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: [RFA] [PATCH] Make VRP understand bitwise OR and AND (better than now)


On Mon, Aug 11, 2008 at 10:19 PM, Denys Vlasenko <dvlasenk@redhat.com> wrote:
> On Mon, 2008-08-11 at 14:40 +0200, Richard Guenther wrote:
>> On Mon, Aug 11, 2008 at 2:14 PM, Denys Vlasenko <dvlasenk@redhat.com> wrote:
>> >> In extract_range_from_binary_expr it looks like the cases for
>> >> BIT_AND_EXPR and BIT_IOR_EXPR can share most (if not all) of
>> >> the code if you use the expression code instead of constant codes.
>> >>
>> >> In bits_always_set and bits_maybe_set it is better to use double_ints
>> >> (see double_int.h) for intermediate calculations, as they do not involve
>> >> allocating new tree constants.
>> >
>> > The updated patch is attached (with instrumentation code present but
>> > #defined out).
>>
>> Uh, please remove the instrumentation code - it makes the code hard
>> to follow.
>>
>> >> integer_nonnegative_range needs a comment.
>> >
>> > Fixed.
>> >
>> > Please review attached patch.
>>
>> This looks much better (with the instrumentation code removed).  Please
>> add a testcase or two, write a proper ChangeLog and specify where
>> you bootstrapped and tested the patch.
>
> Please find it attached. I did run testsuite, twice, once with
> deliberately broken test and one with test as it is in the patch.
> (Boy, that was SLOW).

The patch is ok for mainline if it passed a bootstrap as well.

Thanks,
Richard.


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