Merge branch 'azamat/machines/add-improv' into next (PR #6205)

Add ANL Improv to E3SM machines. Also, disable slow archiving of old tests on all LCRC machines. [BFB]
E3SM-Project · Mar 8, 2024 · 739807d · 739807d
2 parents eb2c1d6 + 52b88bb
commit 739807d
Show file tree

Hide file tree

Showing 7 changed files with 148 additions and 8 deletions.
diff --git a/cime_config/allactive/config_pesall.xml b/cime_config/allactive/config_pesall.xml
@@ -1379,6 +1379,34 @@
         </nthrds>
       </pes>
     </mach>
+    <mach name="improv">
+      <pes compset="any" pesize="any">
+        <comment>improv: any compset on ne30np4 grid</comment>
+        <ntasks>
+          <ntasks_atm>-4</ntasks_atm>
+          <ntasks_lnd>-4</ntasks_lnd>
+          <ntasks_rof>-4</ntasks_rof>
+          <ntasks_ice>-4</ntasks_ice>
+          <ntasks_ocn>-4</ntasks_ocn>
+          <ntasks_cpl>-4</ntasks_cpl>
+          <ntasks_glc>-4</ntasks_glc>
+          <ntasks_wav>-4</ntasks_wav>
+        </ntasks>
+      </pes>
+      <pes compset=".*EAM.+ELM.+MPASSI.+SGLC.+SWAV.+BGC.*" pesize="any">
+        <comment>improv: BGC compset on ne30np4 grid</comment>
+        <ntasks>
+          <ntasks_atm>-8</ntasks_atm>
+          <ntasks_lnd>-8</ntasks_lnd>
+          <ntasks_rof>-8</ntasks_rof>
+          <ntasks_ice>-8</ntasks_ice>
+          <ntasks_ocn>-8</ntasks_ocn>
+          <ntasks_cpl>-8</ntasks_cpl>
+          <ntasks_glc>-8</ntasks_glc>
+          <ntasks_wav>-8</ntasks_wav>
+        </ntasks>
+      </pes>
+    </mach>
   </grid>
   <grid name="a%ne30np4.pg2_l%.+_oi%oEC60to30v3">
     <mach name="compy">

diff --git a/cime_config/machines/Depends.oneapi-ifx.cmake b/cime_config/machines/Depends.oneapi-ifx.cmake
@@ -1,3 +1,5 @@
 
 # compile mpas_seaice_core_interface.f90 with ifort, not ifx
-e3sm_add_flags("${CMAKE_BINARY_DIR}/core_seaice/model_forward/mpas_seaice_core_interface.f90" "-fc=ifort")
+if (NOT MPILIB STREQUAL "openmpi")
+  e3sm_add_flags("${CMAKE_BINARY_DIR}/core_seaice/model_forward/mpas_seaice_core_interface.f90" "-fc=ifort")
+endif()
diff --git a/cime_config/machines/cmake_macros/gnu_improv.cmake b/cime_config/machines/cmake_macros/gnu_improv.cmake
@@ -0,0 +1,8 @@
+if (COMP_NAME STREQUAL gptl)
+  string(APPEND CPPDEFS " -DHAVE_SLASHPROC")
+endif()
+string(APPEND CMAKE_C_FLAGS_RELEASE " -O2")
+string(APPEND CMAKE_Fortran_FLAGS_RELEASE " -O2")
+set(PIO_FILESYSTEM_HINTS "gpfs")
+string(APPEND SLIBS " /lcrc/group/e3sm/soft/improv/netlib-lapack/3.12.0/gcc-12.3.0/liblapack.a /lcrc/group/e3sm/soft/improv/netlib-lapack/3.12.0/gcc-12.3.0/libblas.a")
+
diff --git a/cime_config/machines/config_batch.xml b/cime_config/machines/config_batch.xml
@@ -423,6 +423,16 @@
       </queues>
     </batch_system>
 
