Re: [RFC] Moving TM builtin creation to the machine independent parts of the compiler

[Richard Guenther <richard dot guenther at gmail dot com>]

On Fri, Sep 25, 2009 at 10:50 PM, Michael Meissner
<meissner@linux.vnet.ibm.com> wrote:
> As we were discussing Thursday on IRC, the powerpc port doesn't set up the
> attributes correctly for the builtin functions. ?Richard Henderson suggested
> making the backend builtin creation to use the same infrastructure as the
> normal builtins, so that we can use the types and attributes for the builtins.
>
> I was wondering if the following work in progress is what was suggested.
> Basically, it allows the target machine to define a set of *.def files that are
> handled inside of builtins.def and builtin-types.def and builtin-attrs.def.
> Right now, I don't anticipate needing to do the attributes, but I figure to put
> in the support in case some other backend needs it.
>
> This patch creates tm-builtins-headers.def, tm-builtins-types.def, and
> tm-builtins-attrs.def in the build directory to include the various tm builtin
> files. ?At the moment, I'm including one builtin, before tackling the 900 or so
> powerpc builtins.
>
> I suspect I should use the s-* and moveifchange mecanism in the Makefile.

Will it be possible to initialize the builtins on-demand, like when using
the target or optimize attributes?  OTOH we should switch to lazy
initialization of builtins in general to improve startup time of gcc.

Richard.

> --
> Michael Meissner, IBM
> 4 Technology Place Drive, MS 2203A, Westford, MA, 01886, USA
> meissner@linux.vnet.ibm.com
>