Compare commits

..
268 changed files with 4725 additions and 14450 deletions
+3 -15
View File
@@ -1,13 +1,10 @@
version: '{build}'
# https://www.appveyor.com/docs/build-environment/#build-worker-images
image: Visual Studio 2019
image: Visual Studio 2017
install:
# Start from outside clone directory
- cd ..
# Install MS-MPI
- ps: Start-FileDownload 'https://download.microsoft.com/download/B/2/E/B2EB83FE-98C2-4156-834A-E1711E6884FB/MSMpiSetup.exe'
- MSMpiSetup.exe -unattend
@@ -18,11 +15,6 @@ install:
- msmpisdk.msi /passive
- set PATH=C:\Program Files\Microsoft MPI\Bin;%PATH%
# Set MSMPI environment variables needed for CMake detection
- set MSMPI_LIB32=C:\Program Files (x86)\Microsoft SDKs\MPI\Lib\x86
- set MSMPI_LIB64=C:\Program Files (x86)\Microsoft SDKs\MPI\Lib\x64
- set MSMPI_INC=C:\Program Files (x86)\Microsoft SDKs\MPI\Include
# Install METIS, use a mirror because the original source server is not always
# up. Original url:
# http://glaros.dtc.umn.edu/gkhome/fetch/sw/metis/metis-5.1.0.tar.gz
@@ -33,24 +25,20 @@ install:
- cmake -H. -Bbuild
# -DCMAKE_BUILD_TYPE=Release
- cmake --build build
- set METIS_PATH=%cd%
- cd ..
# Install hypre
- ps: Start-FileDownload 'https://github.com/hypre-space/hypre/archive/v2.19.0.tar.gz'
- 7z x v2.19.0.tar.gz -so | 7z x -si -ttar > nul
- cd hypre-2.19.0/src
- cmake -H. -Bbuild
- cmake -H. -Bbuild -DMPI_C_INCLUDE_PATH="C:\Program Files (x86)\Microsoft SDKs\MPI\Include" -DMPI_C_LIBRARIES="C:\Program Files (x86)\Microsoft SDKs\MPI\Lib\x86\msmpi.lib" -DMPI_CXX_LIBRARIES="C:\Program Files (x86)\Microsoft SDKs\MPI\Lib\x86\msmpi.lib" -DMPI_CXX_INCLUDE_PATH="C:\Program Files (x86)\Microsoft SDKs\MPI\Include"
- cmake --build build
- cmake --build build --target install
- cd ../..
# Return to clone directory
- cd %APPVEYOR_BUILD_FOLDER%
# MFEM
before_build:
- cmake -H. -DCMAKE_INSTALL_PREFIX=install -Bbuild_parallel -DMFEM_USE_MPI=TRUE -DMFEM_USE_METIS_5=TRUE -DHYPRE_DIR=%cd%\..\hypre-2.19.0\src\hypre -DMETIS_LIBRARIES=%METIS_PATH%\build\libmetis\Debug\metis.lib -DMETIS_INCLUDE_DIRS=%METIS_PATH%\include
- cmake -H. -DCMAKE_INSTALL_PREFIX=install -Bbuild_parallel -DMFEM_USE_MPI=TRUE -DMFEM_USE_METIS_5=TRUE -DMPI_CXX_LIBRARIES="C:\Program Files (x86)\Microsoft SDKs\MPI\Lib\x86\msmpi.lib" -DMPI_CXX_INCLUDE_PATH="C:\Program Files (x86)\Microsoft SDKs\MPI\Include" -DHYPRE_DIR=%cd%\hypre-2.19.0\src\hypre -DMETIS_LIBRARIES=%cd%\metis-5.1.0\build\libmetis\Debug\metis.lib -DMETIS_INCLUDE_DIRS=%cd%\metis-5.1.0\include
- cmake -H. -DCMAKE_INSTALL_PREFIX=install -Bbuild_serial -DMFEM_USE_MPI=FALSE
build_script:
+2 -4
View File
@@ -1,4 +1,4 @@
name: "Docker"
name: Build Deploy Container
on:
@@ -20,7 +20,6 @@ on:
jobs:
build:
if: github.repository == 'mfem/mfem' # Don't run in forks
permissions:
packages: write
strategy:
@@ -28,8 +27,7 @@ jobs:
matrix:
# Dockerfiles to build, a matrix supports future expanded builds
container: [["config/docker/Dockerfile.base", "ghcr.io/mfem/mfem-ubuntu-base"],
["config/docker/Dockerfile", "ghcr.io/mfem/mfem-ubuntu"]]
container: [["config/docker/Dockerfile", "ghcr.io/mfem/mfem-ubuntu-base"]]
runs-on: ubuntu-latest
name: Build
+21 -85
View File
@@ -10,7 +10,7 @@
# CONTRIBUTING.md for details.
# In this CI section, we build different variants of mfem and run test on them.
name: "Tests"
name: builds-and-tests
# Github actions can use the default "GITHUB_TOKEN". By default, this token
# is set to have permissive access. However, this is not a good practice
@@ -27,7 +27,6 @@ on:
- master
- next
pull_request:
workflow_dispatch:
env:
HYPRE_ARCHIVE: v2.19.0.tar.gz
@@ -47,18 +46,11 @@ jobs:
builds-and-tests:
strategy:
matrix:
os: [ubuntu-latest, macos-latest, windows-latest]
os: [ubuntu-18.04, macos-10.15]
target: [dbg, opt]
mpi: [seq, par]
build-system: [make, cmake]
build-system: [make]
hypre-target: [int32]
exclude:
- os: ubuntu-latest
build-system: cmake
- os: macos-latest
build-system: cmake
- os: windows-latest
build-system: make
# 'include' allows us to:
# - Add a variable to all jobs without creating a new matrix dimension.
# Codecov is defined that way.
@@ -72,15 +64,13 @@ jobs:
codecov: NO
- target: opt
codecov: YES
- os: windows-latest
codecov: NO
- os: ubuntu-latest
- os: ubuntu-18.04
target: opt
codecov: NO
mpi: par
build-system: cmake
hypre-target: int32
- os: ubuntu-latest
- os: ubuntu-18.04
target: opt
codecov: NO
mpi: par
@@ -112,37 +102,33 @@ jobs:
# TODO: It would be nice to have only one step, e.g. with a dedicated
# action, but I (@adrienbernede) don't see how at the moment.
- name: get MPI (Linux)
if: matrix.mpi == 'par' && matrix.os == 'ubuntu-latest'
if: matrix.mpi == 'par' && matrix.os == 'ubuntu-18.04'
run: |
sudo apt-get install mpich libmpich-dev
export MAKE_CXX_FLAG="MPICXX=mpic++"
- name: get lcov (Linux)
if: matrix.codecov == 'YES' && matrix.os == 'ubuntu-latest'
if: matrix.codecov == 'YES' && matrix.os == 'ubuntu-18.04'
run: |
sudo apt-get install lcov
- name: Set up Homebrew
if: ( matrix.mpi == 'par' || matrix.codecov == 'YES' ) && matrix.os == 'macos-latest'
if: ( matrix.mpi == 'par' || matrix.codecov == 'YES' ) && matrix.os == 'macos-10.15'
uses: Homebrew/actions/setup-homebrew@c4aafe8c4620bf08883dd4679c374f11e73329d3
- name: get MPI (MacOS)
if: matrix.mpi == 'par' && matrix.os == 'macos-latest'
if: matrix.mpi == 'par' && matrix.os == 'macos-10.15'
run: |
export HOMEBREW_NO_INSTALL_CLEANUP=1
brew install openmpi
export MAKE_CXX_FLAG="MPICXX=mpic++"
- name: get MPI (MacOS)
if: matrix.codecov == 'YES' && matrix.os == 'macos-latest'
if: matrix.codecov == 'YES' && matrix.os == 'macos-10.15'
run: |
export HOMEBREW_NO_INSTALL_CLEANUP=1
brew install lcov
- name: get MPI (Windows)
if: matrix.mpi == 'par' && matrix.os == 'windows-latest'
uses: mpi4py/setup-mpi@v1.0.3
# Get Hypre through cache, or build it.
# Install will only run on cache miss.
- name: cache hypre
@@ -151,67 +137,36 @@ jobs:
uses: actions/cache@v2
with:
path: ${{ env.HYPRE_TOP_DIR }}
key: ${{ runner.os }}-build-${{ env.HYPRE_TOP_DIR }}-${{ matrix.hypre-target }}-v2.2
key: ${{ runner.os }}-build-${{ env.HYPRE_TOP_DIR }}-${{ matrix.hypre-target }}-v2.0
- name: get hypre
if: matrix.mpi == 'par' && steps.hypre-cache.outputs.cache-hit != 'true' && matrix.os != 'windows-latest'
uses: mfem/github-actions/build-hypre@v2.2
if: matrix.mpi == 'par' && steps.hypre-cache.outputs.cache-hit != 'true'
uses: mfem/github-actions/build-hypre@v2.0
with:
archive: ${{ env.HYPRE_ARCHIVE }}
dir: ${{ env.HYPRE_TOP_DIR }}
target: ${{ matrix.hypre-target }}
build-system: make
- name: get hypre (Windows)
if: matrix.mpi == 'par' && steps.hypre-cache.outputs.cache-hit != 'true' && matrix.os == 'windows-latest'
uses: mfem/github-actions/build-hypre@v2.2
with:
archive: ${{ env.HYPRE_ARCHIVE }}
dir: ${{ env.HYPRE_TOP_DIR }}
target: ${{ matrix.hypre-target }}
build-system: cmake
# Get Metis through cache, or build it.
# Install will only run on cache miss.
- name: cache metis
id: metis-cache
if: matrix.mpi == 'par' && matrix.os != 'windows-latest'
if: matrix.mpi == 'par'
uses: actions/cache@v2
with:
path: ${{ env.METIS_TOP_DIR }}
key: ${{ runner.os }}-build-${{ env.METIS_TOP_DIR }}-v2.2
key: ${{ runner.os }}-build-${{ env.METIS_TOP_DIR }}-v2.0
- name: install metis
if: matrix.mpi == 'par' && matrix.os != 'windows-latest' && steps.metis-cache.outputs.cache-hit != 'true'
uses: mfem/github-actions/build-metis@v2.2
if: matrix.mpi == 'par' && steps.metis-cache.outputs.cache-hit != 'true'
uses: mfem/github-actions/build-metis@v2.0
with:
archive: ${{ env.METIS_ARCHIVE }}
dir: ${{ env.METIS_TOP_DIR }}
- name: cache vcpkg (Windows)
id: vcpkg-cache
uses: actions/cache@v3
with:
path: vcpkg_cache
key: ${{ runner.os }}-${{ matrix.mpi }}-vcpkg-v1
- name: prepare binary cache location
if: matrix.os == 'windows-latest' && steps.vcpkg-cache.outputs.cache-hit != 'true'
run: |
mkdir -p vcpkg_cache
- name: install metis (Windows)
if: matrix.mpi == 'par' && matrix.os == 'windows-latest'
env:
VCPKG_DEFAULT_BINARY_CACHE: ${{ github.workspace }}/vcpkg_cache
run: |
vcpkg install metis-mfem --triplet=x64-windows-static --overlay-ports=${{ env.MFEM_TOP_DIR }}/config/vcpkg/ports
# MFEM build and test
- name: build
uses: mfem/github-actions/build-mfem@v2.2
env:
VCPKG_DEFAULT_BINARY_CACHE: ${{ github.workspace }}/vcpkg_cache
uses: mfem/github-actions/build-mfem@v2.1
with:
os: ${{ matrix.os }}
target: ${{ matrix.target }}
@@ -221,8 +176,6 @@ jobs:
hypre-dir: ${{ env.HYPRE_TOP_DIR }}
metis-dir: ${{ env.METIS_TOP_DIR }}
mfem-dir: ${{ env.MFEM_TOP_DIR }}
config-options: ${{ env.MFEM_EXTRA_CONFIG }}
library-only: ${{ matrix.target == 'dbg' }}
# Run checks (and only checks) on debug targets
- name: checks
@@ -240,27 +193,10 @@ jobs:
run: |
cd ${{ env.MFEM_TOP_DIR }} && make test
- name: cmake checks
if: matrix.build-system == 'cmake' && matrix.target == 'dbg'
- name: cmake unit tests
if: matrix.build-system == 'cmake'
run: |
CTEST_CONFIG="Debug"
cd ${{ env.MFEM_TOP_DIR }} && cmake --build build --target check --config ${CTEST_CONFIG}
shell: bash
- name: cmake unit tests (Ubuntu 20.04)
if: matrix.build-system == 'cmake' && matrix.target == 'opt' && matrix.os == 'ubuntu-latest'
run: |
CTEST_CONFIG="Release"
[[ ${{ matrix.target }} == 'dbg' ]] && CTEST_CONFIG="Debug"
cd ${{ env.MFEM_TOP_DIR }}/build/tests/unit && ctest --output-on-failure -C ${CTEST_CONFIG}
shell: bash
- name: cmake tests
if: matrix.build-system == 'cmake' && matrix.target == 'opt' && matrix.os != 'ubuntu-latest'
run: |
CTEST_CONFIG="Release"
cd ${{ env.MFEM_TOP_DIR }}/build && ctest --output-on-failure -C ${CTEST_CONFIG}
shell: bash
cd ${{ env.MFEM_TOP_DIR }}/build/tests/unit && ctest --output-on-failure
# Code coverage (process and upload reports)
- name: codecov
-71
View File
@@ -1,71 +0,0 @@
# For most projects, this workflow file will not need changing; you simply need
# to commit it to your repository.
#
# You may wish to alter this file to override the set of languages analyzed,
# or to provide custom queries or build logic.
#
# ******** NOTE ********
# We have attempted to detect the languages in your repository. Please check
# the `language` matrix defined below to confirm you have the correct set of
# supported CodeQL languages.
#
name: "Static Analysis"
on:
push:
branches: [ "master", "next"]
pull_request:
# The branches below must be a subset of the branches above
branches: [ "master" ]
jobs:
analyze:
name: Analyze
runs-on: ubuntu-latest
permissions:
actions: read
contents: read
security-events: write
strategy:
fail-fast: false
matrix:
language: [ 'cpp' ]
# CodeQL supports [ 'cpp', 'csharp', 'go', 'java', 'javascript', 'python', 'ruby' ]
# Learn more about CodeQL language support at https://aka.ms/codeql-docs/language-support
steps:
- name: Checkout repository
uses: actions/checkout@v3
# Initializes the CodeQL tools for scanning.
- name: Initialize CodeQL
uses: github/codeql-action/init@v2
with:
languages: ${{ matrix.language }}
# If you wish to specify custom queries, you can do so here or in a config file.
# By default, queries listed here will override any specified in a config file.
# Prefix the list here with "+" to use these queries and those in the config file.
# Details on CodeQL's query packs refer to : https://docs.github.com/en/code-security/code-scanning/automatically-scanning-your-code-for-vulnerabilities-and-errors/configuring-code-scanning#using-queries-in-ql-packs
# queries: security-extended,security-and-quality
queries: lgtm
# Autobuild attempts to build any compiled languages (C/C++, C#, or Java).
# If this step fails, then you should remove it and run the build manually (see below)
- name: Autobuild
uses: github/codeql-action/autobuild@v2
# ️ Command-line programs to run using the OS shell.
# 📚 See https://docs.github.com/en/actions/using-workflows/workflow-syntax-for-github-actions#jobsjob_idstepsrun
# If the Autobuild fails above, remove it and uncomment the following three lines.
# modify them (or add more) to build your code if your project, please refer to the EXAMPLE below for guidance.
# - run: |
# echo "Run, Build Application using script"
# ./location_of_script_within_repo/buildscript.sh
- name: Perform CodeQL Analysis
uses: github/codeql-action/analyze@v2
+7 -8
View File
@@ -9,7 +9,7 @@
# terms of the BSD-3 license. We welcome feedback and contributions, see file
# CONTRIBUTING.md for details.
name: "Build Analysis"
name: build-analysis
permissions:
actions: write
@@ -20,7 +20,6 @@ on:
- master
- next
pull_request:
workflow_dispatch:
env:
HYPRE_ARCHIVE: v2.19.0.tar.gz
@@ -31,7 +30,7 @@ env:
jobs:
gitignore:
runs-on: ubuntu-latest
runs-on: ubuntu-18.04
steps:
- name: Cancel Previous Runs
@@ -54,11 +53,11 @@ jobs:
uses: actions/cache@v2
with:
path: ${{ env.HYPRE_TOP_DIR }}
key: ${{ runner.os }}-build-${{ env.HYPRE_TOP_DIR }}-v2.2
key: ${{ runner.os }}-build-${{ env.HYPRE_TOP_DIR }}-v2.0
- name: Get Hypre
if: steps.hypre-cache.outputs.cache-hit != 'true'
uses: mfem/github-actions/build-hypre@v2.2
uses: mfem/github-actions/build-hypre@v2.0
with:
archive: ${{ env.HYPRE_ARCHIVE }}
dir: ${{ env.HYPRE_TOP_DIR }}
@@ -69,18 +68,18 @@ jobs:
uses: actions/cache@v2
with:
path: ${{ env.METIS_TOP_DIR }}
key: ${{ runner.os }}-build-${{ env.METIS_TOP_DIR }}-v2.2
key: ${{ runner.os }}-build-${{ env.METIS_TOP_DIR }}-v2.0
- name: Install Metis
if: steps.metis-cache.outputs.cache-hit != 'true'
uses: mfem/github-actions/build-metis@v2.2
uses: mfem/github-actions/build-metis@v2.0
with:
archive: ${{ env.METIS_ARCHIVE }}
dir: ${{ env.METIS_TOP_DIR }}
# MFEM build and test
- name: build-mfem
uses: mfem/github-actions/build-mfem@v2.2
uses: mfem/github-actions/build-mfem@v2.0
with:
os: ${{ runner.os }}
target: opt
+12 -36
View File
@@ -9,35 +9,23 @@
# terms of the BSD-3 license. We welcome feedback and contributions, see file
# CONTRIBUTING.md for details.
name: "Checks"
name: repo-check
permissions:
actions: write
on:
push:
pull_request:
workflow_dispatch:
# This workflow is run on pushes to any branch in the MFEM repo (with or without
# PRs), as well as on updates to PRs from forks. In particular, we do not
# duplicate work by running on both pushes and updates to local PRs. We do that
# by checking if the workflow trigger is 'push' ("github.event_name == 'push'")
# and if we are in a fork ("github.event.pull_request.head.repo.full_name !=
# github.repository").
jobs:
file-headers-check:
runs-on: ubuntu-latest
if: |
(github.event_name == 'push' ||
github.event.pull_request.head.repo.full_name != github.repository)
runs-on: ubuntu-18.04
steps:
- name: Cancel Previous Runs
uses: styfle/cancel-workflow-action@0.9.0
with:
access_token: ${{ github.token }}
- name: checkout mfem
uses: actions/checkout@v2
@@ -61,10 +49,7 @@ jobs:
continue-on-error: true
- name: wrap-up
if: |
steps.copyright.outcome != 'success' ||
steps.license.outcome != 'success' ||
steps.release.outcome != 'success'
if: steps.copyright.outcome != 'success' || steps.license.outcome != 'success' || steps.release.outcome != 'success'
run: |
if [[ "${{ steps.copyright.outcome }}" != "success" ]]; then
echo "copyright check failed, unroll log for details"
@@ -78,27 +63,23 @@ jobs:
exit 1
code-style:
runs-on: ubuntu-latest
if: |
(github.event_name == 'push' ||
github.event.pull_request.head.repo.full_name != github.repository)
runs-on: ubuntu-18.04
steps:
- name: checkout mfem
uses: actions/checkout@v2
- name: get astyle
run: |
sudo apt-get install astyle
sudo apt-get install astyle=3.1-1ubuntu2
- name: style check
run: |
./config/githooks/pre-push --style
documentation:
runs-on: ubuntu-latest
if: |
(github.event_name == 'push' ||
github.event.pull_request.head.repo.full_name != github.repository)
runs-on: ubuntu-18.04
steps:
- name: checkout mfem
uses: actions/checkout@v2
@@ -106,8 +87,6 @@ jobs:
- name: get doxygen and graphviz
run: |
sudo apt-get install doxygen graphviz
cd doc
doxygen -u CodeDocumentation.conf.in 2>/dev/null
- name: build documentation
run: |
@@ -115,12 +94,9 @@ jobs:
./runtest documentation
branch-history:
if: |
github.ref != 'refs/heads/next' &&
github.ref != 'refs/heads/master' &&
(github.event_name == 'push' ||
github.event.pull_request.head.repo.full_name != github.repository)
runs-on: ubuntu-latest
if: github.ref != 'refs/heads/next' && github.ref != 'refs/heads/master'
runs-on: ubuntu-18.04
steps:
- name: checkout mfem
uses: actions/checkout@v2
-4
View File
@@ -261,14 +261,12 @@ miniapps/performance/sol.*
miniapps/shifted/distance
miniapps/shifted/ParaViewDistance
miniapps/shifted/ParaViewLSF
miniapps/shifted/extrapolate
miniapps/shifted/ParaViewExtrapolate
miniapps/shifted/diffusion
miniapps/shifted/diffusion.mesh
miniapps/shifted/diffusion.gf
miniapps/shifted/ParaViewDiffusion
miniapps/shifted/lsf_integral
miniapps/tools/display-basis
miniapps/tools/load-dc
@@ -307,8 +305,6 @@ miniapps/solvers/sol.*
miniapps/parelag/MultilevelHcurlHdivSolver
miniapps/parelag/*.mesh
miniapps/hooke/hooke
# Unit test binary and outputs
tests/unit/output_meshes
tests/unit/unit_tests
+11 -78
View File
@@ -10,83 +10,31 @@
Version 4.4.1 (development)
===========================
Meshing improvements
--------------------
- Added support for mixed meshes and pyramids in GSLIB-FindPoints.
Discretization improvements
---------------------------
- Added support for assembling low-order-refined matrices using a GPU-enabled
"batched" algorithm. The lor_solvers and plor_solvers now fully support GPU
acceleration.
- Added support for partial assembly and fully matrix-free operators on mixed
meshes (different element types and p-adaptivity) through libCEED, including
device acceleration, e.g. with NVIDIA and AMD GPUs. The p-adaptivity is
currently limited by MFEM capabilities, i.e. 2D serial meshes. All mixed
element topologies are supported in serial and parallel: segment, triangle,
square, tetrahedron, cube, prism, and pyramid.
- Added full assembly and device support for several LinearForm integrators:
* DomainLF: (f, v)
* VectorDomainLF: ((f1,...,fn), (v1,...,vn))
* DomainLFGrad: (f, grad(v))
* VectorDomainLFGrad: ((f1x,f1y,f1z,...,fnx,fny,fnz), grad(v1,...,vn))
- Added WhiteGaussianNoiseDomainLFIntegrator: a LinearFormIntegrator class for
spatial Gaussian white noise.
- Added a new Zienkiewicz-Zhu patch recovery-based a posteriori error estimator.
See fem/estimators.hpp.
Linear and nonlinear solvers
----------------------------
New and updated examples and miniapps
-------------------------------------
- Added a new elasticity miniapp, Hooke, that showcases a low-level approach of
using MFEM to solve a nonlinear elasticity problem based on the fundamental
finite element operator decomposition. The miniapp also integrates with
automatic differentiation tools like a native dual number implementation or a
third party library such as Enzyme. See miniapps/elasticity for more details.
- Add a new example code, Example 33/33p, to demonstrate the solution of
spectral fractional PDEs with MFEM.
Integrations, testing and documentation
---------------------------------------
- Added a Dockerfile for a simple MFEM container, see config/docker/README.md.
- Added support for ParMoonolith, https://bitbucket.org/zulianp/par_moonolith,
which provides parallel non-conforming, non-matching, variational, volumetric
mesh information transfer. With ParMortarAssember, fields can be exchanged
between arbitrarily distributed and unrelated finite element meshes in a
variationally consistent way.
- Added support for the LLVM-based automatic differentiation tool Enzyme, see
https://github.com/EnzymeAD/Enzyme. Build system flags and a convenience
header are provided. The functionality and interaction are demonstrated in a
new miniapp in miniapps/elasticity.
- Added full assembly and device support for several LinearForm integrators:
* DomainLF: (f, v)
* VectorDomainLF: ((f1,...,fn), (v1,...,vn))
* DomainLFGrad: (f, grad(v))
* VectorDomainLFGrad: ((f1x,f1y,f1z,...,fnx,fny,fnz), grad(v1,...,vn))
- Added example for body-fitted volumetric and shape integration using the
Algoim library.
- Add a new example code, Example 33/33p, to demonstrate the solution of
spectral fractional PDEs with MFEM.
- Added Windows 2022 CI testing with GitHub actions.
Miscellaneous
-------------
- Various other simplifications, extensions, and bugfixes in the code.
- Added boundary elimination with device support for `SparseMatrix` and
`HypreParMatrix`.
- When using `AssemblyLevel::FULL`, `FABilinearFormExtension::FormSystemMatrix`
outputs an `OperatorHandle` containing a `SparseMatrix` in serial, and an
`HypreParMatrix` in parallel (instead of a `ConstrainedOperator`).
- Added TMOP metrics for mesh untangling and worst-case quality improvement.
- Added a Dockerfile for a simple MFEM container, see config/docker/README.md
- Added support for assembling low-order-refined matrices using a GPU-enabled
"batched" algorithm. The lor_solvers and plor_solvers now fully support GPU
acceleration.
Version 4.4, released on March 21, 2022
=======================================
@@ -119,11 +67,6 @@ Meshing improvements
- Added a simpler interface to access mesh face information, see FaceInformation
and GetFaceInformation in the Mesh class.
- Added the method ParMesh::GetSerialMesh() that reconstructs a partitioned
parallel mesh on a given single rank. Also, added the method
ParMesh::PrintAsSerial() that saves the reconstructed serial mesh to a C++
stream on rank 0.
- Gmsh meshes where all elements have zero physical tag (the default Gmsh output
format if no physical groups are defined) are now successfully loaded, and
elements are reassigned attribute number 1.
@@ -213,9 +156,6 @@ Integrations, testing and documentation
- Switched from Artistic Style (astyle) version 2.05.1 to version 3.1 for code
formatting. See the "make style" target.
- New benchmark for the different assembly levels inspired by the CEED
Bake-Off Problems, see tests/benchmarks/bench_assembly_levels.cpp.
Miscellaneous
-------------
- Added a simple singleton class, Mpi, as a replacement for MPI_Session. New
@@ -226,13 +166,6 @@ Miscellaneous
- Fixed several MinGW build issues on Windows.
- In various places in the library, replace the use of 'long' with 'long long'
to better support Win64 builds where 'long' is 32-bit and 'long long' is
64-bit. On Linux and MacOS, both types are typically 64-bit.
- Update various "MemoryUsage" methods to return 'std::size_t' instead of 'long'
since the latter is 32-bit in Win64 builds.
- Added 'double' atomicAdd implementation for previous versions of CUDA.
