RFC: Putting target-dependent global state into switchable structures

Richard Guenther richard.guenther@gmail.com
Sun Jun 27 03:11:00 GMT 2010


On Sun, Jun 27, 2010 at 12:25 AM, Joseph S. Myers
<joseph@codesourcery.com> wrote:
> On Sun, 27 Jun 2010, Richard Guenther wrote:
>
>> On Sat, Jun 26, 2010 at 10:02 PM, Mark Mitchell <mark@codesourcery.com> wrote:
>> > Richard Sandiford wrote:
>> >
>> >> I've got a series of patches that speed up the switching process by
>> >> avoiding target_reinit.  Rather than spam the list with the entire
>> >> series now, I thought I'd better ask for feedback on the general approach.
>> >
>> > I think this is entirely sensible.  It's a step towards a more modular
>> > compiler, and we all know that we need that.  I'll be happy to help
>> > review the patches as necessary once you're ready.
>>
>> Well.  I don't like the idea of using .def files and a generator program.
>> In fact we are happily using no .def files for target hooks.
>
> I believe we're waiting for build system and middle-end review of Joern's
> patch <http://gcc.gnu.org/ml/gcc-patches/2010-06/msg01863.html> to
> introduce just such a .def file.  (If that can get in, Joern should then
> produce the patch showing the fixed body of existing hook documentation we
> want to license under both GPL and GFDL, for RMS to review; he can deal
> with relicensing requests for fixed concrete pieces of text, where the
> licenses in question already exist, reasonably efficiently.)

Ok.

That doesn't make me like .def files for globals any more though.  In
fact it's exactly included once and the point of .def files is to avoid
duplication of data or to put related data together.

Richard.



More information about the Gcc-patches mailing list