Several tree-ssa issues

Several tree-ssa issues

[Gerald Pfeifer]

The tree-ssa branch seems to have improved recently, however I still note
several problems:

1) Who might be interested in
     internal compiler error: in gimplify_expr, at gimplify.c:662
happening in proprietary source? (I cannot publish that in Bugzilla but
will happily provide it by private mail.)


2) /sw/gcc-current/bin/../lib/gcc/i386-unknown-freebsd4.8/3.5-tree-ssa/../../../../include/c++/3.5-tree-ssa/i386-unknown-freebsd4.8/bits/
atomicity.h:50: warning: '__tmp' might be used uninitialized in this function

(Given that atomicity.h is a GCC header, we should not warn; either the
header or GCC should be fixed.)


3) Finally, why does -O3 consume 118s and 600+MB for the testcase in
PR8361, whereas GCC 3.2.3 only consumes 72s 110MB?

Gerald

Re: Several tree-ssa issues

[Gerald Pfeifer]

On Mon, 29 Sep 2003, Loren James Rittle wrote:
>> 3.5-tree-ssa/i386-unknown-freebsd4.8/bits/atomicity.h:50: \
>> warning: '__tmp' might be used uninitialized in this function
>
> Assuming you are speaking about the tip of "tree-ssa-20020619-branch"
> or "tree-ssa-cfg-branch" (sorry, I test neither and do not know which
> is definitive, if either, based on context), I think the warning
> itself must be in error based on the contents of line 50:
>
> 	register _Atomic_word __result, __tmp = 1;

Thanks for the analysis, Loren.  This is now PR12525.

The other two bugs I reported against tree-ssa one week ago are now
PR12524 and PR12526.

Gerald

Re: Several tree-ssa issues

[Loren James Rittle]

Hi Gerald,

> 3.5-tree-ssa/i386-unknown-freebsd4.8/bits/atomicity.h:50: \
> warning: '__tmp' might be used uninitialized in this function

Assuming you are speaking about the tip of "tree-ssa-20020619-branch"
or "tree-ssa-cfg-branch" (sorry, I test neither and do not know which
is definitive, if either, based on context), I think the warning
itself must be in error based on the contents of line 50:

	register _Atomic_word __result, __tmp = 1;

(I assume that no later C-level statement could invalidate __tmp in
 any manner which produces that warning.  In FD, there is a
 semi-vetted __asm__ construct but it looks legit to me...)

Regards,
Loren