Compiler Internal Error: with icc-16

[bjoo]

ICC-16 suffers from 'internal errors':

Seen in: icpc (ICC) 16.0.0 20150815 icpc (ICC) 16.0.2 20160204

Not necessarily a bug, but may need code-fix to workaround.

[ddkalamk]

I think I know what the problem is here. I think there is some change required in shuffle/permute intrinsic in code-generator. Probably I have it fixed in ks-codegen branch. Will take a look and try to fix it.

Dhiraj

From: Balint Joo [mailto:notifications@github.com] Sent: Monday, February 29, 2016 9:13 PM To: JeffersonLab/qphix Subject: [qphix] Compiler Internal Error: with icc-16 (#16)

ICC-16 suffers from 'internal errors':

Seen in: icpc (ICC) 16.0.0 20150815 icpc (ICC) 16.0.2 20160204

Not necessarily a bug, but may need code-fix to workaround.

— Reply to this email directly or view it on GitHubhttps://github.com/JeffersonLab/qphix/issues/16.

[bjoo]

Hi Dhiraj, Thanks. In the meantime, I have filed a bug with Intel Premier support.

Best, B

On February 29, 2016 12:18:40 PM EST, Dhiraj D Kalamkar notifications@github.com wrote:

I think I know what the problem is here. I think there is some change required in shuffle/permute intrinsic in code-generator. Probably I have it fixed in ks-codegen branch. Will take a look and try to fix it.

Dhiraj

From: Balint Joo [mailto:notifications@github.com] Sent: Monday, February 29, 2016 9:13 PM To: JeffersonLab/qphix Subject: [qphix] Compiler Internal Error: with icc-16 (#16)

ICC-16 suffers from 'internal errors':

Seen in: icpc (ICC) 16.0.0 20150815 icpc (ICC) 16.0.2 20160204

Not necessarily a bug, but may need code-fix to workaround.

— Reply to this email directly or view it on GitHubhttps://github.com/JeffersonLab/qphix/issues/16.

---

Reply to this email directly or view it on GitHub: https://github.com/JeffersonLab/qphix/issues/16#issuecomment-190297155

Balint Joó, Scientific Computing Group, Jefferson Lab Email: bjoo@jlab.org Tel: +1 757 269 5339 Sent form my mobile phone

[ddkalamk]

Ok . I tried with ks-codegen branch. Even there I get internal error so my guess seems wrong. So we will have to follow up with Premier support.

Dhiraj

From: Balint Joo [mailto:notifications@github.com] Sent: Tuesday, March 01, 2016 12:56 AM To: JeffersonLab/qphix Cc: Kalamkar, Dhiraj D Subject: Re: [qphix] Compiler Internal Error: with icc-16 (#16)

Hi Dhiraj, Thanks. In the meantime, I have filed a bug with Intel Premier support.

Best, B

On February 29, 2016 12:18:40 PM EST, Dhiraj D Kalamkar notifications@github.com<mailto:notifications@github.com> wrote:

I think I know what the problem is here. I think there is some change required in shuffle/permute intrinsic in code-generator. Probably I have it fixed in ks-codegen branch. Will take a look and try to fix it.

Dhiraj

From: Balint Joo [mailto:notifications@github.com] Sent: Monday, February 29, 2016 9:13 PM To: JeffersonLab/qphix Subject: [qphix] Compiler Internal Error: with icc-16 (#16)

ICC-16 suffers from 'internal errors':

Seen in: icpc (ICC) 16.0.0 20150815 icpc (ICC) 16.0.2 20160204

Not necessarily a bug, but may need code-fix to workaround.

— Reply to this email directly or view it on GitHubhttps://github.com/JeffersonLab/qphix/issues/16.

---

Reply to this email directly or view it on GitHub: https://github.com/JeffersonLab/qphix/issues/16#issuecomment-190297155

Balint Joó, Scientific Computing Group, Jefferson Lab Email: bjoo@jlab.orgmailto:bjoo@jlab.org Tel: +1 757 269 5339 Sent form my mobile phone

— Reply to this email directly or view it on GitHubhttps://github.com/JeffersonLab/qphix/issues/16#issuecomment-190345559.

[bjoo]

I heard back about this, the issue seems to be the combination of

pragma simd

pragma omp parallel for

in that order, which occurs in some of the BLAS like code. Workarounds suggested were reversing the order of the pragmas, but perhaps a better approach would be to track these down and replace them with OpenMP 4.0 SIMD and / or SIMD parallel for pragmas

Many thanks to Intel Premier Support and Compiler Engineering for their help with this issue.

[azrael417]

Hi Balint,

thanks for digging this out. It should not crash though. However, we should use #pragma tmp parallel for sims instead.

Best Thorsten

Am 21.03.2016 um 20:27 schrieb Balint Joo notifications@github.com:

I heard back about this, the issue seems to be the combination of

pragma simd

pragma omp parallel for

in that order, which occurs in some of the BLAS like code. Workarounds suggested were reversing the order of the pragmas, but perhaps a better approach would be to track these down and replace them with OpenMP 4.0 SIMD and / or SIMD parallel for pragmas

Many thanks to Intel Premier Support and Compiler Engineering for their help with this issue.

— You are receiving this because you are subscribed to this thread. Reply to this email directly or view it on GitHub https://github.com/JeffersonLab/qphix/issues/16#issuecomment-199614243

[azrael417]

I will take care of the OMP 4 conversion and create a new icc16-hotfix branch one that is done.

Am 21.03.2016 um 20:28 schrieb Thorsten Kurth thorstenkurth@me.com:

Hi Balint,

thanks for digging this out. It should not crash though. However, we should use #pragma tmp parallel for sims instead.

Best Thorsten

Am 21.03.2016 um 20:27 schrieb Balint Joo <notifications@github.com mailto:notifications@github.com>:

I heard back about this, the issue seems to be the combination of

pragma simd

pragma omp parallel for

in that order, which occurs in some of the BLAS like code. Workarounds suggested were reversing the order of the pragmas, but perhaps a better approach would be to track these down and replace them with OpenMP 4.0 SIMD and / or SIMD parallel for pragmas

Many thanks to Intel Premier Support and Compiler Engineering for their help with this issue.

— You are receiving this because you are subscribed to this thread. Reply to this email directly or view it on GitHub https://github.com/JeffersonLab/qphix/issues/16#issuecomment-199614243

[bjoo]

Thanks Thorsten!