- HypreParVector and Vector now support C++ move semantics, and the copy
+13 -31
View File
@@ -136,8 +136,6 @@ if (MFEM_USE_CUDA)
"CUDA flags set for MFEM" FORCE)
set(CUSPARSE_FOUND TRUE)
set(CUSPARSE_LIBRARIES "cusparse")
set(CUBLAS_FOUND TRUE)
set(CUSBLAS_LIBRARIES "cublas")
endif()
if (XSDK_ENABLE_C)
@@ -270,7 +268,7 @@ if (MFEM_USE_OPENMP OR MFEM_USE_LEGACY_OPENMP)
if(APPLE)
# On macOS, the compiler needs additional help to find the <omp.h> header.
# See issue #2642 for more information.
set(OPENMP_INCLUDE_DIRS ${OpenMP_CXX_INCLUDE_DIRS})
include_directories(${OpenMP_CXX_INCLUDE_DIRS})
endif(APPLE)
endif()
@@ -433,11 +431,6 @@ if (MFEM_USE_CALIPER)
find_package(Caliper REQUIRED)
endif()
# Algoim
if (MFEM_USE_ALGOIM)
find_package(Algoim REQUIRED)
endif()
# ADIOS2 for parallel I/O
if (MFEM_USE_ADIOS2)
find_package(ADIOS2 REQUIRED)
@@ -454,11 +447,6 @@ if (MFEM_USE_PARELAG)
find_package(PARELAG REQUIRED)
endif()
# Enzyme
if (MFEM_USE_ENZYME)
find_package(ENZYME REQUIRED)
endif()
# MFEM_TIMER_TYPE
if (NOT DEFINED MFEM_TIMER_TYPE)
if (APPLE)
@@ -485,8 +473,8 @@ endif()
set(MFEM_TPLS OPENMP HYPRE BLAS LAPACK SuperLUDist METIS SuiteSparse SUNDIALS
PETSC SLEPC MESQUITE MUMPS STRUMPACK AXOM FMS CONDUIT Ginkgo GNUTLS GSLIB
NETCDF MPFR PUMI HIOP POSIXCLOCKS MFEMBacktrace ZLIB OCCA CEED RAJA UMPIRE
ADIOS2 CUBLAS CUSPARSE MKL_CPARDISO AMGX CALIPER CODIPACK BENCHMARK PARELAG
MPI_CXX HIP HIPSPARSE MOONOLITH BLITZ ALGOIM ENZYME)
ADIOS2 CUSPARSE MKL_CPARDISO AMGX CALIPER CODIPACK BENCHMARK PARELAG
MPI_CXX HIP HIPSPARSE MOONOLITH)
# Add all *_FOUND libraries in the variable TPL_LIBRARIES.
set(TPL_LIBRARIES "")
@@ -501,6 +489,7 @@ endforeach(TPL)
list(REMOVE_DUPLICATES TPL_LIBRARIES)
list(REMOVE_DUPLICATES TPL_INCLUDE_DIRS)
# message(STATUS "TPL_INCLUDE_DIRS = ${TPL_INCLUDE_DIRS}")
include_directories(${TPL_INCLUDE_DIRS})
if (OPENMP_FOUND)
message(STATUS "MFEM: using package OpenMP")
@@ -557,15 +546,6 @@ target_link_libraries(mfem PUBLIC ${TPL_LIBRARIES})
if (MINGW)
target_link_libraries(mfem PRIVATE ws2_32)
endif()
if (MSVC)
target_compile_options(mfem PUBLIC "/wd4819")
endif()
message(STATUS "TPL_INCLUDE_DIRS = ${TPL_INCLUDE_DIRS}")
target_include_directories(mfem
PUBLIC
$<BUILD_INTERFACE:${CMAKE_CURRENT_BINARY_DIR}>
$<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR}>
${TPL_INCLUDE_DIRS})
set_target_properties(mfem PROPERTIES VERSION "${mfem_VERSION}")
set_target_properties(mfem PROPERTIES SOVERSION "${mfem_VERSION}")
@@ -694,10 +674,6 @@ set(INSTALL_LIB_DIR lib
set(INSTALL_CMAKE_DIR lib/cmake/mfem
CACHE PATH "Relative path for installing cmake config files.")
target_include_directories(mfem
PUBLIC
$<INSTALL_INTERFACE:${INSTALL_INCLUDE_DIR}>)
# The 'install' target will not depend on 'all'.
# set(CMAKE_SKIP_INSTALL_ALL_DEPENDENCY TRUE)
@@ -731,7 +707,7 @@ endif()
if (MFEM_USE_CEED)
install(DIRECTORY ${MFEM_SOURCE_DIRS}
DESTINATION ${INSTALL_INCLUDE_DIR}/mfem
FILES_MATCHING PATTERN "fem/ceed/integrators/*/*.h")
FILES_MATCHING PATTERN "fem/ceed/*.h")
endif()
# Install ${HEADERS}
@@ -762,8 +738,14 @@ export(TARGETS ${PROJECT_NAME}
# TODO: How do we register the install-tree? Replacing the build-tree?
export(PACKAGE ${PROJECT_NAME})
# Extract the include directories required to use MFEM
get_target_property(MFEM_TPL_INCLUDE_DIRS mfem INCLUDE_DIRECTORIES)
if (NOT MFEM_TPL_INCLUDE_DIRS)
set(MFEM_TPL_INCLUDE_DIRS "")
endif()
# This is the build-tree version
set(INCLUDE_INSTALL_DIRS ${PROJECT_BINARY_DIR} ${TPL_INCLUDE_DIRS})
set(INCLUDE_INSTALL_DIRS ${PROJECT_BINARY_DIR} ${MFEM_TPL_INCLUDE_DIRS})
set(LIB_INSTALL_DIR ${PROJECT_BINARY_DIR})
configure_package_config_file(config/cmake/MFEMConfig.cmake.in
${CMAKE_CURRENT_BINARY_DIR}/MFEMConfig.cmake
@@ -771,7 +753,7 @@ configure_package_config_file(config/cmake/MFEMConfig.cmake.in
PATH_VARS INCLUDE_INSTALL_DIRS LIB_INSTALL_DIR)
# This is the version that will be installed
set(INCLUDE_INSTALL_DIRS ${INSTALL_INCLUDE_DIR} ${TPL_INCLUDE_DIRS})
set(INCLUDE_INSTALL_DIRS ${INSTALL_INCLUDE_DIR} ${MFEM_TPL_INCLUDE_DIRS})
set(LIB_INSTALL_DIR ${INSTALL_LIB_DIR})
configure_package_config_file(config/cmake/MFEMConfig.cmake.in
${CMAKE_CURRENT_BINARY_DIR}${CMAKE_FILES_DIRECTORY}/MFEMConfig.cmake
-1
View File
@@ -131,7 +131,6 @@ The MFEM source code has the following structure:
│ ├── common
│ ├── electromagnetics
│ ├── gslib
│ ├── hooke
│ ├── meshing
│ ├── mtop
│ ├── navier
-38
View File
@@ -471,14 +471,6 @@ MFEM_USE_CODIPACK = YES/NO
Enable automatic differentiation using the CoDiPack library.
www.scicomp.uni-kl.de/codi/
MFEM_USE_ALGOIM = YES/NO
Enable the usage of Algoim - a collection of high-order accurate numerical
methods and C++ algorithms for working with implicitly-defined geometry and
level set methods. The Algoim library requires the Blitz++ library. The MFEM
provides interface to Algoim v1. Thus, to check out the specific state use:
git checkout 9c9ca0ef094d8ab0390ed36367a1151b459bbe0a
https://algoim.github.io
MFEM_USE_ADFORWARD = YES/NO
Enable forward mode for AD packages. This option is valid
only if the AD package supports two modes (backward/forward).
@@ -558,14 +550,6 @@ MFEM_USE_PARELAG = YES/NO
use ParELAG. In fact, ParELAG is dependent on MFEM. Therefore, this option
currently only concerns the miniapps.
MFEM_USE_ENZYME = YES/NO
Enables automatic differentiation support through the LLVM plugin Enzyme.
This requires the compiler to be set to clang (>=14.0.0). We also advise to
use the link time optimization (LTO) plugin, to enable functions that you
define over multiple files (compilation units) and want to be differentiated
automatically, to work. This requires to also use LLVM/LLD for linking.
Recommended options are in config/defaults.mk.
MFEM_BUILD_TAG = (any value)
An optional tag to characterize the build. Exported to config/config.mk.
Can be used to identify the MFEM build from other makefiles.
@@ -754,20 +738,6 @@ The specific libraries and their options are:
Options: GSLIB_OPT, GSLIB_LIB.
Versions: GSLIB >= 1.0.7.
- ALGOIM (optional), used when MFE_USE_ALGOIM=YES. The library provides only
headers so it just needs to be downloaded at the same level as MFEM. Download
the specific version we use as:
"git clone https://github.com/algoim/algoim.git;
git checkout 9c9ca0ef094d8ab0390ed36367a1151b459bbe0a"
ALGOIM depends on BLITZ and rhe library must be built prior to the MFEM build.
Download v1.0.2, untar it at the same level as MFEM and create a symbolic link:
"ln -s blitz-1.0.2 blitz".
Build Blitz using CMake as:
"cmake . -DCMAKE_INSTALL_PREFIX=.; make lib; make install"
URL: https://github.com/blitzpp/blitz/archive/refs/tags/1.0.2.tar.gz
Options: BLITZ_OPT, BLITZ_LIB
Versions: BLITZ = 1.0.2
- MKL CPardiso (optional), used when MFEM_USE_MKL_CPARDISO = YES.
URL: https://software.intel.com/content/www/us/en/develop/tools/math-kernel-library.html
Options: MKL_CPARDISO_OPT, MKL_CPARDISO_LIB.
@@ -844,12 +814,6 @@ The specific libraries and their options are:
URL: https://github.com/LLNL/parelag
Options: PARELAG_DIR, PARELAG_OPT, PARELAG_LIB.
- Enzyme, used when MFEM_USE_ENZYME = YES. Requires LLVM/Clang >= 14.0.0.
URL: https://github.com/EnzymeAD/Enzyme
Options: ENZYME_DIR, ENZYME_OPT, ENZYME_LIB.
Versions: Enzyme >= v0.0.33.
Building with CMake
===================
The MFEM build system consists of two steps: configuration and compilation.
@@ -988,7 +952,6 @@ MFEM_USE_CALIPER
MFEM_USE_FMS
MFEM_USE_BENCHMARK
MFEM_USE_PARELAG
MFEM_USE_ENZYME
The following options are CMake specific:
@@ -1048,7 +1011,6 @@ The CMake build system adds auto-detection for the following packages/libraries:
- FMS
- BENCHMARK
- ParELAG
- Enzyme
The following built-in CMake packages are also used:
+14 -14
View File
@@ -1,3 +1,4 @@
```
Finite Element Discretization Library
__
_ __ ___ / _| ___ _ __ ___
@@ -7,18 +8,19 @@
https://mfem.org
[MFEM](https://mfem.org) is a modular parallel C++ library for finite element
methods. Its goal is to enable high-performance scalable finite element
discretization research and application development on a wide variety of
platforms, ranging from laptops to supercomputers.
```
MFEM is a modular parallel C++ library for finite element methods. Its goal is
to enable high-performance scalable finite element discretization research and
application development on a wide variety of platforms, ranging from laptops to
supercomputers.
We welcome contributions and feedback from the community. Please see the file
[CONTRIBUTING.md](CONTRIBUTING.md) for additional details about our development
process.
[CONTRIBUTING.md](CONTRIBUTING.md) for additional details about our development process.
* For building instructions, see the file [INSTALL](INSTALL), or type "make help".
* Copyright and licensing information can be found in files [LICENSE](LICENSE) and [NOTICE](NOTICE).
* Copyright and licensing information can be found in files LICENSE and NOTICE.
* The best starting point for new users interested in MFEM's features is to
review the examples and miniapps at https://mfem.org/examples.
@@ -61,16 +63,14 @@ solvers from the hypre library. Comprehensive support for other external
packages, e.g. PETSc, SUNDIALS and libCEED is also included, giving access to
additional linear and nonlinear solvers, preconditioners, time integrators, etc.
For examples of using MFEM, see the [examples/](examples) and [miniapps/](miniapps)
directories, as well as the OpenGL visualization tool GLVis which is available
at https://glvis.org.
For examples of using MFEM, see the examples/ and miniapps/ directories, as well
as the OpenGL visualization tool GLVis which is available at https://glvis.org.
## License
MFEM is distributed under the terms of the BSD-3 license. All new contributions
must be made under this license. See [LICENSE](LICENSE) and [NOTICE](NOTICE) for
details.
must be made under this license. See LICENSE and NOTICE for details.
SPDX-License-Identifier: BSD-3-Clause <br>
LLNL Release Number: LLNL-CODE-806117 <br>
SPDX-License-Identifier: BSD-3-Clause
LLNL Release Number: LLNL-CODE-806117
DOI: 10.11578/dc.20171025.1248
-2
View File
@@ -58,10 +58,8 @@ set(MFEM_USE_MOONOLITH @MFEM_USE_MOONOLITH@)
set(MFEM_USE_CODIPACK @MFEM_USE_CODIPACK@)
set(MFEM_USE_ADFORWARD @MFEM_USE_ADFORWARD@)
set(MFEM_USE_CALIPER @MFEM_USE_CALIPER@)
set(MFEM_USE_ALGOIM @MFEM_USE_ALGOIM@)
set(MFEM_USE_BENCHMARK @MFEM_USE_BENCHMARK@)
set(MFEM_USE_PARELAG @MFEM_USE_PARELAG@)
set(MFEM_USE_ENZYME @MFEM_USE_ENZYME@)
set(MFEM_CXX_COMPILER "@CMAKE_CXX_COMPILER@")
set(MFEM_CXX_FLAGS "@CMAKE_CXX_FLAGS@")
-6
View File
@@ -160,9 +160,6 @@
// Enable MFEM functionality based on the Caliper library
#cmakedefine MFEM_USE_CALIPER
// Enable MFEM functionality based on the Algoim library
#cmakedefine MFEM_USE_ALGOIM
// Which library functions to use in class StopWatch for measuring time.
// For a list of the available options, see INSTALL.
// If not defined, an option is selected automatically.
@@ -190,7 +187,4 @@
// Enable MFEM functionality based on the Google Benchmark library.
#cmakedefine MFEM_USE_BENCHMARK
// Enable Enzyme for AD
#cmakedefine MFEM_USE_ENZYME
#endif // MFEM_CONFIG_HEADER
-22
View File
@@ -1,22 +0,0 @@
# Copyright (c) 2010-2022, Lawrence Livermore National Security, LLC. Produced
# at the Lawrence Livermore National Laboratory. All Rights reserved. See files
# LICENSE and NOTICE for details. LLNL-CODE-806117.
#
# This file is part of the MFEM library. For more information and source code
# availability visit https://mfem.org.
#
# MFEM is free software; you can redistribute it and/or modify it under the
# terms of the BSD-3 license. We welcome feedback and contributions, see file
# CONTRIBUTING.md for details.
# Defines the following variables:
# - ALGOIM_FOUND
# - ALGOIM_LIBRARIES
# - ALGOIM_INCLUDE_DIRS
include(MfemCmakeUtilities)
mfem_find_package(Algoim ALGOIM ALGOIM_DIR
"include" "algoim_quad.hpp"
"" ""
"Paths to headers required by Algoim."
"Libraries required by Algoim.")
-22
View File
@@ -1,22 +0,0 @@
# Copyright (c) 2010-2022, Lawrence Livermore National Security, LLC. Produced
# at the Lawrence Livermore National Laboratory. All Rights reserved. See files
# LICENSE and NOTICE for details. LLNL-CODE-806117.
#
# This file is part of the MFEM library. For more information and source code
# availability visit https://mfem.org.
#
# MFEM is free software; you can redistribute it and/or modify it under the
# terms of the BSD-3 license. We welcome feedback and contributions, see file
# CONTRIBUTING.md for details.
# Defines the following variables:
# - BLITZ_FOUND
# - BLITZ_LIBRARIES
# - BLITZ_INCLUDE_DIRS
include(MfemCmakeUtilities)
mfem_find_package(Blitz BLITZ BLITZ_DIR
"include" "blitz/blitz.h"
"lib" "blitz"
"Paths to headers required by Blitz."
"Libraries required by Blitz.")
-27
View File
@@ -1,27 +0,0 @@
# Copyright (c) 2010-2022, Lawrence Livermore National Security, LLC. Produced
# at the Lawrence Livermore National Laboratory. All Rights reserved. See files
# LICENSE and NOTICE for details. LLNL-CODE-806117.
#
# This file is part of the MFEM library. For more information and source code
# availability visit https://mfem.org.
#
# MFEM is free software; you can redistribute it and/or modify it under the
# terms of the BSD-3 license. We welcome feedback and contributions, see file
# CONTRIBUTING.md for details.
message(STATUS "Looking for ENZYME ...")
message(STATUS " in ENZYME_DIR = ${ENZYME_DIR}")
# Make sure the directory and version combination works. Do nothing otherwise.
if(EXISTS "${ENZYME_DIR}/ClangEnzyme-${ENZYME_VERSION}.so")
message(STATUS "Found ENZYME: ${ENZYME_DIR}/ClangEnzyme-${ENZYME_VERSION}.so")
# Set ENZYME_FOUND
set(ENZYME_FOUND TRUE CACHE BOOL "ENZYME was found." FORCE)
# Set CXX flags to accomodate the Enzyme Clang plugin
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Xclang -load -Xclang ${ENZYME_DIR}/ClangEnzyme-${ENZYME_VERSION}.so -mllvm -enzyme-loose-types=1")
set(MFEM_USE_ENZYME YES)
else()
endif()
@@ -894,7 +894,7 @@ function(mfem_export_mk_files)
MFEM_USE_HIP MFEM_USE_RAJA MFEM_USE_OCCA MFEM_USE_CEED MFEM_USE_CALIPER
MFEM_USE_UMPIRE MFEM_USE_SIMD MFEM_USE_ADIOS2 MFEM_USE_MKL_CPARDISO
MFEM_USE_ADFORWARD MFEM_USE_CODIPACK MFEM_USE_BENCHMARK MFEM_USE_PARELAG
MFEM_USE_MOONOLITH MFEM_USE_ALGOIM MFEM_USE_ENZYME)
MFEM_USE_MOONOLITH)
foreach(var ${CONFIG_MK_BOOL_VARS})
if (${var})
set(${var} YES)
-6
View File
@@ -167,9 +167,6 @@
// Enable functionality based on the Caliper library.
// #define MFEM_USE_CALIPER
// Enable functionality based on the Algoim library.
// #define MFEM_USE_ALGOIM
// Enable functionality based on the Umpire library.
// #define MFEM_USE_UMPIRE
@@ -195,7 +192,4 @@
// Enable functionality based on the Google Benchmark library.
// #define MFEM_USE_BENCHMARK
// Enable the Enzyme LLVM plugin
// #define MFEM_USE_ENZYME
#endif // MFEM_CONFIG_HEADER
-1
View File
@@ -63,7 +63,6 @@ MFEM_USE_ADFORWARD = @MFEM_USE_ADFORWARD@
MFEM_USE_CODIPACK = @MFEM_USE_CODIPACK@
MFEM_USE_BENCHMARK = @MFEM_USE_BENCHMARK@
MFEM_USE_PARELAG = @MFEM_USE_PARELAG@
MFEM_USE_ENZYME = @MFEM_USE_ENZYME@
# Compiler, compile options, and link options
MFEM_CXX = @MFEM_CXX@
-7
View File
@@ -58,13 +58,11 @@ option(MFEM_USE_UMPIRE "Enable Umpire" OFF)
option(MFEM_USE_SIMD "Enable use of SIMD intrinsics" OFF)
option(MFEM_USE_ADIOS2 "Enable ADIOS2" OFF)
option(MFEM_USE_CALIPER "Enable Caliper support" OFF)
option(MFEM_USE_ALGOIM "Enable Algoim support" OFF)
option(MFEM_USE_MKL_CPARDISO "Enable MKL CPardiso" OFF)
option(MFEM_USE_ADFORWARD "Enable forward mode for AD" OFF)
option(MFEM_USE_CODIPACK "Enable automatic differentiation (AD) using CoDiPack" OFF)
option(MFEM_USE_BENCHMARK "Enable Google Benchmark" OFF)
option(MFEM_USE_PARELAG "Enable ParELAG" OFF)
option(MFEM_USE_ENZYME "Enable Enzyme" OFF)
# Optional overrides for autodetected MPIEXEC and MPIEXEC_NUMPROC_FLAG
# set(MFEM_MPIEXEC "mpirun" CACHE STRING "Command for running MPI tests")
@@ -238,11 +236,6 @@ set(RAJA_DIR "${MFEM_DIR}/../raja" CACHE PATH "Path to RAJA")
set(CEED_DIR "${MFEM_DIR}/../libCEED" CACHE PATH "Path to libCEED")
set(UMPIRE_DIR "${MFEM_DIR}/../umpire" CACHE PATH "Path to Umpire")
set(CALIPER_DIR "${MFEM_DIR}/../caliper" CACHE PATH "Path to Caliper")
set(BLITZ_DIR "${MFEM_DIR}/../blitz" CACHE PATH "Path to Blitz")
set(ALGOIM_DIR "${MFEM_DIR}/../algoim" CACHE PATH "Path to Algoim")
set(ALGOIM_REQUIRED_PACKAGES "BLITZ" CACHE STRING
"Packages that ALGOIM depends on.")
set(BENCHMARK_DIR "${MFEM_DIR}/../google-benchmark" CACHE PATH
"Path to Google Benchmark")
+2 -33
View File
@@ -24,7 +24,7 @@ EGREP_BIN = $(shell command -v egrep 2> /dev/null)
CXX = g++
MPICXX = mpicxx
BASE_FLAGS = -std=c++11
BASE_FLAGS = -std=c++17
OPTIM_FLAGS = -O3 $(BASE_FLAGS)
DEBUG_FLAGS = -g $(XCOMPILER)-Wall $(BASE_FLAGS)
@@ -42,9 +42,6 @@ STATIC = YES
SHARED = NO
# CUDA configuration options
#
# If you set MFEM_USE_ENZYME=YES, CUDA_CXX has to be configured to use cuda with
# clang as its host compiler.
CUDA_CXX = nvcc
CUDA_ARCH = sm_60
CUDA_FLAGS = -x=cu --expt-extended-lambda -arch=$(CUDA_ARCH)
@@ -156,7 +153,6 @@ MFEM_USE_RAJA = NO
MFEM_USE_OCCA = NO
MFEM_USE_CEED = NO
MFEM_USE_CALIPER = NO
MFEM_USE_ALGOIM = NO
MFEM_USE_UMPIRE = NO
MFEM_USE_SIMD = NO
MFEM_USE_ADIOS2 = NO
@@ -166,7 +162,6 @@ MFEM_USE_ADFORWARD = NO
MFEM_USE_CODIPACK = NO
MFEM_USE_BENCHMARK = NO
MFEM_USE_PARELAG = NO
MFEM_USE_ENZYME = NO
# MPI library compile and link flags
# These settings are used only when building MFEM with MPI + HIP
@@ -207,7 +202,7 @@ HYPRE_OPT = -I$(HYPRE_DIR)/include
HYPRE_LIB = -L$(HYPRE_DIR)/lib -lHYPRE
ifeq (YES,$(MFEM_USE_CUDA))
# This is only necessary when hypre is built with cuda:
HYPRE_LIB += -lcusparse -lcurand -lcublas
HYPRE_LIB += -lcusparse -lcurand
endif
ifeq (YES,$(MFEM_USE_HIP))
# This is only necessary when hypre is built with hip:
@@ -474,16 +469,6 @@ CALIPER_DIR = @MFEM_DIR@/../caliper
CALIPER_OPT = -I$(CALIPER_DIR)/include
CALIPER_LIB = $(XLINKER)-rpath,$(CALIPER_DIR)/lib64 -L$(CALIPER_DIR)/lib64 -lcaliper
# BLITZ library configuration
BLITZ_DIR = @MFEM_DIR@/../blitz
BLITZ_OPT = -I$(BLITZ_DIR)/include
BLITZ_LIB = $(XLINKER)-rpath,$(BLITZ_DIR)/lib -L$(BLITZ_DIR)/lib -lblitz
# ALGOIM library configuration
ALGOIM_DIR = @MFEM_DIR@/../algoim
ALGOIM_OPT = -I$(ALGOIM_DIR)/src $(BLITZ_OPT)
ALGOIM_LIB = $(BLITZ_LIB)
# BENCHMARK library configuration
BENCHMARK_DIR = @MFEM_DIR@/../google-benchmark
BENCHMARK_OPT = -I$(BENCHMARK_DIR)/include
@@ -524,22 +509,6 @@ PARELAG_DIR = @MFEM_DIR@/../parelag
PARELAG_OPT = -I$(PARELAG_DIR)/src -I$(PARELAG_DIR)/build/src
PARELAG_LIB = -L$(PARELAG_DIR)/build/src -lParELAG
# Enzyme configuration
# If you want to enable automatic differentiation at compile time, use the
# options below, adapted to your configuration. To be more flexible, we
# recommend using the Enzyme plugin during link time optimization. One option is
# to add your options to the global compiler/linker flags like
#
# BASE_FLAGS += -flto
# CXX_XLINKER += -fuse-ld=lld -Wl,--lto-legacy-pass-manager\
# -Wl,-mllvm=-load=$(ENZYME_DIR)/LLDEnzyme-$(ENZYME_VERSION).so -Wl,
#
ENZYME_DIR ?= @MFEM_DIR@/../enzyme
ENZYME_VERSION ?= 14
ENZYME_OPT = -fno-experimental-new-pass-manager -Xclang -load -Xclang $(ENZYME_DIR)/ClangEnzyme-$(ENZYME_VERSION).so
ENZYME_LIB = ""
# If YES, enable some informational messages
VERBOSE = NO
+22 -19
View File
@@ -1,27 +1,30 @@
FROM ghcr.io/mfem/mfem-ubuntu-base:latest as builder
FROM ghcr.io/rse-ops/cuda-ubuntu-20.04:cuda-11.0.3
# docker build -t ghcr.io/mfem/mfem-ubuntu .
COPY ./config/docker/spack.yaml /opt/mfem-env/spack.yaml
RUN apt-get install -y python3 && \
cd /opt/mfem-env && \
. /opt/spack/share/spack/setup-env.sh && \
spack env activate . && \
spack env view regenerate
FROM ubuntu:22.04
COPY --from=builder /opt/view /opt/view
COPY --from=builder /opt/mfem-view /opt/mfem-view
# docker build -t ghcr.io/mfem/mfem-ubuntu-base .
RUN apt-get update && \
apt-get install -y unzip gfortran && \
spack compiler find && \
apt-get install -y libcurl4-openssl-dev libssl-dev
ENV PATH=$PATH:/opt/mfem-view/bin
ENV LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/opt/mfem-view/lib:/opt/mfem-view/lib64
ENV DEBIAN_FRONTEND=noninteractive
# /code is the working directory for code
WORKDIR /code
COPY . /code
# This is for a spack environment/view to install from there
WORKDIR /opt/mfem-env
RUN . /opt/spack/share/spack/setup-env.sh && \
spack env create -d . && \
echo " concretization: together" >> spack.yaml && \
spack env activate . && \
spack develop --path /code mfem@master+examples+miniapps && \
spack add mfem@master+examples+miniapps && \
spack install
# ensure mfem always on various paths
RUN cd /opt/mfem-env && \
spack env activate --sh -d . >> /etc/profile.d/z10_spack_environment.sh
# The user will see the view on shell into the container
WORKDIR /opt/mfem-view
ENTRYPOINT ["/bin/bash"]
WORKDIR /opt/mfem-env/.spack-env/view/
ENTRYPOINT ["/bin/bash", "--rcfile", "/etc/profile", "-l", "-c"]
-47
View File
@@ -1,47 +0,0 @@
FROM ghcr.io/rse-ops/cuda-ubuntu-20.04:cuda-11.0.3
# docker build -f Dockerfile.base -t ghcr.io/mfem/mfem-ubuntu-base .
RUN apt-get update && \
apt-get install -y unzip gfortran && \
spack compiler find && \
apt-get install -y libcurl4-openssl-dev libssl-dev
# /code is the working directory for code
WORKDIR /code
COPY . /code
# This is for a spack environment/view to install from there
RUN mkdir -p /opt/mfem-env \
&& (echo "spack:" \
&& echo " view:" \
&& echo " mfem:" \
&& echo " root: /opt/mfem-view" \
&& echo " link_type: copy" \
&& echo " packages:" \
&& echo " all:" \
&& echo " target:" \
&& echo " - x86_64_v3" \
&& echo " config:" \
&& echo " concretizer: clingo" \
&& echo " compiler:" \
&& echo " target:" \
&& echo " - x86_64_v3" \
&& echo " install_missing_compilers: true" \
&& echo " concretization: together") > /opt/mfem-env/spack.yaml
RUN cd /opt/mfem-env && \
. /opt/spack/share/spack/setup-env.sh && \
spack env activate . && \
spack develop --path /code mfem@master+examples+miniapps && \
spack add mfem@master+examples+miniapps # && \
# spack install
# ensure mfem always on various paths
#RUN cd /opt/mfem-env && \
# spack env activate --sh -d . >> /etc/profile.d/z10_spack_environment.sh
# Present the software install when we shell in
# The view is at /opt/mfem-env/.spack-env/view
#WORKDIR /opt/software
#ENTRYPOINT ["/bin/bash", "--rcfile", "/etc/profile", "-l", "-c"]
+7 -24
View File
@@ -1,8 +1,7 @@
# mfem Docker
We provide a [Dockerfile.base](Dockerfile.base) to build an ubuntu base image,
and a [Dockerfile](Dockerfile) to build a smaller one with a multi-stage build.
