1# libCEED: Efficient Extensible Discretization 2 3```{image} https://github.com/CEED/libCEED/workflows/C/Fortran/badge.svg 4:alt: GitHub Actions 5:target: https://github.com/CEED/libCEED/actions 6``` 7 8```{image} https://gitlab.com/libceed/libCEED/badges/main/pipeline.svg?key_text=GitLab-CI 9:alt: GitLab-CI 10:target: https://gitlab.com/libceed/libCEED/-/pipelines?page=1&scope=all&ref=main 11``` 12 13```{image} https://dev.azure.com/CEED-ECP/libCEED/_apis/build/status/CEED.libCEED?branchName=main 14:alt: Azure Pipelines 15:target: https://dev.azure.com/CEED-ECP/libCEED/_build?definitionId=2 16``` 17 18```{image} https://codecov.io/gh/CEED/libCEED/branch/main/graphs/badge.svg 19:alt: Code Coverage 20:target: https://codecov.io/gh/CEED/libCEED/ 21``` 22 23```{image} https://img.shields.io/badge/License-BSD%202--Clause-orange.svg 24:alt: License 25:target: https://opensource.org/licenses/BSD-2-Clause 26``` 27 28```{image} https://readthedocs.org/projects/libceed/badge/?version=latest 29:alt: Read the Docs 30:target: https://libceed.readthedocs.io/en/latest/?badge=latest 31``` 32 33```{image} https://joss.theoj.org/papers/10.21105/joss.02945/status.svg 34:alt: JOSS 35:target: https://doi.org/10.21105/joss.02945 36``` 37 38```{image} http://mybinder.org/badge_logo.svg 39:alt: Binder 40:target: https://mybinder.org/v2/gh/CEED/libCEED/main?urlpath=lab/tree/examples/tutorials/tutorial-0-ceed.ipynb 41``` 42 43## Summary and Purpose 44 45libCEED provides fast algebra for element-based discretizations, designed for 46performance portability, run-time flexibility, and clean embedding in higher 47level libraries and applications. It offers a C99 interface as well as bindings 48for Fortran, Python, Julia, and Rust. 49While our focus is on high-order finite elements, the approach is mostly 50algebraic and thus applicable to other discretizations in factored form, as 51explained in the [user manual](https://libceed.readthedocs.io/en/latest/) and 52API implementation portion of the 53[documentation](https://libceed.readthedocs.io/en/latest/api/). 54 55One of the challenges with high-order methods is that a global sparse matrix is 56no longer a good representation of a high-order linear operator, both with 57respect to the FLOPs needed for its evaluation, as well as the memory transfer 58needed for a matvec. Thus, high-order methods require a new "format" that still 59represents a linear (or more generally non-linear) operator, but not through a 60sparse matrix. 61 62The goal of libCEED is to propose such a format, as well as supporting 63implementations and data structures, that enable efficient operator evaluation 64on a variety of computational device types (CPUs, GPUs, etc.). This new operator 65description is based on algebraically 66[factored form](https://libceed.readthedocs.io/en/latest/libCEEDapi/#finite-element-operator-decomposition), 67which is easy to incorporate in a wide variety of applications, without significant 68refactoring of their own discretization infrastructure. 69 70The repository is part of the 71[CEED software suite](http://ceed.exascaleproject.org/software/), a collection of 72software benchmarks, miniapps, libraries and APIs for efficient exascale 73discretizations based on high-order finite element and spectral element methods. 74See <http://github.com/ceed> for more information and source code availability. 75 76The CEED research is supported by the 77[Exascale Computing Project](https://exascaleproject.org/exascale-computing-project) 78(17-SC-20-SC), a collaborative effort of two U.S. Department of Energy 79organizations (Office of Science and the National Nuclear Security 80Administration) responsible for the planning and preparation of a 81[capable exascale ecosystem](https://exascaleproject.org/what-is-exascale), including 82software, applications, hardware, advanced system engineering and early testbed 83platforms, in support of the nation’s exascale computing imperative. 