x86-64 bootstrap broken (was Re: [PATCH] tree level if-conversion for vectorizer)

[Devang Patel]

On Sep 4, 2004, at 5:40 PM, Devang Patel wrote:

>
> On Sep 4, 2004, at 1:49 PM, Devang Patel wrote:
>
>>
>> On Sep 4, 2004, at 1:39 PM, Jan Hubicka wrote:
>>
>>>>
>>>> Would it be possible to send me preprocessed source ?
>>>> I do not have access to x86-64 box and ppc-linux box
>>>> in the lab has decided to not respond at the moment.
>>>
>>> Well, the problem is that the same source works fine with stage1
>>> compiler and crash in stage2, so what we are facing is an
>>> misscompilation of gcc itself.
>>
>> When -fdump-tree-ifcvt-stats is enabled, GCC emits
>> "Applying if-conversion" string in status file (.ifcvt).
>> It does not if-convert many loops, so it would be helpful
>> you can identify mis-compiled object file.
>>
>> Meanwhile I'll try if I can reproduce it on powerpc-darwin.
>> I have started couple of bootstrap. Lets see. I'll be
>> offline couple for hours.
>
> I checked out fresh sources as of today morning. Top of ChangeLog...
>
>       1 2004-09-04  H.J. Lu  <hongjiu.lu@intel.com>
>       2
>       3         * configure: Regenerated.
>       4
>       5 2004-09-04  Richard Kenner  <kenner@vlsi1.ultra.nyu.edu>
>       6
>       7         * tree.c (save_expr): No longer TREE_READONLY.
>       8
>       9         * fold-const.c (operand_equal_p): Remove code to allow 
> null ARG0/1.
>      10         Define locals macros OP_SAME and OP_SAME_WITH_NULL and 
> use throughout.
>      11
>      12 2004-09-04  Jan Hubicka  <jh@suse.cz>
>
>
> 	....
>
> It bootstraps just fine on powerpc-darwin.

(BTW, this is after I locally reverted ftrapping-math patch.
	http://gcc.gnu.org/ml/gcc-regression/2004-09/msg00006.html )

> It'd be helpful if someone tries bootstrap on broken platform by 
> disabling
> just if-conversion not entire optimization.
>
> Thanks,
> -
> Devang