Imported from daliwang/uELM (AGENTS.md). Install upstream with npx skills add daliwang/uELM. Copyright stays with the author.

AGENTS.md

E3SM (Energy Exascale Earth System Model) is a state-of-the-art fully coupled Earth System Model. This guide helps you work effectively with this codebase.

Project Overview

E3SM (Energy Exascale Earth System Model) is a state-of-the-art fully coupled Earth System Model developed for DOE Leadership Computing Facilities. It includes biogeochemical and cryospheric processes and is designed for high-performance computing clusters.

• Website: https://e3sm.org
• Documentation: https://docs.e3sm.org/E3SM/
• DOI: 10.11578/E3SM/dc.20240930.1

Build Commands

E3SM uses the CIME (Common Infrastructure for Modeling the Earth) case control system. Build is done out-of-source and is not under the source directory. First a CASEDIR is created using create_newcase. In the CASEDIR, you can run ./case.build after running ./case.setup. Supported machines are in cime_config/machines/config_machines.xml and grouped by hostname. Do not try and create a case or test unless on a supported machine.

#list supported machines
cd cime/scripts
./query_config --machines

Creating and Building a Case

Ask the user to specify a CASENAME. Create the case in the top level project directory unless the case name includes a absolute or relative path.

# Navigate to CIME scripts directory
cd cime/scripts

# Create a new case
./create_newcase --case ../../CASENAME --compset COMPSET --res RESOLUTION --mach MACHINE

# Setup the case
cd /path/to/case
./case.setup

# Build the code for the case
./case.build

#Find the directory of the build output
./xmlquery EXEROOT

# Submit the case
./case.submit

# Find the directory with executable output
./xmlquery RUNDIR

Rebuild a case

#Do not clean just build again
./case.build

#if only components or driver were modified its faster to do
./case.build -m

#if the pelayout is modified, do
./case.setup --reset
./case.build

Rerun a case

# Submit the case
./case.submit

Common Compsets and Resolutions

Use these when testing changes to code made only in the specified component

• elm: COMPSET I1850CNPRDCTCBCTOP, RESOLUTION ne4pg2_ne4pg2
• eam: COMPSET F1850, RESOLUTION ne4pg2_oQU480
• mosart: COMPSET RMOSGPCC, RESOLUTION r05_r05
• eamxx: COMPSET F2010-SCREAMv1, RESOLUTION ne4pg2_ne4pg2
• mpas-ocean: COMPSET CMPASO-NYF, RESOLUTION T62_oQU120
• mpas-seaice: COMPSET DTESTM, RESOLUTION T62_oQU240

If code has been changed in elm and eam, use the F1850 compeset. If code has been changed in 2 or more other components, use the coupled case: COMPSET WCYCL1850NS, RESOLUTION ne4pg2_r05_oQU480

Compsets

Compsets define the component configuration for a simulation. The longname format is: TIME_ATM[%phys]_LND[%phys]_ICE[%phys]_OCN[%phys]_ROF[%phys]_GLC[%phys]_WAV[%phys]

Compset definitions are in cime_config/allactive/config_compsets.xml and component-specific files in components/*/cime_config/config_compsets.xml.

#list all compsets
cd cime/scripts
./query_config --compsets all

For a created case, the longname is in the README.case file in the case directory.

Component Abbreviations in Compset Names

• ATM: EAM (atmosphere), EAMXX/SCREAM (next-gen), DATM (data), SATM (stub)
• LND: ELM (land), SLND (stub)
• ICE: MPASSI (MPAS sea ice), CICE (legacy), DICE (data), SICE (stub)
• OCN: MPASO (MPAS ocean), DOCN (data), SOCN (stub)
• ROF: MOSART (river routing), SROF (stub)
• GLC: MALI (land ice), SGLC (stub)
• WAV: WW3 (waves), SWAV (stub)

Grid Resolutions

Grid definitions are in cime_config/config_grids.xml. The grid alias format is: a%name_l%name_oi%name_r%name_m%mask_g%name_w%name (atm_lnd_ocn/ice_rof_mask_glc_wav)

#list all grids
cd cime/scripts
./query_config --grids

Testing

While doing development, create a single case that uses the component being modified and run ./case.build and ./case.submit as needed to test code modifications.

Most of E3SM does not have unit tests.

