Backport of MkFitCore vectorization PR 37868 #38259

gartung · 2022-06-06T14:41:26Z

commit 18abd3a
Author: Patrick Gartung [email protected]
Date: Wed May 25 02:12:21 2022 +0200

Code format

commit f067f25
Author: Patrick Gartung [email protected]
Date: Wed May 25 02:08:47 2022 +0200

Fix icc compilation errors

commit b265ce2
Author: Patrick Gartung [email protected]
Date: Fri May 20 23:22:10 2022 +0200

Add line to RecoTracker/MkFitCore/standalone/README_buildFromCMSSW.txt

commit 8618997
Author: Patrick Gartung [email protected]
Date: Fri May 20 18:18:50 2022 +0200

Add CMS_UNROLL_LOOP defines for Intel compilers and use with pragme omp simd instructions that cause internal icc error

commit 45977ea
Author: Patrick Gartung [email protected]
Date: Fri May 20 18:17:54 2022 +0200

Squashed commit of the following:

commit a565f528861deb96fb549f12c8ec619dc60fde1d
Author: Patrick Gartung <[email protected]>
Date:   Thu May 19 15:23:44 2022 +0200

    Remove fast-math flag and attribute needed to prevent segfault when fast-math is used

commit c3fbce9e16987ce4f152b3ae1197c543f988a363
Author: Patrick Gartung <[email protected]>
Date:   Mon May 16 17:28:41 2022 +0200

    Code format

commit c388ab36ccecad460b22042ca882b0f572ce7c6e
Author: Patrick Gartung <[email protected]>
Date:   Mon May 16 17:27:00 2022 +0200

    Update location of #endif after enabling debug. #pragma omp simd not needed for debug statements. Move debug statements to own loop.

commit 468b4760a1a6df6bf8545903339b6312e8420c31
Author: Patrick Gartung <[email protected]>
Date:   Sat May 14 19:51:00 2022 +0200

    No need to link to libmvec. It is called from libm on el8. Always use -ffastmath

commit 0ef90368b2b7c5345b417d9e99fb37acef9f8685
Author: Patrick Gartung <[email protected]>
Date:   Fri May 13 20:01:43 2022 +0200

    Only works on el8_amd64 arch

commit 8866ca64dc913d274f25da401c321cc5d2193790
Author: Patrick Gartung <[email protected]>
Date:   Fri May 13 16:10:23 2022 +0200

    Update for macro name change

commit fe4a38fa1fca00e3746b14c1d72a9e35c675360c
Author: Patrick Gartung <[email protected]>
Date:   Thu May 12 21:43:02 2022 +0200

    code format?

commit 8326129cd2333b8fcc256880f70a5deb9e5d899f
Author: Patrick Gartung <[email protected]>
Date:   Thu May 12 21:18:44 2022 +0200

	    Add CMS_UNROLL_LOOP defines for Intel compilers and use with pragme omp simd instructions that cause internal icc error

commit 056f7ef4fcc745107ff55365f0a1d9e0a2634349
Author: Patrick Gartung <[email protected]>
Date:   Thu May 12 03:42:01 2022 +0200

    Add CMS_UNROLL_LOOP_COUNT(Config::Niter) and initialize arrays with += and -= assignment statements to 0. first

commit 31e2c8351f4f988e97b04ec39eb52547654c3923
Author: Patrick Gartung <[email protected]>
Date:   Wed May 11 05:16:27 2022 +0200

    Add CMS_UNROLL_LOOP_COUNT(Config::Niter)
    and #include "FWCore/Utilities/interface/CMSUnrollLoop.h"

commit 8eab5b1ea5f6d29b90d4670ef7aedbec0ff8af72
Author: Patrick Gartung <[email protected]>
Date:   Wed May 11 04:57:47 2022 +0200

    Change atrribute from no-fast-math to match-errno

commit 688e905a36b83cf8d5da1313b703098598cd74ae
Author: Patrick Gartung <[email protected]>
Date:   Tue May 10 17:37:25 2022 +0200

    Code format

commit ed58228b16f3b86686725556739cc93508d3af14
Author: Patrick Gartung <[email protected]>
Date:   Tue May 10 17:23:08 2022 +0200

    Change attribute from no-inline to no-fast-math. Back out changes to applyMaterialEffects since it does not vectorize with the conditional on radL.

