#!/bin/csh unalias rm unalias cp unalias mv # This script runs a large number of WRF regression tests. On the # IBM machines (which at NCAR allow queueing), the jobs are sent to # the queue system via load leveler commands. On all other machines, # the jobs are sent in sequentially as soon as the process returns # from completing the previous task. # HOW TO RUN # ---------- # Linux # joshua1 or joshua3 # mkdir /data3/mp/$USER/`hostname` # put all_reg.csh, regtest.csh, and wrf.tar in dir # execute all_reg.csh # takes about 36-48 h # flex lm errors show up as fails to compile # AIX # bluevista or blueice # put all_reg.csh, regtest.csh, and wrf.tar in ~ # execute all_reg.csh # takes about 8-10 h # Unless you are editing the script, no changes are required below #======================================================================= #======================================================================= # The only really important thing. if ( ! -e wrf.tar ) then echo " " echo "Whoa there pardner, where is that wrf.tar file" echo " " exit ( 1 ) endif # What these tests do, must be a single string. set NAME = ( "Standard" "NESTED=FALSE" "NESTED=FALSE" "NONE" 1 \ "Moving_Nest1" "NESTED1=FALSE" "NESTED1=TRUE" "NONE" 2 \ "Moving_Nest2" "NESTED2=FALSE" "NESTED2=TRUE" "NONE" 3 \ "NMM_Nest" "NESTED2=FALSE" "NESTED2=FALSE" "NONE" 4 \ "Full_Optimization" "REG_TYPE=BIT4BIT" "REG_TYPE=OPTIMIZED" "NONE" 5 \ "Chemistry" "CHEM=FALSE" "CHEM=TRUE" "NONE" 6 \ "Chemistry2" "KPP=FALSE" "KPP=TRUE" "NONE" 7 \ "Quilting" "QUILT=FALSE" "QUILT=TRUE" "NONE" 8 \ "Binary_IO" "IO_FORM=2" "IO_FORM=1" "NONE" 9 \ "GriB1_Output" "IO_FORM=2" "IO_FORM=5" "NONE" 10 \ "REAL8_Floats" "REAL8=FALSE" "REAL8=TRUE" "NONE" 11 \ "FDDA" "FDDA=FALSE" "FDDA=TRUE" "NONE" 12 \ "FDDA2" "FDDA2=FALSE" "FDDA2=TRUE" "NONE" 13 \ "ESMF_Library" "ESMF_LIB=FALSE" "ESMF_LIB=TRUE" "ONLY_AIX" 14 \ "Global" "GLOBAL=FALSE" "GLOBAL=TRUE" "NONE" 15 \ "Adaptive" "ADAPTIVE=FALSE" "ADAPTIVE=TRUE" "NONE" 16 \ ) # Where are we located. set starting_dir = `pwd` # Get the tag manually. This is for the auto-report that gets # sent to the WRF web page on the history of the regression tests. set current_year4 = `date -u +"%Y"` set current_month = `date -u +"%m"` set current_day = `date -u +"%d"` set current_hour = `date -u +"%H"` set current_minute = `date -u +"%M"` set current_second = `date -u +"%S"` set datehms = ${current_year4}-${current_month}-${current_day}_${current_hour}:${current_minute}:${current_second}_UTC set dateh = ${current_year4}${current_month}${current_day}${current_hour} set date = ${current_year4}${current_month}${current_day} if ( ! $?TAG ) then if ( $user == michalak ) then set initials = jm else if ( $user == hender ) then set initials = th else if ( $user == gill ) then set initials = dg else if ( $user == dudhia ) then set initials = jd else if ( $user == weiwang ) then set initials = ww else set initials = XX endif echo the TAG is NOT defined echo Please define an environment variable that is the WRFV3 tag echo Something such as: setenv TAG