+  <batch_system MACH="improv" type="pbspro">
+    <directives>
+      <directive> -l select={{ num_nodes }}:mpiprocs={{ tasks_per_node }} </directive>
+    </directives>
+    <queues>
+      <queue walltimemax="00:30:00" jobmin="1" jobmax="8" strict="true" default="true">debug</queue>
+      <queue walltimemax="00:30:00" jobmin="9" jobmax="737" strict="true">compute</queue>
+    </queues>
+  </batch_system>
+
   <batch_system MACH="miller" type="miller_slurm">
     <queues>
       <queue walltimemax="06:00:00" nodemax="800" default="true">collaboration</queue>

diff --git a/cime_config/machines/config_machines.xml b/cime_config/machines/config_machines.xml
@@ -2616,7 +2616,7 @@
     <EXEROOT>$CIME_OUTPUT_ROOT/$CASE/bld</EXEROOT>
     <TEST_TPUT_TOLERANCE>0.05</TEST_TPUT_TOLERANCE>
     <TEST_MEMLEAK_TOLERANCE>0.05</TEST_MEMLEAK_TOLERANCE>
-    <MAX_GB_OLD_TEST_DATA>1000</MAX_GB_OLD_TEST_DATA>
+    <MAX_GB_OLD_TEST_DATA>0</MAX_GB_OLD_TEST_DATA>
     <environment_variables>
       <env name="PERL5LIB">/lcrc/group/e3sm/soft/perl/chrys/lib/perl5</env>
       <env name="NETCDF_C_PATH">$SHELL{dirname $(dirname $(which nc-config))}</env>
@@ -2698,7 +2698,7 @@
     <RUNDIR>$CIME_OUTPUT_ROOT/$CASE/run</RUNDIR>
     <EXEROOT>$CIME_OUTPUT_ROOT/$CASE/bld</EXEROOT>
     <TEST_TPUT_TOLERANCE>0.1</TEST_TPUT_TOLERANCE>
-    <MAX_GB_OLD_TEST_DATA>1000</MAX_GB_OLD_TEST_DATA>
+    <MAX_GB_OLD_TEST_DATA>0</MAX_GB_OLD_TEST_DATA>
     <environment_variables>
       <env name="NETCDF_C_PATH">$SHELL{dirname $(dirname $(which nc-config))}</env>
       <env name="NETCDF_FORTRAN_PATH">$SHELL{dirname $(dirname $(which nf-config))}</env>
@@ -2777,7 +2777,7 @@
     <RUNDIR>$CIME_OUTPUT_ROOT/$CASE/run</RUNDIR>
     <EXEROOT>$CIME_OUTPUT_ROOT/$CASE/bld</EXEROOT>
     <TEST_TPUT_TOLERANCE>0.1</TEST_TPUT_TOLERANCE>
-    <MAX_GB_OLD_TEST_DATA>1000</MAX_GB_OLD_TEST_DATA>
+    <MAX_GB_OLD_TEST_DATA>0</MAX_GB_OLD_TEST_DATA>
     <environment_variables>
       <env name="NETCDF_C_PATH">$SHELL{dirname $(dirname $(which nc-config))}</env>
       <env name="NETCDF_FORTRAN_PATH">$SHELL{dirname $(dirname $(which nf-config))}</env>
@@ -2897,6 +2897,68 @@
     </environment_variables>
   </machine>
 