commit c19f9c8b6bb718ae110f914177788bb760a5e59a
Author: Patrick Gartung <[email protected]>
Date:   Mon May 9 19:49:15 2022 +0200

    code format

commit 373ea0d28c50e3b7d949604eddae5ac0859e0a93
Author: Patrick Gartung <[email protected]>
Date:   Mon May 9 19:02:48 2022 +0200

    missed a few initialization of variable length arrays

commit 753934cb32362ecf846b24e726ec4c1b49cf644c
Author: Patrick Gartung <[email protected]>
Date:   Mon May 9 18:51:13 2022 +0200

    for all el8 archs

commit 2eb25ef5bf7c45353d4cae9c852d5b75fd848a85
Author: Patrick Gartung <[email protected]>
Date:   Mon May 9 18:31:35 2022 +0200

    clang does not allow initialization of variable length arrays

commit 7f237622ef7b3c216e6ed3ce1a0a03f978b03817
Author: Patrick Gartung <[email protected]>
Date:   Mon May 9 17:41:29 2022 +0200

    Add fast-math and mvec when compiling under el8

commit 5f6efe223081fe7a17becd39d5f27088c227136b
Author: Patrick Gartung <[email protected]>
Date:   Mon May 9 16:07:08 2022 +0200

    Comment out changes that are CERN specific

commit 1ba6e069887f2ca38a3318f292d409a5c19b8406
Author: Patrick Gartung <[email protected]>
Date:   Thu May 5 21:25:33 2022 +0200

    Try to break loop in applyMaterialEffects into smaller loops
    Combine arithmetic ops into one loop.
    Add -lmvec which is avaiable on RHEL8

commit dbc748cdd84b82a3faa3d621eb3ea9cb2621bc7c
Author: Patrick Gartung <[email protected]>
Date:   Wed May 4 20:02:05 2022 +0200

    Incorporate Dan Riley's suggested changes

commit d6ae2c9c7bb50e03e3e9f187a1413568f5aeb7e0
Author: Patrick Gartung <[email protected]>
Date:   Tue May 3 20:20:47 2022 +0200

    Update helixAtZ to use small calculation loops

commit 3d704ba96e6cefc4d7d1959f621c0d47fc57879e
Author: Patrick Gartung <[email protected]>
Date:   Fri Apr 29 21:43:00 2022 +0200

    Work in progress

commit 31550b4446669845eb696676f47a257726923532
Author: Patrick Gartung <[email protected]>
Date:   Fri Apr 29 20:03:09 2022 +0200

    Fixups

commit f10ef52690510ba9ad9ad00dc9285bc4951e651e
Author: Patrick Gartung <[email protected]>
Date:   Fri Apr 29 00:57:47 2022 +0200

    WIP

PR description:

PR validation:

if this PR is a backport please specify the original PR and why you need to backport that PR:

Before submitting your pull requests, make sure you followed this checklist:

verify that the PR is really intended for the chosen branch
verify that changes follow CMS Naming, Coding, And Style Rules
verify that the PR passes the basic test procedure suggested in the CMSSW PR instructions

