Re: RFA: Fix optab reinitialisation when $(CC) is GCC 4.0 or above

[Jakub Jelinek <jakub at redhat dot com>]

On Mon, Nov 26, 2007 at 09:22:26AM +0000, Richard Sandiford wrote:
> After Jakub's recent (and very welcome) patch to speed up optabs
> initialisation, target_reinit no longer reinitialises the optab
> table properly when compiled with GCC 4.0 or above.  This breaks
> the mips16 attribute.

Oops, didn't know that something is trying to reinitialize optabs.

> The change to init_optabs itself was search-and-replace, so I don't
> think it makes the function harder to maintain.  All lines stay
> within the 80-column limit.
> 
> Bootstrapped & regression-tested on x86_64-linux-gnu.  Also
> regression-tested on mipsisa64-elfoabi, where it fixes several
> gcc.target/mips failures.  I also checked that the file still
> works when compiled with gcc 3.4.  OK to install?

Actually, I'm afraid this will create bigger code even on targets that never
need this, since the init_insn_codes argument is passed down.
Perhaps it would be better to remove new_optab altogether and insn_codes
setup code from init_convert_optab and instead add a new function,

static void
init_insn_codes (void)
{
  int i, j, op;

  for (op = 0; op < OTI_MAX; op++)
    for (i = 0; i < NUM_MACHINE_MODES; i++)
      optab_handler (&optab_table[op], i)->insn_code = CODE_FOR_nothing;
  for (op = 0; op < COI_MAX; op++)
    for (i = 0; i < NUM_MACHINE_MODES; i++)
      for (j = 0; j < NUM_MACHINE_MODES; j++)
        convert_optab_handler (&convert_optab_table[op], i, j)->insn_code
	  = CODE_FOR_nothing;
}

and call this either when not compiled with GCC 4.0 and above, or
when initializing for the second and later time.

	Jakub