llvm dev - Feb 2015 - [LLVMdev] [3.6 Release] Bugfixes in Masked Load/Store

Hi Hans,

I fixed 2 bugs in the trunk branch related to Masked Load / Store.
Since these intrinsics are generated by Loop Vectorizer on AVX2, a wrong code
may be generated. One of the bugs was detected while gcc benchmark testing.
http://llvm.org/bugs/show_bug.cgi?id=22225

I think that the bugs should be fixed in 3.6.
I have 2 options
(1) promote changes of revisions 226791 and 226808 from trunk to 3.6
(2) disable the Masked Load / Store support for AVX2 in 3.6 

Please let me know what do you think?

-  Elena

---------------------------------------------------------------------
Intel Israel (74) Limited

This e-mail and any attachments may contain confidential material for
the sole use of the intended recipient(s). Any review or distribution
by others is strictly prohibited. If you are not the intended
recipient, please contact the sender and delete all copies.

On Tue, Feb 17, 2015 at 6:55 AM, Demikhovsky, Elena
<elena.demikhovsky at intel.com> wrote:
> I fixed 2 bugs in the trunk branch related to Masked Load / Store.
> Since these intrinsics are generated by Loop Vectorizer on AVX2, a wrong
code may be generated. One of the bugs was detected while gcc benchmark testing.
> http://llvm.org/bugs/show_bug.cgi?id=22225
>
> I think that the bugs should be fixed in 3.6.
> I have 2 options
> (1) promote changes of revisions 226791 and 226808 from trunk to 3.6
> (2) disable the Masked Load / Store support for AVX2 in 3.6
>
> Please let me know what do you think?
We'll be doing another release candidate of 3.6 so I guess we could
potentially squeeze these in (though I wish you'd told me sooner, the
patches are almost 4 weeks old).

Nadav, you're the X86 owner. Do you think r226791 and r226808 are safe to
merge?

Thanks,
Hans

Both r226791 and r226808 are safe to merge because they only effect AVX2 masked
loads/stores. We should take these changes into the release branch.

Elena, can you run the test suite with AVX2 enabled using 3.6?

> On Feb 17, 2015, at 11:00 AM, Hans Wennborg <hans at chromium.org>
wrote:
> 
> On Tue, Feb 17, 2015 at 6:55 AM, Demikhovsky, Elena
> <elena.demikhovsky at intel.com> wrote:
>> I fixed 2 bugs in the trunk branch related to Masked Load / Store.
>> Since these intrinsics are generated by Loop Vectorizer on AVX2, a
wrong code may be generated. One of the bugs was detected while gcc benchmark
testing.
>> http://llvm.org/bugs/show_bug.cgi?id=22225
>> 
>> I think that the bugs should be fixed in 3.6.
>> I have 2 options
>> (1) promote changes of revisions 226791 and 226808 from trunk to 3.6
>> (2) disable the Masked Load / Store support for AVX2 in 3.6
>> 
>> Please let me know what do you think?
> 
> We'll be doing another release candidate of 3.6 so I guess we could
> potentially squeeze these in (though I wish you'd told me sooner, the
> patches are almost 4 weeks old).
> 
> Nadav, you're the X86 owner. Do you think r226791 and r226808 are safe
to merge?
> 
> Thanks,
> Hans