commit 18abd3a Author: Patrick Gartung <[email protected]> Date: Wed May 25 02:12:21 2022 +0200 Code format commit f067f25 Author: Patrick Gartung <[email protected]> Date: Wed May 25 02:08:47 2022 +0200 Fix icc compilation errors commit b265ce2 Author: Patrick Gartung <[email protected]> Date: Fri May 20 23:22:10 2022 +0200 Add line to RecoTracker/MkFitCore/standalone/README_buildFromCMSSW.txt commit 8618997 Author: Patrick Gartung <[email protected]> Date: Fri May 20 18:18:50 2022 +0200 Add CMS_UNROLL_LOOP defines for Intel compilers and use with pragme omp simd instructions that cause internal icc error commit 45977ea Author: Patrick Gartung <[email protected]> Date: Fri May 20 18:17:54 2022 +0200 Squashed commit of the following: commit a565f52 Author: Patrick Gartung <[email protected]> Date: Thu May 19 15:23:44 2022 +0200 Remove fast-math flag and attribute needed to prevent segfault when fast-math is used commit c3fbce9 Author: Patrick Gartung <[email protected]> Date: Mon May 16 17:28:41 2022 +0200 Code format commit c388ab3 Author: Patrick Gartung <[email protected]> Date: Mon May 16 17:27:00 2022 +0200 Update location of #endif after enabling debug. #pragma omp simd not needed for debug statements. Move debug statements to own loop. commit 468b476 Author: Patrick Gartung <[email protected]> Date: Sat May 14 19:51:00 2022 +0200 No need to link to libmvec. It is called from libm on el8. Always use -ffastmath commit 0ef9036 Author: Patrick Gartung <[email protected]> Date: Fri May 13 20:01:43 2022 +0200 Only works on el8_amd64 arch commit 8866ca6 Author: Patrick Gartung <[email protected]> Date: Fri May 13 16:10:23 2022 +0200 Update for macro name change commit fe4a38f Author: Patrick Gartung <[email protected]> Date: Thu May 12 21:43:02 2022 +0200 code format? commit 8326129 Author: Patrick Gartung <[email protected]> Date: Thu May 12 21:18:44 2022 +0200 Add CMS_UNROLL_LOOP defines for Intel compilers and use with pragme omp simd instructions that cause internal icc error commit 056f7ef Author: Patrick Gartung <[email protected]> Date: Thu May 12 03:42:01 2022 +0200 Add CMS_UNROLL_LOOP_COUNT(Config::Niter) and initialize arrays with += and -= assignment statements to 0. first commit 31e2c83 Author: Patrick Gartung <[email protected]> Date: Wed May 11 05:16:27 2022 +0200 Add CMS_UNROLL_LOOP_COUNT(Config::Niter) and #include "FWCore/Utilities/interface/CMSUnrollLoop.h" commit 8eab5b1 Author: Patrick Gartung <[email protected]> Date: Wed May 11 04:57:47 2022 +0200 Change atrribute from no-fast-math to match-errno commit 688e905 Author: Patrick Gartung <[email protected]> Date: Tue May 10 17:37:25 2022 +0200 Code format commit ed58228 Author: Patrick Gartung <[email protected]> Date: Tue May 10 17:23:08 2022 +0200 Change attribute from no-inline to no-fast-math. Back out changes to applyMaterialEffects since it does not vectorize with the conditional on radL. commit c19f9c8 Author: Patrick Gartung <[email protected]> Date: Mon May 9 19:49:15 2022 +0200 code format commit 373ea0d Author: Patrick Gartung <[email protected]> Date: Mon May 9 19:02:48 2022 +0200 missed a few initialization of variable length arrays commit 753934c Author: Patrick Gartung <[email protected]> Date: Mon May 9 18:51:13 2022 +0200 for all el8 archs commit 2eb25ef Author: Patrick Gartung <[email protected]> Date: Mon May 9 18:31:35 2022 +0200 clang does not allow initialization of variable length arrays commit 7f23762 Author: Patrick Gartung <[email protected]> Date: Mon May 9 17:41:29 2022 +0200 Add fast-math and mvec when compiling under el8 commit 5f6efe2 Author: Patrick Gartung <[email protected]> Date: Mon May 9 16:07:08 2022 +0200 Comment out changes that are CERN specific commit 1ba6e06 Author: Patrick Gartung <[email protected]> Date: Thu May 5 21:25:33 2022 +0200 Try to break loop in applyMaterialEffects into smaller loops Combine arithmetic ops into one loop. Add -lmvec which is avaiable on RHEL8 commit dbc748c Author: Patrick Gartung <[email protected]> Date: Wed May 4 20:02:05 2022 +0200 Incorporate Dan Riley's suggested changes commit d6ae2c9 Author: Patrick Gartung <[email protected]> Date: Tue May 3 20:20:47 2022 +0200 Update helixAtZ to use small calculation loops commit 3d704ba Author: Patrick Gartung <[email protected]> Date: Fri Apr 29 21:43:00 2022 +0200 Work in progress commit 31550b4 Author: Patrick Gartung <[email protected]> Date: Fri Apr 29 20:03:09 2022 +0200 Fixups commit f10ef52 Author: Patrick Gartung <[email protected]> Date: Fri Apr 29 00:57:47 2022 +0200 WIP

cmsbuild · 2022-06-06T14:41:48Z

A new Pull Request was created by @gartung (Patrick Gartung) for CMSSW_12_4_X.

It involves the following packages:

FWCore/Utilities (core)
RecoTracker/MkFitCore (reconstruction)

@smuzaffar, @Dr15Jones, @makortel, @clacaputo, @cmsbuild, @slava77, @jpata can you please review it and eventually sign? Thanks.
@makortel, @felicepantaleo, @GiacomoSguazzoni, @JanFSchulte, @wddgit, @VinInn, @ebrondol, @rovere, @gpetruc, @mmusich, @mtosi, @dgulhan this is something you requested to watch as well.
@perrotta, @dpiparo, @qliphy you are the release manager for this.

cms-bot commands are listed here

gartung · 2022-06-06T14:47:18Z

enable profiling

gartung · 2022-06-06T14:47:41Z

please test

cmsbuild · 2022-06-07T02:44:44Z

+1

Summary: https://cmssdt.cern.ch/SDT/jenkins-artifacts/pull-request-integration/PR-640ff4/25310/summary.html
COMMIT: 05e1025
CMSSW: CMSSW_12_4_X_2022-06-06-1100/el8_amd64_gcc10
Additional Tests: PROFILING
User test area: For local testing, you can use /cvmfs/cms-ci.cern.ch/week0/cms-sw/cmssw/38259/25310/install.sh to create a dev area with all the needed externals and cmssw changes.

