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: [Patch] Adjust diag-scans in vect-tests to fix fails on AVX/AVX2


> I think there is a difference between different vector sizes, and calling
> it vect_X_vector_size_available is not sufficient. Your patch will cause
> failures on ARM. It has two vector sizes, 16 and 8 bytes. E.g., vect-33.c
> gets vectorized with the default vector size, and the alignment message
> should be printed only once, and not twice as with your patch. So, it looks
> like you need several vect_multiple_sizes_X.

Probably we really need to have something like
vect_multiple_sizes_32B_16B - it'll help in this test. In this case we
could check if specific size is available, and it's crucial when array
sizes are so small, that they define the vector width to be used,
vect_multiple_sizes isn't sufficient for this, because the specific
size matters. I'll prepare a patch with such changes.

By the way, how could we check if '-mprefer-avx128' was specified from
target-supports.exp? Is there any global-variable for command line
options or something similar?

On 12 December 2011 15:36, Ira Rosen <IRAR@il.ibm.com> wrote:
>
>
> gcc-patches-owner@gcc.gnu.org wrote on 12/12/2011 01:00:52 PM:
>
>> I changed xfails to target-checks - for now I use common
>> vect_multiple_sizes (though it'll fail when wider vectors emerge).
>> Also, I changed AVX-check to the version Uros suggested. Please check
>> updated patch (attached).
>>
>> As for vect_multiple_sizes_32B_16B and similar - isn't it too
>> target-specific? I think if we want to keep everything as general as
>> possible, we should have something like vect_1_vector_size_available,
>> vect_2_vector_sizes_available, etc.
>
> I think there is a difference between different vector sizes, and calling
> it vect_X_vector_size_available is not sufficient. Your patch will cause
> failures on ARM. It has two vector sizes, 16 and 8 bytes. E.g., vect-33.c
> gets vectorized with the default vector size, and the alignment message
> should be printed only once, and not twice as with your patch. So, it looks
> like you need several vect_multiple_sizes_X.
>
> Ira
>
>
>



-- 
---
Best regards,
Michael V. Zolotukhin,
Software Engineer
Intel Corporation.


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