1# libCEED: Efficient Extensible Discretization 2 3[![GitHub Actions][github-badge]][github-link] 4[![GitLab-CI][gitlab-badge]][gitlab-link] 5[![Azure Pipelines][azure-badge]][azure-link] 6[![Code coverage][codecov-badge]][codecov-link] 7[![BSD-2-Clause][license-badge]][license-link] 8[![Documentation][doc-badge]][doc-link] 9[![JOSS paper][joss-badge]][joss-link] 10[![Binder][binder-badge]][binder-link] 11 12## Summary and Purpose 13 14libCEED provides fast algebra for element-based discretizations, designed for 15performance portability, run-time flexibility, and clean embedding in higher 16level libraries and applications. It offers a C99 interface as well as bindings 17for Fortran, Python, Julia, and Rust. 18While our focus is on high-order finite elements, the approach is mostly 19algebraic and thus applicable to other discretizations in factored form, as 20explained in the [user manual](https://libceed.readthedocs.io/en/latest/) and 21API implementation portion of the 22[documentation](https://libceed.readthedocs.io/en/latest/api/). 23 24One of the challenges with high-order methods is that a global sparse matrix is 25no longer a good representation of a high-order linear operator, both with 26respect to the FLOPs needed for its evaluation, as well as the memory transfer 27needed for a matvec. Thus, high-order methods require a new "format" that still 28represents a linear (or more generally non-linear) operator, but not through a 29sparse matrix. 30 31The goal of libCEED is to propose such a format, as well as supporting 32implementations and data structures, that enable efficient operator evaluation 33on a variety of computational device types (CPUs, GPUs, etc.). This new operator 34description is based on algebraically 35[factored form](https://libceed.readthedocs.io/en/latest/libCEEDapi/#finite-element-operator-decomposition), 36which is easy to incorporate in a wide variety of applications, without significant 37refactoring of their own discretization infrastructure. 38 39The repository is part of the 40[CEED software suite](http://ceed.exascaleproject.org/software/), a collection of 41software benchmarks, miniapps, libraries and APIs for efficient exascale 42discretizations based on high-order finite element and spectral element methods. 43See <http://github.com/ceed> for more information and source code availability. 44 45The CEED research is supported by the 46[Exascale Computing Project](https://exascaleproject.org/exascale-computing-project) 47(17-SC-20-SC), a collaborative effort of two U.S. Department of Energy 48organizations (Office of Science and the National Nuclear Security 49Administration) responsible for the planning and preparation of a 50[capable exascale ecosystem](https://exascaleproject.org/what-is-exascale), including 51software, applications, hardware, advanced system engineering and early testbed 52platforms, in support of the nation’s exascale computing imperative. 53 54For more details on the CEED API see the [user manual](https://libceed.readthedocs.io/en/latest/). 55 56% gettingstarted-inclusion-marker 57 58## Building 59 60The CEED library, `libceed`, is a C99 library with no required dependencies, and 61with Fortran, Python, Julia, and Rust interfaces. It can be built using: 62 63``` 64make 65``` 66 67or, with optimization flags: 68 69``` 70make OPT='-O3 -march=skylake-avx512 -ffp-contract=fast' 71``` 72 73These optimization flags are used by all languages (C, C++, Fortran) and this 74makefile variable can also be set for testing and examples (below). 