You can use this image for a demo of using mfem! 🎉️
We provide a [Dockerfile](Dockerfile) to build an ubuntu base image. You can use
this image for a demo of using mfem! 🎉️
Updated containers are built and deployed on merges to the main branch and releases.
If you want to request a build on demand, you can [manually run the workflow](https://docs.github.com/en/actions/managing-workflow-runs/manually-running-a-workflow) thanks to the workflow dispatch event.
@@ -15,33 +14,18 @@ is the [GitHub packages](https://github.com/features/packages) registry that sup
Docker images and other OCI artifacts. From the root of the repository:
```bash
$ docker build -f config/docker/Dockerfile -t ghcr.io/mfem/mfem-ubuntu .
$ docker build -f config/docker/Dockerfile.base -t ghcr.io/mfem/mfem-ubuntu-base .
$ docker build -f config/docker/Dockerfile -t ghcr.io/mfem/mfem-ubuntu-base .
```
### Shell Ubuntu
To shell into the container:
or this directory:
```bash
$ docker run -it ghcr.io/mfem/mfem-ubuntu
$ docker build -f Dockerfile -t ghcr.io/mfem/mfem-ubuntu-base ../../
```
This smaller image has a view where everything is installed.
### Shell
```bash
$ ls
bin etc include lib libexec sbin share var
```
- Examples are in share/mfem/examples
- Examples are in share/mfem/miniapps
You can read more about interaction with these examples and miniapps below.
### Shell Ubuntu Base
To shell into the container:
To shell into a container (here is an example with ubuntu):
```bash
$ docker run -it ghcr.io/mfem/mfem-ubuntu-base bash
@@ -144,4 +128,3 @@ $ docker run -it ghcr.io/mfem/mfem-ubuntu-base -v $PWD:/src bash
In the above, we can pretend your project is in the present working directory (PWD) and we are
binding to source. You can then use the mfem in the container for development, and if you
want to distribute your library or app in a container, you can use the mfem container as the base.
-11
View File
@@ -1,11 +0,0 @@
spack:
specs: [mfem@master+examples+miniapps]
view:
mfem:
root: /opt/mfem-view
link_type: copy
concretization: together
develop:
mfem:
path: /code
spec: mfem@master+examples+miniapps
@@ -1,8 +0,0 @@
--- a/CMakeLists.txt Wed Dec 21 18:24:22 2016
+++ b/CMakeLists.txt Wed Dec 21 18:24:26 2016
@@ -20,4 +20,4 @@
# Recursively look for CMakeLists.txt in subdirs.
add_subdirectory("include")
add_subdirectory("libmetis")
-add_subdirectory("programs")
+# add_subdirectory("programs")
@@ -1,15 +0,0 @@
--- a/CMakeLists.txt Sat Mar 30 17:24:45 2013
+++ b/CMakeLists.txt Wed Dec 21 18:23:43 2016
@@ -4,11 +4,7 @@
set(GKLIB_PATH "GKlib" CACHE PATH "path to GKlib")
set(SHARED FALSE CACHE BOOL "build a shared library")
-if(MSVC)
- set(METIS_INSTALL FALSE)
-else()
- set(METIS_INSTALL TRUE)
-endif()
+set(METIS_INSTALL TRUE)
# Configure libmetis library.
if(SHARED)
@@ -1,34 +0,0 @@
diff --git a/include/metis.h b/include/metis.h
index dc5406a..7732437 100644
--- a/include/metis.h
+++ b/include/metis.h
@@ -72,10 +72,14 @@ typedef __int64 int64_t;
#define PRId64 "I64d"
#define SCNd32 "ld"
#define SCNd64 "I64d"
+#ifdef _WIN32
+#include <stdint.h>
+#else
#define INT32_MIN ((int32_t)_I32_MIN)
#define INT32_MAX _I32_MAX
#define INT64_MIN ((int64_t)_I64_MIN)
#define INT64_MAX _I64_MAX
+#endif
#else
#include <inttypes.h>
#endif
diff --git a/GKlib/gk_arch.h b/GKlib/gk_arch.h
index 78b1431..7258763 100644
--- a/GKlib/gk_arch.h
+++ b/GKlib/gk_arch.h
@@ -32,8 +32,8 @@
#ifdef __MSC__
- #include "ms_stdint.h"
- #include "ms_inttypes.h"
+ #include <stdint.h>
+ #include <inttypes.h>
#include "ms_stat.h"
#else
#ifndef SUNOS
@@ -1,11 +0,0 @@
--- a/GKlib/gk_arch.h Wed Dec 21 18:34:18 2016
+++ b/GKlib/gk_arch.h Wed Dec 21 18:30:49 2016
@@ -58,7 +58,7 @@
#define PTRDIFF_MAX INT64_MAX
#endif
-#ifdef __MSC__
+#if defined(__MSC__) && (_MSC_VER < 1900)
/* MSC does not have rint() function */
#define rint(x) ((int)((x)+0.5))
@@ -1,14 +0,0 @@
diff --git a/CMakeLists.txt b/CMakeLists.txt
index e94f050..b9613a7 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -1,7 +1,8 @@
cmake_minimum_required(VERSION 2.8)
project(METIS)
-set(GKLIB_PATH "GKlib" CACHE PATH "path to GKlib")
+set(GKLIB_PATH "${CMAKE_SOURCE_DIR}/GKlib" CACHE PATH "path to GKlib")
+
set(SHARED FALSE CACHE BOOL "build a shared library")
set(METIS_INSTALL TRUE)
@@ -1,11 +0,0 @@
--- a/libmetis/metislib.h Sat Mar 30 17:24:45 2013
+++ b/libmetis/metislib.h Wed Dec 21 18:30:59 2016
@@ -31,7 +31,7 @@
#include <proto.h>
-#if defined(COMPILER_MSC)
+#if defined(COMPILER_MSC) && (_MSC_VER < 1900)
#if defined(rint)
#undef rint
#endif
@@ -1,10 +0,0 @@
--- a/libmetis/CMakeLists.txt Sat Mar 30 17:24:45 2013
+++ b/libmetis/CMakeLists.txt Wed Dec 21 17:41:37 2016
@@ -11,6 +11,6 @@
if(METIS_INSTALL)
install(TARGETS metis
LIBRARY DESTINATION lib
- RUNTIME DESTINATION lib
+ RUNTIME DESTINATION bin
ARCHIVE DESTINATION lib)
endif()
@@ -1,44 +0,0 @@
diff --git a/CMakeLists.txt b/CMakeLists.txt
index b9613a7..e43ffee 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -22,3 +22,23 @@ include_directories(include)
add_subdirectory("include")
add_subdirectory("libmetis")
# add_subdirectory("programs")
+
+if(METIS_INSTALL)
+ set(PRJ_NAME metis)
+ set(PRJ_VER 5.1.0)
+ install(EXPORT metisTargets
+ FILE ${PRJ_NAME}Targets.cmake
+ DESTINATION lib/cmake/${PRJ_NAME})
+ include(CMakePackageConfigHelpers)
+ write_basic_package_version_file(
+ ${CMAKE_CURRENT_BINARY_DIR}/${PRJ_NAME}ConfigVersion.cmake
+ VERSION ${PRJ_VER}
+ COMPATIBILITY SameMajorVersion)
+ file(WRITE ${CMAKE_CURRENT_BINARY_DIR}/${PRJ_NAME}Config.cmake
+ "include(\${CMAKE_CURRENT_LIST_DIR}/${PRJ_NAME}Targets.cmake)")
+ install(FILES
+ ${CMAKE_CURRENT_BINARY_DIR}/${PRJ_NAME}ConfigVersion.cmake
+ ${CMAKE_CURRENT_BINARY_DIR}/${PRJ_NAME}Config.cmake
+ DESTINATION lib/cmake/${PRJ_NAME})
+endif()
+
diff --git a/libmetis/CMakeLists.txt b/libmetis/CMakeLists.txt
index 7a5fc74..5a68cf0 100644
--- a/libmetis/CMakeLists.txt
+++ b/libmetis/CMakeLists.txt
@@ -9,8 +9,9 @@ if(UNIX)
endif()
if(METIS_INSTALL)
- install(TARGETS metis
+ install(TARGETS metis EXPORT metisTargets
LIBRARY DESTINATION lib
RUNTIME DESTINATION bin
- ARCHIVE DESTINATION lib)
+ ARCHIVE DESTINATION lib
+ INCLUDES DESTINATION include)
endif()
@@ -1,41 +0,0 @@
vcpkg_check_linkage(ONLY_STATIC_LIBRARY)
set(OPTIONS -DSHARED=OFF)
set(METIS_VERSION 5.1.0)
vcpkg_download_distfile(ARCHIVE
URLS "https://github.com/mfem/tpls/raw/gh-pages/metis-${METIS_VERSION}.tar.gz"
FILENAME "metis-${METIS_VERSION}.tar.gz"
SHA512 deea47749d13bd06fbeaf98a53c6c0b61603ddc17a43dae81d72c8015576f6495fd83c11b0ef68d024879ed5415c14ebdbd87ce49c181bdac680573bea8bdb25
)
vcpkg_extract_source_archive_ex(
OUT_SOURCE_PATH SOURCE_PATH
ARCHIVE ${ARCHIVE}
REF ${METIS_VERSION}
PATCHES
enable-install.patch
disable-programs.patch
fix-runtime-install-destination.patch
fix-metis-vs14-math.patch
fix-gklib-vs14-math.patch
fix-linux-build-error.patch
install-metisConfig.patch
fix-INT_MIN_define.patch
)
vcpkg_configure_cmake(
SOURCE_PATH ${SOURCE_PATH}
PREFER_NINJA
OPTIONS ${OPTIONS}
)
vcpkg_install_cmake()
vcpkg_copy_pdbs()
vcpkg_fixup_cmake_targets(CONFIG_PATH lib/cmake/metis)
file(REMOVE_RECURSE ${CURRENT_PACKAGES_DIR}/debug/include)
# Handle copyright
file(COPY ${SOURCE_PATH}/LICENSE.txt DESTINATION ${CURRENT_PACKAGES_DIR}/share/metis)
file(INSTALL ${SOURCE_PATH}/LICENSE.txt DESTINATION ${CURRENT_PACKAGES_DIR}/share/${PORT} RENAME copyright)
-7
View File
@@ -1,7 +0,0 @@
{
"name": "metis-mfem",
"version-string": "5.1.0",
"port-version": 0,
"description": "Serial Graph Partitioning and Fill-reducing Matrix Ordering",
"homepage": "https://glaros.dtc.umn.edu/gkhome/metis/metis/overview"
}
-56
View File
@@ -1,56 +0,0 @@
MFEM mesh v1.0
#
# MFEM Geomety Types (see mesh/geom.hpp):
#
# POINT = 0
# SEGMENT = 1
# TRIANGLE = 2
# SQUARE = 3
# TETRAHEDRON = 4
# CUBE = 5
# PRISM = 6
# PYRAMID = 7
dimension
3
elements
4
1 5 0 1 4 3 6 7 10 9
1 6 4 1 5 10 7 11
1 7 11 7 1 5 8
1 4 2 5 1 8
boundary
14
1 3 0 3 4 1
1 3 6 7 10 9
1 3 0 6 9 3
1 3 0 1 7 6
1 3 3 9 10 4
1 2 1 5 2
1 2 1 4 5
1 2 1 8 7
1 2 1 2 8
1 2 2 5 8
1 2 5 11 8
1 2 7 8 11
1 2 7 11 10
1 3 4 10 11 5
vertices
12
3
0 0 0
1 0 0
2 0 0
0 1 0
1 1 0
2 1 0
0 0 1
1 0 1
2 0 1
0 1 1
1 1 1
2 1 1
+1 -1
View File
@@ -2349,7 +2349,7 @@ PLANTUML_INCLUDE_PATH =
# Minimum value: 0, maximum value: 10000, default value: 50.
# This tag requires that the tag HAVE_DOT is set to YES.
DOT_GRAPH_MAX_NODES = 100
DOT_GRAPH_MAX_NODES = 50
# The MAX_DOT_GRAPH_DEPTH tag can be used to set the maximum depth of the graphs
# generated by dot. A depth value of 3 means that only nodes reachable from the
-9
View File
@@ -30,7 +30,6 @@
//
// Device sample runs:
// ex1 -pa -d cuda
// * ex1 -fa -d cuda
// ex1 -pa -d raja-cuda
// * ex1 -pa -d raja-hip
// ex1 -pa -d occa-cuda
@@ -38,13 +37,9 @@
// ex1 -pa -d occa-omp
// ex1 -pa -d ceed-cpu
// ex1 -pa -d ceed-cpu -o 4 -a
// ex1 -pa -d ceed-cpu -m ../data/square-mixed.mesh
// ex1 -pa -d ceed-cpu -m ../data/fichera-mixed.mesh
// * ex1 -pa -d ceed-cuda
// * ex1 -pa -d ceed-hip
// ex1 -pa -d ceed-cuda:/gpu/cuda/shared
// ex1 -pa -d ceed-cuda:/gpu/cuda/shared -m ../data/square-mixed.mesh
// ex1 -pa -d ceed-cuda:/gpu/cuda/shared -m ../data/fichera-mixed.mesh
// ex1 -m ../data/beam-hex.mesh -pa -d cuda
// ex1 -m ../data/beam-tet.mesh -pa -d ceed-cpu
// ex1 -m ../data/beam-tet.mesh -pa -d ceed-cuda:/gpu/cuda/ref
@@ -78,7 +73,6 @@ int main(int argc, char *argv[])
int order = 1;
bool static_cond = false;
bool pa = false;
bool fa = false;
const char *device_config = "cpu";
bool visualization = true;
bool algebraic_ceed = false;
@@ -93,8 +87,6 @@ int main(int argc, char *argv[])
"--no-static-condensation", "Enable static condensation.");
args.AddOption(&pa, "-pa", "--partial-assembly", "-no-pa",
"--no-partial-assembly", "Enable Partial Assembly.");
args.AddOption(&fa, "-fa", "--full-assembly", "-no-fa",
"--no-full-assembly", "Enable Full Assembly.");
args.AddOption(&device_config, "-d", "--device",
"Device configuration string, see Device::Configure().");
#ifdef MFEM_USE_CEED
@@ -192,7 +184,6 @@ int main(int argc, char *argv[])
// domain integrator.
BilinearForm a(&fespace);
if (pa) { a.SetAssemblyLevel(AssemblyLevel::PARTIAL); }
if (fa) { a.SetAssemblyLevel(AssemblyLevel::FULL); }
a.AddDomainIntegrator(new DiffusionIntegrator(one));
// 10. Assemble the bilinear form and the corresponding linear system,
+8
View File
@@ -222,6 +222,10 @@ int main(int argc, char *argv[])
visit_dc.RegisterField("solution", &x);
int vis_cycle = 0;
MFEMDataCollection mfem_dc("Example15-internal", pmesh, true);
mfem_dc.RegisterField("solution", &x);
mfem_dc.ResetMetadata();
// 10. As in Example 6p, we set up an estimator that will be used to obtain
// element error indicators. The integrator needs to provide the method
// ComputeElementFlux. We supply an L2 space for the discontinuous flux
@@ -344,6 +348,10 @@ int main(int argc, char *argv[])
}
if (visit)
{
mfem_dc.SetCycle(vis_cycle);
mfem_dc.SetTime(time);
mfem_dc.Save();
visit_dc.SetCycle(vis_cycle++);
visit_dc.SetTime(time);
visit_dc.Save();
+2 -2
View File
@@ -43,7 +43,7 @@
#include <sstream>
#include <iostream>
// Classes FE_Evolution, RiemannSolver, and FaceIntegrator
// Classes FE_Evolution, RiemannSolver, DomainIntegrator and FaceIntegrator
// shared between the serial and parallel version of the example.
#include "ex18.hpp"
@@ -189,7 +189,7 @@ int main(int argc, char *argv[])
// 7. Set up the nonlinear form corresponding to the DG discretization of the
// flux divergence, and assemble the corresponding mass matrix.
MixedBilinearForm Aflux(&dfes, &fes);
Aflux.AddDomainIntegrator(new TransposeIntegrator(new GradientIntegrator()));
Aflux.AddDomainIntegrator(new DomainIntegrator(dim));
Aflux.Assemble();
NonlinearForm A(&vfes);
+75
View File
@@ -56,6 +56,27 @@ public:
const Vector &nor, Vector &flux);
};
// Constant (in time) mixed bilinear form multiplying the flux grid function.
// The form is (vec(v), grad(w)) where the trial space = vector L2 space (mesh
// dim) and test space = scalar L2 space.
class DomainIntegrator : public BilinearFormIntegrator
{
private:
Vector shape;
DenseMatrix flux;
DenseMatrix dshapedr;
DenseMatrix dshapedx;
public:
DomainIntegrator(const int dim);
virtual void AssembleElementMatrix2(const FiniteElement &trial_fe,
const FiniteElement &test_fe,
ElementTransformation &Tr,
DenseMatrix &elmat);
};
// Interior face term: <F.n(u),[w]>
class FaceIntegrator : public NonlinearFormIntegrator
{
@@ -297,6 +318,60 @@ double RiemannSolver::Eval(const Vector &state1, const Vector &state2,
return maxE;
}
// Implementation of class DomainIntegrator
DomainIntegrator::DomainIntegrator(const int dim) : flux(num_equation, dim) { }
void DomainIntegrator::AssembleElementMatrix2(const FiniteElement &trial_fe,
const FiniteElement &test_fe,
ElementTransformation &Tr,
DenseMatrix &elmat)
{
// Assemble the form (vec(v), grad(w))
// Trial space = vector L2 space (mesh dim)
// Test space = scalar L2 space
const int dof_trial = trial_fe.GetDof();
const int dof_test = test_fe.GetDof();
const int dim = trial_fe.GetDim();
shape.SetSize(dof_trial);
dshapedr.SetSize(dof_test, dim);
dshapedx.SetSize(dof_test, dim);
elmat.SetSize(dof_test, dof_trial * dim);
elmat = 0.0;
const int maxorder = max(trial_fe.GetOrder(), test_fe.GetOrder());
const int intorder = 2 * maxorder;
const IntegrationRule *ir = &IntRules.Get(trial_fe.GetGeomType(), intorder);
for (int i = 0; i < ir->GetNPoints(); i++)
{
const IntegrationPoint &ip = ir->IntPoint(i);
// Calculate the shape functions
trial_fe.CalcShape(ip, shape);
shape *= ip.weight;
// Compute the physical gradients of the test functions
Tr.SetIntPoint(&ip);
test_fe.CalcDShape(ip, dshapedr);
Mult(dshapedr, Tr.AdjugateJacobian(), dshapedx);
for (int d = 0; d < dim; d++)
{
for (int j = 0; j < dof_test; j++)
{
for (int k = 0; k < dof_trial; k++)
{
elmat(j, k + d * dof_trial) += shape(k) * dshapedx(j, d);
}
}
}
}
}
// Implementation of class FaceIntegrator
FaceIntegrator::FaceIntegrator(RiemannSolver &rsolver_, const int dim) :
rsolver(rsolver_),
+2 -2
View File
@@ -43,7 +43,7 @@
#include <sstream>
#include <iostream>
// Classes FE_Evolution, RiemannSolver, and FaceIntegrator
// Classes FE_Evolution, RiemannSolver, DomainIntegrator and FaceIntegrator
// shared between the serial and parallel version of the example.
#include "ex18.hpp"
@@ -219,7 +219,7 @@ int main(int argc, char *argv[])
// 9. Set up the nonlinear form corresponding to the DG discretization of the
// flux divergence, and assemble the corresponding mass matrix.
MixedBilinearForm Aflux(&dfes, &fes);
Aflux.AddDomainIntegrator(new TransposeIntegrator(new GradientIntegrator()));
Aflux.AddDomainIntegrator(new DomainIntegrator(dim));
Aflux.Assemble();
ParNonlinearForm A(&vfes);
-9
View File
@@ -30,18 +30,13 @@
//
// Device sample runs:
// mpirun -np 4 ex1p -pa -d cuda
// * mpirun -np 4 ex1p -fa -d cuda
// mpirun -np 4 ex1p -pa -d occa-cuda
// mpirun -np 4 ex1p -pa -d raja-omp
// mpirun -np 4 ex1p -pa -d ceed-cpu
// mpirun -np 4 ex1p -pa -d ceed-cpu -o 4 -a
// mpirun -np 4 ex1p -pa -d ceed-cpu -m ../data/square-mixed.mesh
// mpirun -np 4 ex1p -pa -d ceed-cpu -m ../data/fichera-mixed.mesh
// * mpirun -np 4 ex1p -pa -d ceed-cuda
// * mpirun -np 4 ex1p -pa -d ceed-hip
// mpirun -np 4 ex1p -pa -d ceed-cuda:/gpu/cuda/shared
// mpirun -np 4 ex1p -pa -d ceed-cuda:/gpu/cuda/shared -m ../data/square-mixed.mesh
// mpirun -np 4 ex1p -pa -d ceed-cuda:/gpu/cuda/shared -m ../data/fichera-mixed.mesh
// mpirun -np 4 ex1p -m ../data/beam-tet.mesh -pa -d ceed-cpu
//
// Description: This example code demonstrates the use of MFEM to define a
@@ -79,7 +74,6 @@ int main(int argc, char *argv[])
int order = 1;
bool static_cond = false;
bool pa = false;
bool fa = false;
const char *device_config = "cpu";
bool visualization = true;
bool algebraic_ceed = false;
@@ -94,8 +88,6 @@ int main(int argc, char *argv[])
"--no-static-condensation", "Enable static condensation.");
args.AddOption(&pa, "-pa", "--partial-assembly", "-no-pa",
"--no-partial-assembly", "Enable Partial Assembly.");
args.AddOption(&fa, "-fa", "--full-assembly", "-no-fa",
"--no-full-assembly", "Enable Full Assembly.");
args.AddOption(&device_config, "-d", "--device",
"Device configuration string, see Device::Configure().");
#ifdef MFEM_USE_CEED
@@ -219,7 +211,6 @@ int main(int argc, char *argv[])
// Diffusion domain integrator.
ParBilinearForm a(&fespace);
if (pa) { a.SetAssemblyLevel(AssemblyLevel::PARTIAL); }
if (fa) { a.SetAssemblyLevel(AssemblyLevel::FULL); }
a.AddDomainIntegrator(new DiffusionIntegrator(one));
// 12. Assemble the parallel bilinear form and the corresponding linear
+1 -1
View File
@@ -182,7 +182,7 @@ int main(int argc, char *argv[])
}
for (int level = 0; level < order_refinements; ++level)
{
collections.Append(new H1_FECollection((int)std::pow(2, level+1), dim));
collections.Append(new H1_FECollection(std::pow(2, level+1), dim));
fespaces.AddOrderRefinedLevel(collections.Last());
}
+1 -1
View File
@@ -219,7 +219,7 @@ int main(int argc, char *argv[])
}
for (int level = 0; level < order_refinements; ++level)
{
collections.Append(new H1_FECollection((int)std::pow(2, level+1), dim));
collections.Append(new H1_FECollection(std::pow(2, level+1), dim));
fespaces->AddOrderRefinedLevel(collections.Last());
}
+63 -275
View File
@@ -3,63 +3,34 @@
// Compile with: make ex33
//
// Sample runs: ex33 -m ../data/square-disc.mesh -alpha 0.33 -o 2
// ex33 -m ../data/square-disc.mesh -alpha 4.5 -o 3
// ex33 -m ../data/star.mesh -alpha 1.4 -o 3
// ex33 -m ../data/star.mesh -alpha 0.99 -o 3
// ex33 -m ../data/inline-quad.mesh -alpha 0.5 -o 3
// ex33 -m ../data/amr-quad.mesh -alpha 1.5 -o 3
// ex33 -m ../data/disc-nurbs.mesh -alpha 0.33 -o 3
// ex33 -m ../data/disc-nurbs.mesh -alpha 2.4 -o 3 -r 4
// ex33 -m ../data/l-shape.mesh -alpha 0.33 -o 3 -r 4
// ex33 -m ../data/l-shape.mesh -alpha 1.7 -o 3 -r 5
//
// Verification runs:
// ex33 -m ../data/inline-segment.mesh -ver -alpha 1.7 -o 2 -r 2
// ex33 -m ../data/inline-quad.mesh -ver -alpha 1.2 -o 2 -r 2
// ex33 -m ../data/amr-quad.mesh -ver -alpha 2.6 -o 2 -r 2
// ex33 -m ../data/inline-hex.mesh -ver -alpha 0.3 -o 2 -r 1
//
// Note: the analytic solution to this problem is u = ∏_{i=0}^{dim-1} sin(π x_i)
// for all alpha.
