-
Notifications
You must be signed in to change notification settings - Fork 545
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Output OMP threading info to screen [ww3_shel, ww3_multi] #1191
Output OMP threading info to screen [ww3_shel, ww3_multi] #1191
Conversation
Thanks for sharing this @ukmo-ccbunney ! I'll run some tests on my end with this. |
Update with regtest results. |
@ukmo-ccbunney I was finally able to run some tests with this, and I do at least get the expected number of threads that I request, although I'm not seeing any impact on model performance using various thread counts... |
How strange! Do you have another machine you can test on? I sometimes run stuff on my linux desktop to check non-MPI code (including OMP runs). |
I do, but it's on maintenance today... I'll try that tomorrow! |
@ukmo-ccbunney I made a PR to your branch because I was running this and realized I wasn't getting all the build information to the log file that I wanted to see. It's here: ukmo-waves#46 I can put it in a new PR if you'd like but I was on this branch so just pushed here. The good news is my tests this afternoon are finally showing OMP speed-up closer to what's expected so maybe the other day was just an off day... I don't know. But this is a nice update to WW3 for a sanity check. |
Thanks - that's a really useful update. I've merged it into this branch.
That's good news. Odd how you were experiencing such a lot of variability. It almost sounds like CPU over-subscription on your HPC? Is there a chance that other jobs are grabbing resources from a node you are running on? Also, it could be a result of file system contention too (although our WW3 files are not huge...) When our HPC is running it's weekly RAID check the I/O throughput is degraded and I can see an impact on model runtimes. Just a thought! Perhaps it's just gremlins... |
Would not be surprised by the gremlins!!! I have no idea what is going on to be honest but am happy to see what I was expecting finally. File system or other things would not surprise me at all... |
@ukmo-ccbunney @JessicaMeixner-NOAA I wanted to see if either of you are still looking at this, or is it ready for final review? |
Hi @MatthewMasarik-NOAA |
Awesome, thanks @ukmo-ccbunney, I'll get the tests going! |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Code review
Pass
Testing
Pass
**********************************************************************
********************* non-identical cases ****************************
**********************************************************************
mww3_test_03/./work_PR3_UQ_MPI_e_c (1 files differ)
mww3_test_03/./work_PR3_UNO_MPI_e (1 files differ)
mww3_test_03/./work_PR2_UQ_MPI_e (1 files differ)
mww3_test_03/./work_PR2_UNO_MPI_e (1 files differ)
mww3_test_03/./work_PR2_UNO_MPI_d2 (16 files differ)
mww3_test_03/./work_PR1_MPI_d2 (10 files differ)
mww3_test_03/./work_PR3_UNO_MPI_d2_c (16 files differ)
mww3_test_03/./work_PR3_UQ_MPI_d2_c (15 files differ)
mww3_test_03/./work_PR3_UNO_MPI_d2 (12 files differ)
mww3_test_03/./work_PR2_UQ_MPI_d2 (15 files differ)
mww3_test_03/./work_PR3_UQ_MPI_e (1 files differ)
mww3_test_03/./work_PR3_UNO_MPI_e_c (1 files differ)
mww3_test_03/./work_PR3_UQ_MPI_d2 (16 files differ)
mww3_test_09/./work_MPI_ASCII (0 files differ)
ww3_tp2.10/./work_MPI_OMPH (7 files differ)
ww3_tp2.16/./work_MPI_OMPH (4 files differ)
ww3_tp2.6/./work_ST4_ASCII (0 files differ)
ww3_ufs1.3/./work_a (3 files differ)
**********************************************************************
************************ identical cases *****************************
**********************************************************************
Approved.
