This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH] x86-64: memcmp-avx2-movbe.S needs saturating subtraction [BZ #21662]
On 06/23/2017 05:43 PM, H.J. Lu wrote:
> On Fri, Jun 23, 2017 at 8:12 AM, Florian Weimer <fweimer@redhat.com> wrote:
>> On 06/23/2017 05:08 PM, H.J. Lu wrote:
>>> On Fri, Jun 23, 2017 at 6:51 AM, Florian Weimer <fweimer@redhat.com> wrote:
>>>> I think this revised patch is an improvement because it avoids adding a
>>>> branch.
>>>>
>>>> Thanks,
>>>> Florian
>>>
>>> The memcmp-avx2-movbe.S part is OK.
>>
>> Thanks, I will commit it shortly.
>>
>>> For
>>>
>>> (check_result, do_random_tests): Write error messages to standard
>>> output.
>>>
>>> any particular reason to replace error with printf?
>>
>> The error function writes to standard error, so the test harness does
>> not capture it output in the .out file.
>
> All string tests are like this.
Right.
> We still got
>
> FAIL: string/test-memcmp
>
> [hjl@gnu-6 build-x86_64-linux]$ string/test-memcmp
> string/test-memcmp: Wrong result in function __memcmp_avx2_movbe 1488148660 -168
> string/test-memcmp: Wrong result in function __memcmp_avx2_movbe 1488237565 -168
> string/test-memcmp: Wrong result in function __memcmp_avx2_movbe -1275218947 180
> string/test-memcmp: Wrong result in function __memcmp_avx2_movbe -1275265792 180
> simple_memcmp __memcmp_avx2_movbe
> __memcmp_sse4_1__memcmp_ssse3 __memcmp_sse2
> [hjl@gnu-6 build-x86_64-linux]$
>
> If we want to change this, it should be in a separate patch.
Agreed, and we should fix all the instances in one go. I'm not going to
work on this immediately.
Florian