//
// Description:
//
// In this example we solve the following fractional PDE with MFEM:
//
// ( - Δ )^α u = f in Ω, u = 0 on ∂Ω, 0 < α,
// ( - Δ )^α u = f in Ω, u = 0 on ∂Ω, 0 < α < 1,
//
// To solve this FPDE, we apply the operator ( - Δ )^(-N), where the integer
// N is given by floor(α). By doing so, we obtain
// To solve this FPDE, we rely on a rational approximation [2] of the normal
// linear operator A^{-α}, where A = - Δ (with associated homogeneous
// boundary conditions). Namely, we first approximate the operator
//
// ( - Δ )^(α-N) u = ( - Δ )^(-N) f in Ω, u = 0 on ∂Ω, 0 < α.
//
// We first compute the right hand side by solving the integer order PDE
//
// ( - Δ )^N g = f in Ω, g = ( - Δ )^k g = 0 on ∂Ω, k = 1,..,N-1
//
// The remaining FPDE is then given by
//
// ( - Δ )^(α-N) u = g in Ω, u = 0 on ∂Ω.
//
// We rely on a rational approximation [2] of the normal linear operator
// A^{-α + N}, where A = - Δ (with associated homogeneous boundary conditions)
// and (a-N) in (0,1). We approximate the operator
//
// A^{-α+N} ≈ Σ_{i=0}^M c_i (A + d_i I)^{-1}, d_0 = 0, d_i > 0,
// A^{-α} ≈ Σ_{i=0}^N c_i (A + d_i I)^{-1}, d_0 = 0, d_i > 0,
//
// where I is the L2-identity operator and the coefficients c_i and d_i
// are generated offline to a prescribed accuracy in a pre-processing step.
// We use the triple-A algorithm [1] to generate the rational approximation
// that this partial fractional expansion derives from. We then solve M+1
// that this partial fractional expansion derives from. We then solve N+1
// independent integer-order PDEs,
//
// A u_i + d_i u_i = c_i g in Ω, u_i = 0 on ∂Ω, i=0,...,M,
// A u_i + d_i u_i = c_i f in Ω, u_i = 0 on ∂Ω, i=0,...,N,
//
// using MFEM and sum u_i to arrive at an approximate solution of the FPDE
//
// u ≈ Σ_{i=0}^M u_i.
//
// (If alpha is an integer, we stop after the first PDE was solved.)
// u ≈ Σ_{i=0}^N u_i.
//
// References:
//
@@ -76,8 +47,6 @@
#include "mfem.hpp"
#include <fstream>
#include <iostream>
#include <math.h>
#include <string>
#include "ex33.hpp"
@@ -90,9 +59,8 @@ int main(int argc, char *argv[])
const char *mesh_file = "../data/star.mesh";
int order = 1;
int num_refs = 3;
double alpha = 0.5;
bool visualization = true;
bool verification = false;
double alpha = 0.5;
OptionsParser args(argc, argv);
args.AddOption(&mesh_file, "-m", "--mesh",
@@ -107,9 +75,6 @@ int main(int argc, char *argv[])
args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
"--no-visualization",
"Enable or disable GLVis visualization.");
args.AddOption(&verification, "-ver", "--verification", "-no-ver",
"--no-verification",
"Use sinusoidal function (f) for analytic comparison.");
args.Parse();
if (!args.Good())
{
@@ -119,31 +84,9 @@ int main(int argc, char *argv[])
args.PrintOptions(cout);
Array<double> coeffs, poles;
int progress_steps = 1;
// 2. Compute the rational expansion coefficients that define the
// integer-order PDEs.
const int power_of_laplace = floor(alpha);
double exponent_to_approximate = alpha - power_of_laplace;
bool integer_order = false;
// Check if alpha is an integer or not.
if (abs(exponent_to_approximate) > 1e-12)
{
mfem::out << "Approximating the fractional exponent "
<< exponent_to_approximate
<< endl;
ComputePartialFractionApproximation(exponent_to_approximate, coeffs,
poles);
// If the example is build without LAPACK, the exponent_to_approximate
// might be modified by the function call above.
alpha = exponent_to_approximate + power_of_laplace;
}
else
{
integer_order = true;
mfem::out << "Treating integer order PDE." << endl;
}
// 2. Compute the coefficients that define the integer-order PDEs.
ComputePartialFractionApproximation(alpha,coeffs,poles);
// 3. Read the mesh from the given mesh file.
Mesh mesh(mesh_file, 1, 1);
@@ -156,8 +99,8 @@ int main(int argc, char *argv[])
}
// 5. Define a finite element space on the mesh.
H1_FECollection fec(order, dim);
FiniteElementSpace fespace(&mesh, &fec);
FiniteElementCollection *fec = new H1_FECollection(order, dim);
FiniteElementSpace fespace(&mesh, fec);
cout << "Number of finite element unknowns: "
<< fespace.GetTrueVSize() << endl;
@@ -171,234 +114,79 @@ int main(int argc, char *argv[])
}
// 7. Define diffusion coefficient, load, and solution GridFunction.
auto func = [&alpha](const Vector &x)
{
double val = 1.0;
for (int i=0; i<x.Size(); i++)
{
val *= sin(M_PI*x(i));
}
return pow(x.Size()*pow(M_PI,2), alpha) * val;
};
FunctionCoefficient f(func);
ConstantCoefficient f(1.0);
ConstantCoefficient one(1.0);
GridFunction u(&fespace);
GridFunction x(&fespace);
GridFunction g(&fespace);
u = 0.0;
x = 0.0;
g = 0.0;
u = 0.;
// 8. Prepare for visualization.
char vishost[] = "localhost";
int visport = 19916;
// 9. Set up the linear form b(.) for integer-order PDE solves.
LinearForm b(&fespace);
if (verification)
socketstream xout, uout;
ostringstream oss_x, oss_u;
if (visualization)
{
// This statement is only relevant for the verification of the code. It
// uses a different f such that an analytic solution is known and easy
// to compare with the numerical one. The FPDE becomes:
// (-Δ)^α u = (2\pi ^2)^α sin(\pi x) sin(\pi y) on [0,1]^2
// -> u(x,y) = sin(\pi x) sin(\pi y)
b.AddDomainIntegrator(new DomainLFIntegrator(f));
xout.open(vishost, visport);
xout.precision(8);
uout.open(vishost, visport);
uout.precision(8);
}
else
for (int i = 0; i < coeffs.Size(); i++)
{
b.AddDomainIntegrator(new DomainLFIntegrator(one));
}
b.Assemble();
// 9. Set up the linear form b(.) for integer-order PDE solve.
LinearForm b(&fespace);
ProductCoefficient cf(coeffs[i], f);
b.AddDomainIntegrator(new DomainLFIntegrator(cf));
b.Assemble();
// ------------------------------------------------------------------------
// 10. Solve the PDE (-Δ)^N g = f, i.e. compute g = (-Δ)^{-1}^N f.
// ------------------------------------------------------------------------
// 10. Define GridFunction for integer-order PDE solve.
GridFunction x(&fespace);
x = 0.0;
if (power_of_laplace > 0)
{
// 10.1 Compute Stiffnes Matrix
BilinearForm k(&fespace);
k.AddDomainIntegrator(new DiffusionIntegrator(one));
k.Assemble();
// 11. Set up the bilinear form a(.,.) for integer-order PDE solve.
BilinearForm a(&fespace);
a.AddDomainIntegrator(new DiffusionIntegrator(one));
ConstantCoefficient c2(-poles[i]);
a.AddDomainIntegrator(new MassIntegrator(c2));
a.Assemble();
// 10.2 Compute Mass Matrix
BilinearForm m(&fespace);
m.AddDomainIntegrator(new MassIntegrator(one));
m.Assemble();
SparseMatrix mass;
Array<int> empty;
m.FormSystemMatrix(empty, mass);
// 10.3 Form the system of equations
// 12. Assemble the bilinear form and the corresponding linear system.
OperatorPtr A;
Vector B, X;
OperatorPtr Op;
k.FormLinearSystem(ess_tdof_list, g, b, Op, X, B);
GSSmoother M((SparseMatrix&)(*Op));
a.FormLinearSystem(ess_tdof_list, x, b, A, X, B);
mfem::out << "\nComputing (-Δ) ^ -" << power_of_laplace
<< " ( f ) " << endl;
for (int i = 0; i < power_of_laplace; i++)
{
// 10.4 Solve the linear system Op X = B (N times).
PCG(*Op, M, B, X, 3, 300, 1e-12, 0.0);
// 13. Solve the linear system A X = B.
GSSmoother M((SparseMatrix&)(*A));
// 10.5 Visualize the solution g of -Δ ^ N g = f in the last step
if (i == power_of_laplace - 1)
{
// Needed for visualization and solution verification.
k.RecoverFEMSolution(X, b, g);
if (integer_order && verification)
{
// For an integer order PDE, g is also our solution u.
u+=g;
}
if (visualization)
{
socketstream fout;
ostringstream oss_f;
fout.open(vishost, visport);
fout.precision(8);
oss_f.str(""); oss_f.clear();
oss_f << "Step " << progress_steps++ << ": Solution of PDE -Δ ^ "
<< power_of_laplace
<< " g = f";
fout << "solution\n" << mesh << g
<< "window_title '" << oss_f.str() << "'" << flush;
}
}
mfem::out << "\nSolving PDE -Δ u + " << -poles[i]
<< " u = " << coeffs[i] << " f " << endl;
PCG(*A, M, B, X, 3, 200, 1e-12, 0.0);
// 10.6 Prepare for next iteration (primal / dual space)
mass.Mult(X, B);
X.SetSubVectorComplement(ess_tdof_list,0.0);
}
// 14. Recover the solution as a finite element grid function.
a.RecoverFEMSolution(X, b, x);
// 10.7 Extract solution for the next step. The b now corresponds to the
// function g in the PDE.
const SparseMatrix * R = fespace.GetRestrictionMatrix();
if (R)
{
R->MultTranspose(B,b);
}
else
{
b = B;
}
}
// 15. Accumulate integer-order PDE solutions.
u+=x;
// ------------------------------------------------------------------------
// 11. Solve the fractional PDE by solving M integer order PDEs and adding
// up the solutions.
// ------------------------------------------------------------------------
if (!integer_order)
{
// Setup visualization.
socketstream xout, uout;
ostringstream oss_x, oss_u;
// 16. Send the solutions by socket to a GLVis server.
if (visualization)
{
xout.open(vishost, visport);
xout.precision(8);
uout.open(vishost, visport);
uout.precision(8);
}
// Iterate over all expansion coefficient that contribute to the
// solution.
for (int i = 0; i < coeffs.Size(); i++)
{
mfem::out << "\nSolving PDE -Δ u + " << -poles[i]
<< " u = " << coeffs[i] << " g " << endl;
oss_x.str(""); oss_x.clear();
oss_x << "Solution of PDE -Δ u + " << -poles[i]
<< " u = " << coeffs[i] << " f";
xout << "solution\n" << mesh << x
<< "window_title '" << oss_x.str() << "'" << flush;
// 11.1 Reset GridFunction for integer-order PDE solve.
x = 0.0;
// 11.2 Set up the bilinear form a(.,.) for integer-order PDE solve.
BilinearForm a(&fespace);
a.AddDomainIntegrator(new DiffusionIntegrator(one));
ConstantCoefficient d_i(-poles[i]);
a.AddDomainIntegrator(new MassIntegrator(d_i));
a.Assemble();
// 11.3 Assemble the bilinear form and the corresponding linear system.
OperatorPtr A;
Vector B, X;
a.FormLinearSystem(ess_tdof_list, x, b, A, X, B);
// 11.4 Solve the linear system A X = B.
GSSmoother M((SparseMatrix&)(*A));
PCG(*A, M, B, X, 3, 300, 1e-12, 0.0);
// 11.5 Recover the solution as a finite element grid function.
a.RecoverFEMSolution(X, b, x);
// 11.6 Accumulate integer-order PDE solutions.
x *= coeffs[i];
u += x;
// 11.7 Send fractional PDE solution to a GLVis server.
if (visualization)
{
oss_x.str(""); oss_x.clear();
oss_x << "Step " << progress_steps
<< ": Solution of PDE -Δ u + " << -poles[i]
<< " u = " << coeffs[i] << " g";
xout << "solution\n" << mesh << x
<< "window_title '" << oss_x.str() << "'" << flush;
oss_u.str(""); oss_u.clear();
oss_u << "Step " << progress_steps + 1
<< ": Solution of fractional PDE (-Δ)^" << alpha
<< " u = f";
uout << "solution\n" << mesh << u
<< "window_title '" << oss_u.str() << "'"
<< flush;
}
oss_u.str(""); oss_u.clear();
oss_u << "Solution of fractional PDE -Δ^" << alpha
<< " u = f";
uout << "solution\n" << mesh << u
<< "window_title '" << oss_u.str() << "'" << flush;
}
}
// ------------------------------------------------------------------------
// 12. (optional) Verify the solution.
// ------------------------------------------------------------------------
if (verification)
{
auto solution = [] (const Vector &x)
{
double val = 1.0;
for (int i=0; i<x.Size(); i++)
{
val *= sin(M_PI*x(i));
}
return val;
};
FunctionCoefficient sol(solution);
double l2_error = u.ComputeL2Error(sol);
string analytic_solution,expected_mesh;
switch (dim)
{
case 1:
analytic_solution = "sin(π x)";
expected_mesh = "inline_segment.mesh";
break;
case 2:
analytic_solution = "sin(π x) sin(π y)";
expected_mesh = "inline_quad.mesh";
break;
default:
analytic_solution = "sin(π x) sin(π y) sin(π z)";
expected_mesh = "inline_hex.mesh";
break;
}
mfem::out << "\n" << string(80,'=')
<< "\n\nSolution Verification in "<< dim << "D \n\n"
<< "Analytic solution : " << analytic_solution << "\n"
<< "Expected mesh : " << expected_mesh <<"\n"
<< "Your mesh : " << mesh_file << "\n"
<< "L2 error : " << l2_error << "\n\n"
<< string(80,'=') << endl;
}
// 17. Free the used memory.
delete fec;
return 0;
}
+4 -15
View File
@@ -32,7 +32,6 @@
#include "mfem.hpp"
#include <fstream>
#include <iostream>
#include <string>
using namespace std;
using namespace mfem;
@@ -250,13 +249,6 @@ void PartialFractionExpansion(double scale, Array<double> & poles,
coeffs.SetSize(psize);
coeffs = scale;
// Note: C p(z)/q(z) = Σ_i c_i / (z - p_i) results in an system of equations
// where the N unknowns are the coefficients c_i. After multiplying the
// system with q(z), the coefficients c_i can be computed analytically by
// choosing N values for z. Choosing z_j = = p_j diagonalizes the system and
// one can obtain an analytic form for the c_i coefficients. The result is
// implemented in the code block below.
for (int i=0; i<psize; i++)
{
double tmp_numer=1.0;
@@ -313,12 +305,9 @@ void ComputePartialFractionApproximation(double & alpha,
if (print_warning)
{
mfem::out
<< "\n" << string(80, '=')
<< "\nMFEM is compiled without LAPACK."
<< "\nUsing precomputed values for PartialFractionApproximation."
<< "\nOnly alpha = 0.33, 0.5, and 0.99 are available."
<< "\nThe default is alpha = 0.5.\n" << string(80, '=') << "\n"
<< endl;
<< "\nMFEM is compiled without LAPACK.\nUsing precomputed values for PartialFractionApproximation. \n"
<< "Only alpha = 0.33, 0.5, and 0.99 are available.\nThe default is alpha = 0.5."
<< std::endl;
}
const double eps = std::numeric_limits<double>::epsilon();
@@ -362,7 +351,7 @@ void ComputePartialFractionApproximation(double & alpha,
if (print_warning)
{
mfem::out << "=> Using precomputed values for alpha = "
mfem::out << "Using precomputed values for alpha = "
<< alpha << "\n" << std::endl;
}
+142 -293
View File
@@ -3,63 +3,34 @@
// Compile with: make ex33p
//
// Sample runs: mpirun -np 4 ex33p -m ../data/square-disc.mesh -alpha 0.33 -o 2
// mpirun -np 4 ex33p -m ../data/square-disc.mesh -alpha 4.5 -o 3
// mpirun -np 4 ex33p -m ../data/star.mesh -alpha 1.4 -o 3
// mpirun -np 4 ex33p -m ../data/star.mesh -alpha 0.99 -o 3
// mpirun -np 4 ex33p -m ../data/inline-quad.mesh -alpha 0.5 -o 3
// mpirun -np 4 ex33p -m ../data/amr-quad.mesh -alpha 1.5 -o 3
// mpirun -np 4 ex33p -m ../data/disc-nurbs.mesh -alpha 0.33 -o 3 -r 2
// mpirun -np 4 ex33p -m ../data/disc-nurbs.mesh -alpha 2.4 -o 3 -r 4
// mpirun -np 4 ex33p -m ../data/disc-nurbs.mesh -alpha 0.33 -o 3
// mpirun -np 4 ex33p -m ../data/l-shape.mesh -alpha 0.33 -o 3 -r 4
// mpirun -np 4 ex33p -m ../data/l-shape.mesh -alpha 1.7 -o 3 -r 5
//
// Verification runs:
// mpirun -np 4 ex33p -m ../data/inline-segment.mesh -ver -alpha 1.7 -o 2 -r 2
// mpirun -np 4 ex33p -m ../data/inline-quad.mesh -ver -alpha 1.2 -o 2 -r 2
// mpirun -np 4 ex33p -m ../data/amr-quad.mesh -ver -alpha 2.6 -o 2 -r 2
// mpirun -np 4 ex33p -m ../data/inline-hex.mesh -ver -alpha 0.3 -o 2 -r 1
// Note: the analytic solution to this problem is u = ∏_{i=0}^{dim-1} sin(π x_i)
// for all alpha.
//
// Description:
//
// In this example we solve the following fractional PDE with MFEM:
//
// ( - Δ )^α u = f in Ω, u = 0 on ∂Ω, 0 < α,
// ( - Δ )^α u = f in Ω, u = 0 on ∂Ω, 0 < α < 1,
//
// To solve this FPDE, we apply the operator ( - Δ )^(-N), where the integer
// N is given by floor(α). By doing so, we obtain
// To solve this FPDE, we rely on a rational approximation [2] of the normal
// linear operator A^{-α}, where A = - Δ (with associated homogeneous
// boundary conditions). Namely, we first approximate the operator
//
// ( - Δ )^(α-N) u = ( - Δ )^(-N) f in Ω, u = 0 on ∂Ω, 0 < α.
//
// We first compute the right hand side by solving the integer order PDE
//
// ( - Δ )^N g = f in Ω, g = ( - Δ )^k g = 0 on ∂Ω, k = 1,..,N-1
//
// The remaining FPDE is then given by
//
// ( - Δ )^(α-N) u = g in Ω, u = 0 on ∂Ω.
//
// We rely on a rational approximation [2] of the normal linear operator
// A^{-α + N}, where A = - Δ (with associated homogeneous boundary conditions)
// and (a-N) in (0,1). We approximate the operator
//
// A^{-α+N} ≈ Σ_{i=0}^M c_i (A + d_i I)^{-1}, d_0 = 0, d_i > 0,
// A^{-α} ≈ Σ_{i=0}^N c_i (A + d_i I)^{-1}, d_0 = 0, d_i > 0,
//
// where I is the L2-identity operator and the coefficients c_i and d_i
// are generated offline to a prescribed accuracy in a pre-processing step.
// We use the triple-A algorithm [1] to generate the rational approximation
// that this partial fractional expansion derives from. We then solve M+1
// that this partial fractional expansion derives from. We then solve N+1
// independent integer-order PDEs,
//
// A u_i + d_i u_i = c_i g in Ω, u_i = 0 on ∂Ω, i=0,...,M,
// A u_i + d_i u_i = c_i f in Ω, u_i = 0 on ∂Ω, i=0,...,N,
//
// using MFEM and sum u_i to arrive at an approximate solution of the FPDE
//
// u ≈ Σ_{i=0}^M u_i.
//
// (If alpha is an integer, we stop after the first PDE was solved.)
// u ≈ Σ_{i=0}^N u_i.
//
// References:
//
@@ -76,8 +47,6 @@
#include "mfem.hpp"
#include <fstream>
#include <iostream>
#include <math.h>
#include <string>
#include "ex33.hpp"
@@ -96,9 +65,9 @@ int main(int argc, char *argv[])
const char *mesh_file = "../data/star.mesh";
int order = 1;
int num_refs = 3;
double alpha = 0.5;
bool visualization = true;
bool verification = false;
bool visualize_x = false;
double alpha = 0.5;
OptionsParser args(argc, argv);
args.AddOption(&mesh_file, "-m", "--mesh",
@@ -110,12 +79,12 @@ int main(int argc, char *argv[])
"Number of uniform refinements");
args.AddOption(&alpha, "-alpha", "--alpha",
"Fractional exponent");
args.AddOption(&visualize_x, "-vis_x", "--visualize_x", "-no-vis_x",
"--no-visualization_x",
"Enable or disable GLVis visualization of each integer-order PDE solution.");
args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
"--no-visualization",
"Enable or disable GLVis visualization.");
args.AddOption(&verification, "-ver", "--verification", "-no-ver",
"--no-verification",
"Use sinusoidal function (f) for analytic comparison.");
"Enable or disable GLVis visualization of the fractional PDE solution.");
args.Parse();
if (!args.Good())
{
@@ -128,51 +97,61 @@ int main(int argc, char *argv[])
}
Array<double> coeffs, poles;
int progress_steps = 1;
// 2. Compute the rational expansion coefficients that define the
// integer-order PDEs.
const int power_of_laplace = floor(alpha);
double exponent_to_approximate = alpha - power_of_laplace;
bool integer_order = false;
// Check if alpha is an integer or not.
if (abs(exponent_to_approximate) > 1e-12)
{
if (Mpi::Root())
{
mfem::out << "Approximating the fractional exponent "
<< exponent_to_approximate
<< endl;
}
ComputePartialFractionApproximation(exponent_to_approximate, coeffs,
poles);
// 2. Compute the coefficients that define the integer-order PDEs.
ComputePartialFractionApproximation(alpha,coeffs,poles);
// If the example is build without LAPACK, the exponent_to_approximate
// might be modified by the function call above.
alpha = exponent_to_approximate + power_of_laplace;
}
else
int num_par_solves;
int max_par_solves = max(1,num_procs/2);
for (num_par_solves=max_par_solves; num_par_solves>0; num_par_solves--)
{
integer_order = true;
if (Mpi::Root())
if (num_procs%num_par_solves==0 && num_par_solves<coeffs.Size())
{
mfem::out << "Treating integer order PDE." << endl;
break;
}
}
if (num_par_solves == 1) {num_par_solves = num_procs;}
// 3. Read the mesh from the given mesh file.
int solver_ranks = num_procs/num_par_solves;
// 3. Split the MPI communicator:
// row_comm is used for parallel partition of the mesh
// col_comm is used for independent integer-order solves
int row_color = myid / solver_ranks; // Determine color based on row
int col_color = myid % solver_ranks; // Determine color based on col
MPI_Comm row_comm, col_comm;
MPI_Comm_split(MPI_COMM_WORLD, row_color, myid, &row_comm);
MPI_Comm_split(MPI_COMM_WORLD, col_color, myid, &col_comm);
int row_rank, row_size, col_rank, col_size;
MPI_Comm_rank(row_comm, &row_rank);
MPI_Comm_size(row_comm, &row_size);
MPI_Comm_rank(col_comm, &col_rank);
MPI_Comm_size(col_comm, &col_size);
if (Mpi::Root())
{
mfem::out << "\nTotal number of MPI ranks = " << num_procs << endl;
mfem::out << "Number of independent parallel solves = " << col_size << endl;
mfem::out << "Number of MPI ranks within each solve = " << row_size
<<"\n" << endl;
}
// 4. Read the mesh from the given mesh file.
Mesh mesh(mesh_file, 1, 1);
int dim = mesh.Dimension();
// 4. Refine the mesh to increase the resolution.
// 5. Refine the mesh to increase the resolution.
for (int i = 0; i < num_refs; i++)
{
mesh.UniformRefinement();
}
ParMesh pmesh(MPI_COMM_WORLD, mesh);
ParMesh pmesh(row_comm, mesh);
mesh.Clear();
// 5. Define a finite element space on the mesh.
// 6. Define a finite element space on the mesh.
H1_FECollection fec(order, dim);
ParFiniteElementSpace fespace(&pmesh, &fec);
if (Mpi::Root())
@@ -181,7 +160,7 @@ int main(int argc, char *argv[])
<< fespace.GetTrueVSize() << endl;
}
// 6. Determine the list of true (i.e. conforming) essential boundary dofs.
// 7. Determine the list of true (i.e. conforming) essential boundary dofs.
Array<int> ess_tdof_list;
if (pmesh.bdr_attributes.Size())
{
@@ -190,250 +169,120 @@ int main(int argc, char *argv[])
fespace.GetEssentialTrueDofs(ess_bdr, ess_tdof_list);
}
// 7. Define diffusion coefficient, load, and solution GridFunction.
auto func = [&alpha](const Vector &x)
{
double val = 1.0;
for (int i=0; i<x.Size(); i++)
{
val *= sin(M_PI*x(i));
}
return pow(x.Size()*pow(M_PI,2), alpha) * val;
};
FunctionCoefficient f(func);
// 8. Define diffusion coefficient, load, and solution GridFunction.
ConstantCoefficient f(1.0);
ConstantCoefficient one(1.0);
ParGridFunction u(&fespace);
ParGridFunction x(&fespace);
ParGridFunction g(&fespace);
u = 0.0;
x = 0.0;
g = 0.0;
// 8. Prepare for visualization.
char vishost[] = "localhost";
int visport = 19916;
// 9. Set up the linear form b(.) for integer-order PDE solves.
ParLinearForm b(&fespace);
if (verification)
{
// This statement is only relevant for the verification of the code. It
// uses a different f such that an analytic solution is known and easy
// to compare with the numerical one. The FPDE becomes:
// (-Δ)^α u = (2\pi ^2)^α sin(\pi x) sin(\pi y) on [0,1]^2
// -> u(x,y) = sin(\pi x) sin(\pi y)
b.AddDomainIntegrator(new DomainLFIntegrator(f));
}
else
{
b.AddDomainIntegrator(new DomainLFIntegrator(one));
}
b.AddDomainIntegrator(new DomainLFIntegrator(f));
b.Assemble();
// ------------------------------------------------------------------------
// 10. Solve the PDE (-Δ)^N g = f, i.e. compute g = (-Δ)^{-1}^N f.
// ------------------------------------------------------------------------
if (power_of_laplace > 0)
int my_coeff_size = max(coeffs.Size()/col_size,1);
int ibeg = col_rank*my_coeff_size;
if (ibeg + 2*my_coeff_size > coeffs.Size())
{
// 10.1 Compute Stiffnes Matrix
ParBilinearForm k(&fespace);
k.AddDomainIntegrator(new DiffusionIntegrator(one));
k.Assemble();
my_coeff_size = coeffs.Size()-col_rank*my_coeff_size;
}
else if (ibeg > coeffs.Size() - 1)
{
my_coeff_size = 0;
}
// 10.2 Compute Mass Matrix
ParBilinearForm m(&fespace);
m.AddDomainIntegrator(new MassIntegrator(one));
m.Assemble();
HypreParMatrix mass;
Array<int> empty;
m.FormSystemMatrix(empty, mass);
int iend = ibeg+my_coeff_size;
// 10.3 Form the system of equations
for (int i = ibeg; i < iend; i++)
{
// 10. Reset GridFunction for integer-order PDE solve.
x = 0.0;
// 11. Set up the bilinear form a(.,.) for integer-order PDE solve.