84 85For more details on the CEED API see the [user manual](https://libceed.readthedocs.io/en/latest/). 86 87% gettingstarted-inclusion-marker 88 89## Building 90 91The CEED library, `libceed`, is a C99 library with no required dependencies, and 92with Fortran, Python, Julia, and Rust interfaces. It can be built using: 93 94``` 95make 96``` 97 98or, with optimization flags: 99 100``` 101make OPT='-O3 -march=skylake-avx512 -ffp-contract=fast' 102``` 103 104These optimization flags are used by all languages (C, C++, Fortran) and this 105makefile variable can also be set for testing and examples (below). 106 107The library attempts to automatically detect support for the AVX 108instruction set using gcc-style compiler options for the host. 109Support may need to be manually specified via: 110 111``` 112make AVX=1 113``` 114 115or: 116 117``` 118make AVX=0 119``` 120 121if your compiler does not support gcc-style options, if you are cross 122compiling, etc. 123 124To enable CUDA support, add `CUDA_DIR=/opt/cuda` or an appropriate directory 125to your `make` invocation. To enable HIP support, add `HIP_DIR=/opt/rocm` or 126an appropriate directory. To store these or other arguments as defaults for 127future invocations of `make`, use: 128 129``` 130make configure CUDA_DIR=/usr/local/cuda HIP_DIR=/opt/rocm OPT='-O3 -march=znver2' 131``` 132 133which stores these variables in `config.mk`. 134 135## Additional Language Interfaces 136 137The Fortran interface is built alongside the library automatically. 138 139Python users can install using: 140 141``` 142pip install libceed 143``` 144 145or in a clone of the repository via `pip install .`. 146 147Julia users can install using: 148 149``` 150$ julia 151julia> ] 152pkg> add LibCEED 153``` 154 155in the Julia package manager or in a clone of the repository via: 156 157``` 158JULIA_LIBCEED_LIB=/path/to/libceed.so julia 159julia> # press ] to enter package manager 160(env) pkg> build LibCEED 161``` 162 163Rust users can include libCEED via `Cargo.toml`: 164 165```toml 166[dependencies] 167libceed = { git = "https://github.com/CEED/libCEED", branch = "main" } 168``` 169 170See the [Cargo documentation](https://doc.rust-lang.org/cargo/reference/specifying-dependencies.html#specifying-dependencies-from-git-repositories) for details. 171 172## Testing 173 174The test suite produces [TAP](https://testanything.org) output and is run by: 175 176``` 177make test 178``` 179 180or, using the `prove` tool distributed with Perl (recommended): 181 182``` 183make prove 184``` 185 186## Backends 187 188There are multiple supported backends, which can be selected at runtime in the examples: 189 190```{eval-rst} 191+----------------------------+---------------------------------------------------+-----------------------+ 192| CEED resource | Backend | Deterministic Capable | 193+----------------------------+---------------------------------------------------+-----------------------+ 194| CPU Native Backends | 195+----------------------------+---------------------------------------------------+-----------------------+ 196| ``/cpu/self/ref/serial`` | Serial reference implementation | Yes | 197+----------------------------+---------------------------------------------------+-----------------------+ 198| ``/cpu/self/ref/blocked`` | Blocked reference implementation | Yes | 199+----------------------------+---------------------------------------------------+-----------------------+ 200| ``/cpu/self/opt/serial`` | Serial optimized C implementation | Yes | 201+----------------------------+---------------------------------------------------+-----------------------+ 