75 76The library attempts to automatically detect support for the AVX 77instruction set using gcc-style compiler options for the host. 78Support may need to be manually specified via: 79 80``` 81make AVX=1 82``` 83 84or: 85 86``` 87make AVX=0 88``` 89 90if your compiler does not support gcc-style options, if you are cross 91compiling, etc. 92 93To enable CUDA support, add `CUDA_DIR=/opt/cuda` or an appropriate directory 94to your `make` invocation. To enable HIP support, add `HIP_DIR=/opt/rocm` or 95an appropriate directory. To store these or other arguments as defaults for 96future invocations of `make`, use: 97 98``` 99make configure CUDA_DIR=/usr/local/cuda HIP_DIR=/opt/rocm OPT='-O3 -march=znver2' 100``` 101 102which stores these variables in `config.mk`. 103 104## Additional Language Interfaces 105 106The Fortran interface is built alongside the library automatically. 107 108Python users can install using: 109 110``` 111pip install libceed 112``` 113 114or in a clone of the repository via `pip install .`. 115 116Julia users can install using: 117 118``` 119$ julia 120julia> ] 121pkg> add LibCEED 122``` 123 124See the [LibCEED.jl documentation](http://ceed.exascaleproject.org/libCEED-julia-docs/dev/) 125for more information. 126 127Rust users can include libCEED via `Cargo.toml`: 128 129```toml 130[dependencies] 131libceed = { git = "https://github.com/CEED/libCEED", branch = "main" } 132``` 133 134See the [Cargo documentation](https://doc.rust-lang.org/cargo/reference/specifying-dependencies.html#specifying-dependencies-from-git-repositories) for details. 135 136## Testing 137 138The test suite produces [TAP](https://testanything.org) output and is run by: 139 140``` 141make test 142``` 143 144or, using the `prove` tool distributed with Perl (recommended): 145 146``` 147make prove 148``` 149 150## Backends 151 152There are multiple supported backends, which can be selected at runtime in the examples: 153 154| CEED resource | Backend | Deterministic Capable | 155| :--- | :--- | :---: | 156|| 157| **CPU Native** | 158| `/cpu/self/ref/serial` | Serial reference implementation | Yes | 159| `/cpu/self/ref/blocked` | Blocked reference implementation | Yes | 160| `/cpu/self/opt/serial` | Serial optimized C implementation | Yes | 161| `/cpu/self/opt/blocked` | Blocked optimized C implementation | Yes | 162| `/cpu/self/avx/serial` | Serial AVX implementation | Yes | 163| `/cpu/self/avx/blocked` | Blocked AVX implementation | Yes | 164|| 165| **CPU Valgrind** | 166| `/cpu/self/memcheck/*` | Memcheck backends, undefined value checks | Yes | 167|| 168| **CPU LIBXSMM** | 169| `/cpu/self/xsmm/serial` | Serial LIBXSMM implementation | Yes | 170| `/cpu/self/xsmm/blocked` | Blocked LIBXSMM implementation | Yes | 171|| 172| **CUDA Native** | 173| `/gpu/cuda/ref` | Reference pure CUDA kernels | Yes | 174| `/gpu/cuda/shared` | Optimized pure CUDA kernels using shared memory | Yes | 175| `/gpu/cuda/gen` | Optimized pure CUDA kernels using code generation | No | 176|| 177| **HIP Native** | 178| `/gpu/hip/ref` | Reference pure HIP kernels | Yes | 179| `/gpu/hip/shared` | Optimized pure HIP kernels using shared memory | Yes | 180| `/gpu/hip/gen` | Optimized pure HIP kernels using code generation | No | 181|| 182| **MAGMA** | 183| `/gpu/cuda/magma` | CUDA MAGMA kernels | No | 184| `/gpu/cuda/magma/det` | CUDA MAGMA kernels | Yes | 185| `/gpu/hip/magma` | HIP MAGMA kernels | No | 186| `/gpu/hip/magma/det` | HIP MAGMA kernels | Yes | 187|| 188| **OCCA** | 189| `/*/occa` | Selects backend based on available OCCA modes | Yes | 190| `/cpu/self/occa` | OCCA backend with serial CPU kernels | Yes | 191| `/cpu/openmp/occa` | OCCA backend with OpenMP kernels | Yes | 192| `/gpu/cuda/occa` | OCCA backend with CUDA kernels | Yes | 193| `/gpu/hip/occa`~ | OCCA backend with HIP kernels | Yes | 194 195The `/cpu/self/*/serial` backends process one element at a time and are intended for meshes 196with a smaller number of high order elements. The `/cpu/self/*/blocked` backends process 197blocked batches of eight interlaced elements and are intended for meshes with higher numbers 198of elements. 199 200The `/cpu/self/ref/*` backends are written in pure C and provide basic functionality. 