ParBilinearForm a(&fespace);
a.AddDomainIntegrator(new DiffusionIntegrator(one));
ConstantCoefficient d_i(-poles[i]);
a.AddDomainIntegrator(new MassIntegrator(d_i));
a.Assemble();
// 12. Assemble the bilinear form and the corresponding linear system.
OperatorPtr A;
Vector B, X;
OperatorPtr Op;
k.FormLinearSystem(ess_tdof_list, g, b, Op, X, B);
HypreBoomerAMG prec;
prec.SetPrintLevel(-1);
CGSolver cg(MPI_COMM_WORLD);
a.FormLinearSystem(ess_tdof_list, x, b, A, X, B);
// 13. Solve the linear system A X = B.
HypreBoomerAMG * prec = new HypreBoomerAMG;
prec->SetPrintLevel(-1);
int print_level = (col_rank==0) ? 3 : 0;
if (Mpi::Root())
{
mfem::out << "\nMPI rank " << myid
<< ": Solving PDE -Δ u + " << -poles[i]
<< " u = " << coeffs[i] << " f " << endl;
}
CGSolver cg(row_comm);
cg.SetRelTol(1e-12);
cg.SetMaxIter(2000);
cg.SetPrintLevel(3);
cg.SetPreconditioner(prec);
cg.SetOperator(*Op);
cg.SetPrintLevel(print_level);
cg.SetPreconditioner(*prec);
cg.SetOperator(*A);
cg.Mult(B, X);
delete prec;
if (Mpi::Root())
// 14. Recover the solution as a finite element grid function.
a.RecoverFEMSolution(X, b, x);
// 15. Accumulate integer-order PDE solutions.
x *= coeffs[i];
u += x;
// 16. Send integer-order PDE solutions to a GLVis server.
if (visualize_x)
{
mfem::out << "\nComputing (-Δ) ^ -" << power_of_laplace
<< " ( f ) " << endl;
}
for (int i = 0; i < power_of_laplace; i++)
{
// 10.4 Solve the linear system Op X = B (N times).
cg.Mult(B, X);
// 10.5 Visualize the solution g of -Δ ^ N g = f in the last step
if (i == power_of_laplace - 1)
if (col_rank > 0 && i < iend-1)
{
// Needed for visualization and solution verification.
k.RecoverFEMSolution(X, b, g);
if (integer_order && verification)
{
// For an integer order PDE, g is also our solution u.
u+=g;
}
if (visualization)
{
socketstream fout;
ostringstream oss_f;
fout.open(vishost, visport);
fout.precision(8);
oss_f.str(""); oss_f.clear();
oss_f << "Step " << progress_steps++ << ": Solution of PDE -Δ ^ "
<< power_of_laplace
<< " g = f";
fout << "parallel " << num_procs << " " << myid << "\n"
<< "solution\n" << pmesh << g
<< "window_title '" << oss_f.str() << "'" << flush;
}
MPI_Status status;
MPI_Recv(nullptr,0,MPI_INT, col_rank-1,0,col_comm,&status);
}
// 10.6 Prepare for next iteration (primal / dual space)
mass.Mult(X, B);
X.SetSubVectorComplement(ess_tdof_list,0.0);
}
// 10.7 Extract solution for the next step. The b now corresponds to the
// function g in the PDE.
const SparseMatrix* rm = fespace.GetRestrictionMatrix();
rm->MultTranspose(B, b);
}
// ------------------------------------------------------------------------
// 11. Solve the fractional PDE by solving M integer order PDEs and adding
// up the solutions.
// ------------------------------------------------------------------------
if (!integer_order)
{
// Setup visualization.
socketstream xout, uout;
ostringstream oss_x, oss_u;
if (visualization)
{
xout.open(vishost, visport);
char vishost[] = "localhost";
int visport = 19916;
socketstream xout(vishost, visport);
xout.precision(8);
uout.open(vishost, visport);
uout.precision(8);
}
// Iterate over all expansion coefficient that contribute to the
// solution.
for (int i = 0; i < coeffs.Size(); i++)
{
if (Mpi::Root())
ostringstream oss;
oss << "Solution of PDE -Δ u + " << -poles[i]
<< " u = " << coeffs[i] << " f" ;
xout << "parallel " << row_size << " " << row_rank << "\n";
xout << "solution\n" << pmesh << x
<< "window_title '" << oss.str() << "'" << flush;
if (col_rank < col_size-1)
{
mfem::out << "\nSolving PDE -Δ u + " << -poles[i]
<< " u = " << coeffs[i] << " g " << endl;
}
// 11.1 Reset GridFunction for integer-order PDE solve.
x = 0.0;
// 11.2 Set up the bilinear form a(.,.) for integer-order PDE solve.
ParBilinearForm a(&fespace);
a.AddDomainIntegrator(new DiffusionIntegrator(one));
ConstantCoefficient d_i(-poles[i]);
a.AddDomainIntegrator(new MassIntegrator(d_i));
a.Assemble();
// 11.3 Assemble the bilinear form and the corresponding linear system.
OperatorPtr A;
Vector B, X;
a.FormLinearSystem(ess_tdof_list, x, b, A, X, B);
// 11.4 Solve the linear system A X = B.
HypreBoomerAMG prec;
prec.SetPrintLevel(-1);
CGSolver cg(MPI_COMM_WORLD);
cg.SetRelTol(1e-12);
cg.SetMaxIter(2000);
cg.SetPrintLevel(3);
cg.SetPreconditioner(prec);
cg.SetOperator(*A);
cg.Mult(B, X);
// 11.5 Recover the solution as a finite element grid function.
a.RecoverFEMSolution(X, b, x);
// 11.6 Accumulate integer-order PDE solutions.
x *= coeffs[i];
u += x;
// 11.7 Send fractional PDE solution to a GLVis server.
if (visualization)
{
oss_x.str(""); oss_x.clear();
oss_x << "Step " << progress_steps
<< ": Solution of PDE -Δ u + " << -poles[i]
<< " u = " << coeffs[i] << " g";
xout << "parallel " << num_procs << " " << myid << "\n"
<< "solution\n" << pmesh << x
<< "window_title '" << oss_x.str() << "'" << flush;
oss_u.str(""); oss_u.clear();
oss_u << "Step " << progress_steps + 1
<< ": Solution of fractional PDE (-Δ)^" << alpha
<< " u = f";
uout << "parallel " << num_procs << " " << myid << "\n"
<< "solution\n" << pmesh << u
<< "window_title '" << oss_u.str() << "'"
<< flush;
MPI_Send(nullptr,0,MPI_INT,col_rank+1,0,col_comm);
}
}
}
// ------------------------------------------------------------------------
// 12. (optional) Verify the solution.
// ------------------------------------------------------------------------
if (verification)
// 17. Accumulate for the fractional PDE solution
MPI_Allreduce(MPI_IN_PLACE, u.GetData(), u.Size(),
MPI_DOUBLE, MPI_SUM,col_comm);
// 18. Send fractional PDE solution to a GLVis server.
if (visualization)
{
auto solution = [] (const Vector &x)
if (col_rank == 0)
{
double val = 1.0;
for (int i=0; i<x.Size(); i++)
{
val *= sin(M_PI*x(i));
}
return val;
};
FunctionCoefficient sol(solution);
double l2_error = u.ComputeL2Error(sol);
if (Mpi::Root())
{
string analytic_solution,expected_mesh;
switch (dim)
{
case 1:
analytic_solution = "sin(π x)";
expected_mesh = "inline_segment.mesh";
break;
case 2:
analytic_solution = "sin(π x) sin(π y)";
expected_mesh = "inline_quad.mesh";
break;
default:
analytic_solution = "sin(π x) sin(π y) sin(π z)";
expected_mesh = "inline_hex.mesh";
break;
}
mfem::out << "\n" << string(80,'=')
<< "\n\nSolution Verification in "<< dim << "D \n\n"
<< "Analytic solution : " << analytic_solution << "\n"
<< "Expected mesh : " << expected_mesh <<"\n"
<< "Your mesh : " << mesh_file << "\n"
<< "L2 error : " << l2_error << "\n\n"
<< string(80,'=') << endl;
char vishost[] = "localhost";
int visport = 19916;
socketstream uout(vishost, visport);
uout.precision(8);
ostringstream oss;
oss << "Solution of fractional PDE -Δ^" << alpha
<< " u = f" ;
uout << "parallel " << row_size << " " << row_rank << "\n";
uout << "solution\n" << pmesh << u
<< "window_title '" << oss.str() << "'" << flush;
}
}
+24
View File
@@ -252,6 +252,7 @@ int main(int argc, char *argv[])
double t_final = 10.0;
double dt = 0.01;
bool visualization = true;
bool internal = false;
bool visit = false;
bool paraview = false;
bool adios2 = false;
@@ -300,6 +301,9 @@ int main(int argc, char *argv[])
args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
"--no-visualization",
"Enable or disable GLVis visualization.");
args.AddOption(&internal, "-idf", "--mfem-datafiles", "-no-idf",
"--no-mfem-datafiles",
"Save data files in the internal formal.");
args.AddOption(&visit, "-visit", "--visit-datafiles", "-no-visit",
"--no-visit-datafiles",
"Save data files for VisIt (visit.llnl.gov) visualization.");
@@ -505,6 +509,18 @@ int main(int argc, char *argv[])
pd->Save();
}
MFEMDataCollection *idc = NULL;
if (internal)
{
idc = new MFEMDataCollection("Example9P-internal", *pmesh, false);
idc->SetPrefixPath("internal");
idc->RegisterField("solution", u);
idc->ResetMetadata();
idc->SetCycle(0);
idc->SetTime(0.0);
idc->Save();
}
// Optionally output a BP (binary pack) file using ADIOS2. This can be
// visualized with the ParaView VTX reader.
#ifdef MFEM_USE_ADIOS2
@@ -606,6 +622,13 @@ int main(int argc, char *argv[])
pd->Save();
}
if (internal)
{
idc->SetCycle(ti);
idc->SetTime(t);
idc->Save();
}
#ifdef MFEM_USE_ADIOS2
// transient solutions can be visualized with ParaView
if (adios2)
@@ -639,6 +662,7 @@ int main(int argc, char *argv[])
delete fes;
delete pmesh;
delete ode_solver;
delete idc;
delete pd;
#ifdef MFEM_USE_ADIOS2
if (adios2)
+20 -25
View File
@@ -58,17 +58,15 @@ set(SRCS
gridfunc.cpp
hybridization.cpp
intrules.cpp
ceed/interface/basis.cpp
ceed/interface/restriction.cpp
ceed/interface/operator.cpp
ceed/interface/util.cpp
ceed/integrators/convection/convection.cpp
ceed/integrators/diffusion/diffusion.cpp
ceed/integrators/nlconvection/nlconvection.cpp
ceed/integrators/mass/mass.cpp
ceed/solvers/algebraic.cpp
ceed/solvers/full-assembly.cpp
ceed/solvers/solvers-atpmg.cpp
ceed/algebraic.cpp
ceed/full-assembly.cpp
ceed/solvers-atpmg.cpp
ceed/convection.cpp
ceed/diffusion.cpp
ceed/nlconvection.cpp
ceed/mass.cpp
ceed/operator.cpp
ceed/util.cpp
linearform.cpp
linearform_ext.cpp
lininteg.cpp
@@ -160,20 +158,17 @@ set(HDRS
gridfunc.hpp
hybridization.hpp
intrules.hpp
ceed/interface/basis.hpp
ceed/interface/integrator.hpp
ceed/interface/interface.hpp
ceed/interface/operator.hpp
ceed/interface/restriction.hpp
ceed/interface/util.hpp
ceed/integrators/convection/convection.hpp
ceed/integrators/diffusion/diffusion.hpp
ceed/integrators/mass/mass.hpp
ceed/integrators/nlconvection/nlconvection.hpp
ceed/interface/coefficient.hpp
ceed/solvers/algebraic.hpp
ceed/solvers/full-assembly.hpp
ceed/solvers/solvers-atpmg.hpp
ceed/algebraic.hpp
ceed/full-assembly.hpp
ceed/solvers-atpmg.hpp
ceed/coefficient.hpp
ceed/convection.hpp
ceed/diffusion.hpp
ceed/integrator.hpp
ceed/mass.hpp
ceed/nlconvection.hpp
ceed/operator.hpp
ceed/util.hpp
linearform.hpp
linearform_ext.hpp
lininteg.hpp
+4 -40
View File
@@ -305,66 +305,38 @@ public:
/// Finalizes the matrix initialization.
virtual void Finalize(int skip_zeros = 1);
/** @brief Returns a const reference to the sparse matrix: \f$ M \f$
This will fail if HasSpMat() is false. */
/// Returns a const reference to the sparse matrix.
const SparseMatrix &SpMat() const
{
MFEM_VERIFY(mat, "mat is NULL and can't be dereferenced");
return *mat;
}
/** @brief Returns a reference to the sparse matrix: \f$ M \f$
This will fail if HasSpMat() is false. */
/// Returns a reference to the sparse matrix: \f$ M \f$
SparseMatrix &SpMat()
{
MFEM_VERIFY(mat, "mat is NULL and can't be dereferenced");
return *mat;
}
/** @brief Returns true if the sparse matrix is not null, false otherwise.
@sa SpMat(). */
bool HasSpMat()
{
return mat != nullptr;
}
/** @brief Nullifies the internal matrix \f$ M \f$ and returns a pointer
to it. Used for transfering ownership. */
SparseMatrix *LoseMat() { SparseMatrix *tmp = mat; mat = NULL; return tmp; }
/** @brief Returns a const reference to the sparse matrix of eliminated b.c.:
\f$ M_e \f$
This will fail if HasSpMatElim() is false. */
/// Returns a const reference to the sparse matrix of eliminated b.c.: \f$ M_e \f$
const SparseMatrix &SpMatElim() const
{
MFEM_VERIFY(mat_e, "mat_e is NULL and can't be dereferenced");
return *mat_e;
}
/** @brief Returns a reference to the sparse matrix of eliminated b.c.:
\f$ M_e \f$
This will fail if HasSpMatElim() is false. */
/// Returns a reference to the sparse matrix of eliminated b.c.: \f$ M_e \f$
SparseMatrix &SpMatElim()
{
MFEM_VERIFY(mat_e, "mat_e is NULL and can't be dereferenced");
return *mat_e;
}
/** @brief Returns true if the sparse matrix of eliminated b.c.s is not null,
false otherwise.
@sa SpMatElim(). */
bool HasSpMatElim()
{
return mat_e != nullptr;
}
/// Adds new Domain Integrator. Assumes ownership of @a bfi.
void AddDomainIntegrator(BilinearFormIntegrator *bfi);
/// Adds new Domain Integrator restricted to certain elements specified by
@@ -438,14 +410,6 @@ public:
virtual const Operator *GetOutputRestriction() const
{ return GetRestriction(); }
/// @brief Compute serial RAP operator and store it in @a A as a SparseMatrix.
void SerialRAP(OperatorHandle &A)
{
MFEM_ASSERT(mat, "SerialRAP requires the SparseMatrix to be assembled.");
ConformingAssemble();
A.Reset(mat, false);
}
/** @brief Form the linear system A X = B, corresponding to this bilinear
form and the linear form @a b(.). */
/** This method applies any necessary transformations to the linear system
+1 -53
View File
@@ -16,7 +16,7 @@
#include "bilinearform.hpp"
#include "pbilinearform.hpp"
#include "pgridfunc.hpp"
#include "ceed/interface/util.hpp"
#include "ceed/util.hpp"
namespace mfem
{
@@ -251,7 +251,6 @@ PABilinearFormExtension::PABilinearFormExtension(BilinearForm *form)
void PABilinearFormExtension::SetupRestrictionOperators(const L2FaceValues m)
{
if ( Device::Allows(Backend::CEED_MASK) ) { return; }
ElementDofOrdering ordering = UsesTensorBasis(*a->FESpace())?
ElementDofOrdering::LEXICOGRAPHIC:
ElementDofOrdering::NATIVE;
@@ -957,57 +956,6 @@ void FABilinearFormExtension::Assemble()
}
}
void FABilinearFormExtension::RAP(OperatorHandle &A)
{
#ifdef MFEM_USE_MPI
if ( auto pa = dynamic_cast<ParBilinearForm*>(a) )
{
pa->ParallelRAP(*pa->mat, A);
}
else
#endif
{
a->SerialRAP(A);
}
}
void FABilinearFormExtension::EliminateBC(const Array<int> &ess_dofs,
OperatorHandle &A)
{
#ifdef MFEM_USE_MPI
if ( dynamic_cast<ParBilinearForm*>(a) )
{
A.As<HypreParMatrix>()->EliminateBC(ess_dofs,
DiagonalPolicy::DIAG_ONE);
}
else
#endif
{
A.As<SparseMatrix>()->EliminateBC(ess_dofs,
DiagonalPolicy::DIAG_ONE);
}
}
void FABilinearFormExtension::FormSystemMatrix(const Array<int> &ess_dofs,
OperatorHandle &A)
{
RAP(A);
EliminateBC(ess_dofs, A);
}
void FABilinearFormExtension::FormLinearSystem(const Array<int> &ess_tdof_list,
Vector &x, Vector &b,
OperatorHandle &A,
Vector &X, Vector &B,
int copy_interior)
{
Operator *A_out;
Operator::FormLinearSystem(ess_tdof_list, x, b, A_out, X, B, copy_interior);
delete A_out;
FormSystemMatrix(ess_tdof_list, A);
}
void FABilinearFormExtension::DGMult(const Vector &x, Vector &y) const
{
#ifdef MFEM_USE_MPI
-9
View File
@@ -125,15 +125,6 @@ public:
FABilinearFormExtension(BilinearForm *form);
void Assemble();
void RAP(OperatorHandle &A);
/** @note Always does `DIAG_ONE` policy to be consistent with
`Operator::FormConstrainedSystemOperator`. */
void EliminateBC(const Array<int> &ess_dofs, OperatorHandle &A);
void FormSystemMatrix(const Array<int> &ess_tdof_list, OperatorHandle &A);
void FormLinearSystem(const Array<int> &ess_tdof_list,
Vector &x, Vector &b,
OperatorHandle &A, Vector &X, Vector &B,
int copy_interior = 0);
void Mult(const Vector &x, Vector &y) const;
void MultTranspose(const Vector &x, Vector &y) const;
+1 -2
View File
@@ -1672,8 +1672,7 @@ void VectorFEDivergenceIntegrator::AssembleElementMatrix2(
{
const IntegrationPoint &ip = ir->IntPoint(i);
trial_fe.CalcDivShape(ip, divshape);
Trans.SetIntPoint(&ip);
test_fe.CalcPhysShape(Trans, shape);
test_fe.CalcShape(ip, shape);
double w = ip.weight;
if (Q)
{
+3 -6
View File
@@ -15,7 +15,6 @@
#include "../config/config.hpp"
#include "nonlininteg.hpp"
#include "fespace.hpp"
#include "ceed/interface/util.hpp"
namespace mfem
{
@@ -2398,9 +2397,8 @@ public:
scalar function given by FiniteElement through standard transformation.
Here, u is the trial function and p is the test function.
Note: if the test space does not have map type INTEGRAL, then the element
matrix returned by AssembleElementMatrix2 will not depend on the
ElementTransformation Trans. */
Note: the element matrix returned by AssembleElementMatrix2 does NOT depend
on the ElementTransformation Trans. */
class VectorFEDivergenceIntegrator : public BilinearFormIntegrator
{
protected:
@@ -2737,8 +2735,7 @@ private:
public:
DivDivIntegrator() { Q = NULL; }
DivDivIntegrator(Coefficient &q, const IntegrationRule *ir = NULL) :
BilinearFormIntegrator(ir), Q(&q) { }
DivDivIntegrator(Coefficient &q) : Q(&q) { }
virtual void AssembleElementMatrix(const FiniteElement &el,
ElementTransformation &Trans,
+2 -11
View File
@@ -12,7 +12,7 @@
#include "../general/forall.hpp"
#include "bilininteg.hpp"
#include "gridfunc.hpp"
#include "ceed/integrators/convection/convection.hpp"
#include "ceed/convection.hpp"
using namespace std;
@@ -30,16 +30,7 @@ void ConvectionIntegrator::AssembleMF(const FiniteElementSpace &fes)
if (DeviceCanUseCeed())
{
delete ceedOp;
const bool mixed = mesh->GetNumGeometries(mesh->Dimension()) > 1 ||
fes.IsVariableOrder();
if (mixed)
{
ceedOp = new ceed::MixedMFConvectionIntegrator(*this, fes, Q, alpha);
}
else
{
ceedOp = new ceed::MFConvectionIntegrator(fes, *ir, Q, alpha);
}
ceedOp = new ceed::MFConvectionIntegrator(fes, *ir, Q, alpha);
return;
}
MFEM_ABORT("Error: ConvectionIntegrator::AssembleMF only implemented with"
+2 -13
View File
@@ -12,7 +12,7 @@
#include "../general/forall.hpp"
#include "bilininteg.hpp"
#include "gridfunc.hpp"
#include "ceed/integrators/convection/convection.hpp"
#include "ceed/convection.hpp"
#include "quadinterpolator.hpp"
namespace mfem
@@ -1386,16 +1386,7 @@ void ConvectionIntegrator::AssemblePA(const FiniteElementSpace &fes)
if (DeviceCanUseCeed())
{
delete ceedOp;
const bool mixed = mesh->GetNumGeometries(mesh->Dimension()) > 1 ||
fes.IsVariableOrder();
if (mixed)
{
ceedOp = new ceed::MixedPAConvectionIntegrator(*this, fes, Q, alpha);
}
else
{
ceedOp = new ceed::PAConvectionIntegrator(fes, *ir, Q, alpha);
}
ceedOp = new ceed::PAConvectionIntegrator(fes, *ir, Q, alpha);
return;
}
const int dims = el.GetDim();
@@ -1506,7 +1497,6 @@ static void PAConvectionApply(const int dim,
{
switch ((D1D << 4 ) | Q1D)
{
case 0x22: return SmemPAConvectionApply3D<2,2>(NE,B,G,Bt,Gt,op,x,y);
case 0x23: return SmemPAConvectionApply3D<2,3>(NE,B,G,Bt,Gt,op,x,y);
case 0x24: return SmemPAConvectionApply3D<2,4>(NE,B,G,Bt,Gt,op,x,y);
case 0x26: return SmemPAConvectionApply3D<2,6>(NE,B,G,Bt,Gt,op,x,y);
@@ -1558,7 +1548,6 @@ static void PAConvectionApplyT(const int dim,
{
switch ((D1D << 4 ) | Q1D)
{
case 0x22: return SmemPAConvectionApplyT3D<2,2>(NE,B,G,Bt,Gt,op,x,y);
case 0x23: return SmemPAConvectionApplyT3D<2,3>(NE,B,G,Bt,Gt,op,x,y);
case 0x24: return SmemPAConvectionApplyT3D<2,4>(NE,B,G,Bt,Gt,op,x,y);
case 0x26: return SmemPAConvectionApplyT3D<2,6>(NE,B,G,Bt,Gt,op,x,y);
+1 -6
View File
@@ -136,9 +136,6 @@ static void PADGTraceSetup(const int dim,
void DGTraceIntegrator::SetupPA(const FiniteElementSpace &fes, FaceType type)
{
const MemoryType mt = (pa_mt == MemoryType::DEFAULT) ?