202| ``/cpu/self/opt/blocked`` | Blocked optimized C implementation | Yes | 203+----------------------------+---------------------------------------------------+-----------------------+ 204| ``/cpu/self/avx/serial`` | Serial AVX implementation | Yes | 205+----------------------------+---------------------------------------------------+-----------------------+ 206| ``/cpu/self/avx/blocked`` | Blocked AVX implementation | Yes | 207+----------------------------+---------------------------------------------------+-----------------------+ 208| CPU Valgrind Backends | 209+----------------------------+---------------------------------------------------+-----------------------+ 210| ``/cpu/self/memcheck/*`` | Memcheck backends, undefined value checks | Yes | 211+----------------------------+---------------------------------------------------+-----------------------+ 212| CPU LIBXSMM Backends | 213+----------------------------+---------------------------------------------------+-----------------------+ 214| ``/cpu/self/xsmm/serial`` | Serial LIBXSMM implementation | Yes | 215+----------------------------+---------------------------------------------------+-----------------------+ 216| ``/cpu/self/xsmm/blocked`` | Blocked LIBXSMM implementation | Yes | 217+----------------------------+---------------------------------------------------+-----------------------+ 218| CUDA Native Backends | 219+----------------------------+---------------------------------------------------+-----------------------+ 220| ``/gpu/cuda/ref`` | Reference pure CUDA kernels | Yes | 221+----------------------------+---------------------------------------------------+-----------------------+ 222| ``/gpu/cuda/shared`` | Optimized pure CUDA kernels using shared memory | Yes | 223+----------------------------+---------------------------------------------------+-----------------------+ 224| ``/gpu/cuda/gen`` | Optimized pure CUDA kernels using code generation | No | 225+----------------------------+---------------------------------------------------+-----------------------+ 226| HIP Native Backends | 227+----------------------------+---------------------------------------------------+-----------------------+ 228| ``/gpu/hip/ref`` | Reference pure HIP kernels | Yes | 229+----------------------------+---------------------------------------------------+-----------------------+ 230| ``/gpu/hip/shared`` | Optimized pure HIP kernels using shared memory | Yes | 231+----------------------------+---------------------------------------------------+-----------------------+ 232| ``/gpu/hip/gen`` | Optimized pure HIP kernels using code generation | No | 233+----------------------------+---------------------------------------------------+-----------------------+ 234| MAGMA Backends | 235+----------------------------+---------------------------------------------------+-----------------------+ 236| ``/gpu/cuda/magma`` | CUDA MAGMA kernels | No | 237+----------------------------+---------------------------------------------------+-----------------------+ 238| ``/gpu/cuda/magma/det`` | CUDA MAGMA kernels | Yes | 239+----------------------------+---------------------------------------------------+-----------------------+ 240| ``/gpu/hip/magma`` | HIP MAGMA kernels | No | 241+----------------------------+---------------------------------------------------+-----------------------+ 242| ``/gpu/hip/magma/det`` | HIP MAGMA kernels | Yes | 243+----------------------------+---------------------------------------------------+-----------------------+ 244| OCCA Backends | 245+----------------------------+---------------------------------------------------+-----------------------+ 246| ``/*/occa`` | Selects backend based on available OCCA modes | Yes | 247+----------------------------+---------------------------------------------------+-----------------------+ 