Re: target specific builtin expansion (middle end and back end definition inconsistence problem?).

[Ian Lance Taylor <iant at google dot com>]

Feng LI <nemokingdom@gmail.com> writes:

> I generate builtin function directly in the middle end and and expand
> the builtin function in the x86 backend to certain set of
> instructions.
>
> I've defined x86 builtin functions in the gcc backend like this:
>
> { OPTION_MASK_ISA_TSTAR | OPTION_MASK_ISA_64BIT,
> CODE_FOR_tstar_create, "__builtin_ia32_tcreate", IX86_BUILTIN_TCREATE,
> UNKNOWN, (int) PVOID_FTYPE_UINT_UINT },
> and corresponding define_insn in .md file.
>
> This works with .c file as a input, and it'll generate the assemble
> code I expect.
>
> But if I try to generate the builtin functions directly in GCC middle
> end, by using DEF_BUILTIN
>
>   DEF_BUILTIN (BUILTIN_TCREATE, "__builtin_ia32_tcreate", BUILT_IN_MD,
>                          BT_FN_PTR_SIZE_SIZE, BT_FN_PTR_SIZE_SIZE,    \
>                          false, true, true, ATTRS, false, flags)

It doesn't make sense to add a x86-specific builtin function to
builtins.def.  So I'm not sure what you are trying to accomplish with
this change.

> #0  fancy_abort (file=0xb45b0f "../../gcc/c-decl.c", line=3569,
> function=0xb45a10 "c_builtin_function") at ../../gcc/diagnostic.c:892
> #1  0x000000000049634f in c_builtin_function (decl=0x7ffff7fb4500) at
> ../../gcc/c-decl.c:3569

This abort is

  /* Should never be called on a symbol with a preexisting meaning.  */
  gcc_assert (!I_SYMBOL_BINDING (id));

That makes it sound like you are defining the function in both the x86
backend and also in builtins.def.  Don't do that.

Ian