Device::GetDeviceMemoryType() : pa_mt;
nf = fes.GetNFbyType(type);
if (nf==0) { return; }
// Assumes tensor-product elements
@@ -156,7 +153,7 @@ void DGTraceIntegrator::SetupPA(const FiniteElementSpace &fes, FaceType type)
geom = mesh->GetFaceGeometricFactors(
*ir,
FaceGeometricFactors::DETERMINANTS |
FaceGeometricFactors::NORMALS, type, mt);
FaceGeometricFactors::NORMALS, type);
maps = &el.GetDofToQuad(*ir, DofToQuad::TENSOR);
dofs1D = maps->ndof;
quad1D = maps->nqpt;
@@ -698,7 +695,6 @@ static void PADGTraceApply(const int dim,
{
switch ((D1D << 4 ) | Q1D)
{
case 0x22: return SmemPADGTraceApply3D<2,2,1>(NF,B,Bt,op,x,y);
case 0x23: return SmemPADGTraceApply3D<2,3,1>(NF,B,Bt,op,x,y);
case 0x34: return SmemPADGTraceApply3D<3,4,2>(NF,B,Bt,op,x,y);
case 0x45: return SmemPADGTraceApply3D<4,5,2>(NF,B,Bt,op,x,y);
@@ -1128,7 +1124,6 @@ static void PADGTraceApplyTranspose(const int dim,
{
switch ((D1D << 4 ) | Q1D)
{
case 0x22: return SmemPADGTraceApplyTranspose3D<2,2>(NF,B,Bt,op,x,y);
case 0x23: return SmemPADGTraceApplyTranspose3D<2,3>(NF,B,Bt,op,x,y);
case 0x34: return SmemPADGTraceApplyTranspose3D<3,4>(NF,B,Bt,op,x,y);
case 0x45: return SmemPADGTraceApplyTranspose3D<4,5>(NF,B,Bt,op,x,y);
+2 -11
View File
@@ -12,7 +12,7 @@
#include "../general/forall.hpp"
#include "bilininteg.hpp"
#include "gridfunc.hpp"
#include "ceed/integrators/diffusion/diffusion.hpp"
#include "ceed/diffusion.hpp"
using namespace std;
@@ -33,16 +33,7 @@ void DiffusionIntegrator::AssembleMF(const FiniteElementSpace &fes)
MFEM_VERIFY(!VQ && !MQ,
"Only scalar coefficient supported for DiffusionIntegrator"
" with libCEED");
const bool mixed = mesh->GetNumGeometries(mesh->Dimension()) > 1 ||
fes.IsVariableOrder();
if (mixed)
{
ceedOp = new ceed::MixedMFDiffusionIntegrator(*this, fes, Q);
}
else
{
ceedOp = new ceed::MFDiffusionIntegrator(fes, *ir, Q);
}
ceedOp = new ceed::MFDiffusionIntegrator(fes, *ir, Q);
return;
}
MFEM_ABORT("Error: DiffusionIntegrator::AssembleMF only implemented with"
+5 -17
View File
@@ -12,7 +12,7 @@
#include "../general/forall.hpp"
#include "bilininteg.hpp"
#include "gridfunc.hpp"
#include "ceed/integrators/diffusion/diffusion.hpp"
#include "ceed/diffusion.hpp"
using namespace std;
@@ -271,21 +271,18 @@ void PADiffusionSetup3D(const int Q1D,
D(qx,qy,qz,1,e) = D12; // 1,2
D(qx,qy,qz,2,e) = w_detJ * (A11*R13 + A12*R23 + A13*R33); // 1,3
const double D21 = w_detJ * (A21*R11 + A22*R21 + A23*R31);
const double D22 = w_detJ * (A21*R12 + A22*R22 + A23*R32);
const double D23 = w_detJ * (A21*R13 + A22*R23 + A23*R33);
const double D33 = w_detJ * (A31*R13 + A32*R23 + A33*R33);
D(qx,qy,qz,3,e) = symmetric ? D22 : D21; // 2,2 or 2,1
D(qx,qy,qz,4,e) = symmetric ? D23 : D22; // 2,3 or 2,2
D(qx,qy,qz,5,e) = symmetric ? D33 : D23; // 3,3 or 2,3
if (symmetric)
if (!symmetric)
{
D(qx,qy,qz,3,e) = D22; // 2,2
}
else
{
D(qx,qy,qz,3,e) = w_detJ * (A21*R11 + A22*R21 + A23*R31); // 2,1
D(qx,qy,qz,6,e) = w_detJ * (A31*R11 + A32*R21 + A33*R31); // 3,1
D(qx,qy,qz,7,e) = w_detJ * (A31*R12 + A32*R22 + A33*R32); // 3,2
D(qx,qy,qz,8,e) = D33; // 3,3
@@ -368,16 +365,7 @@ void DiffusionIntegrator::AssemblePA(const FiniteElementSpace &fes)
MFEM_VERIFY(!VQ && !MQ,
"Only scalar coefficient supported for DiffusionIntegrator"
" with libCEED");
const bool mixed = mesh->GetNumGeometries(mesh->Dimension()) > 1 ||
fes.IsVariableOrder();
if (mixed)
{
ceedOp = new ceed::MixedPADiffusionIntegrator(*this, fes, Q);
}
else
{
ceedOp = new ceed::PADiffusionIntegrator(fes, *ir, Q);
}
ceedOp = new ceed::PADiffusionIntegrator(fes, *ir, Q);
return;
}
const int dims = el.GetDim();
+31 -626
View File
@@ -24,20 +24,18 @@ namespace mfem
// PA H(div) Mass Assemble 2D kernel
void PAHdivSetup2D(const int Q1D,
const int coeffDim,
const int NE,
const Array<double> &w,
const Vector &j,
Vector &coeff_,
Vector &op)
{
const bool symmetric = (coeffDim != 4);
const int NQ = Q1D*Q1D;
auto W = w.Read();
auto J = Reshape(j.Read(), NQ, 2, 2, NE);
auto C = Reshape(coeff_.Read(), coeffDim, NQ, NE);
auto y = Reshape(op.Write(), NQ, symmetric ? 3 : 4, NE);
auto coeff = Reshape(coeff_.Read(), NQ, NE);
auto y = Reshape(op.Write(), NQ, 3, NE);
MFEM_FORALL(e, NE,
{
@@ -47,60 +45,28 @@ void PAHdivSetup2D(const int Q1D,
const double J21 = J(q,1,0,e);
const double J12 = J(q,0,1,e);
const double J22 = J(q,1,1,e);
const double c_detJ = W[q] / ((J11*J22)-(J21*J12));
// (1/detJ) J^T C J
if (coeffDim == 3 || coeffDim == 4) // Matrix coefficient
{
const double C11 = C(0,q,e);
const double C12 = C(1,q,e);
const double C21 = symmetric ? C12 : C(2,q,e);
const double C22 = symmetric ? C(2,q,e) : C(3,q,e);
const double R11 = C11*J11 + C12*J21;
const double R21 = C21*J11 + C22*J21;
const double R12 = C11*J12 + C12*J22;
const double R22 = C21*J12 + C22*J22;
y(q,0,e) = c_detJ * (J11*R11 + J21*R21); // 1,1
y(q,1,e) = c_detJ * (J11*R12 + J21*R22); // 1,2
if (symmetric)
{
y(q,2,e) = c_detJ * (J12*R12 + J22*R22); // 2,2
}
else
{
y(q,2,e) = c_detJ * (J12*R11 + J22*R21); // 2,1
y(q,3,e) = c_detJ * (J12*R12 + J22*R22); // 2,2
}
}
else // Vector or scalar coefficient
{
const double C1 = C(0,q,e);
const double C2 = (coeffDim == 2 ? C(1,q,e) : C1);
y(q,0,e) = c_detJ * (J11*C1*J11 + J21*C2*J21); // 1,1
y(q,1,e) = c_detJ * (J11*C1*J12 + J21*C2*J22); // 1,2
y(q,2,e) = c_detJ * (J12*C1*J12 + J22*C2*J22); // 2,2
}
const double c_detJ = W[q] * coeff(q, e) / ((J11*J22)-(J21*J12));
// (c/detJ) J^T J
y(q,0,e) = c_detJ * (J11*J11 + J21*J21); // 1,1
y(q,1,e) = c_detJ * (J11*J12 + J21*J22); // 1,2
y(q,2,e) = c_detJ * (J12*J12 + J22*J22); // 2,2
}
});
}
// PA H(div) Mass Assemble 3D kernel
void PAHdivSetup3D(const int Q1D,
const int coeffDim,
const int NE,
const Array<double> &w,
const Vector &j,
Vector &coeff_,
Vector &op)
{
const bool symmetric = (coeffDim != 9);
const int NQ = Q1D*Q1D*Q1D;
auto W = w.Read();
auto J = Reshape(j.Read(), NQ, 3, 3, NE);
auto C = Reshape(coeff_.Read(), coeffDim, NQ, NE);
auto y = Reshape(op.Write(), NQ, symmetric ? 6 : 9, NE);
auto coeff = Reshape(coeff_.Read(), NQ, NE);
auto y = Reshape(op.Write(), NQ, 6, NE);
MFEM_FORALL(e, NE,
{
@@ -118,58 +84,14 @@ void PAHdivSetup3D(const int Q1D,
const double detJ = J11 * (J22 * J33 - J32 * J23) -
/* */ J21 * (J12 * J33 - J32 * J13) +
/* */ J31 * (J12 * J23 - J22 * J13);
const double c_detJ = W[q] / detJ;
// (1/detJ) J^T C J
if (coeffDim == 6 || coeffDim == 9) // Matrix coefficient version
{
double M[3][3];
M[0][0] = C(0, q, e);
M[0][1] = C(1, q, e);
M[0][2] = C(2, q, e);
M[1][0] = (!symmetric) ? C(3, q, e) : M[0][1];
M[1][1] = (!symmetric) ? C(4, q, e) : C(3, q, e);
M[1][2] = (!symmetric) ? C(5, q, e) : C(4, q, e);
M[2][0] = (!symmetric) ? C(6, q, e) : M[0][2];
M[2][1] = (!symmetric) ? C(7, q, e) : M[1][2];
M[2][2] = (!symmetric) ? C(8, q, e) : C(5, q, e);
int idx = 0;
for (int i=0; i<3; ++i)
for (int j = (symmetric ? i : 0); j<3; ++j)
{
y(q,idx,e) = 0.0;
for (int k=0; k<3; ++k)
{
double MJ_kj = 0.0;
for (int l=0; l<3; ++l)
{
MJ_kj += M[k][l] * J(q,l,j,e);
}
y(q,idx,e) += J(q,k,i,e) * MJ_kj;
}
y(q,idx,e) *= c_detJ;
idx++;
}
}
else // Vector or scalar coefficient version
{
int idx = 0;
for (int i=0; i<3; ++i)
for (int j=i; j<3; ++j)
{
y(q,idx,e) = 0.0;
for (int k=0; k<3; ++k)
{
y(q,idx,e) += J(q,k,i,e) * C(coeffDim == 3 ? k : 0, q, e) * J(q,k,j,e);
}
y(q,idx,e) *= c_detJ;
idx++;
}
}
const double c_detJ = W[q] * coeff(q, e) / detJ;
// (c/detJ) J^T J
y(q,0,e) = c_detJ * (J11*J11 + J21*J21 + J31*J31); // 1,1
y(q,1,e) = c_detJ * (J12*J11 + J22*J21 + J32*J31); // 2,1
y(q,2,e) = c_detJ * (J13*J11 + J23*J21 + J33*J31); // 3,1
y(q,3,e) = c_detJ * (J12*J12 + J22*J22 + J32*J32); // 2,2
y(q,4,e) = c_detJ * (J13*J12 + J23*J22 + J33*J32); // 3,2
y(q,5,e) = c_detJ * (J13*J13 + J23*J23 + J33*J33); // 3,3
}
});
}
@@ -177,7 +99,6 @@ void PAHdivSetup3D(const int Q1D,
void PAHdivMassApply2D(const int D1D,
const int Q1D,
const int NE,
const bool symmetric,
const Array<double> &Bo_,
const Array<double> &Bc_,
const Array<double> &Bot_,
@@ -194,7 +115,7 @@ void PAHdivMassApply2D(const int D1D,
auto Bc = Reshape(Bc_.Read(), Q1D, D1D);
auto Bot = Reshape(Bot_.Read(), D1D-1, Q1D);
auto Bct = Reshape(Bct_.Read(), D1D, Q1D);
auto op = Reshape(op_.Read(), Q1D, Q1D, symmetric ? 3 : 4, NE);
auto op = Reshape(op_.Read(), Q1D, Q1D, 3, NE);
auto x = Reshape(x_.Read(), 2*(D1D-1)*D1D, NE);
auto y = Reshape(y_.ReadWrite(), 2*(D1D-1)*D1D, NE);
@@ -257,12 +178,11 @@ void PAHdivMassApply2D(const int D1D,
{
const double O11 = op(qx,qy,0,e);
const double O12 = op(qx,qy,1,e);
const double O21 = symmetric ? O12 : op(qx,qy,2,e);
const double O22 = symmetric ? op(qx,qy,2,e) : op(qx,qy,3,e);
const double O22 = op(qx,qy,2,e);
const double massX = mass[qy][qx][0];
const double massY = mass[qy][qx][1];
mass[qy][qx][0] = (O11*massX)+(O12*massY);
mass[qy][qx][1] = (O21*massX)+(O22*massY);
mass[qy][qx][1] = (O12*massX)+(O22*massY);
}
}
@@ -305,179 +225,9 @@ void PAHdivMassApply2D(const int D1D,
}); // end of element loop
}
template<int T_D1D = 0, int T_Q1D = 0>
void SmemPAHdivMassApply2D(const int NE,
const bool symmetric,
const Array<double> &Bo_,
const Array<double> &Bc_,
const Array<double> &Bot_,
const Array<double> &Bct_,
const Vector &op_,
const Vector &x_,
Vector &y_,
const int d1d = 0,
const int q1d = 0)
{
MFEM_CONTRACT_VAR(Bot_);
MFEM_CONTRACT_VAR(Bct_);
static constexpr int VDIM = 2;
const int D1D = T_D1D ? T_D1D : d1d;
const int Q1D = T_Q1D ? T_Q1D : q1d;
const auto bo = Reshape(Bo_.Read(), Q1D, D1D-1);
const auto bc = Reshape(Bc_.Read(), Q1D, D1D);
const auto D = Reshape(op_.Read(), Q1D, Q1D, symmetric ? 3 : 4, NE);
const auto x = Reshape(x_.Read(), D1D*(D1D-1), VDIM, NE);
auto y = y_.ReadWrite();
MFEM_FORALL_3D(e, NE, Q1D, Q1D, VDIM,
{
const int tidz = MFEM_THREAD_ID(z);
const int D1D = T_D1D ? T_D1D : d1d;
const int Q1D = T_Q1D ? T_Q1D : q1d;
constexpr int MQ1 = T_Q1D ? T_Q1D : HDIV_MAX_Q1D;
constexpr int MD1 = T_D1D ? T_D1D : HDIV_MAX_D1D;
constexpr int MDQ = (MQ1 > MD1) ? MQ1 : MD1;
MFEM_SHARED double smo[MQ1*(MD1-1)];
DeviceMatrix Bo(smo, D1D-1, Q1D);
MFEM_SHARED double smc[MQ1*MD1];
DeviceMatrix Bc(smc, D1D, Q1D);
MFEM_SHARED double sm0[VDIM*MDQ*MDQ];
MFEM_SHARED double sm1[VDIM*MDQ*MDQ];
DeviceMatrix X(sm0, D1D*(D1D-1), VDIM);
DeviceCube QD(sm1, Q1D, D1D, VDIM);
DeviceCube QQ(sm0, Q1D, Q1D, VDIM);
// Load X, Bo and Bc into shared memory
MFEM_FOREACH_THREAD(vd,z,VDIM)
{
MFEM_FOREACH_THREAD(dy,y,D1D)
{
MFEM_FOREACH_THREAD(qx,x,Q1D)
{
if (qx < D1D && dy < (D1D-1)) { X(qx + dy*D1D,vd) = x(qx+dy*D1D,vd,e); }
if (tidz == 0)
{
if (dy < (D1D-1)) { Bo(dy,qx) = bo(qx,dy); }
Bc(dy,qx) = bc(qx,dy);
}
}
}
}
MFEM_SYNC_THREAD;
// Apply B operator
MFEM_FOREACH_THREAD(vd,z,VDIM)
{
const int nx = (vd == 0) ? D1D : D1D-1;
const int ny = (vd == 1) ? D1D : D1D-1;
DeviceCube Xxy(X, nx, ny, VDIM);
DeviceMatrix Bx = (vd == 0) ? Bc : Bo;
MFEM_FOREACH_THREAD(dy,y,ny)
{
MFEM_FOREACH_THREAD(qx,x,Q1D)
{
double dq = 0.0;
for (int dx = 0; dx < nx; ++dx)
{
dq += Xxy(dx,dy,vd) * Bx(dx,qx);
}
QD(qx,dy,vd) = dq;
}
}
}
MFEM_SYNC_THREAD;
MFEM_FOREACH_THREAD(vd,z,VDIM)
{
const int ny = (vd == 1) ? D1D : D1D-1;
DeviceMatrix By = (vd == 1) ? Bc : Bo;
MFEM_FOREACH_THREAD(qy,y,Q1D)
{
MFEM_FOREACH_THREAD(qx,x,Q1D)
{
double qq = 0.0;
for (int dy = 0; dy < ny; ++dy)
{
qq += QD(qx,dy,vd) * By(dy,qy);
}
QQ(qx,qy,vd) = qq;
}
}
}
MFEM_SYNC_THREAD;
// Apply D operator
if (tidz == 0)
{
MFEM_FOREACH_THREAD(qy,y,Q1D)
{
MFEM_FOREACH_THREAD(qx,x,Q1D)
{
const double Qx = QQ(qx,qy,0);
const double Qy = QQ(qx,qy,1);
const double D11 = D(qx,qy,0,e);
const double D12 = D(qx,qy,1,e);
const double D21 = symmetric ? D12 : D(qx,qy,2,e);
const double D22 = symmetric ? D(qx,qy,2,e) : D(qx,qy,3,e);
QQ(qx,qy,0) = D11*Qx + D12*Qy;
QQ(qx,qy,1) = D21*Qx + D22*Qy;
}
}
}
MFEM_SYNC_THREAD;
// Apply Bt operator
MFEM_FOREACH_THREAD(vd,z,VDIM)
{
const int nx = (vd == 0) ? D1D : D1D-1;
DeviceMatrix Btx = (vd == 0) ? Bc : Bo;
MFEM_FOREACH_THREAD(qy,y,Q1D)
{
MFEM_FOREACH_THREAD(dx,x,nx)
{
double qd = 0.0;
for (int qx = 0; qx < Q1D; ++qx)
{
qd += QQ(qx,qy,vd) * Btx(dx,qx);
}
QD(dx,qy,vd) = qd;
}
}
}
MFEM_SYNC_THREAD;
MFEM_FOREACH_THREAD(vd,z,VDIM)
{
const int nx = (vd == 0) ? D1D : D1D-1;
const int ny = (vd == 1) ? D1D : D1D-1;
DeviceMatrix Bty = (vd == 1) ? Bc : Bo;
DeviceTensor<4> Yxy(y, nx, ny, VDIM, NE);
MFEM_FOREACH_THREAD(dy,y,ny)
{
MFEM_FOREACH_THREAD(dx,x,nx)
{
double dd = 0.0;
for (int qy = 0; qy < Q1D; ++qy)
{
dd += QD(dx,qy,vd) * Bty(dy,qy);
}
Yxy(dx,dy,vd,e) += dd;
}
}
}
MFEM_SYNC_THREAD;
});
}
void PAHdivMassAssembleDiagonal2D(const int D1D,
const int Q1D,
const int NE,
const bool symmetric,
const Array<double> &Bo_,
const Array<double> &Bc_,
const Vector &op_,
@@ -488,7 +238,7 @@ void PAHdivMassAssembleDiagonal2D(const int D1D,
auto Bo = Reshape(Bo_.Read(), Q1D, D1D-1);
auto Bc = Reshape(Bc_.Read(), Q1D, D1D);
auto op = Reshape(op_.Read(), Q1D, Q1D, symmetric ? 3 : 4, NE);
auto op = Reshape(op_.Read(), Q1D, Q1D, 3, NE);
auto diag = Reshape(diag_.ReadWrite(), 2*(D1D-1)*D1D, NE);
MFEM_FORALL(e, NE,
@@ -509,7 +259,7 @@ void PAHdivMassAssembleDiagonal2D(const int D1D,
for (int qy = 0; qy < Q1D; ++qy)
{
const double wy = (c == 1) ? Bc(qy,dy) : Bo(qy,dy);
mass[qx] += wy*wy*((c == 0) ? op(qx,qy,0,e) : op(qx,qy,symmetric ? 2 : 3,e));
mass[qx] += wy*wy*((c == 0) ? op(qx,qy,0,e) : op(qx,qy,2,e));
}
}
@@ -533,7 +283,6 @@ void PAHdivMassAssembleDiagonal2D(const int D1D,
void PAHdivMassAssembleDiagonal3D(const int D1D,
const int Q1D,
const int NE,
const bool symmetric,
const Array<double> &Bo_,
const Array<double> &Bc_,
const Vector &op_,
@@ -545,7 +294,7 @@ void PAHdivMassAssembleDiagonal3D(const int D1D,
auto Bo = Reshape(Bo_.Read(), Q1D, D1D-1);
auto Bc = Reshape(Bc_.Read(), Q1D, D1D);
auto op = Reshape(op_.Read(), Q1D, Q1D, Q1D, symmetric ? 6 : 9, NE);
auto op = Reshape(op_.Read(), Q1D, Q1D, Q1D, 6, NE);
auto diag = Reshape(diag_.ReadWrite(), 3*(D1D-1)*(D1D-1)*D1D, NE);
MFEM_FORALL(e, NE,
@@ -558,8 +307,7 @@ void PAHdivMassAssembleDiagonal3D(const int D1D,
const int D1Dy = (c == 1) ? D1D : D1D - 1;
const int D1Dx = (c == 0) ? D1D : D1D - 1;
const int opc = (c == 0) ? 0 : ((c == 1) ? (symmetric ? 3 : 4) :
(symmetric ? 5 : 8));
const int opc = (c == 0) ? 0 : ((c == 1) ? 3 : 5);
double mass[HDIV_MAX_Q1D];
@@ -602,7 +350,6 @@ void PAHdivMassAssembleDiagonal3D(const int D1D,
void PAHdivMassApply3D(const int D1D,
const int Q1D,
const int NE,
const bool symmetric,
const Array<double> &Bo_,
const Array<double> &Bc_,
const Array<double> &Bot_,
@@ -619,7 +366,7 @@ void PAHdivMassApply3D(const int D1D,
auto Bc = Reshape(Bc_.Read(), Q1D, D1D);
auto Bot = Reshape(Bot_.Read(), D1D-1, Q1D);
auto Bct = Reshape(Bct_.Read(), D1D, Q1D);
auto op = Reshape(op_.Read(), Q1D, Q1D, Q1D, symmetric ? 6 : 9, NE);
auto op = Reshape(op_.Read(), Q1D, Q1D, Q1D, 6, NE);
auto x = Reshape(x_.Read(), 3*(D1D-1)*(D1D-1)*D1D, NE);
auto y = Reshape(y_.ReadWrite(), 3*(D1D-1)*(D1D-1)*D1D, NE);
@@ -714,19 +461,15 @@ void PAHdivMassApply3D(const int D1D,
const double O11 = op(qx,qy,qz,0,e);
const double O12 = op(qx,qy,qz,1,e);
const double O13 = op(qx,qy,qz,2,e);
const double O21 = symmetric ? O12 : op(qx,qy,qz,3,e);
const double O22 = symmetric ? op(qx,qy,qz,3,e) : op(qx,qy,qz,4,e);
const double O23 = symmetric ? op(qx,qy,qz,4,e) : op(qx,qy,qz,5,e);
const double O31 = symmetric ? O13 : op(qx,qy,qz,6,e);
const double O32 = symmetric ? O23 : op(qx,qy,qz,7,e);
const double O33 = symmetric ? op(qx,qy,qz,5,e) : op(qx,qy,qz,8,e);
const double O22 = op(qx,qy,qz,3,e);
const double O23 = op(qx,qy,qz,4,e);
const double O33 = op(qx,qy,qz,5,e);
const double massX = mass[qz][qy][qx][0];
const double massY = mass[qz][qy][qx][1];
const double massZ = mass[qz][qy][qx][2];
mass[qz][qy][qx][0] = (O11*massX)+(O12*massY)+(O13*massZ);
mass[qz][qy][qx][1] = (O21*massX)+(O22*massY)+(O23*massZ);
mass[qz][qy][qx][2] = (O31*massX)+(O32*massY)+(O33*massZ);
mass[qz][qy][qx][1] = (O12*massX)+(O22*massY)+(O23*massZ);
mass[qz][qy][qx][2] = (O13*massX)+(O23*massY)+(O33*massZ);
}
}
}
@@ -794,337 +537,6 @@ void PAHdivMassApply3D(const int D1D,
}); // end of element loop
}
template<int T_D1D = 0, int T_Q1D = 0>
void SmemPAHdivMassApply3D(const int NE,
const bool symmetric,
const Array<double> &Bo_,
const Array<double> &Bc_,
const Array<double> &Bot_,
const Array<double> &Bct_,
const Vector &op_,
const Vector &x_,
Vector &y_,
const int d1d = 0,
const int q1d = 0)
{
MFEM_CONTRACT_VAR(Bot_);
MFEM_CONTRACT_VAR(Bct_);
static constexpr int VDIM = 3;
const int D1D = T_D1D ? T_D1D : d1d;
const int Q1D = T_Q1D ? T_Q1D : q1d;
const auto bo = Reshape(Bo_.Read(), Q1D, D1D-1);
const auto bc = Reshape(Bc_.Read(), Q1D, D1D);
const auto D = Reshape(op_.Read(), Q1D, Q1D, Q1D, symmetric ? 6 : 9, NE);
const auto x = Reshape(x_.Read(), D1D*(D1D-1)*(D1D-1), VDIM, NE);
auto y = y_.ReadWrite();
MFEM_FORALL_3D(e, NE, Q1D, Q1D, VDIM,
{
const int tidz = MFEM_THREAD_ID(z);
const int D1D = T_D1D ? T_D1D : d1d;
const int Q1D = T_Q1D ? T_Q1D : q1d;
constexpr int MQ1 = T_Q1D ? T_Q1D : HDIV_MAX_Q1D;
constexpr int MD1 = T_D1D ? T_D1D : HDIV_MAX_D1D;
constexpr int MDQ = (MQ1 > MD1) ? MQ1 : MD1;
MFEM_SHARED double smo[MQ1*(MD1-1)];
DeviceMatrix Bo(smo, D1D-1, Q1D);
MFEM_SHARED double smc[MQ1*MD1];
DeviceMatrix Bc(smc, D1D, Q1D);
MFEM_SHARED double sm0[VDIM*MDQ*MDQ*MDQ];
MFEM_SHARED double sm1[VDIM*MDQ*MDQ*MDQ];
DeviceMatrix X(sm0, D1D*(D1D-1)*(D1D-1), VDIM);
DeviceTensor<4> QDD(sm1, Q1D, D1D, D1D, VDIM);
DeviceTensor<4> QQD(sm0, Q1D, Q1D, D1D, VDIM);
DeviceTensor<4> QQQ(sm1, Q1D, Q1D, Q1D, VDIM);
DeviceTensor<4> DQQ(sm0, D1D, Q1D, Q1D, VDIM);
DeviceTensor<4> DDQ(sm1, D1D, D1D, Q1D, VDIM);
// Load X into shared memory
MFEM_FOREACH_THREAD(vd,z,VDIM)
{
MFEM_FOREACH_THREAD(dz,y,D1D-1)
{
MFEM_FOREACH_THREAD(dy,x,D1D-1)
{
MFEM_UNROLL(MD1)
for (int dx = 0; dx < D1D; ++dx)
{
X(dx+(dy+dz*(D1D-1))*D1D,vd) = x(dx+(dy+dz*(D1D-1))*D1D,vd,e);
}
}
}
}
// Load Bo and Bc into shared memory
if (tidz == 0)
{
MFEM_FOREACH_THREAD(d,y,D1D-1)
{
MFEM_FOREACH_THREAD(q,x,Q1D)
{
Bo(d,q) = bo(q,d);
}
}
MFEM_FOREACH_THREAD(d,y,D1D)
{
MFEM_FOREACH_THREAD(q,x,Q1D)
{
Bc(d,q) = bc(q,d);
}
}
}
MFEM_SYNC_THREAD;
// Apply B operator
MFEM_FOREACH_THREAD(vd,z,VDIM)
{
const int nx = (vd == 0) ? D1D : D1D-1;
const int ny = (vd == 1) ? D1D : D1D-1;
const int nz = (vd == 2) ? D1D : D1D-1;
DeviceTensor<4> Xxyz(X, nx, ny, nz, VDIM);
DeviceMatrix Bx = (vd == 0) ? Bc : Bo;
MFEM_FOREACH_THREAD(dy,y,ny)
{
MFEM_FOREACH_THREAD(qx,x,Q1D)
{
double u[D1D];
MFEM_UNROLL(MD1)
for (int dz = 0; dz < nz; ++dz) { u[dz] = 0.0; }
MFEM_UNROLL(MD1)
for (int dx = 0; dx < nx; ++dx)
{
MFEM_UNROLL(MD1)
for (int dz = 0; dz < nz; ++dz)
{
u[dz] += Xxyz(dx,dy,dz,vd) * Bx(dx,qx);
}
}
MFEM_UNROLL(MD1)
for (int dz = 0; dz < nz; ++dz) { QDD(qx,dy,dz,vd) = u[dz]; }
}
}
}
MFEM_SYNC_THREAD;
MFEM_FOREACH_THREAD(vd,z,VDIM)
{
const int ny = (vd == 1) ? D1D : D1D-1;