trunk_${date}_${initials} echo " " exit ( 1 ) endif # Are we only interested in the PASS/FAIL report? if ( $#argv == 1 ) then if ( $argv[1] == PASSFAIL ) then goto PASSFAIL endif endif # If there are any command line args, they are processed, else # we run all of the regression tests without a generate or # compare flag being set. if ( $#argv == 0 ) then set BASELINE = RUN_ONLY set OLD_TEXT set NEW_TEXT set TOAST set tests set TEST_NUM set count_test = 1 while ( $count_test < $#NAME ) set tests = ( $tests "$NAME[$count_test]" ) @ count_test ++ set OLD_TEXT = ( $OLD_TEXT "$NAME[$count_test]" ) @ count_test ++ set NEW_TEXT = ( $NEW_TEXT "$NAME[$count_test]" ) @ count_test ++ set TOAST = ( $TOAST "$NAME[$count_test]" ) @ count_test ++ set TEST_NUM = ( $TEST_NUM "$NAME[$count_test]" ) @ count_test ++ end # We have some command line args. They are either a request to run # the test with a generate/compare flag, or a list of tests to perform. else set INIT_OPTS = ( $* ) # First, find the baseline type. This is going to be one of three # possibilities: GENERATE, COMPARE, or a RUN_ONLY option. The # default is RUN_ONLY. Only the first baseline option found is # used. set count = 0 while ( $count < $#INIT_OPTS ) @ count ++ set arg = $INIT_OPTS[$count] if ( ( $arg == GENERATE ) || \ ( $arg == COMPARE ) || \ ( $arg == RUN_ONLY ) ) then set BASELINE = $arg goto FINISHED_BASELINE_TYPE endif end set BASELINE = RUN_ONLY FINISHED_BASELINE_TYPE: # If there was only one input, and it was telling us to do which type of baseline # option (GENERATE, COMPARE, RUN_ONLY), we assume they want all of the tests # conducted. if ( $#INIT_OPTS == 1 ) then if ( ( $INIT_OPTS == GENERATE ) || ( $INIT_OPTS == COMPARE ) || ( $INIT_OPTS == RUN_ONLY ) ) then set OLD_TEXT set NEW_TEXT set TOAST set tests set TEST_NUM set count_test = 1 while ( $count_test < $#NAME ) set tests = ( $tests "$NAME[$count_test]" ) @ count_test ++ set OLD_TEXT = ( $OLD_TEXT "$NAME[$count_test]" ) @ count_test ++ set NEW_TEXT = ( $NEW_TEXT "$NAME[$count_test]" ) @ count_test ++ set TOAST = ( $TOAST "$NAME[$count_test]" ) @ count_test ++ set TEST_NUM = ( $TEST_NUM "$NAME[$count_test]" ) @ count_test ++ end goto FINISHED_TEST_LIST endif endif # Find which tests are to be conducted. Loop over all of the # input, and compare each of the input fields with the list # of available test names. When a match occurs, increment the # test found counter, and save the test name. set count = 0 set OLD_TEXT set NEW_TEXT set TOAST set tests set TEST_NUM while ( $count < $#INIT_OPTS ) @ count ++ set count_test = 1 while ( $count_test < $#NAME ) if ( $INIT_OPTS[$count] == $NAME[$count_test] ) then set tests = ( $tests "$NAME[$count_test]" ) @ count_test ++ set OLD_TEXT = ( $OLD_TEXT "$NAME[$count_test]" ) @ count_test ++ set NEW_TEXT = ( $NEW_TEXT "$NAME[$count_test]" ) @ count_test ++ set TOAST = ( $TOAST "$NAME[$count_test]" ) @ count_test ++ set TEST_NUM = ( $TEST_NUM "$NAME[$count_test]" ) @ count_test ++ else @ count_test += 5 endif end end # If there are no