Comparison Summary

Summary:

No significant changes to the logs found
Reco comparison results: 63 differences found in the comparisons
DQMHistoTests: Total files compared: 50
DQMHistoTests: Total histograms compared: 3671502
DQMHistoTests: Total failures: 1120
DQMHistoTests: Total nulls: 0
DQMHistoTests: Total successes: 3670360
DQMHistoTests: Total skipped: 22
DQMHistoTests: Total Missing objects: 0
DQMHistoSizes: Histogram memory added: 0.0 KiB( 49 files compared)
Checked 208 log files, 45 edm output root files, 50 DQM output files
TriggerResults: no differences found

clacaputo · 2022-06-07T13:30:33Z

Hi @gartung, could you please edit the title by removing the URL? Thanks

clacaputo · 2022-06-08T08:42:19Z

+reconstruction

this PR is a backport of RecoTracker/MkFitCore: factorize loops to allow gcc -msse3 to vectorize loops #37868: code changes are in line with the original PR
reco changes are in some slimmedMET uncertanties_ and corrections_ quantities. This is related to the MET code, as discussed in Electron and photon masses are set to zero #36954 (comment)

gartung · 2022-06-09T01:09:33Z

@Dr15Jones can you sign for core?

Dr15Jones · 2022-06-09T13:53:32Z

+1

cmsbuild · 2022-06-09T13:53:56Z

This pull request is fully signed and it will be integrated in one of the next CMSSW_12_4_X IBs (tests are also fine) and once validation in the development release cycle CMSSW_12_5_X is complete. This pull request will now be reviewed by the release team before it's merged. @perrotta, @dpiparo, @qliphy (and backports should be raised in the release meeting by the corresponding L2)

qliphy · 2022-06-09T14:05:22Z

+1

cmsbuild added this to the CMSSW_12_4_X milestone Jun 6, 2022

cmsbuild added core-pending orp-pending pending-signatures reconstruction-pending tests-pending labels Jun 6, 2022

cmsbuild added tests-started and removed tests-pending labels Jun 6, 2022

cmsbuild added tests-approved and removed tests-started labels Jun 7, 2022

gartung changed the title ~~Backport of MkFitCore vectorization PR https://github.com/cms-sw/cmssw/pull/37868~~ Backport of MkFitCore vectorization PR 37868 Jun 7, 2022

cmsbuild added reconstruction-approved and removed reconstruction-pending labels Jun 8, 2022

cmsbuild added core-approved fully-signed and removed core-pending pending-signatures labels Jun 9, 2022

cmsbuild added orp-approved and removed orp-pending labels Jun 9, 2022

cmsbuild merged commit e83b01d into cms-sw:CMSSW_12_4_X Jun 9, 2022

gartung deleted the gartung-backport-MkfitCore-37868 branch August 22, 2022 20:26

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Backport of MkFitCore vectorization PR 37868 #38259

Backport of MkFitCore vectorization PR 37868 #38259

gartung commented Jun 6, 2022

cmsbuild commented Jun 6, 2022

gartung commented Jun 6, 2022

gartung commented Jun 6, 2022

cmsbuild commented Jun 7, 2022

clacaputo commented Jun 7, 2022

clacaputo commented Jun 8, 2022

gartung commented Jun 9, 2022

Dr15Jones commented Jun 9, 2022

cmsbuild commented Jun 9, 2022

qliphy commented Jun 9, 2022

Backport of MkFitCore vectorization PR 37868 #38259

Backport of MkFitCore vectorization PR 37868 #38259

Conversation

gartung commented Jun 6, 2022

PR description:

PR validation:

if this PR is a backport please specify the original PR and why you need to backport that PR:

cmsbuild commented Jun 6, 2022

gartung commented Jun 6, 2022

gartung commented Jun 6, 2022

cmsbuild commented Jun 7, 2022

Comparison Summary

clacaputo commented Jun 7, 2022

clacaputo commented Jun 8, 2022

gartung commented Jun 9, 2022

Dr15Jones commented Jun 9, 2022

cmsbuild commented Jun 9, 2022

qliphy commented Jun 9, 2022