Thank you @ukmo-ccbunney and @JessicaMeixner-NOAA, having the OMP thread count and full build log output will be very useful for troubleshooting build and runtime issues. |
* Bugfix - initialised VD and VS to zero in w3srcemd. (NOAA-EMC#1037) * More efficient test for binary files in matrix.comp (NOAA-EMC#1035) * Tidy up of pre-processor directives and unused variables in w3srcemd.F90 (NOAA-EMC#1010) * Correct typo in w3srcemd.F90 pre-processor directive. (NOAA-EMC#1039) * minor bugfix for matrix grepping on keywords (NOAA-EMC#1049) * Stop masking group 1 output where icec > icen (NOAA-EMC#1019) * Doxygen documentation added, 8th subset.(NOAA-EMC#1046) * NC4 ,F90 ,XX0 switches removed from ww3_tp2.19 regtest (NOAA-EMC#1054) * CI: Fix for Intel scripts. GNU scripts updated. (NOAA-EMC#1064) * correct the computation of QP parameter, add QKK output parameter, change UST scale factor (NOAA-EMC#1050) * correct issue with ww3_multi when requesting restart2 and using nml file instead of inp file (NOAA-EMC#1070) * correct calendar for track netcdf output (NOAA-EMC#1079) * Fix missing mod_def.ww3 file in multigrid regression tests for track output (NOAA-EMC#1091) * STAB3: fix cmake build for ST4 or ST3 (NOAA-EMC#1086) * new feature to output out_grd.ww3, out_pnt.ww3 and mod_def.ww3 both in binary and ascii format using switch ASCII. (NOAA-EMC#1089) * Update local unit number arrays (NDS, MDS) to be same size of array defined in w3odatmd (size=15). Also, defined unit numbers for NDS(14) and NDS(15). (NOAA-EMC#1098) * Removed code referencing PHIOC in output section for PHICE in ww3_ounf (NOAA-EMC#1093) * implementation of the GQM (Gaussian Quadrature Method) to replace the DIA in NL1 or NL2. (NOAA-EMC#1083) * update logic to ensure you are not accessing uninitialized dates (NOAA-EMC#1114) * Initialised S and D arrays in W3SDB1 before potential early return if zero energy. (NOAA-EMC#1115) * ww3_ounp.F90: x/y units attribute corrected from 'm' to 'km' (NOAA-EMC#1088) * Bugfix: Assign unit numbers to ASCII gridded/point output in multi-grid mode. (NOAA-EMC#1118) * correct bugs to run correctly GQM implementation (NOAA-EMC#1127) * Adding documentation to w3iopo() in preparation for code for NOAA-EMC#682. (NOAA-EMC#1131) * NCEP regtest module updates: uses spack-stack/1.5.0, includes scotch/7.0.4 (NOAA-EMC#1137) * Minor update to ncep regtests (NOAA-EMC#1138) * Updated intel workflow to install oneapi compilers from new location. (NOAA-EMC#1157) * Add unit test for points I/O code. (NOAA-EMC#1158) * Update Intel CI (relocate /usr/local; ensure intel-oneapi-mpi; use ubuntu-latest) (NOAA-EMC#1161) * remove lookup table for ST4 to speed up computation and clean up the ST4 code (NOAA-EMC#1124) Co-authored-by: Fabrice Ardhuin <[email protected]> * initialize USSP_WN for mod_def (NOAA-EMC#1165) * Introduce IC4M8 and IC4M9 to WW3 (NOAA-EMC#1176) * clean up and add ST4 variables (NOAA-EMC#1181) * w3fld1md.F90: fix divide by zero in CRIT2 parameter (NOAA-EMC#1184) * ww3_prnc.F90: fix out-of-scope grid index write statement (NOAA-EMC#1185) * Bugfix: address potential divide-by-zero in APPENDTAIL (NOAA-EMC#1188) Co-authored-by: Denise Worthen <[email protected]> * Provide initial drying of cells with depth < ZLIM for SMC grid. (NOAA-EMC#1192) * Output OMP threading info to screen when running ww3_shel/ww3_multi compiled with the OMPG switch. Also fixes truncation of build.log when running run_cmake_build. (NOAA-EMC#1191) * Added screen output showing number of threads when OMP enabled. * update build to get more info in logs (NOAA-EMC#46) --------- Co-authored-by: Jessica Meixner <[email protected]> * update run_cmake_test to catch build errors and exit (NOAA-EMC#1194) * fix merge conflicts * Fix gustiness bug, as suggst by Pieter * Change USTARsigma to WAM implementation --------- Co-authored-by: Chris Bunney <[email protected]> Co-authored-by: Mickael Accensi <[email protected]> Co-authored-by: Benoit Pouliot <[email protected]> Co-authored-by: Matthew Masarik <[email protected]> Co-authored-by: Ghazal-Mohammadpour <[email protected]> Co-authored-by: Jessica Meixner <[email protected]> Co-authored-by: Biao Zhao <[email protected]> Co-authored-by: Edward Hartnett <[email protected]> Co-authored-by: Alex Richert <[email protected]> Co-authored-by: Fabrice Ardhuin <[email protected]> Co-authored-by: W. Erick Rogers <[email protected]> Co-authored-by: Denise Worthen <[email protected]> Co-authored-by: Camille Teicheira <[email protected]>
Pull Request Summary
Added screen output in
ww3_shel
andww3_multi
showing number of (available) threads when OMP enabled.Also includes a fix to avoid truncation of the
build.log
file (see: ukmo-waves#46)Description
When running an OMP enabled version of
ww3_shel
orww3_multi
(i.e. compiled withOMPG
), the program will output a message to the screen stating that OMP is enabled and the number of threads available to the parallel regions.Note, in MPI/OMP hybrid mode only the "screen" processor will output this message, so the number of OMP threads will be relevant to that processor. I imagine that most people are not running a different number of threads per processor, but that is something to bear in mind!
No change to the regtests outputs are expected, other than the new line in the screen output for
ww3_shel
andww3_multi
in theOMP
/OMPH
regtests.Issue(s) addressed
Fixes: #1032
Commit Message
Output OMP threading info to screen when running ww3_shel/ww3_multi compiled with the OMPG switch.
Also fixes truncation of build.log when running
run_cmake_build
.Check list
Testing
Regtest results
There are some differences in the
ww3_shel.out
andww3_multi.out
files for OMP/OMPH tests, but these differences are filtered out by thematrix.comp
script. The differences are all like this: