This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [lno] symbolic number of iterations


On Mon, May 24, 2004 at 12:48 AM +0300, Sebastian Pop wrote:

>
> Maybe.
>
> I have replaced the code handling the computation of loop counts with
> a call to number_of_iterations_cond.

Sebastian, I'm sorry if it sounded as if it's your patch responsible for
the recent regression I have seen; this is not what I meant, just trying to
track down which is the relevant change.

I believe it's actually the following patch that's causing the change of
behavior:
http://gcc.gnu.org/ml/gcc-patches/2004-05/msg01234.html

With this patch, number_of_iterations_in_loop returns 'chrec_top' for the
loop bound in tree-ssa-vect-8.c;

When I disable the patch, I get the original (good!) behavior again - the
loop bound returned is the following:
'(<unnamed type>)(n_6 - 1) + 1'

> The tests on the loop counts of
> the scev testsuite passed after this change.  The problem is that the
> testsuite doesn't contains enough cases.

For example, the following vectorizer test fails now:
gcc/gcc/testsuite/gcc.dg/tree-ssa-vect/tree-ssa-vect-8.c

thanks, and sorry for any misunderstanding,

Olga




Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]