+  <machine MACH="improv">
+    <DESC>ANL LCRC cluster 825-node AMD 7713 2-sockets 128-cores per node</DESC>
+    <NODENAME_REGEX>ilogin(1|2|3|4).lcrc.anl.gov</NODENAME_REGEX>
+    <OS>LINUX</OS>
+    <COMPILERS>gnu</COMPILERS>
+    <MPILIBS>openmpi</MPILIBS>
+    <PROJECT>e3sm</PROJECT>
+    <CIME_OUTPUT_ROOT>/lcrc/group/e3sm/$USER/scratch/improv</CIME_OUTPUT_ROOT>
+    <DIN_LOC_ROOT>/lcrc/group/e3sm/data/inputdata</DIN_LOC_ROOT>
+    <DIN_LOC_ROOT_CLMFORC>/lcrc/group/e3sm/data/inputdata/atm/datm7</DIN_LOC_ROOT_CLMFORC>
+    <DOUT_S_ROOT>/lcrc/group/e3sm/$USER/scratch/improv/archive/$CASE</DOUT_S_ROOT>
+    <BASELINE_ROOT>/lcrc/group/e3sm/baselines/improv/$COMPILER</BASELINE_ROOT>
+    <CCSM_CPRNC>/lcrc/group/e3sm/tools/cprnc/cprnc.improv</CCSM_CPRNC>
+    <GMAKE_J>8</GMAKE_J>
+    <TESTS>e3sm_integration</TESTS>
+    <NTEST_PARALLEL_JOBS>8</NTEST_PARALLEL_JOBS>
+    <BATCH_SYSTEM>pbspro</BATCH_SYSTEM>
+    <SUPPORTED_BY>E3SM</SUPPORTED_BY>
+    <MAX_TASKS_PER_NODE>128</MAX_TASKS_PER_NODE>
+    <MAX_MPITASKS_PER_NODE>128</MAX_MPITASKS_PER_NODE>
+    <PROJECT_REQUIRED>FALSE</PROJECT_REQUIRED>
+    <mpirun mpilib="openmpi">
+      <executable>mpirun</executable>
+      <arguments>
+        <arg name="num_tasks">--tag-output -n {{ total_tasks }}</arg>
+        <arg name="tasks_per_node"> --map-by ppr:1:core:PE=$ENV{OMP_NUM_THREADS} --bind-to core --oversubscribe </arg>
+      </arguments>
+    </mpirun>
+    <module_system type="module">
+      <init_path lang="sh">/gpfs/fs1/soft/chrysalis/spack/opt/spack/linux-centos8-x86_64/gcc-9.3.0/lmod-8.3-5be73rg/lmod/lmod/init/sh</init_path>
+      <init_path lang="csh">/gpfs/fs1/soft/chrysalis/spack/opt/spack/linux-centos8-x86_64/gcc-9.3.0/lmod-8.3-5be73rg/lmod/lmod/init/csh</init_path>
+      <init_path lang="python">/gpfs/fs1/soft/chrysalis/spack/opt/spack/linux-centos8-x86_64/gcc-9.3.0/lmod-8.3-5be73rg/lmod/lmod/init/env_modules_python.py</init_path>
+      <cmd_path lang="python">/gpfs/fs1/soft/chrysalis/spack/opt/spack/linux-centos8-x86_64/gcc-9.3.0/lmod-8.3-5be73rg/lmod/lmod/libexec/lmod python</cmd_path>
+      <cmd_path lang="sh">module</cmd_path>
+      <cmd_path lang="csh">module</cmd_path>
+      <modules>
+        <command name="purge"/>
+        <command name="load">cmake/3.27.4</command>
+      </modules>
+      <modules compiler="gnu">
+        <command name="load">gcc/12.3.0</command>
+      </modules>
+    </module_system>
+    <RUNDIR>$CIME_OUTPUT_ROOT/$CASE/run</RUNDIR>
+    <EXEROOT>$CIME_OUTPUT_ROOT/$CASE/bld</EXEROOT>
+    <TEST_TPUT_TOLERANCE>0.05</TEST_TPUT_TOLERANCE>
+    <MAX_GB_OLD_TEST_DATA>0</MAX_GB_OLD_TEST_DATA>
+    <environment_variables compiler="gnu" mpilib="openmpi">
+      <env name="NETCDF_C_PATH">/lcrc/group/e3sm/soft/improv/netcdf-c/4.9.2b/gcc-12.3.0/openmpi-4.1.6</env>
+      <env name="NETCDF_FORTRAN_PATH">/lcrc/group/e3sm/soft/improv/netcdf-fortran/4.6.1b/gcc-12.3.0/openmpi-4.1.6</env>
+      <env name="PNETCDF_PATH">/lcrc/group/e3sm/soft/improv/pnetcdf/1.12.3/gcc-12.3.0/openmpi-4.1.6</env>
+      <env name="PATH">/lcrc/group/e3sm/soft/improv/pnetcdf/1.12.3/gcc-12.3.0/openmpi-4.1.6/bin:/lcrc/group/e3sm/soft/improv/netcdf-fortran/4.6.1b/gcc-12.3.0/openmpi-4.1.6/bin:/lcrc/group/e3sm/soft/improv/netcdf-c/4.9.2b/gcc-12.3.0/openmpi-4.1.6/bin:/lcrc/group/e3sm/soft/improv/openmpi/4.1.6/gcc-12.3.0/bin:/lcrc/group/e3sm/soft/perl/improv/bin:$ENV{PATH}</env>
+      <env name="LD_LIBRARY_PATH">$SHELL{lp=/lcrc/group/e3sm/soft/improv/netlib-lapack/3.12.0/gcc-12.3.0:/lcrc/group/e3sm/soft/improv/pnetcdf/1.12.3/gcc-12.3.0/openmpi-4.1.6/lib:/lcrc/group/e3sm/soft/improv/netcdf-fortran/4.6.1b/gcc-12.3.0/openmpi-4.1.6/lib:/lcrc/group/e3sm/soft/improv/netcdf-c/4.9.2b/gcc-12.3.0/openmpi-4.1.6/lib:/opt/pbs/lib:/lcrc/group/e3sm/soft/improv/openmpi/4.1.6/gcc-12.3.0/lib; if [ -z "$LD_LIBRARY_PATH" ]; then echo $lp; else echo "$lp:$LD_LIBRARY_PATH"; fi}</env>
+    </environment_variables>
+    <environment_variables BUILD_THREADED="TRUE">
+      <env name="OMP_STACKSIZE">128M</env>
+    </environment_variables>
+    <environment_variables BUILD_THREADED="TRUE" compiler="gnu">
+      <env name="OMP_PLACES">cores</env>
+    </environment_variables>
+  </machine>
+
   <machine MACH="lassen">
     <DESC>LLNL Linux Cluster, Linux, 4 V100 GPUs/node, 44 IBM P9 cpu cores/node</DESC>
     <NODENAME_REGEX>lassen.*</NODENAME_REGEX>