const int nz = (vd == 2) ? D1D : D1D-1;
DeviceMatrix By = (vd == 1) ? Bc : Bo;
MFEM_FOREACH_THREAD(qy,y,Q1D)
{
MFEM_FOREACH_THREAD(qx,x,Q1D)
{
double u[D1D];
MFEM_UNROLL(MD1)
for (int dz = 0; dz < nz; ++dz) { u[dz] = 0.0; }
MFEM_UNROLL(MD1)
for (int dy = 0; dy < ny; ++dy)
{
MFEM_UNROLL(MD1)
for (int dz = 0; dz < nz; ++dz)
{
u[dz] += QDD(qx,dy,dz,vd) * By(dy,qy);
}
}
MFEM_UNROLL(MD1)
for (int dz = 0; dz < nz; ++dz) { QQD(qx,qy,dz,vd) = u[dz]; }
}
}
}
MFEM_SYNC_THREAD;
MFEM_FOREACH_THREAD(vd,z,VDIM)
{
const int nz = (vd == 2) ? D1D : D1D-1;
DeviceMatrix Bz = (vd == 2) ? Bc : Bo;
MFEM_FOREACH_THREAD(qy,y,Q1D)
{
MFEM_FOREACH_THREAD(qx,x,Q1D)
{
double u[Q1D];
MFEM_UNROLL(MQ1)
for (int qz = 0; qz < Q1D; ++qz) { u[qz] = 0.0; }
MFEM_UNROLL(MD1)
for (int dz = 0; dz < nz; ++dz)
{
MFEM_UNROLL(MQ1)
for (int qz = 0; qz < Q1D; ++qz)
{
u[qz] += QQD(qx,qy,dz,vd) * Bz(dz,qz);
}
}
MFEM_UNROLL(MQ1)
for (int qz = 0; qz < Q1D; ++qz) { QQQ(qx,qy,qz,vd) = u[qz]; }
}
}
}
MFEM_SYNC_THREAD;
// Apply D operator
if (tidz == 0)
{
MFEM_FOREACH_THREAD(qy,y,Q1D)
{
MFEM_FOREACH_THREAD(qx,x,Q1D)
{
MFEM_UNROLL(MQ1)
for (int qz = 0; qz < Q1D; ++qz)
{
const double Qx = QQQ(qx,qy,qz,0);
const double Qy = QQQ(qx,qy,qz,1);
const double Qz = QQQ(qx,qy,qz,2);
const double D11 = D(qx,qy,qz,0,e);
const double D12 = D(qx,qy,qz,1,e);
const double D13 = D(qx,qy,qz,2,e);
const double D21 = symmetric ? D12 : D(qx,qy,qz,3,e);
const double D22 = symmetric ? D(qx,qy,qz,3,e) : D(qx,qy,qz,4,e);
const double D23 = symmetric ? D(qx,qy,qz,4,e) : D(qx,qy,qz,5,e);
const double D31 = symmetric ? D13 : D(qx,qy,qz,6,e);
const double D32 = symmetric ? D23 : D(qx,qy,qz,7,e);
const double D33 = symmetric ? D(qx,qy,qz,5,e) : D(qx,qy,qz,8,e);
QQQ(qx,qy,qz,0) = D11*Qx + D12*Qy + D13*Qz;
QQQ(qx,qy,qz,1) = D21*Qx + D22*Qy + D23*Qz;
QQQ(qx,qy,qz,2) = D31*Qx + D32*Qy + D33*Qz;
}
}
}
}
MFEM_SYNC_THREAD;
// Apply Bt operator
MFEM_FOREACH_THREAD(vd,z,VDIM)
{
const int nx = (vd == 0) ? D1D : D1D-1;
DeviceMatrix Btx = (vd == 0) ? Bc : Bo;
MFEM_FOREACH_THREAD(qy,y,Q1D)
{
MFEM_FOREACH_THREAD(dx,x,nx)
{
double u[Q1D];
MFEM_UNROLL(MQ1)
for (int qz = 0; qz < Q1D; ++qz) { u[qz] = 0.0; }
MFEM_UNROLL(MQ1)
for (int qx = 0; qx < Q1D; ++qx)
{
MFEM_UNROLL(MQ1)
for (int qz = 0; qz < Q1D; ++qz)
{
u[qz] += QQQ(qx,qy,qz,vd) * Btx(dx,qx);
}
}
MFEM_UNROLL(MQ1)
for (int qz = 0; qz < Q1D; ++qz) { DQQ(dx,qy,qz,vd) = u[qz]; }
}
}
}
MFEM_SYNC_THREAD;
MFEM_FOREACH_THREAD(vd,z,VDIM)
{
const int nx = (vd == 0) ? D1D : D1D-1;
const int ny = (vd == 1) ? D1D : D1D-1;
DeviceMatrix Bty = (vd == 1) ? Bc : Bo;
MFEM_FOREACH_THREAD(dy,y,ny)
{
MFEM_FOREACH_THREAD(dx,x,nx)
{
double u[Q1D];
MFEM_UNROLL(MQ1)
for (int qz = 0; qz < Q1D; ++qz) { u[qz] = 0.0; }
MFEM_UNROLL(MQ1)
for (int qy = 0; qy < Q1D; ++qy)
{
MFEM_UNROLL(MQ1)
for (int qz = 0; qz < Q1D; ++qz)
{
u[qz] += DQQ(dx,qy,qz,vd) * Bty(dy,qy);
}
}
MFEM_UNROLL(MQ1)
for (int qz = 0; qz < Q1D; ++qz) { DDQ(dx,dy,qz,vd) = u[qz]; }
}
}
}
MFEM_SYNC_THREAD;
MFEM_FOREACH_THREAD(vd,z,VDIM)
{
const int nx = (vd == 0) ? D1D : D1D-1;
const int ny = (vd == 1) ? D1D : D1D-1;
const int nz = (vd == 2) ? D1D : D1D-1;
DeviceTensor<5> Yxyz(y, nx, ny, nz, VDIM, NE);
DeviceMatrix Btz = (vd == 2) ? Bc : Bo;
MFEM_FOREACH_THREAD(dy,y,ny)
{
MFEM_FOREACH_THREAD(dx,x,nx)
{
double u[D1D];
MFEM_UNROLL(MD1)
for (int dz = 0; dz < nz; ++dz) { u[dz] = 0.0; }
MFEM_UNROLL(MQ1)
for (int qz = 0; qz < Q1D; ++qz)
{
MFEM_UNROLL(MD1)
for (int dz = 0; dz < nz; ++dz)
{
u[dz] += DDQ(dx,dy,qz,vd) * Btz(dz,qz);
}
}
MFEM_UNROLL(MD1)
for (int dz = 0; dz < nz; ++dz) { Yxyz(dx,dy,dz,vd,e) += u[dz]; }
}
}
}
MFEM_SYNC_THREAD;
});
}
void PAHdivMassApply(const int dim,
const int D1D,
const int Q1D,
const int NE,
const bool symmetric,
const Array<double> &Bo,
const Array<double> &Bc,
const Array<double> &Bot,
const Array<double> &Bct,
const Vector &op,
const Vector &x,
Vector &y)
{
const int id = (D1D << 4) | Q1D;
if (dim == 2)
{
switch (id)
{
case 0x22: return SmemPAHdivMassApply2D<2,2>(NE,symmetric,Bo,Bc,Bot,Bct,op,x,y);
case 0x33: return SmemPAHdivMassApply2D<3,3>(NE,symmetric,Bo,Bc,Bot,Bct,op,x,y);
case 0x44: return SmemPAHdivMassApply2D<4,4>(NE,symmetric,Bo,Bc,Bot,Bct,op,x,y);
case 0x55: return SmemPAHdivMassApply2D<5,5>(NE,symmetric,Bo,Bc,Bot,Bct,op,x,y);
default: // fallback
return PAHdivMassApply2D(D1D,Q1D,NE,symmetric,Bo,Bc,Bot,Bct,op,x,y);
}
}
else if (dim == 3)
{
switch (id)
{
case 0x23: return SmemPAHdivMassApply3D<2,3>(NE,symmetric,Bo,Bc,Bot,Bct,op,x,y);
case 0x34: return SmemPAHdivMassApply3D<3,4>(NE,symmetric,Bo,Bc,Bot,Bct,op,x,y);
case 0x45: return SmemPAHdivMassApply3D<4,5>(NE,symmetric,Bo,Bc,Bot,Bct,op,x,y);
case 0x56: return SmemPAHdivMassApply3D<5,6>(NE,symmetric,Bo,Bc,Bot,Bct,op,x,y);
case 0x67: return SmemPAHdivMassApply3D<6,7>(NE,symmetric,Bo,Bc,Bot,Bct,op,x,y);
case 0x78: return SmemPAHdivMassApply3D<7,8>(NE,symmetric,Bo,Bc,Bot,Bct,op,x,y);
default: // fallback
return PAHdivMassApply3D(D1D,Q1D,NE,symmetric,Bo,Bc,Bot,Bct,op,x,y);
}
}
}
// PA H(div) div-div assemble 2D kernel
// NOTE: this is identical to PACurlCurlSetup3D
static void PADivDivSetup2D(const int Q1D,
@@ -1214,7 +626,7 @@ static void PADivDivApply2D(const int D1D,
{
double div[MAX_Q1D][MAX_Q1D];
// div[qy][qx] will be computed as du_x/dx + du_y/dy
// div[qy][qx] will be computed as du_x/dx + duy_/dy
for (int qy = 0; qy < Q1D; ++qy)
{
@@ -1797,13 +1209,6 @@ VectorFEDivergenceIntegrator::AssemblePA(const FiniteElementSpace &trial_fes,
}
}
if (test_el->GetMapType() == FiniteElement::INTEGRAL)
{
const GeometricFactors *geom =
mesh->GetGeometricFactors(*ir, GeometricFactors::DETERMINANTS);
coeff /= geom->detJ;
}
if (trial_el->GetDerivType() == mfem::FiniteElement::DIV && dim == 3)
{
PADivL2Setup3D(quad1D, ne, ir->GetWeights(), coeff, pa_data);
+2 -11
View File
@@ -12,7 +12,7 @@
#include "../general/forall.hpp"
#include "bilininteg.hpp"
#include "gridfunc.hpp"
#include "ceed/integrators/mass/mass.hpp"
#include "ceed/mass.hpp"
using namespace std;
@@ -31,16 +31,7 @@ void MassIntegrator::AssembleMF(const FiniteElementSpace &fes)
if (DeviceCanUseCeed())
{
delete ceedOp;
const bool mixed = mesh->GetNumGeometries(mesh->Dimension()) > 1 ||
fes.IsVariableOrder();
if (mixed)
{
ceedOp = new ceed::MixedMFMassIntegrator(*this, fes, Q);
}
else
{
ceedOp = new ceed::MFMassIntegrator(fes, *ir, Q);
}
ceedOp = new ceed::MFMassIntegrator(fes, *ir, Q);
return;
}
MFEM_ABORT("Error: MassIntegrator::AssembleMF only implemented with"
+2 -11
View File
@@ -12,7 +12,7 @@
#include "../general/forall.hpp"
#include "bilininteg.hpp"
#include "gridfunc.hpp"
#include "ceed/integrators/mass/mass.hpp"
#include "ceed/mass.hpp"
using namespace std;
@@ -38,16 +38,7 @@ void MassIntegrator::AssemblePA(const FiniteElementSpace &fes)
if (DeviceCanUseCeed())
{
delete ceedOp;
const bool mixed = mesh->GetNumGeometries(mesh->Dimension()) > 1 ||
fes.IsVariableOrder();
if (mixed)
{
ceedOp = new ceed::MixedPAMassIntegrator(*this, fes, Q);
}
else
{
ceedOp = new ceed::PAMassIntegrator(fes, *ir, Q);
}
ceedOp = new ceed::PAMassIntegrator(fes, *ir, Q);
return;
}
int map_type = el.GetMapType();
+2 -11
View File
@@ -12,7 +12,7 @@
#include "../general/forall.hpp"
#include "bilininteg.hpp"
#include "gridfunc.hpp"
#include "ceed/integrators/diffusion/diffusion.hpp"
#include "ceed/diffusion.hpp"
using namespace std;
@@ -149,16 +149,7 @@ void VectorDiffusionIntegrator::AssemblePA(const FiniteElementSpace &fes)
if (DeviceCanUseCeed())
{
delete ceedOp;
const bool mixed = mesh->GetNumGeometries(mesh->Dimension()) > 1 ||
fes.IsVariableOrder();
if (mixed)
{
ceedOp = new ceed::MixedPADiffusionIntegrator(*this, fes, Q);
}
else
{
ceedOp = new ceed::PADiffusionIntegrator(fes, *ir, Q);
}
ceedOp = new ceed::PADiffusionIntegrator(fes, *ir, Q);
return;
}
const int dims = el.GetDim();
+2 -14
View File
@@ -12,7 +12,7 @@
#include "../general/forall.hpp"
#include "bilininteg.hpp"
#include "gridfunc.hpp"
#include "ceed/integrators/diffusion/diffusion.hpp"
#include "ceed/diffusion.hpp"
using namespace std;
@@ -30,19 +30,7 @@ void VectorDiffusionIntegrator::AssembleMF(const FiniteElementSpace &fes)
if (DeviceCanUseCeed())
{
delete ceedOp;
MFEM_VERIFY(!VQ && !MQ,
"Only scalar coefficient supported for DiffusionIntegrator"
" with libCEED");
const bool mixed = mesh->GetNumGeometries(mesh->Dimension()) > 1 ||
fes.IsVariableOrder();
if (mixed)
{
ceedOp = new ceed::MixedMFDiffusionIntegrator(*this, fes, Q);
}
else
{
ceedOp = new ceed::MFDiffusionIntegrator(fes, *ir, Q);
}
ceedOp = new ceed::MFDiffusionIntegrator(fes, *ir, Q);
return;
}
MFEM_ABORT("Error: VectorDiffusionIntegrator::AssembleMF only implemented"
+2 -11
View File
@@ -12,7 +12,7 @@
#include "../general/forall.hpp"
#include "bilininteg.hpp"
#include "gridfunc.hpp"
#include "ceed/integrators/mass/mass.hpp"
#include "ceed/mass.hpp"
using namespace std;
@@ -34,16 +34,7 @@ void VectorMassIntegrator::AssemblePA(const FiniteElementSpace &fes)
if (DeviceCanUseCeed())
{
delete ceedOp;
const bool mixed = mesh->GetNumGeometries(mesh->Dimension()) > 1 ||
fes.IsVariableOrder();
if (mixed)
{
ceedOp = new ceed::MixedPAMassIntegrator(*this, fes, Q);
}
else
{
ceedOp = new ceed::PAMassIntegrator(fes, *ir, Q);
}
ceedOp = new ceed::PAMassIntegrator(fes, *ir, Q);
return;
}
dim = mesh->Dimension();
+2 -11
View File
@@ -12,7 +12,7 @@
#include "../general/forall.hpp"
#include "bilininteg.hpp"
#include "gridfunc.hpp"
#include "ceed/integrators/mass/mass.hpp"
#include "ceed/mass.hpp"
using namespace std;
@@ -34,16 +34,7 @@ void VectorMassIntegrator::AssembleMF(const FiniteElementSpace &fes)
if (DeviceCanUseCeed())
{
delete ceedOp;
const bool mixed = mesh->GetNumGeometries(mesh->Dimension()) > 1 ||
fes.IsVariableOrder();
if (mixed)
{
ceedOp = new ceed::MixedMFMassIntegrator(*this, fes, Q);
}
else
{
ceedOp = new ceed::MFMassIntegrator(fes, *ir, Q);
}
ceedOp = new ceed::MFMassIntegrator(fes, *ir, Q);
return;
}
MFEM_ABORT("Error: VectorMassIntegrator::AssembleMF only implemented with"
+83 -95
View File
@@ -11,7 +11,6 @@
#include "../general/forall.hpp"
#include "bilininteg.hpp"
#include "gridfunc.hpp"
namespace mfem
{
@@ -90,7 +89,6 @@ void SmemPAHcurlMassApply3D(const int D1D,
Vector &y);
void PAHdivSetup2D(const int Q1D,
const int coeffDim,
const int NE,
const Array<double> &w,
const Vector &j,
@@ -98,7 +96,6 @@ void PAHdivSetup2D(const int Q1D,
Vector &op);
void PAHdivSetup3D(const int Q1D,
const int coeffDim,
const int NE,
const Array<double> &w,
const Vector &j,
@@ -152,7 +149,6 @@ void PAHcurlH1ApplyTranspose3D(const int D1D,
void PAHdivMassAssembleDiagonal2D(const int D1D,
const int Q1D,
const int NE,
const bool symmetric,
const Array<double> &Bo_,
const Array<double> &Bc_,
const Vector &op_,
@@ -161,24 +157,32 @@ void PAHdivMassAssembleDiagonal2D(const int D1D,
void PAHdivMassAssembleDiagonal3D(const int D1D,
const int Q1D,
const int NE,
const bool symmetric,
const Array<double> &Bo_,
const Array<double> &Bc_,
const Vector &op_,
Vector &diag_);
void PAHdivMassApply(const int dim,
const int D1D,
const int Q1D,
const int NE,
const bool symmetric,
const Array<double> &Bo,
const Array<double> &Bc,
const Array<double> &Bot,
const Array<double> &Bct,
const Vector &op,
const Vector &x,
Vector &y);
void PAHdivMassApply2D(const int D1D,
const int Q1D,
const int NE,
const Array<double> &Bo_,
const Array<double> &Bc_,
const Array<double> &Bot_,
const Array<double> &Bct_,
const Vector &op_,
const Vector &x_,
Vector &y_);
void PAHdivMassApply3D(const int D1D,
const int Q1D,
const int NE,
const Array<double> &Bo_,
const Array<double> &Bc_,
const Array<double> &Bot_,
const Array<double> &Bct_,
const Vector &op_,
const Vector &x_,
Vector &y_);
void PAHcurlL2Setup(const int NQ,
const int coeffDim,
@@ -814,80 +818,69 @@ void VectorFEMassIntegrator::AssemblePA(const FiniteElementSpace &trial_fes,
pa_data.SetSize((symmetric ? symmDims : MQfullDim) * nq * ne,
Device::GetMemoryType());
Vector coeff;
Vector coeff(coeffDim * ne * nq);
coeff = 1.0;
auto coeffh = Reshape(coeff.HostWrite(), coeffDim, nq, ne);
if (Q || DQ || MQ)
{
Vector DM(DQ ? coeffDim : 0);
DenseMatrix M;
DenseSymmetricMatrix SM;
auto *qf_c = dynamic_cast<QuadratureFunctionCoefficient*>(Q);
if (qf_c)
{
const QuadratureFunction &qf = qf_c->GetQuadFunction();
qf.Read();
coeff.MakeRef(const_cast<QuadratureFunction&>(qf), 0);
}
else
{
coeff.SetSize(coeffDim * ne * nq);
coeff = 1.0;
auto coeffh = Reshape(coeff.HostWrite(), coeffDim, nq, ne);
if (Q || DQ || MQ)
if (DQ)
{
Vector DM(DQ ? coeffDim : 0);
DenseMatrix M;
DenseSymmetricMatrix SM;
MFEM_VERIFY(coeffDim == dim, "");
}
if (SMQ)
{
MFEM_VERIFY(SMQ->GetSize() == dim, "");
SM.SetSize(dim);
}
else if (MQ)
{
MFEM_VERIFY(coeffDim == MQdim, "");
MFEM_VERIFY(MQ->GetHeight() == dim && MQ->GetWidth() == dim, "");
M.SetSize(dim);
}
if (DQ)
{
MFEM_VERIFY(coeffDim == dim, "");
}
if (SMQ)
{
MFEM_VERIFY(SMQ->GetSize() == dim, "");
SM.SetSize(dim);
}
else if (MQ)
{
MFEM_VERIFY(coeffDim == MQdim, "");
MFEM_VERIFY(MQ->GetHeight() == dim && MQ->GetWidth() == dim, "");
M.SetSize(dim);
}
for (int e=0; e<ne; ++e)
for (int e=0; e<ne; ++e)
{
ElementTransformation *tr = mesh->GetElementTransformation(e);
for (int p=0; p<nq; ++p)
{
ElementTransformation *tr = mesh->GetElementTransformation(e);
for (int p=0; p<nq; ++p)
if (SMQ)
{
if (SMQ)
{
SMQ->Eval(SM, *tr, ir->IntPoint(p));
int cnt = 0;
for (int i=0; i<dim; ++i)
for (int j=i; j<dim; ++j, ++cnt)
{
coeffh(cnt, p, e) = SM(i,j);
}
}
else if (MQ)
{
MQ->Eval(M, *tr, ir->IntPoint(p));
for (int i=0; i<dim; ++i)
for (int j=0; j<dim; ++j)
{
coeffh(j+(i*dim), p, e) = M(i,j);
}
}
else if (DQ)
{
DQ->Eval(DM, *tr, ir->IntPoint(p));
for (int i=0; i<coeffDim; ++i)
SMQ->Eval(SM, *tr, ir->IntPoint(p));
int cnt = 0;
for (int i=0; i<dim; ++i)
for (int j=i; j<dim; ++j, ++cnt)
{
coeffh(i, p, e) = DM[i];
coeffh(cnt, p, e) = SM(i,j);
}
}
else
}
else if (MQ)
{
MQ->Eval(M, *tr, ir->IntPoint(p));
for (int i=0; i<dim; ++i)
for (int j=0; j<dim; ++j)
{
coeffh(j+(i*dim), p, e) = M(i,j);
}
}
else if (DQ)
{
DQ->Eval(DM, *tr, ir->IntPoint(p));
for (int i=0; i<coeffDim; ++i)
{
coeffh(0, p, e) = Q->Eval(*tr, ir->IntPoint(p));
coeffh(i, p, e) = DM[i];
}
}
else
{
coeffh(0, p, e) = Q->Eval(*tr, ir->IntPoint(p));
}
}
}
}
@@ -904,12 +897,12 @@ void VectorFEMassIntegrator::AssemblePA(const FiniteElementSpace &trial_fes,
}
else if (trial_div && test_div && dim == 3)
{
PAHdivSetup3D(quad1D, coeffDim, ne, ir->GetWeights(), geom->J,
PAHdivSetup3D(quad1D, ne, ir->GetWeights(), geom->J,
coeff, pa_data);
}
else if (trial_div && test_div && dim == 2)
{
PAHdivSetup2D(quad1D, coeffDim, ne, ir->GetWeights(), geom->J,
PAHdivSetup2D(quad1D, ne, ir->GetWeights(), geom->J,
coeff, pa_data);
}
else if (((trial_curl && test_div) || (trial_div && test_curl)) &&
@@ -970,7 +963,7 @@ void VectorFEMassIntegrator::AssembleDiagonalPA(Vector& diag)
else if (trial_fetype == mfem::FiniteElement::DIV &&
test_fetype == trial_fetype)
{
PAHdivMassAssembleDiagonal3D(dofs1D, quad1D, ne, symmetric,
PAHdivMassAssembleDiagonal3D(dofs1D, quad1D, ne,
mapsO->B, mapsC->B, pa_data, diag);
}
else
@@ -978,7 +971,7 @@ void VectorFEMassIntegrator::AssembleDiagonalPA(Vector& diag)
MFEM_ABORT("Unknown kernel.");
}
}
else // 2D
else
{
if (trial_fetype == mfem::FiniteElement::CURL && test_fetype == trial_fetype)
{
@@ -988,7 +981,7 @@ void VectorFEMassIntegrator::AssembleDiagonalPA(Vector& diag)
else if (trial_fetype == mfem::FiniteElement::DIV &&
test_fetype == trial_fetype)
{
PAHdivMassAssembleDiagonal2D(dofs1D, quad1D, ne, symmetric,
PAHdivMassAssembleDiagonal2D(dofs1D, quad1D, ne,
mapsO->B, mapsC->B, pa_data, diag);
}
else
@@ -1041,8 +1034,8 @@ void VectorFEMassIntegrator::AddMultPA(const Vector &x, Vector &y) const
}
else if (trial_div && test_div)
{
PAHdivMassApply(3, dofs1D, quad1D, ne, symmetric, mapsO->B, mapsC->B, mapsO->Bt,
mapsC->Bt, pa_data, x, y);
PAHdivMassApply3D(dofs1D, quad1D, ne, mapsO->B, mapsC->B, mapsO->Bt,
mapsC->Bt, pa_data, x, y);
}
else if (trial_curl && test_div)
{
@@ -1063,7 +1056,7 @@ void VectorFEMassIntegrator::AddMultPA(const Vector &x, Vector &y) const
MFEM_ABORT("Unknown kernel.");
}
}
else // 2D
else
{
if (trial_curl && test_curl)
{
@@ -1072,8 +1065,8 @@ void VectorFEMassIntegrator::AddMultPA(const Vector &x, Vector &y) const
}
else if (trial_div && test_div)
{
PAHdivMassApply(2, dofs1D, quad1D, ne, symmetric, mapsO->B, mapsC->B, mapsO->Bt,
mapsC->Bt, pa_data, x, y);
PAHdivMassApply2D(dofs1D, quad1D, ne, mapsO->B, mapsC->B, mapsO->Bt,
mapsC->Bt, pa_data, x, y);
}
else if ((trial_curl && test_div) || (trial_div && test_curl))
{
@@ -1118,11 +1111,6 @@ void VectorFEMassIntegrator::AddMultTransposePA(const Vector &x,
if (symmetricSpaces)
{
if (MQ && dynamic_cast<SymmetricMatrixCoefficient*>(MQ) == NULL)
{
MFEM_ABORT("VectorFEMassIntegrator transpose not implemented for asymmetric MatrixCoefficient");
}
this->AddMultPA(x, y);
}
}
@@ -11,14 +11,12 @@
#include "algebraic.hpp"
#include "../../bilinearform.hpp"
#include "../../fespace.hpp"
#include "../../pfespace.hpp"
#include "../../../general/forall.hpp"
#include "solvers-atpmg.hpp"
#include "full-assembly.hpp"
#include "../interface/restriction.hpp"
#include "../interface/ceed.hpp"
#include "../bilinearform.hpp"
#include "../fespace.hpp"
#include "../ceed/solvers-atpmg.hpp"
#include "../ceed/full-assembly.hpp"
#include "../pfespace.hpp"
#include "../../general/forall.hpp"
namespace mfem
{
@@ -633,7 +631,7 @@ AlgebraicSpaceHierarchy::AlgebraicSpaceHierarchy(FiniteElementSpace &fes)
current_order = order;
Ceed ceed = internal::ceed;
InitRestriction(fes, ceed, &fine_er);
InitTensorRestriction(fes, ceed, &fine_er);
CeedElemRestriction er = fine_er;
int dim = fes.GetMesh()->Dimension();
@@ -12,10 +12,12 @@
#ifndef MFEM_CEED_ALGEBRAIC_HPP
#define MFEM_CEED_ALGEBRAIC_HPP
#include "../../fespacehierarchy.hpp"
#include "../../multigrid.hpp"
#include "../interface/operator.hpp"
#include "../interface/ceed.hpp"
#include "../../config/config.hpp"
#include "../fespacehierarchy.hpp"
#include "../multigrid.hpp"
#include "util.hpp"
#include "operator.hpp"
namespace mfem
{
@@ -134,6 +136,7 @@ public:
}
~AlgebraicSpaceHierarchy()
{
CeedElemRestrictionDestroy(&fine_er);
for (int i=0; i<R_tr.Size(); ++i)
{
delete R_tr[i];
+222
View File
@@ -0,0 +1,222 @@
// Copyright (c) 2010-2022, Lawrence Livermore National Security, LLC. Produced
// at the Lawrence Livermore National Laboratory. All Rights reserved. See files
// LICENSE and NOTICE for details. LLNL-CODE-806117.
//
// This file is part of the MFEM library. For more information and source code
// availability visit https://mfem.org.
//
// MFEM is free software; you can redistribute it and/or modify it under the
// terms of the BSD-3 license. We welcome feedback and contributions, see file
// CONTRIBUTING.md for details.
