[Bug target/59625] New: asm goto and TARGET_FOUR_JUMP_LIMIT

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59625

            Bug ID: 59625
           Summary: asm goto and TARGET_FOUR_JUMP_LIMIT
           Product: gcc
           Version: 4.8.3
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: jakub at gcc dot gnu.org
                CC: hubicka at gcc dot gnu.org, rth at gcc dot gnu.org

On the following testcase (with 4.8 with say -O2 -mtune=generic, with 4.9
starting with r203012 with e.g. -O2 -mtune=atom) we generate extra .p2align
directives in between the asm goto.

#define EFAULT 14

#define put_user_try do {\
  __label__ put_user_fail;\
  stac();\
  barrier();

#define put_user_catch(err)\
  clac();\
  if (0) {\
   put_user_fail:\
    clac();\
    (err) = -EFAULT;\
  }\
} while (0)

# define _ASM_EXTABLE_EX(from,to)                            \
     " .pushsection \"__ex_table\",\"a\"\n"                  \
     " .balign 8\n"                                          \
     " .long (" #from ") - .\n"                              \
     " .long (" #to ") - . + 0x7ffffff0\n"                   \
     " .popsection\n"

#define __put_user_asm_ex(x, addr, itype, rtype, ltype)\
asm goto("1:mov"itype" %"rtype"0,%1\n"\
     _ASM_EXTABLE_EX(1b, %l[put_user_fail])\
     : : ltype(x), "m" (addr) : : put_user_fail)

int main(int argc, char **argv)
{
  int err = 0;

  put_user_try {
    __put_user_asm_ex(0, argv[0], "q", "", "er");
    __put_user_asm_ex(0, argv[1], "q", "", "er");
    __put_user_asm_ex(0, argv[2], "q", "", "er");
    __put_user_asm_ex(0, argv[3], "q", "", "er");
    __put_user_asm_ex(0, argv[4], "q", "", "er");
  } put_user_catch(err);
  return err;
}

This is from TARGET_FOUR_JUMP_LIMIT ix86_avoid_jump_mispredicts.  Perhaps we
shouldn't count asm goto as a jump there (i.e. add
"&& asm_noperands (PATTERN (insn)) < 0"
next to every JUMP_P (insn) in that routine)?  Because while asm goto could
contain a jump, it can contain something completely different too (as in this
case), not to mention that we have to conservatively count the asm goto as
occupying 0 bytes in the cache line.