
void frobulate_for_gcc(uint32_t) (uint32_t v)
{
  const char * s;

  <bb 2> [0.00%]:
  switch (v_2(D)) <default: <L9> [0.00%], case 0: <L7> [0.00%], case 1: <L1>
[0.00%], case 2: <L2> [0.00%]>

<L1> [0.00%]:
  goto <bb 5> (<L7>); [0.00%]

<L2> [0.00%]:

  # s_1 = PHI <"foo"(2), "bar"(3), "spam"(4)>
<L7> [0.00%]:
  __builtin_puts (s_1);
  return;

<L9> [0.00%]:
  abort ();

}

Looks that early jump threading transforms frobulate_for_gcc case to frobulate.
Well, I guess we can improve switchconv to handle cases, where a case contains
a noreturn call.
Honza?