diff --git a/components/eam/cime_config/config_pes.xml b/components/eam/cime_config/config_pes.xml
@@ -771,6 +771,21 @@
         </nthrds>
       </pes>
     </mach>
+    <mach name="improv">
+      <pes compset=".*EAM.+ELM.+MPASSI.+DOCN.+SGLC_SWAV_SIAC_SESP_BGC.*" pesize="any">
+        <comment>improv pelayout for tri-grid BGC tests with EAM+DOCN </comment>
+        <ntasks>
+          <ntasks_atm>-4</ntasks_atm>
+          <ntasks_cpl>-4</ntasks_cpl>
+          <ntasks_lnd>-4</ntasks_lnd>
+          <ntasks_rof>-4</ntasks_rof>
+          <ntasks_ocn>-4</ntasks_ocn>
+          <ntasks_ice>-4</ntasks_ice>
+          <ntasks_glc>-4</ntasks_glc>
+          <ntasks_wav>-4</ntasks_wav>
+        </ntasks>
+      </pes>
+    </mach>
   </grid>
   <grid name="a%ne30np4.pg2_l%.+_oi%oEC60to30v3">
     <mach name="anvil">
@@ -1093,6 +1108,21 @@
         </ntasks>
       </pes>
     </mach>
+    <mach name="improv">
+      <pes compset=".*EAM.+ELM.+MPASSI.+DOCN.+MOSART.+" pesize="M">
+        <comment>--res conusx4v1_r05_oECv3 --compset F2010 </comment>
+        <ntasks>
+          <ntasks_atm>-4</ntasks_atm>
+          <ntasks_cpl>-4</ntasks_cpl>
+          <ntasks_lnd>-4</ntasks_lnd>
+          <ntasks_rof>-4</ntasks_rof>
+          <ntasks_ocn>-4</ntasks_ocn>
+          <ntasks_ice>-4</ntasks_ice>
+          <ntasks_glc>-4</ntasks_glc>
+          <ntasks_wav>-4</ntasks_wav>
+        </ntasks>
+      </pes>
+    </mach>
     <mach name="summit|ascent|pm-gpu|muller-gpu">
       <pes compset=".*EAM.+ELM.+MPASSI.+DOCN.+MOSART.+" pesize="M">
         <comment>--res conusx4v1_r05_oECv3 --compset F2010 </comment>

diff --git a/components/elm/cime_config/config_pes.xml b/components/elm/cime_config/config_pes.xml
@@ -218,9 +218,9 @@
         </ntasks>
       </pes>
     </mach>
-    <mach name="ascent|summit">
+    <mach name="ascent|summit|improv">
       <pes compset="any" pesize="any">
-        <comment>elm: ascent|summit PEs for grid l%1.9x2.5|l%0.9x1.25|l%360x720cru</comment>
+        <comment>elm: ascent|summit|improv PEs for grid l%1.9x2.5|l%0.9x1.25|l%360x720cru</comment>
         <ntasks>
           <ntasks_atm>-2</ntasks_atm>
           <ntasks_lnd>-2</ntasks_lnd>
@@ -509,9 +509,9 @@
         </ntasks>
       </pes>
     </mach>
-    <mach name="ascent|summit">
+    <mach name="ascent|summit|improv">
       <pes compset="any" pesize="any">
-        <comment>elm: ascent|summit PEs for grid l%r05_*r%r05</comment>
+        <comment>elm: ascent|summit|improv PEs for grid l%r05_*r%r05</comment>
         <ntasks>
           <ntasks_atm>-2</ntasks_atm>
           <ntasks_lnd>-2</ntasks_lnd>