recognizable tests requested, let them know our concern. if ( $#tests == 0 ) then echo No valid test requested in argument list exit ( 4 ) endif endif FINISHED_TEST_LIST: # A friendly check for the baseline directory existence, and locations for known # NCAR machines. if ( ( $BASELINE == GENERATE ) || ( $BASELINE == COMPARE ) ) then if ( ( `uname` == AIX ) && ( ( `hostname | cut -c 1-2` != bs ) && \ ( `hostname | cut -c 1-2` != bv ) && ( `hostname | cut -c 1-2` != be ) ) ) then set SAVE_DIR = /ptmp/${USER}/BASELINE/`uname` else if ( `uname` == AIX ) then set SAVE_DIR = /ptmp/${USER}/BASELINE/`uname` else if ( ( `uname` == Linux ) && ( `hostname` == basswood ) ) then set SAVE_DIR = /basswood/${user}/Regression_Tests/BASELINE/`uname` else if ( ( `uname` == Darwin ) && ( `hostname` == stink ) ) then set SAVE_DIR = /stink/${user}/Regression_Tests/BASELINE/`uname` else if ( ( `uname` == Linux ) && ( `hostname` == bay-mmm ) ) then set SAVE_DIR = /data3/mp/${USER}/BASELINE/`uname` else echo Hmm, no idea where to put/get this baseline data, stopping exit ( 10 ) endif if ( ( -d $SAVE_DIR ) && ( $BASELINE == GENERATE ) ) then echo "Troubles with SAVE_DIR logic." echo "$SAVE_DIR should not exist for a $BASELINE run." # exit ( 2 ) else if ( ( ! -d $SAVE_DIR ) && ( $BASELINE == COMPARE ) ) then echo "Troubles with SAVE_DIR logic." echo "$SAVE_DIR should exist for a $BASELINE run." exit ( 3 ) endif endif # We need the regtest.csh file, badly. if ( ! -e regtest.csh ) then echo we need regtest.csh in this dir echo "supply-ez vous, s'il vous plait" exit ( 1 ) endif # The regtest.csh file is treated as a template. The following # strings (first occurrence) is sought (OLD_TEXT) and replaced # with the modified string (NEW_TEXT). This edited regression # script is then processed. # Any exceptions to where they can run? NONE means no # exceptions, should run on all machines. AIX/Linux # means it will NOT run on that single machine. The # option ONLY_AIX/ONLY_Linux means that the option # ONLY works on that specific architecture. # Loop over all selected tests. set count_test = 0 while ( $count_test < $#tests ) @ count_test ++ set count = 1 while ( $count < $#NAME ) if ( "$tests[$count_test]" == "$NAME[$count]" ) then goto FOUND_SELECTED_TEST endif @ count += 5 end echo "Hmmm, no valid test found" exit ( 11 ) FOUND_SELECTED_TEST: # Specifically skip this test on this architecture. if ( `uname` == $TOAST[$count_test] ) then echo skipping test $tests[$count_test] for `uname` specifically # Skip this test on this architecture because it ONLY runs # on a different architecture. else if ( ( `echo $TOAST[$count_test] | cut -c 1-5` == ONLY_ ) && ( ONLY_`uname` != $TOAST[$count_test] ) ) then echo skipping test $tests[$count_test] for `uname`, works on $TOAST[$count_test] # OK, we are allowed to run this test on this architecture. else if ( ( $TOAST[$count_test] == NONE ) || \ ( $TOAST[$count_test] != `uname` ) || \ ( ONLY_`uname` == $TOAST[$count_test] ) ) then echo doing test $tests[$count_test] for `uname` # If this is the