201 202The `/cpu/self/opt/*` backends are written in pure C and use partial e-vectors to improve performance. 203 204The `/cpu/self/avx/*` backends rely upon AVX instructions to provide vectorized CPU performance. 205 206The `/cpu/self/memcheck/*` backends rely upon the [Valgrind](http://valgrind.org/) Memcheck tool 207to help verify that user QFunctions have no undefined values. To use, run your code with 208Valgrind and the Memcheck backends, e.g. `valgrind ./build/ex1 -ceed /cpu/self/ref/memcheck`. A 209'development' or 'debugging' version of Valgrind with headers is required to use this backend. 210This backend can be run in serial or blocked mode and defaults to running in the serial mode 211if `/cpu/self/memcheck` is selected at runtime. 212 213The `/cpu/self/xsmm/*` backends rely upon the [LIBXSMM](http://github.com/hfp/libxsmm) package 214to provide vectorized CPU performance. If linking MKL and LIBXSMM is desired but 215the Makefile is not detecting `MKLROOT`, linking libCEED against MKL can be 216forced by setting the environment variable `MKL=1`. 217 218The `/gpu/cuda/*` backends provide GPU performance strictly using CUDA. 219 220The `/gpu/hip/*` backends provide GPU performance strictly using HIP. They are based on 221the `/gpu/cuda/*` backends. ROCm version 4.2 or newer is required. 222 223The `/gpu/*/magma/*` backends rely upon the [MAGMA](https://bitbucket.org/icl/magma) package. 224To enable the MAGMA backends, the environment variable `MAGMA_DIR` must point to the top-level 225MAGMA directory, with the MAGMA library located in `$(MAGMA_DIR)/lib/`. 226By default, `MAGMA_DIR` is set to `../magma`; to build the MAGMA backends 227with a MAGMA installation located elsewhere, create a link to `magma/` in libCEED's parent 228directory, or set `MAGMA_DIR` to the proper location. MAGMA version 2.5.0 or newer is required. 229Currently, each MAGMA library installation is only built for either CUDA or HIP. The corresponding 230set of libCEED backends (`/gpu/cuda/magma/*` or `/gpu/hip/magma/*`) will automatically be built 231for the version of the MAGMA library found in `MAGMA_DIR`. 232 233Users can specify a device for all CUDA, HIP, and MAGMA backends through adding `:device_id=#` 234after the resource name. For example: 235 236> - `/gpu/cuda/gen:device_id=1` 237 238The `/*/occa` backends rely upon the [OCCA](http://github.com/libocca/occa) package to provide 239cross platform performance. To enable the OCCA backend, the environment variable `OCCA_DIR` must point 240to the top-level OCCA directory, with the OCCA library located in the `${OCCA_DIR}/lib` (By default, 241`OCCA_DIR` is set to `../occa`). 242 243Additionally, users can pass specific OCCA device properties after setting the CEED resource. 244For example: 245 246> - `"/*/occa:mode='CUDA',device_id=0"` 247 248Bit-for-bit reproducibility is important in some applications. 249However, some libCEED backends use non-deterministic operations, such as `atomicAdd` for increased performance. 250The backends which are capable of generating reproducible results, with the proper compilation options, are highlighted in the list above. 251 252## Examples 253 254libCEED comes with several examples of its usage, ranging from standalone C 255codes in the `/examples/ceed` directory to examples based on external packages, 256such as MFEM, PETSc, and Nek5000. Nek5000 v18.0 or greater is required. 257 258To build the examples, set the `MFEM_DIR`, `PETSC_DIR`, and 259`NEK5K_DIR` variables and run: 260 261``` 262cd examples/ 263``` 264 265% running-examples-inclusion-marker 266 267```console 268# libCEED examples on CPU and GPU 269cd ceed/ 270make 271./ex1-volume -ceed /cpu/self 272./ex1-volume -ceed /gpu/cuda 273./ex2-surface -ceed /cpu/self 274./ex2-surface -ceed /gpu/cuda 275cd .. 276 277# MFEM+libCEED examples on CPU and GPU 278cd mfem/ 279make 280./bp1 -ceed /cpu/self -no-vis 281./bp3 -ceed /gpu/cuda -no-vis 282cd .. 