248| ``/cpu/self/occa`` | OCCA backend with serial CPU kernels | Yes | 249+----------------------------+---------------------------------------------------+-----------------------+ 250| ``/cpu/openmp/occa`` | OCCA backend with OpenMP kernels | Yes | 251+----------------------------+---------------------------------------------------+-----------------------+ 252| ``/gpu/cuda/occa`` | OCCA backend with CUDA kernels | Yes | 253+----------------------------+---------------------------------------------------+-----------------------+ 254| ``/gpu/hip/occa`` | OCCA backend with HIP kernels | Yes | 255+----------------------------+---------------------------------------------------+-----------------------+ 256``` 257 258The `/cpu/self/*/serial` backends process one element at a time and are intended for meshes 259with a smaller number of high order elements. The `/cpu/self/*/blocked` backends process 260blocked batches of eight interlaced elements and are intended for meshes with higher numbers 261of elements. 262 263The `/cpu/self/ref/*` backends are written in pure C and provide basic functionality. 264 265The `/cpu/self/opt/*` backends are written in pure C and use partial e-vectors to improve performance. 266 267The `/cpu/self/avx/*` backends rely upon AVX instructions to provide vectorized CPU performance. 268 269The `/cpu/self/memcheck/*` backends rely upon the [Valgrind](http://valgrind.org/) Memcheck tool 270to help verify that user QFunctions have no undefined values. To use, run your code with 271Valgrind and the Memcheck backends, e.g. `valgrind ./build/ex1 -ceed /cpu/self/ref/memcheck`. A 272'development' or 'debugging' version of Valgrind with headers is required to use this backend. 273This backend can be run in serial or blocked mode and defaults to running in the serial mode 274if `/cpu/self/memcheck` is selected at runtime. 275 276The `/cpu/self/xsmm/*` backends rely upon the [LIBXSMM](http://github.com/hfp/libxsmm) package 277to provide vectorized CPU performance. If linking MKL and LIBXSMM is desired but 278the Makefile is not detecting `MKLROOT`, linking libCEED against MKL can be 279forced by setting the environment variable `MKL=1`. 280 281The `/gpu/cuda/*` backends provide GPU performance strictly using CUDA. 282 283The `/gpu/hip/*` backends provide GPU performance strictly using HIP. They are based on 284the `/gpu/cuda/*` backends. ROCm version 3.6 or newer is required. 285 286The `/gpu/*/magma/*` backends rely upon the [MAGMA](https://bitbucket.org/icl/magma) package. 287To enable the MAGMA backends, the environment variable `MAGMA_DIR` must point to the top-level 288MAGMA directory, with the MAGMA library located in `$(MAGMA_DIR)/lib/`. 289By default, `MAGMA_DIR` is set to `../magma`; to build the MAGMA backends 290with a MAGMA installation located elsewhere, create a link to `magma/` in libCEED's parent 291directory, or set `MAGMA_DIR` to the proper location. MAGMA version 2.5.0 or newer is required. 292Currently, each MAGMA library installation is only built for either CUDA or HIP. The corresponding 293set of libCEED backends (`/gpu/cuda/magma/*` or `/gpu/hip/magma/*`) will automatically be built 294for the version of the MAGMA library found in `MAGMA_DIR`. 295 296Users can specify a device for all CUDA, HIP, and MAGMA backends through adding `:device_id=#` 297after the resource name. For example: 298 299> - `/gpu/cuda/gen:device_id=1` 300 301The `/*/occa` backends rely upon the [OCCA](http://github.com/libocca/occa) package to provide 302cross platform performance. To enable the OCCA backend, the environment variable `OCCA_DIR` must point 303to the top-level OCCA directory, with the OCCA library located in the `${OCCA_DIR}/lib` (By default, 304`OCCA_DIR` is set to `../occa`). 