EAMxx (SCREAM) Standalone Testing

Code modifications in components/eamxx can be tested in a standalone mode without CIME or a driver.

cd components/eamxx

# Run all tests on a machine (requires compute node on batch systems)
./scripts/test-all-eamxx -m MACHINE

# Run specific test type (sp, dbg, fpe, opt)
./scripts/test-all-eamxx -m MACHINE -t dbg

# Preserve current environment instead of loading machine-specific modules
./scripts/test-all-eamxx --preserve-env -m MACHINE

# Run with local baselines
./scripts/test-all-eamxx -m MACHINE --baseline-dir=LOCAL

CIME-Based System Tests

cd cime/scripts

# Run a single test
./create_test TEST_NAME

# Run a test suite
./create_test SUITE_NAME

# Monitor test progress by going to the test directory and tailing the TestStatus.log file

# Test naming format: TEST_TYPE.RESOLUTION.COMPSET[.MACHINE][.TESTMOD]
# Example: SMS_D_Ln5_P4.ne4pg2_oQU480.F2010.pm-cpu_gnu

Test types for various properties of the model:

• SMS - Short smoke test
• ERS - Exact restart test
• SMS_D - Debug smoke test
• PEM - Test bit-for-bit when changing mpi task count
• PET - Test bit-for-bit when changing thread count
• ERS_Ld - Exact restart test run for lenght of d days

Test Suites (defined in cime_config/tests.py)

• e3sm_land_developer
• e3sm_mosart_developer
• e3sm_land_exeshare

#get list of test suites
cd cime/CIME/Tools
./list_e3sm_tests

#list compsets in a test suite
cd cime/CIME/Tools
./list_e3sm_tests -t compsets SUITE_NAME

#list compset longnames in a test suite
./list_e3sm_tests -t compsets -l SUITE_NAME

Nightly results from test suites appear on https://my.cdash.org/index.php?project=E3SM

Architecture

Primary Languages

• Fortran (~4300 files): Core model physics and dynamics
• Python (~2100 files): CIME infrastructure, testing, configuration
• C++: Modern EAMxx/SCREAM atmosphere component, external libraries

Major Components (in /components/)

Key Infrastructure Directories

• /cime/ - CIME submodule (case control system)
• /cime_config/ - E3SM-specific CIME configuration
  • machines/ - Machine, compiler, batch specifications
  • machines/config_machines.xml - supported machines
  • machines/cmake_macros/ - Machine-specific CMake settings
  • tests.py - Test suite definitions
  • config_grids.xml - Grid definitions
  • config_compilers.xml - Compiler configurations
• /driver-mct/ - MCT coupler driver
• /driver-moab/ - MOAB coupler driver
• /share/ - Shared utilities (timing, RNG, streams)
• /externals/ - External library submodules

External Dependencies (Git Submodules)

Key submodules in /externals/:

• ekat - E3SM Kokkos Application Toolkit
• scorpio - Parallel I/O library
• mam4xx - Modal Aerosol Model (C++)
• haero - Aerosol package
• YAKL - Yet Another Kernel Language
• mct - Model Coupling Toolkit

After cloning, initialize submodules:

git submodule update --init --recursive --depth=1

Build System

• Primary: CMake 3.18+ (top-level CMakeLists.txt in /components)
• Secondary: Make (MPAS components, MCT)
• GPU Support: CUDA and HIP via cmake_macros configuration

CI/CD

GitHub Actions workflows in /.github/workflows/:

• eamxx-sa-testing.yml - EAMxx standalone (gcc-openmp, gcc-cuda)
• eamxx-v1-testing.yml - CIME tests with EAMxx/MAM4xx

Build types tested: sp (single precision), dbg (debug), fpe (floating point exceptions), opt (optimized)

Development Workflow

1. Developing E3SM requires running on supported machines.
2. E3SM uses gitworkflows but without a 'seen' branch https://www.kernel.org/pub/software/scm/git/docs/gitworkflows.html
3. Development guide: https://e3sm.org/model/running-e3sm/developing-e3sm/
4. PR-based workflow with automated testing
5. Feature branch names should use this pattern: <github username>/<source code area or component>/<feature-description>
6. New features should be coordinated via E3SM management and science plan

Testing EAMxx Modifications (for AI Agents)

When modifying files under components/eamxx/**, see components/eamxx/AGENTS.md for standalone testing instructions.

Rules for pull request (PR) description

1. PR description should have two parts. The fist should be a brief description in plain text with no markdown or other formatting. Add a line. The second part should be the full description with markdown formatting.
2. PR description should use imperative tense and start with a verb like 'Fix' or 'Add'