283 284# Nek5000+libCEED examples on CPU and GPU 285cd nek/ 286make 287./nek-examples.sh -e bp1 -ceed /cpu/self -b 3 288./nek-examples.sh -e bp3 -ceed /gpu/cuda -b 3 289cd .. 290 291# PETSc+libCEED examples on CPU and GPU 292cd petsc/ 293make 294./bps -problem bp1 -ceed /cpu/self 295./bps -problem bp2 -ceed /gpu/cuda 296./bps -problem bp3 -ceed /cpu/self 297./bps -problem bp4 -ceed /gpu/cuda 298./bps -problem bp5 -ceed /cpu/self 299./bps -problem bp6 -ceed /gpu/cuda 300cd .. 301 302cd petsc/ 303make 304./bpsraw -problem bp1 -ceed /cpu/self 305./bpsraw -problem bp2 -ceed /gpu/cuda 306./bpsraw -problem bp3 -ceed /cpu/self 307./bpsraw -problem bp4 -ceed /gpu/cuda 308./bpsraw -problem bp5 -ceed /cpu/self 309./bpsraw -problem bp6 -ceed /gpu/cuda 310cd .. 311 312cd petsc/ 313make 314./bpssphere -problem bp1 -ceed /cpu/self 315./bpssphere -problem bp2 -ceed /gpu/cuda 316./bpssphere -problem bp3 -ceed /cpu/self 317./bpssphere -problem bp4 -ceed /gpu/cuda 318./bpssphere -problem bp5 -ceed /cpu/self 319./bpssphere -problem bp6 -ceed /gpu/cuda 320cd .. 321 322cd petsc/ 323make 324./area -problem cube -ceed /cpu/self -degree 3 325./area -problem cube -ceed /gpu/cuda -degree 3 326./area -problem sphere -ceed /cpu/self -degree 3 -dm_refine 2 327./area -problem sphere -ceed /gpu/cuda -degree 3 -dm_refine 2 328 329cd fluids/ 330make 331./navierstokes -ceed /cpu/self -degree 1 332./navierstokes -ceed /gpu/cuda -degree 1 333cd .. 334 335cd solids/ 336make 337./elasticity -ceed /cpu/self -mesh [.exo file] -degree 2 -E 1 -nu 0.3 -problem Linear -forcing mms 338./elasticity -ceed /gpu/cuda -mesh [.exo file] -degree 2 -E 1 -nu 0.3 -problem Linear -forcing mms 339cd .. 340``` 341 342For the last example shown, sample meshes to be used in place of 343`[.exo file]` can be found at <https://github.com/jeremylt/ceedSampleMeshes> 344 345The above code assumes a GPU-capable machine with the OCCA backend 346enabled. Depending on the available backends, other CEED resource 347specifiers can be provided with the `-ceed` option. Other command line 348arguments can be found in [examples/petsc](https://github.com/CEED/libCEED/blob/main/examples/petsc/README.md). 349 350% benchmarks-marker 351 352## Benchmarks 353 354A sequence of benchmarks for all enabled backends can be run using: 355 356``` 357make benchmarks 358``` 359 360The results from the benchmarks are stored inside the `benchmarks/` directory 361and they can be viewed using the commands (requires python with matplotlib): 362 363``` 364cd benchmarks 365python postprocess-plot.py petsc-bps-bp1-*-output.txt 366python postprocess-plot.py petsc-bps-bp3-*-output.txt 367``` 368 369Using the `benchmarks` target runs a comprehensive set of benchmarks which may 370take some time to run. Subsets of the benchmarks can be run using the scripts in the `benchmarks` folder. 371 372For more details about the benchmarks, see the `benchmarks/README.md` file. 373 374## Install 375 376To install libCEED, run: 377 378``` 379make install prefix=/usr/local 380``` 381 382or (e.g., if creating packages): 383 384``` 385make install prefix=/usr DESTDIR=/packaging/path 386``` 387 388The usual variables like `CC` and `CFLAGS` are used, and optimization flags 389for all languages can be set using the likes of `OPT='-O3 -march=native'`. Use 390`STATIC=1` to build static libraries (`libceed.a`). 391 392To install libCEED for Python, run: 393 394``` 395pip install libceed 396``` 397 398with the desired setuptools options, such as `--user`. 399 400### pkg-config 401 402In addition to library and header, libCEED provides a [pkg-config](https://en.wikipedia.org/wiki/Pkg-config) 403file that can be used to easily compile and link. 404[For example](https://people.freedesktop.org/~dbn/pkg-config-guide.html#faq), if 405`$prefix` is a standard location or you set the environment variable 406`PKG_CONFIG_PATH`: 407 408``` 409cc `pkg-config --cflags --libs ceed` -o myapp myapp.c 410``` 411 412will build `myapp` with libCEED. This can be used with the source or 413installed directories. Most build systems have support for pkg-config. 