305 306Additionally, users can pass specific OCCA device properties after setting the CEED resource. 307For example: 308 309> - `"/*/occa:mode='CUDA',device_id=0"` 310 311Bit-for-bit reproducibility is important in some applications. 312However, some libCEED backends use non-deterministic operations, such as `atomicAdd` for increased performance. 313The backends which are capable of generating reproducible results, with the proper compilation options, are highlighted in the list above. 314 315## Examples 316 317libCEED comes with several examples of its usage, ranging from standalone C 318codes in the `/examples/ceed` directory to examples based on external packages, 319such as MFEM, PETSc, and Nek5000. Nek5000 v18.0 or greater is required. 320 321To build the examples, set the `MFEM_DIR`, `PETSC_DIR`, and 322`NEK5K_DIR` variables and run: 323 324``` 325cd examples/ 326``` 327 328% running-examples-inclusion-marker 329 330```console 331# libCEED examples on CPU and GPU 332cd ceed/ 333make 334./ex1-volume -ceed /cpu/self 335./ex1-volume -ceed /gpu/cuda 336./ex2-surface -ceed /cpu/self 337./ex2-surface -ceed /gpu/cuda 338cd .. 339 340# MFEM+libCEED examples on CPU and GPU 341cd mfem/ 342make 343./bp1 -ceed /cpu/self -no-vis 344./bp3 -ceed /gpu/cuda -no-vis 345cd .. 346 347# Nek5000+libCEED examples on CPU and GPU 348cd nek/ 349make 350./nek-examples.sh -e bp1 -ceed /cpu/self -b 3 351./nek-examples.sh -e bp3 -ceed /gpu/cuda -b 3 352cd .. 353 354# PETSc+libCEED examples on CPU and GPU 355cd petsc/ 356make 357./bps -problem bp1 -ceed /cpu/self 358./bps -problem bp2 -ceed /gpu/cuda 359./bps -problem bp3 -ceed /cpu/self 360./bps -problem bp4 -ceed /gpu/cuda 361./bps -problem bp5 -ceed /cpu/self 362./bps -problem bp6 -ceed /gpu/cuda 363cd .. 364 365cd petsc/ 366make 367./bpsraw -problem bp1 -ceed /cpu/self 368./bpsraw -problem bp2 -ceed /gpu/cuda 369./bpsraw -problem bp3 -ceed /cpu/self 370./bpsraw -problem bp4 -ceed /gpu/cuda 371./bpsraw -problem bp5 -ceed /cpu/self 372./bpsraw -problem bp6 -ceed /gpu/cuda 373cd .. 374 375cd petsc/ 376make 377./bpssphere -problem bp1 -ceed /cpu/self 378./bpssphere -problem bp2 -ceed /gpu/cuda 379./bpssphere -problem bp3 -ceed /cpu/self 380./bpssphere -problem bp4 -ceed /gpu/cuda 381./bpssphere -problem bp5 -ceed /cpu/self 382./bpssphere -problem bp6 -ceed /gpu/cuda 383cd .. 384 385cd petsc/ 386make 387./area -problem cube -ceed /cpu/self -degree 3 388./area -problem cube -ceed /gpu/cuda -degree 3 389./area -problem sphere -ceed /cpu/self -degree 3 -dm_refine 2 390./area -problem sphere -ceed /gpu/cuda -degree 3 -dm_refine 2 391 392cd fluids/ 393make 394./navierstokes -ceed /cpu/self -degree 1 395./navierstokes -ceed /gpu/cuda -degree 1 396cd .. 397 398cd solids/ 399make 400./elasticity -ceed /cpu/self -mesh [.exo file] -degree 2 -E 1 -nu 0.3 -problem Linear -forcing mms 401./elasticity -ceed /gpu/cuda -mesh [.exo file] -degree 2 -E 1 -nu 0.3 -problem Linear -forcing mms 402cd .. 403``` 404 405For the last example shown, sample meshes to be used in place of 406`[.exo file]` can be found at <https://github.com/jeremylt/ceedSampleMeshes> 407 408The above code assumes a GPU-capable machine with the OCCA backend 409enabled. Depending on the available backends, other CEED resource 410specifiers can be provided with the `-ceed` option. Other command line 411arguments can be found in [examples/petsc](https://github.com/CEED/libCEED/blob/main/examples/petsc/README.md). 412 413% benchmarks-marker 414 415## Benchmarks 416 417A sequence of benchmarks for all enabled backends can be run using: 418 419``` 420make benchmarks 421``` 422 423The results from the benchmarks are stored inside the `benchmarks/` directory 424and they can be viewed using the commands (requires python with matplotlib): 425 426``` 427cd benchmarks 428python postprocess-plot.py petsc-bps-bp1-*-output.txt 429python postprocess-plot.py petsc-bps-bp3-*-output.txt 430``` 431 432Using the `benchmarks` target runs a comprehensive set of benchmarks which may 433take some time to run. Subsets of the benchmarks can be run using the scripts in the `benchmarks` folder. 