On Thu, Nov 13, 2014 at 6:36 AM, Jakub Jelinek <ja...@redhat.com> wrote:
> On Thu, Nov 13, 2014 at 06:20:16AM -0800, Teresa Johnson wrote:
>> Here is the new patch. Bootstrapped and tested on
>> x86_64-unknown-linux-gnu. OK for trunk?
>>
>> Thanks,
>> Teresa
>>
>> 2014-11-13 <tejohn...@google.com>
>>
>> gcc:
>> PR tree-optimization/63841
>> * tree.c (initializer_zerop): A constructor with no elements
>> does not zero initialize.
>>
>> gcc/testsuite:
>> PR tree-optimization/63841
>> * g++.dg/tree-ssa/pr63841.C: New test.
>>
>> Index: tree.c
>> ===================================================================
>> --- tree.c (revision 217190)
>> +++ tree.c (working copy)
>> @@ -10330,6 +10330,8 @@ initializer_zerop (const_tree init)
>> {
>> unsigned HOST_WIDE_INT idx;
>>
>> + if (TREE_CLOBBER_P (init))
>> + return false;
>
> Wrong formatting.
Sorry, not sure I understand why? My mailer does tend to eat spaces,
but it is indented the correct amount and I think has the right spaces
within the line.
>
> Also, while this perhaps is useful, I'd say the right fix is that
> strlen_optimize_stmt
> should just ignore gimple_clobber_p (stmt) statements (well, call
> the maybe_invalidate at the end for them).
> So
> - else if (is_gimple_assign (stmt))
> + else if (is_gimple_assign (stmt) && !gimple_clobber_p (stmt))
> in strlen_optimize_stmt.
Ok, I have held off on my commit for now. I considered fixing this in
tree-ssa-strlen, but thought this was a potentially larger problem
with initializer_zerop, where it shouldn't be considering a clobber to
be a zero init and might be affecting other callers as well.
If we make the change to initializer_zerop is it still useful to
change tree-strlen?
Teresa
>
> Jakub
--
Teresa Johnson | Software Engineer | tejohn...@google.com | 408-460-2413