414 415## Contact 416 417You can reach the libCEED team by emailing [ceed-users@llnl.gov](mailto:ceed-users@llnl.gov) 418or by leaving a comment in the [issue tracker](https://github.com/CEED/libCEED/issues). 419 420## How to Cite 421 422If you utilize libCEED please cite: 423 424``` 425@article{libceed-joss-paper, 426 author = {Jed Brown and Ahmad Abdelfattah and Valeria Barra and Natalie Beams and Jean Sylvain Camier and Veselin Dobrev and Yohann Dudouit and Leila Ghaffari and Tzanio Kolev and David Medina and Will Pazner and Thilina Ratnayaka and Jeremy Thompson and Stan Tomov}, 427 title = {{libCEED}: Fast algebra for high-order element-based discretizations}, 428 journal = {Journal of Open Source Software}, 429 year = {2021}, 430 publisher = {The Open Journal}, 431 volume = {6}, 432 number = {63}, 433 pages = {2945}, 434 doi = {10.21105/joss.02945} 435} 436 437@misc{libceed-user-manual, 438 author = {Abdelfattah, Ahmad and 439 Barra, Valeria and 440 Beams, Natalie and 441 Brown, Jed and 442 Camier, Jean-Sylvain and 443 Dobrev, Veselin and 444 Dudouit, Yohann and 445 Ghaffari, Leila and 446 Kolev, Tzanio and 447 Medina, David and 448 Pazner, Will and 449 Ratnayaka, Thilina and 450 Thompson, Jeremy L and 451 Tomov, Stanimire}, 452 title = {{libCEED} User Manual}, 453 month = jul, 454 year = 2021, 455 publisher = {Zenodo}, 456 version = {0.9.0}, 457 doi = {10.5281/zenodo.5077489} 458} 459``` 460 461For libCEED's Python interface please cite: 462 463``` 464@InProceedings{libceed-paper-proc-scipy-2020, 465 author = {{V}aleria {B}arra and {J}ed {B}rown and {J}eremy {T}hompson and {Y}ohann {D}udouit}, 466 title = {{H}igh-performance operator evaluations with ease of use: lib{C}{E}{E}{D}'s {P}ython interface}, 467 booktitle = {{P}roceedings of the 19th {P}ython in {S}cience {C}onference}, 468 pages = {85 - 90}, 469 year = {2020}, 470 editor = {{M}eghann {A}garwal and {C}hris {C}alloway and {D}illon {N}iederhut and {D}avid {S}hupe}, 471 doi = {10.25080/Majora-342d178e-00c} 472} 473``` 474 475The BiBTeX entries for these references can be found in the 476`doc/bib/references.bib` file. 477 478## Copyright 479 480The following copyright applies to each file in the CEED software suite, unless 481otherwise stated in the file: 482 483> Copyright (c) 2017, Lawrence Livermore National Security, LLC. Produced at the 484> Lawrence Livermore National Laboratory. LLNL-CODE-734707. All Rights reserved. 485 486See files LICENSE and NOTICE for details. 487 488[github-badge]: https://github.com/CEED/libCEED/workflows/C/Fortran/badge.svg 489[github-link]: https://github.com/CEED/libCEED/actions 490[gitlab-badge]: https://gitlab.com/libceed/libCEED/badges/main/pipeline.svg?key_text=GitLab-CI 491[gitlab-link]: https://gitlab.com/libceed/libCEED/-/pipelines?page=1&scope=all&ref=main 492[azure-badge]: https://dev.azure.com/CEED-ECP/libCEED/_apis/build/status/CEED.libCEED?branchName=main 493[azure-link]: https://dev.azure.com/CEED-ECP/libCEED/_build?definitionId=2 494[codecov-badge]: https://codecov.io/gh/CEED/libCEED/branch/main/graphs/badge.svg 495[codecov-link]: https://codecov.io/gh/CEED/libCEED/ 496[license-badge]: https://img.shields.io/badge/License-BSD%202--Clause-orange.svg 497[license-link]: https://opensource.org/licenses/BSD-2-Clause 498[doc-badge]: https://readthedocs.org/projects/libceed/badge/?version=latest 499[doc-link]: https://libceed.readthedocs.io/en/latest/?badge=latest 500[joss-badge]: https://joss.theoj.org/papers/10.21105/joss.02945/status.svg 501[joss-link]: https://doi.org/10.21105/joss.02945 502[binder-badge]: http://mybinder.org/badge_logo.svg 503[binder-link]: https://mybinder.org/v2/gh/CEED/libCEED/main?urlpath=lab/tree/examples/python/tutorial-0-ceed.ipynb 504
