[llvm-dev] FW: Loop vectorizer Queires

[Demikhovsky, Elena via llvm-dev]

LLVM compiler supports KNL and does not support KNC.
-march=knl will enable KNL instruction set.
In order to see ZMM registers you should have a loop that can be vectorized and use –O2 or –O3 optimization level.

-           Elena

From: llvm-dev [mailto:llvm-dev-bounces at lists.llvm.org] On Behalf Of Jumana Mp via llvm-dev
Sent: Tuesday, June 14, 2016 01:59
To: C Bergström <cbergstrom at pathscale.com<mailto:cbergstrom at pathscale.com>>
Cc: llvm-dev <llvm-dev at lists.llvm.org<mailto:llvm-dev at lists.llvm.org>>
Subject: Re: [llvm-dev] Loop vectorizer Queires

Hello,

Thanks for your response. I would like to run my code on Xeon phi earlier version that is KNC.

Well I asked the basic question because of various reasons.

I found that LLVM emits auto vectorized code for MIC with flags -mavx512f and  -mavx512cd. I used clang 3.8 and my vectorized code did not emit any vector-512 instructions nor zmm registers are used. So I thought, there is a difference between flags for KNL and KNC. The architecture I have got to test is KNC. - mcpu=help does not distinguish this(may be I dont understand) and I got confused.

I think there is a combination of flags that I should use for KNC.

It will be nice to be clarified with this.
[https://ssl.gstatic.com/ui/v1/icons/mail/images/cleardot.gif]

On Tue, Jun 14, 2016 at 12:24 AM, C Bergström <cbergstrom at pathscale.com<mailto:cbergstrom at pathscale.com>> wrote:
On Tue, Jun 14, 2016 at 6:03 AM, Jumana Mp via llvm-dev
<llvm-dev at lists.llvm.org<mailto:llvm-dev at lists.llvm.org>> wrote:
> Hello,
>
> I have a few issues in vectorizing loops using Clang 3.8.
> Will it be ok if I shoot some of my findings and queries here?
>
> Meanwhile, can I please know if LLVM support autovectorized MIC instructions
> for Xeon phi?
> If so, could you please tell me the flags to use?
>
> I am Jumana, a masters student in Embedded system working as a graduate
> research intern with Intel. For my thesis, I am working with bench-marking
> auto-vectorizer for various architectures.
> I have certain issues which I find difficult to explain myself. I tried
> posting in LLVM forum, but looks like I have no access.

You're a researcher and intern, but asking basic questions like this? Umm...

"MIC instructions" - are you referring to the AVX512 supported in KNL
or the pre-cursor to this in KNC? (Which supported a few more and a
few less instructions than what ended up being AVX512)

https://en.wikipedia.org/wiki/AVX-512
-----------
I've seen a bunch of work from @Intel people around here and I'll let
them comment on more of the specifics about the level of support for
Intel PHI products.

---------------------------------------------------------------------
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.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20160614/4f4aba77/attachment.html>