434 435For more details about the benchmarks, see the `benchmarks/README.md` file. 436 437## Install 438 439To install libCEED, run: 440 441``` 442make install prefix=/usr/local 443``` 444 445or (e.g., if creating packages): 446 447``` 448make install prefix=/usr DESTDIR=/packaging/path 449``` 450 451The usual variables like `CC` and `CFLAGS` are used, and optimization flags 452for all languages can be set using the likes of `OPT='-O3 -march=native'`. Use 453`STATIC=1` to build static libraries (`libceed.a`). 454 455To install libCEED for Python, run: 456 457``` 458pip install libceed 459``` 460 461with the desired setuptools options, such as `--user`. 462 463### pkg-config 464 465In addition to library and header, libCEED provides a [pkg-config](https://en.wikipedia.org/wiki/Pkg-config) 466file that can be used to easily compile and link. 467[For example](https://people.freedesktop.org/~dbn/pkg-config-guide.html#faq), if 468`$prefix` is a standard location or you set the environment variable 469`PKG_CONFIG_PATH`: 470 471``` 472cc `pkg-config --cflags --libs ceed` -o myapp myapp.c 473``` 474 475will build `myapp` with libCEED. This can be used with the source or 476installed directories. Most build systems have support for pkg-config. 477 478## Contact 479 480You can reach the libCEED team by emailing [ceed-users@llnl.gov](mailto:ceed-users@llnl.gov) 481or by leaving a comment in the [issue tracker](https://github.com/CEED/libCEED/issues). 482 483## How to Cite 484 485If you utilize libCEED please cite: 486 487``` 488@article{libceed-joss-paper, 489 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}, 490 title = {{libCEED}: Fast algebra for high-order element-based discretizations}, 491 journal = {Journal of Open Source Software}, 492 year = {2021}, 493 publisher = {The Open Journal}, 494 volume = {6}, 495 number = {63}, 496 pages = {2945}, 497 doi = {10.21105/joss.02945} 498} 499 500@misc{libceed-user-manual, 501 author = {Abdelfattah, Ahmad and 502 Barra, Valeria and 503 Beams, Natalie and 504 Brown, Jed and 505 Camier, Jean-Sylvain and 506 Dobrev, Veselin and 507 Dudouit, Yohann and 508 Ghaffari, Leila and 509 Kolev, Tzanio and 510 Medina, David and 511 Pazner, Will and 512 Ratnayaka, Thilina and 513 Thompson, Jeremy L and 514 Tomov, Stanimire}, 515 title = {{libCEED} User Manual}, 516 month = jul, 517 year = 2021, 518 publisher = {Zenodo}, 519 version = {0.9.0}, 520 doi = {10.5281/zenodo.5077489} 521} 522``` 523 524For libCEED's Python interface please cite: 525 526``` 527@InProceedings{libceed-paper-proc-scipy-2020, 528 author = {{V}aleria {B}arra and {J}ed {B}rown and {J}eremy {T}hompson and {Y}ohann {D}udouit}, 529 title = {{H}igh-performance operator evaluations with ease of use: lib{C}{E}{E}{D}'s {P}ython interface}, 530 booktitle = {{P}roceedings of the 19th {P}ython in {S}cience {C}onference}, 531 pages = {85 - 90}, 532 year = {2020}, 533 editor = {{M}eghann {A}garwal and {C}hris {C}alloway and {D}illon {N}iederhut and {D}avid {S}hupe}, 534 doi = {10.25080/Majora-342d178e-00c} 535} 536``` 537 538The BiBTeX entries for these references can be found in the 539`doc/bib/references.bib` file. 540 541## Copyright 542 543The following copyright applies to each file in the CEED software suite, unless 544otherwise stated in the file: 545 546> Copyright (c) 2017, Lawrence Livermore National Security, LLC. Produced at the 547> Lawrence Livermore National Laboratory. LLNL-CODE-734707. All Rights reserved. 548 549See files LICENSE and NOTICE for details. 550