generate or compare baseline test, where do we # save the data to/read the data from. if ( ( $BASELINE == GENERATE ) || ( $BASELINE == COMPARE ) ) then if ( ( `uname` == AIX ) && ( ( `hostname | cut -c 1-2` != bs ) && \ ( `hostname | cut -c 1-2` != bv ) && ( `hostname | cut -c 1-2` != be ) ) ) then set SAVE_DIR = /ptmp/${USER}/BASELINE/`uname`/$tests[$count_test] else if ( `uname` == AIX ) then set SAVE_DIR = /ptmp/${USER}/BASELINE/`uname`/$tests[$count_test] else if ( ( `uname` == Linux ) && ( `hostname` == basswood ) ) then set SAVE_DIR = /basswood/${USER}/Regression_Tests/BASELINE/`uname`/$tests[$count_test] else if ( ( `uname` == Darwin ) && ( `hostname` == stink ) ) then set SAVE_DIR = /stink/${USER}/Regression_Tests/BASELINE/`uname`/$tests[$count_test] else if ( ( `uname` == Linux ) && ( `hostname` == bay-mmm ) ) then set SAVE_DIR = /data3/mp/${USER}/BASELINE/`uname`/$tests[$count_test] else echo No idea where to put the data, stopping exit ( 2 ) endif # Either zap existing stuff (GENERATE), or make sure it is there (COMPARE) if ( $BASELINE == GENERATE ) then /bin/rm -rf $SAVE_DIR else if ( ( $BASELINE == COMPARE ) && ( $tests[$count_test] != Full_Optimization ) ) then if ( ! -d $SAVE_DIR ) then echo $SAVE_DIR does not exist for BASELINE comparison, stopping exit ( 3 ) endif else if ( ( $BASELINE == COMPARE ) && ( $tests[$count_test] == Full_Optimization ) ) then echo No comparison done with baseline since this is an optimized run endif endif # Build the short edit input script for ed and edit the regtest.csh file. set OLDT = `echo $OLD_TEXT[$count_test] | sed 's/=/ = /'` set NEWT = `echo $NEW_TEXT[$count_test] | sed 's/=/ = /'` if ( $tests[$count_test] == "NMM_Nest" ) then cp regtest_nmmnest.csh reg.foo.$TEST_NUM[$count_test].$tests[$count_test] else if ( ( $BASELINE == RUN_ONLY ) || ( $tests[$count_test] == Full_Optimization ) ) then if ( -e ed.in ) rm ed.in cat >! ed_in << EOF ,s/$OLDT/$NEWT/ ,s/1 2 3 4 5 6 7 8 9 10 11 12 13 14/1 2 3 4 5 6 7 8 9 10 11 12/ w reg.foo.$TEST_NUM[$count_test].$tests[$count_test] q EOF else if ( $BASELINE == GENERATE ) then if ( -e ed.in ) rm ed.in cat >! ed_in << EOF ,s/$OLDT/$NEWT/ ,s/1 2 3 4 5 6 7 8 9 10 11 12 13 14/1 2 3 4 5 6 7 8 9 10 11 12/ ,s?GENERATE_BASELINE = FALSE?GENERATE_BASELINE = $SAVE_DIR? w reg.foo.$TEST_NUM[$count_test].$tests[$count_test] q EOF else if ( $BASELINE == COMPARE ) then if ( -e ed.in ) rm ed.in cat >! ed_in << EOF ,s/$OLDT/$NEWT/ ,s/1 2 3 4 5 6 7 8 9 10 11 12 13 14/1 2 3 4 5 6 7 8 9 10 11 12/ ,s?COMPARE_BASELINE = FALSE?COMPARE_BASELINE = $SAVE_DIR? w reg.foo.$TEST_NUM[$count_test].$tests[$count_test] q EOF endif ed regtest.csh < ed_in >& /dev/null endif chmod +x reg.foo.$TEST_NUM[$count_test].$tests[$count_test] # On AIX, we submit jobs to the load leveler queue for bluesky and to the LSF queue for # bluevista. After submission, we wait around until it completes, and then we send in # the next one. if ( ( `uname` == AIX ) && ( `hostname | cut -c 1-2` == bs ) ) then llsubmit reg.foo.$TEST_NUM[$count_test].