On 1/18/18 1:36 PM, Rebecca Cran wrote:
> On 01/18/2018 10:47 AM, Jens Axboe wrote:
>>
>> Adding the memcpy for avx/sse to the test case might be interesting
>> though, just to be able to compare performances with builtin
>> memcpy/memmove on a given system.
>
> I've attached the patch. I think it should work on all Intel systems
> newer than 2012. It causes a segfault on my Atom E3826 (Baytrail)
> because it doesn't support AVX.
>
> I've tested it using Clang 3.8, GCC 4.8 and 7.x on openSUSE, and Clang
> 4.0 on FreeBSD 11.1.
I'm worried that this will break some platforms. A few comments that
might help rectify that:
1) Use fio_memalign() instead of relying on a c11 function
2) Add a configure check for the avx and sse, separately.
This includes -msse for the compiler, includes for the
xmm/imm intrinsics, etc.
I don't care if it segfaults if avx/sse isn't there, but I don't
want to risk failing a build on some rarely tested platform, just
to have this as part of the memcpy test. The risk/reward just
isn't there then.
--
Jens Axboe
--
To unsubscribe from this list: send the line "unsubscribe fio" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
