This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [Patch PR44576]: Don't do further analysis if we could know prefetching is not benefitial
On Sun, Jul 11, 2010 at 00:26, Christian Borntraeger
<borntraeger@de.ibm.com> wrote:
> Am Freitag 09 Juli 2010, 19:06:55 schrieb Fang, Changpeng:
>>
>> Am Freitag 09 Juli 2010, 03:15:26 schrieb Fang, Changpeng:
>> >> + ?/* FIXME: the time should be weighted by the probabilities of the blocks in
>> >> + ? ? the loop body. ?*/
>> >> + ?time = tree_num_loop_insns (loop, &eni_time_weights);
>> >> + ?ahead = (PREFETCH_LATENCY + time - 1) / time;
>>
>> >Here I sometimes get a division by zero.
>>
>> That should not happen (time couldn't be zero). Would you please double check and give a testcase to show the problem
>> if possible?
>
> dependence.c from gcc (spec version) fails in such a way.
> Since you added a check I guess you dont need a testcase any more?
>
I think that testcases are always useful. Please use delta as
described in http://gcc.gnu.org/wiki/A_guide_to_testcase_reduction
to reduce the ICE that you see, then add the reduced testcase for
this problem to the testsuite.
Thanks,
Sebastian