$tests[$count_test] >&! llsub.out set ok = 0 set in_already = 0 while ( $ok == 0 ) sleep 10 ; llq -u $USER >&! llq.report grep `cat llsub.out | grep '"bs1' | cut -d\" -f2` llq.report >& /dev/null set ok = $status if ( ( $ok == 0 ) && ( $in_already == 0 ) ) then set in_already = 1 set joe_id = `cat llsub.out | grep '"bs1' | cut -d\" -f2 | cut -d. -f2` endif end cp /ptmp/$USER/wrf_regression.$joe_id/wrftest.output wrftest.output.$TEST_NUM[$count_test].$tests[$count_test] rm llsub.out llq.report else if ( ( `uname` == AIX ) && ( ( `hostname | cut -c 1-2` == bv ) || ( `hostname | cut -c 1-2` == be ) ) ) then bsub < reg.foo.$TEST_NUM[$count_test].$tests[$count_test] >&! bsub.out set ok = 0 set in_already = 0 while ( $ok == 0 ) sleep 10 ; bjobs >&! bjobs.report grep `cat bsub.out | grep Job | cut -d"<" -f2 | cut -d">" -f1` bjobs.report >& /dev/null set ok = $status if ( ( $ok == 0 ) && ( $in_already == 0 ) ) then set in_already = 1 set joe_id = `cat bsub.out | grep Job | cut -d"<" -f2 | cut -d">" -f1` endif end cp /ptmp/$USER/wrf_regression.$joe_id/wrftest.output wrftest.output.$TEST_NUM[$count_test].$tests[$count_test] rm bsub.out bjobs.report else if ( ( `uname` == AIX ) && ( ( `hostname | cut -c 1-2` != bs ) && \ ( `hostname | cut -c 1-2` != bv ) && ( `hostname | cut -c 1-2` != be ) ) ) then llsubmit reg.foo.$TEST_NUM[$count_test].$tests[$count_test] >&! llsub.out set ok = 0 set in_already = 0 while ( $ok == 0 ) sleep 10 ; llq -u $USER >&! llq.report set llsubmit_name_all = `cat llsub.out | grep '"b' | cut -d\" -f2` set llsubmit_name_front = `echo $llsubmit_name_all | cut -d\. -f1` set llsubmit_name_end = `echo $llsubmit_name_all | cut -d\. -f5` grep $llsubmit_name_front llq.report | grep $llsubmit_name_end >& /dev/null set ok = $status if ( ( $ok == 0 ) && ( $in_already == 0 ) ) then set in_already = 1 set joe_id = `cat llsub.out | grep '"b' | cut -d\" -f2 | cut -d. -f5` endif end cp /ptmp/$USER/wrf_regression.$joe_id/wrftest.output wrftest.output.$TEST_NUM[$count_test].$tests[$count_test] rm llsub.out llq.report # On the "other" non-queued machines, we just execute the script and wait until # we get the process returning control, then we move on. else reg.foo.$TEST_NUM[$count_test].$tests[$count_test] -f wrf.tar >&! output.$TEST_NUM[$count_test].$tests[$count_test] mv wrf_regression/wrftest.output wrftest.output.$TEST_NUM[$count_test].$tests[$count_test] if ( -d wrf_regression ) then mv wrf_regression wrf_regression.$TEST_NUM[$count_test].$tests[$count_test] endif endif endif end if ( ( `uname` == AIX ) && ( ( `hostname | cut -c 1-2` != bs ) && \ ( `hostname | cut -c 1-2` != bv ) && ( `hostname | cut -c 1-2` != be ) ) ) then echo no web page building, stopping exit endif PASSFAIL: # Build the html page. We only need the middle portion. It's # a table with 5 columns: Date of test, WRFV3 tag, Developer # who conducted the test, machine the test was run on, and the # pass/fail status of the all_reg.csh script when compared # to the benchmark results (usually a released code). cat >! history_middle_OK.html << EOF