[debug-early, jit] Port of jit to debug-early

[Aldy Hernandez]

On 06/05/2015 05:17 PM, David Malcolm wrote:
> On Fri, 2015-06-05 at 16:18 -0400, David Malcolm wrote:
>> On Fri, 2015-06-05 at 15:37 -0400, Aldy Hernandez wrote:
>>> On 06/05/2015 03:00 PM, David Malcolm wrote:
>>>> "jit" isn't part of "all" languages, and currently fails to build on
>>>> the aldyh/debug-early branch:
>>>
>>> The debug-early work is now in mainline.  Feel free to commit when you
>>> deem appropriate, since you are the jit maintainer ;-).
>>
>> Given that it fixes the build of the jit, I've gone ahead and committed
>> the patch as-is to trunk (r224170).
>>
>>>>    static void
>>>> -jit_langhook_write_globals (void)
>>>> +jit_langhook_post_compilation_parsing_cleanups (void)
>>>>    {
>>>>      gcc::jit::playback::context *ctxt = gcc::jit::active_playback_ctxt;
>>>>      gcc_assert (ctxt);
>>>>      JIT_LOG_SCOPE (ctxt->get_logger ());
>>>>
>>>> -  ctxt->write_global_decls_1 ();
>>>> -
>>>> -  /* This is the hook that runs the middle and backends: */
>>>> -  symtab->finalize_compilation_unit ();
>>>> -
>>>> -  ctxt->write_global_decls_2 ();
>>>> +  ctxt->finalize_global_decls ();
>>>
>>> This hook gets called after the compilation proper has run, and we're
>>> trying to avoid it.  We're trying to get all the front-ends to finish
>>> their thing and avoid depending on the compilation proper.
>>>
>>> Can you not get this to work without running finalize_global_decls after
>>> the compilation proper?  If not, then I see no problem with your
>>> patch....after all, C++ is still using this hook :(.
>>
>> Am investigating.
>
> All that it was doing was calling varpool_node::finalize_decl on each
> global; it appears that it's OK to instead call that on them as soon as
> they're created (despite the name).

Perfect.  The earlier the better.

In most of the front-ends I think we are calling finalize_decl via 
rest_of_decl_compilation which is where we also call the 
early_global_decl hook.

>
> I've committed the attached patch to trunk (r224173) which removes the
> hook; "make check-jit" gives the same results as before, and I again
> hand-tested (with gdb) jit.dg's test-using-global.c and
> test-fibonacci.c.
>

Excellent.

Thanks.