#ifndef MFEM_LIBCEED_COEFF
#define MFEM_LIBCEED_COEFF
#include "../../config/config.hpp"
#ifdef MFEM_USE_CEED
#include "util.hpp"
#include "../../linalg/vector.hpp"
#include "../../linalg/dtensor.hpp"
#include "../../mesh/mesh.hpp"
#include "../gridfunc.hpp"
#include "../../linalg/dtensor.hpp"
namespace mfem
{
class Mesh;
class IntegrationRule;
class Coefficient;
class VectorCoefficient;
class GridFunction;
namespace ceed
{
struct Coefficient
{
const int ncomp;
Coefficient(int ncomp_) : ncomp(ncomp_) { }
virtual bool IsConstant() const { return true; }
virtual ~Coefficient() { }
};
struct VariableCoefficient : Coefficient
{
CeedVector coeffVector = nullptr;
const CeedEvalMode emode;
VariableCoefficient(int ncomp_, CeedEvalMode emode_)
: Coefficient(ncomp_), emode(emode_) { }
virtual bool IsConstant() const override { return false; }
~VariableCoefficient()
{
CeedVectorDestroy(&coeffVector);
}
};
struct GridCoefficient : VariableCoefficient
{
const mfem::GridFunction &gf;
CeedBasis basis;
CeedElemRestriction restr;
GridCoefficient(const mfem::GridFunction &gf_)
: VariableCoefficient(gf_.VectorDim(), CEED_EVAL_INTERP),
gf(gf_)
{
InitVector(gf, coeffVector);
}
};
struct QuadCoefficient : VariableCoefficient
{
mfem::Vector coeff;
CeedElemRestriction restr;
QuadCoefficient(int ncomp_) : VariableCoefficient(ncomp_, CEED_EVAL_NONE) { }
};
/** @brief Initializes an mfem::ceed::Coefficient @a coeff_ptr from an
mfem::Coefficient @a Q, an mfem::Mesh @a mesh, and an mfem::IntegrationRule
@a ir. */
template <typename Context>
void InitCoefficient(mfem::Coefficient *Q, mfem::Mesh &mesh,
const mfem::IntegrationRule &ir,
Coefficient*& coeff_ptr, Context &ctx)
{
if ( Q == nullptr )
{
Coefficient *ceedCoeff = new Coefficient(1);
ctx.coeff = 1.0;
coeff_ptr = ceedCoeff;
}
else if (ConstantCoefficient *const_coeff =
dynamic_cast<ConstantCoefficient*>(Q))
{
Coefficient *ceedCoeff = new Coefficient(1);
ctx.coeff = const_coeff->constant;
coeff_ptr = ceedCoeff;
}
else if (GridFunctionCoefficient* gf_coeff =
dynamic_cast<GridFunctionCoefficient*>(Q))
{
GridCoefficient *ceedCoeff =
new GridCoefficient(*gf_coeff->GetGridFunction());
coeff_ptr = ceedCoeff;
}
else if (QuadratureFunctionCoefficient *cQ =
dynamic_cast<QuadratureFunctionCoefficient*>(Q))
{
QuadCoefficient *ceedCoeff = new QuadCoefficient(1);
const int ne = mesh.GetNE();
const int nq = ir.GetNPoints();
const mfem::QuadratureFunction &qFun = cQ->GetQuadFunction();
MFEM_VERIFY(qFun.Size() == nq * ne,
"Incompatible QuadratureFunction dimension \n");
MFEM_VERIFY(&ir == &qFun.GetSpace()->GetElementIntRule(0),
"IntegrationRule used within integrator and in"
" QuadratureFunction appear to be different");
qFun.Read();
ceedCoeff->coeff.MakeRef(const_cast<mfem::QuadratureFunction &>(qFun),0);
InitVector(ceedCoeff->coeff, ceedCoeff->coeffVector);
coeff_ptr = ceedCoeff;
}
else
{
QuadCoefficient *ceedCoeff = new QuadCoefficient(1);
const int ne = mesh.GetNE();
const int nq = ir.GetNPoints();
ceedCoeff->coeff.SetSize(nq * ne);
auto C = Reshape(ceedCoeff->coeff.HostWrite(), nq, ne);
for (int e = 0; e < ne; ++e)
{
mfem::ElementTransformation &T = *mesh.GetElementTransformation(e);
for (int q = 0; q < nq; ++q)
{
C(q,e) = Q->Eval(T, ir.IntPoint(q));
}
}
InitVector(ceedCoeff->coeff, ceedCoeff->coeffVector);
coeff_ptr = ceedCoeff;
}
}
/** @brief Initializes an mfem::ceed::Coefficient @a coeff_ptr from an
mfem::VectorCoefficient @a Q, an mfem::Mesh @a mesh, and an
mfem::IntegrationRule @a ir. */
template <typename Context>
void InitCoefficient(mfem::VectorCoefficient *VQ, mfem::Mesh &mesh,
const mfem::IntegrationRule &ir,
Coefficient *&coeff_ptr, Context &ctx)
{
if (VectorConstantCoefficient *const_coeff =
dynamic_cast<VectorConstantCoefficient*>(VQ))
{
const int vdim = const_coeff->GetVDim();
const mfem::Vector &val = const_coeff->GetVec();
Coefficient *ceedCoeff = new Coefficient(vdim);
for (int i = 0; i < vdim; i++)
{
ctx.coeff[i] = val[i];
}
coeff_ptr = ceedCoeff;
}
else if (VectorGridFunctionCoefficient* vgf_coeff =
dynamic_cast<VectorGridFunctionCoefficient*>(VQ))
{
GridCoefficient *ceedCoeff =
new GridCoefficient(*vgf_coeff->GetGridFunction());
coeff_ptr = ceedCoeff;
}
else if (VectorQuadratureFunctionCoefficient *cQ =
dynamic_cast<VectorQuadratureFunctionCoefficient*>(VQ))
{
QuadCoefficient *ceedCoeff = new QuadCoefficient(cQ->GetVDim());
const int dim = mesh.Dimension();
const int ne = mesh.GetNE();
const int nq = ir.GetNPoints();
const mfem::QuadratureFunction &qFun = cQ->GetQuadFunction();
MFEM_VERIFY(qFun.Size() == dim * nq * ne,
"Incompatible QuadratureFunction dimension \n");
MFEM_VERIFY(&ir == &qFun.GetSpace()->GetElementIntRule(0),
"IntegrationRule used within integrator and in"
" QuadratureFunction appear to be different");
qFun.Read();
ceedCoeff->coeff.MakeRef(const_cast<mfem::QuadratureFunction &>(qFun),0);
InitVector(ceedCoeff->coeff, ceedCoeff->coeffVector);
coeff_ptr = ceedCoeff;
}
else
{
const int dim = mesh.Dimension();
QuadCoefficient *ceedCoeff = new QuadCoefficient(dim);
const int ne = mesh.GetNE();
const int nq = ir.GetNPoints();
ceedCoeff->coeff.SetSize(dim * nq * ne);
auto C = Reshape(ceedCoeff->coeff.HostWrite(), dim, nq, ne);
mfem::DenseMatrix Q_ir;
for (int e = 0; e < ne; ++e)
{
mfem::ElementTransformation &T = *mesh.GetElementTransformation(e);
VQ->Eval(Q_ir, T, ir);
for (int q = 0; q < nq; ++q)
{
for (int i = 0; i < dim; ++i)
{
C(i,q,e) = Q_ir(i,q);
}
}
}
InitVector(ceedCoeff->coeff, ceedCoeff->coeffVector);
coeff_ptr = ceedCoeff;
}
}
} // namespace ceed
} // namespace mfem
#endif
#endif // MFEM_LIBCEED_COEFF
@@ -11,7 +11,7 @@
#include "convection.hpp"
#include "../../../../config/config.hpp"
#include "../../config/config.hpp"
#ifdef MFEM_USE_CEED
#include "convection_qf.h"
#endif
@@ -28,7 +28,7 @@ struct ConvectionOperatorInfo : public OperatorInfo
ConvectionContext ctx;
ConvectionOperatorInfo(int dim, double alpha)
{
header = "/integrators/convection/convection_qf.h";
header = "/convection_qf.h";
build_func_const = ":f_build_conv_const";
build_qf_const = &f_build_conv_const;
build_func_quad = ":f_build_conv_quad";
@@ -62,20 +62,6 @@ PAConvectionIntegrator::PAConvectionIntegrator(
#endif
}
MixedPAConvectionIntegrator::MixedPAConvectionIntegrator(
const ConvectionIntegrator &integ,
const mfem::FiniteElementSpace &fes,
mfem::VectorCoefficient *Q,
const double alpha)
{
#ifdef MFEM_USE_CEED
ConvectionOperatorInfo info(fes.GetMesh()->Dimension(), alpha);
Assemble(integ, info, fes, Q);
#else
MFEM_ABORT("MFEM must be built with MFEM_USE_CEED=YES to use libCEED.");
#endif
}
MFConvectionIntegrator::MFConvectionIntegrator(
const mfem::FiniteElementSpace &fes,
const mfem::IntegrationRule &irm,
@@ -91,20 +77,6 @@ MFConvectionIntegrator::MFConvectionIntegrator(
#endif
}
MixedMFConvectionIntegrator::MixedMFConvectionIntegrator(
const ConvectionIntegrator &integ,
const mfem::FiniteElementSpace &fes,
mfem::VectorCoefficient *Q,
const double alpha)
{
#ifdef MFEM_USE_CEED
ConvectionOperatorInfo info(fes.GetMesh()->Dimension(), alpha);
Assemble(integ, info, fes, Q);
#else
MFEM_ABORT("MFEM must be built with MFEM_USE_CEED=YES to use libCEED.");
#endif
}
} // namespace ceed
} // namespace mfem
@@ -12,9 +12,8 @@
#ifndef MFEM_LIBCEED_CONV_HPP
#define MFEM_LIBCEED_CONV_HPP
#include "../../interface/integrator.hpp"
#include "../../interface/mixed_integrator.hpp"
#include "../../../fespace.hpp"
#include "integrator.hpp"
#include "../fespace.hpp"
namespace mfem
{
@@ -27,39 +26,21 @@ class PAConvectionIntegrator : public PAIntegrator
{
public:
PAConvectionIntegrator(const mfem::FiniteElementSpace &fes,
const mfem::IntegrationRule &ir,
const mfem::IntegrationRule &irm,
mfem::VectorCoefficient *Q,
const double alpha);
};
class MixedPAConvectionIntegrator : public MixedIntegrator<PAIntegrator>
{
public:
MixedPAConvectionIntegrator(const ConvectionIntegrator &integ,
const mfem::FiniteElementSpace &fes,
mfem::VectorCoefficient *Q,
const double alpha);
};
/// Represent a ConvectionIntegrator with AssemblyLevel::None using libCEED.
class MFConvectionIntegrator : public MFIntegrator
{
public:
MFConvectionIntegrator(const mfem::FiniteElementSpace &fes,
const mfem::IntegrationRule &ir,
const mfem::IntegrationRule &irm,
mfem::VectorCoefficient *Q,
const double alpha);
};
class MixedMFConvectionIntegrator : public MixedIntegrator<MFIntegrator>
{
public:
MixedMFConvectionIntegrator(const ConvectionIntegrator &integ,
const mfem::FiniteElementSpace &fes,
mfem::VectorCoefficient *Q,
const double alpha);
};
}
}
@@ -11,7 +11,7 @@
#include "diffusion.hpp"
#include "../../../../config/config.hpp"
#include "../../config/config.hpp"
#ifdef MFEM_USE_CEED
#include "diffusion_qf.h"
#endif
@@ -28,7 +28,7 @@ struct DiffusionOperatorInfo : public OperatorInfo
DiffusionContext ctx;
DiffusionOperatorInfo(int dim)
{
header = "/integrators/diffusion/diffusion_qf.h";
header = "/diffusion_qf.h";
build_func_const = ":f_build_diff_const";
build_qf_const = &f_build_diff_const;
build_func_quad = ":f_build_diff_quad";
@@ -60,32 +60,6 @@ PADiffusionIntegrator::PADiffusionIntegrator(
#endif
}
MixedPADiffusionIntegrator::MixedPADiffusionIntegrator(
const DiffusionIntegrator &integ,
const mfem::FiniteElementSpace &fes,
mfem::Coefficient *Q)
{
#ifdef MFEM_USE_CEED
DiffusionOperatorInfo info(fes.GetMesh()->Dimension());
Assemble(integ, info, fes, Q);
#else
MFEM_ABORT("MFEM must be built with MFEM_USE_CEED=YES to use libCEED.");
#endif
}
MixedPADiffusionIntegrator::MixedPADiffusionIntegrator(
const VectorDiffusionIntegrator &integ,
const mfem::FiniteElementSpace &fes,
mfem::Coefficient *Q)
{
#ifdef MFEM_USE_CEED
DiffusionOperatorInfo info(fes.GetMesh()->Dimension());
Assemble(integ, info, fes, Q);
#else
MFEM_ABORT("MFEM must be built with MFEM_USE_CEED=YES to use libCEED.");
#endif
}
MFDiffusionIntegrator::MFDiffusionIntegrator(
const mfem::FiniteElementSpace &fes,
const mfem::IntegrationRule &irm,
@@ -100,32 +74,6 @@ MFDiffusionIntegrator::MFDiffusionIntegrator(
#endif
}
MixedMFDiffusionIntegrator::MixedMFDiffusionIntegrator(
const DiffusionIntegrator &integ,
const mfem::FiniteElementSpace &fes,
mfem::Coefficient *Q)
{
#ifdef MFEM_USE_CEED
DiffusionOperatorInfo info(fes.GetMesh()->Dimension());
Assemble(integ, info, fes, Q);
#else
MFEM_ABORT("MFEM must be built with MFEM_USE_CEED=YES to use libCEED.");
#endif
}
MixedMFDiffusionIntegrator::MixedMFDiffusionIntegrator(
const VectorDiffusionIntegrator &integ,
const mfem::FiniteElementSpace &fes,
mfem::Coefficient *Q)
{
#ifdef MFEM_USE_CEED
DiffusionOperatorInfo info(fes.GetMesh()->Dimension());
Assemble(integ, info, fes, Q);
#else
MFEM_ABORT("MFEM must be built with MFEM_USE_CEED=YES to use libCEED.");
#endif
}
} // namespace ceed
} // namespace mfem
@@ -12,9 +12,8 @@
#ifndef MFEM_LIBCEED_DIFF_HPP
#define MFEM_LIBCEED_DIFF_HPP
#include "../../interface/integrator.hpp"
#include "../../interface/mixed_integrator.hpp"
#include "../../../fespace.hpp"
#include "integrator.hpp"
#include "../fespace.hpp"
namespace mfem
{
@@ -27,43 +26,19 @@ class PADiffusionIntegrator : public PAIntegrator
{
public:
PADiffusionIntegrator(const mfem::FiniteElementSpace &fes,
const mfem::IntegrationRule &ir,
const mfem::IntegrationRule &irm,
mfem::Coefficient *Q);
};
class MixedPADiffusionIntegrator : public MixedIntegrator<PAIntegrator>
{
public:
MixedPADiffusionIntegrator(const DiffusionIntegrator &integ,
const mfem::FiniteElementSpace &fes,
mfem::Coefficient *Q);
MixedPADiffusionIntegrator(const VectorDiffusionIntegrator &integ,
const mfem::FiniteElementSpace &fes,
mfem::Coefficient *Q);
};
/// Represent a DiffusionIntegrator with AssemblyLevel::None using libCEED.
class MFDiffusionIntegrator : public MFIntegrator
{
public:
MFDiffusionIntegrator(const mfem::FiniteElementSpace &fes,
const mfem::IntegrationRule &ir,
const mfem::IntegrationRule &irm,
mfem::Coefficient *Q);
};
class MixedMFDiffusionIntegrator : public MixedIntegrator<MFIntegrator>
{
public:
MixedMFDiffusionIntegrator(const DiffusionIntegrator &integ,
const mfem::FiniteElementSpace &fes,
mfem::Coefficient *Q);
MixedMFDiffusionIntegrator(const VectorDiffusionIntegrator &integ,
const mfem::FiniteElementSpace &fes,
mfem::Coefficient *Q);
};
}
}
@@ -11,11 +11,8 @@
#include "full-assembly.hpp"
#include "../../../linalg/sparsemat.hpp"
#include "../interface/util.hpp"
#include "../interface/ceed.hpp"
#ifdef MFEM_USE_CEED
#include "util.hpp"
namespace mfem
{
@@ -12,9 +12,11 @@
#ifndef MFEM_CEED_ASSEMBLE_HPP
#define MFEM_CEED_ASSEMBLE_HPP
#include "../interface/ceed.hpp"
#include "../../config/config.hpp"
#ifdef MFEM_USE_CEED
#include <ceed.h>
#include "../../linalg/sparsemat.hpp"
namespace mfem
{
@@ -9,17 +9,14 @@
// terms of the BSD-3 license. We welcome feedback and contributions, see file
// CONTRIBUTING.md for details.
#ifndef MFEM_LIBCEED_INTEG
#define MFEM_LIBCEED_INTEG
#ifndef MFEM_LIBCEED_PAINTEG
#define MFEM_LIBCEED_PAINTEG
#include "../../../config/config.hpp"
#include "../../fespace.hpp"
#include "../../gridfunc.hpp"
#include "operator.hpp"
#include "../../config/config.hpp"
#include "coefficient.hpp"
#include "restriction.hpp"
#include "util.hpp"
#include "ceed.hpp"
#include "../fespace.hpp"
#include "../gridfunc.hpp"
namespace mfem
{
@@ -87,7 +84,6 @@ protected:
CeedQFunctionContext build_ctx;
CeedOperator build_oper;
public:
PAIntegrator()
: Operator(),
trial_basis(nullptr), test_basis(nullptr), mesh_basis(nullptr),
@@ -97,51 +93,23 @@ public:
qdata(nullptr), coeff(nullptr), build_ctx(nullptr), build_oper(nullptr)
{ }
/** @brief This method assembles the `PAIntegrator` with the given
`CeedOperatorInfo` @a info, an `mfem::FiniteElementSpace` @a fes, an
`mfem::IntegrationRule` @a ir, and `mfem::Coefficient` or
`mfem::VectorCoefficient` @a Q.
The `CeedOperatorInfo` type is expected to inherit from `OperatorInfo`,
and contain a `Context` type relevant to the qFunctions.
public:
/** This method assembles the PAIntegrator.
@param[in] info is the structure describing the CeedOperator to assemble.
@param[in] fes is the finite element space.
@param[in] ir is the integration rule for the operator.
@param[in] Q is the coefficient from the `Integrator`. */
@param[in] info the `CeedOperatorInfo` describing the `CeedOperator`,
the `CeedOperatorInfo` type is expected to inherit from
`OperatorInfo` and contain a `Context` type relevant to
the qFunctions.
@param[in] fes the `FiniteElementSpace` for the form,
@param[in] ir the `IntegrationRule` for the numerical integration,
@param[in] Q `Coefficient` or `VectorCoefficient`. */
template <typename CeedOperatorInfo, typename CoeffType>
void Assemble(CeedOperatorInfo &info,
const mfem::FiniteElementSpace &fes,
const mfem::IntegrationRule &ir,
const mfem::IntegrationRule &irm,
CoeffType *Q)
{
Assemble(info, fes, ir, fes.GetNE(), nullptr, Q);
}
/** @brief This method assembles the `PAIntegrator` with the given
`CeedOperatorInfo` @a info, an `mfem::FiniteElementSpace` @a fes, an
`mfem::IntegrationRule` @a ir, and `mfem::Coefficient` or
`mfem::VectorCoefficient` @a Q for the elements given by the indices
@a indices.
The `CeedOperatorInfo` type is expected to inherit from `OperatorInfo`,
and contain a `Context` type relevant to the qFunctions.
@param[in] info is the structure describing the CeedOperator to assemble.
@param[in] fes is the finite element space.
@param[in] ir is the integration rule for the operator.
@param[in] nelem The number of elements.
@param[in] indices The indices of the elements of same type in the
`FiniteElementSpace`. If `indices == nullptr`, assumes
that the `FiniteElementSpace` is not mixed.
@param[in] Q is the coefficient from the `Integrator`. */
template <typename CeedOperatorInfo, typename CoeffType>
void Assemble(CeedOperatorInfo &info,
const mfem::FiniteElementSpace &fes,
const mfem::IntegrationRule &ir,
int nelem,
const int* indices,
CoeffType *Q)
{
Assemble(info, fes, fes, ir, nelem, indices, Q);
Assemble(info, fes, fes, irm, Q);
}
/** This method assembles the PAIntegrator for mixed forms.
@@ -158,40 +126,12 @@ public:
void Assemble(CeedOperatorInfo &info,
const mfem::FiniteElementSpace &trial_fes,
const mfem::FiniteElementSpace &test_fes,
const mfem::IntegrationRule &ir,
CoeffType *Q)
{
Assemble(info, trial_fes, test_fes, ir, trial_fes.GetNE(), nullptr, Q);
}
/** This method assembles the PAIntegrator for mixed forms on mixed meshes.
@param[in] info the `CeedOperatorInfo` describing the `CeedOperator`,
the `CeedOperatorInfo` type is expected to inherit from
`OperatorInfo` and contain a `Context` type relevant to
the qFunctions.
@param[in] trial_fes the trial `FiniteElementSpace` for the form,
@param[in] test_fes the test `FiniteElementSpace` for the form,
@param[in] ir the `IntegrationRule` for the numerical integration,
@param[in] nelem The number of elements,
@param[in] indices The indices of the elements of same type in the
`FiniteElementSpace`. If `indices == nullptr`, assumes
that the `FiniteElementSpace` is not mixed,
@param[in] Q `Coefficient` or `VectorCoefficient`. */
template <typename CeedOperatorInfo, typename CoeffType>
void Assemble(CeedOperatorInfo &info,
const mfem::FiniteElementSpace &trial_fes,
const mfem::FiniteElementSpace &test_fes,
const mfem::IntegrationRule &ir,
int nelem,
const int* indices,
const mfem::IntegrationRule &irm,
CoeffType *Q)
{
Ceed ceed(internal::ceed);
mfem::Mesh &mesh = *trial_fes.GetMesh();
MFEM_VERIFY(!(!indices && mesh.GetNumGeometries(mesh.Dimension()) > 1),
"Use ceed::MixedIntegrator on mixed meshes.");
InitCoefficient(Q, mesh, ir, nelem, indices, coeff, info.ctx);
InitCoefficient(Q, mesh, irm, coeff, info.ctx);
bool const_coeff = coeff->IsConstant();
std::string build_func = const_coeff ? info.build_func_const
: info.build_func_quad;
@@ -203,6 +143,7 @@ public:
info.trial_op,
info.test_op
};
CeedInt nqpts, nelem = mesh.GetNE();
CeedInt dim = mesh.SpaceDimension();
CeedInt trial_vdim = trial_fes.GetVDim();
CeedInt test_vdim = test_fes.GetVDim();
@@ -210,23 +151,23 @@ public:
mesh.EnsureNodes();
if ( &trial_fes == &test_fes )
{
InitBasisAndRestriction(trial_fes, ir, nelem, indices,
ceed, &trial_basis, &trial_restr);
InitBasisAndRestriction(trial_fes, irm, ceed,
&trial_basis, &trial_restr);
test_basis = trial_basis;
test_restr = trial_restr;
}
else
{
InitBasisAndRestriction(trial_fes, ir, nelem, indices,
ceed, &trial_basis, &trial_restr);
InitBasisAndRestriction(test_fes, ir, nelem, indices,
ceed, &test_basis, &test_restr);
InitBasisAndRestriction(trial_fes, irm, ceed,
&trial_basis, &trial_restr);
InitBasisAndRestriction(test_fes, irm, ceed,
&test_basis, &test_restr);
}
const mfem::FiniteElementSpace *mesh_fes = mesh.GetNodalFESpace();
MFEM_VERIFY(mesh_fes, "the Mesh has no nodal FE space");
InitBasisAndRestriction(*mesh_fes, ir, nelem, indices,
ceed, &mesh_basis, &mesh_restr);
InitBasisAndRestriction(*mesh_fes, irm, ceed, &mesh_basis,
&mesh_restr);
CeedInt trial_nqpts, test_nqpts;
CeedBasisGetNumQuadraturePoints(trial_basis, &trial_nqpts);
@@ -234,7 +175,7 @@ public:
MFEM_VERIFY(trial_nqpts == test_nqpts,
"Trial and test basis must have the same number of quadrature"
" points.");
CeedInt nqpts = trial_nqpts;
nqpts = trial_nqpts;
const int qdatasize = op.qdatasize;
InitStridedRestriction(*mesh_fes, nelem, nqpts, qdatasize,
@@ -278,10 +219,8 @@ public:
CeedOperatorCreate(ceed, build_qfunc, NULL, NULL, &build_oper);
if (GridCoefficient *gridCoeff = dynamic_cast<GridCoefficient*>(coeff))
{
InitBasisAndRestriction(*gridCoeff->gf.FESpace(), ir,
nelem, indices, ceed,
&gridCoeff->basis,
&gridCoeff->restr);
InitBasisAndRestriction(*gridCoeff->gf.FESpace(), irm, ceed,
&gridCoeff->basis, &gridCoeff->restr);
CeedOperatorSetField(build_oper, "coeff", gridCoeff->restr,
gridCoeff->basis, gridCoeff->coeffVector);
}
@@ -290,8 +229,7 @@ public:
{
const int ncomp = quadCoeff->ncomp;
CeedInt strides[3] = {ncomp, 1, ncomp*nqpts};
InitStridedRestriction(*mesh.GetNodalFESpace(),
nelem, nqpts, ncomp, strides,
InitStridedRestriction(*mesh_fes, nelem, nqpts, ncomp, strides,
&quadCoeff->restr);
CeedOperatorSetField(build_oper, "coeff", quadCoeff->restr,
CEED_BASIS_COLLOCATED, quadCoeff->coeffVector);
@@ -314,17 +252,22 @@ public:
switch (op.trial_op)
{
case EvalMode::None:
CeedQFunctionAddInput(apply_qfunc, "u", trial_vdim, CEED_EVAL_NONE);
CeedQFunctionAddInput(apply_qfunc, "u", trial_vdim,
CEED_EVAL_NONE);
break;
case EvalMode::Interp:
CeedQFunctionAddInput(apply_qfunc, "u", trial_vdim, CEED_EVAL_INTERP);
CeedQFunctionAddInput(apply_qfunc, "u", trial_vdim,
CEED_EVAL_INTERP);
break;
case EvalMode::Grad:
CeedQFunctionAddInput(apply_qfunc, "gu", trial_vdim*dim, CEED_EVAL_GRAD);
CeedQFunctionAddInput(apply_qfunc, "gu", trial_vdim*dim,
CEED_EVAL_GRAD);
break;
case EvalMode::InterpAndGrad:
CeedQFunctionAddInput(apply_qfunc, "u", trial_vdim, CEED_EVAL_INTERP);
CeedQFunctionAddInput(apply_qfunc, "gu", trial_vdim*dim, CEED_EVAL_GRAD);
CeedQFunctionAddInput(apply_qfunc, "u", trial_vdim,
CEED_EVAL_INTERP);
CeedQFunctionAddInput(apply_qfunc, "gu", trial_vdim*dim,
CEED_EVAL_GRAD);
break;
}
// qdata
@@ -333,17 +276,22 @@ public:
switch (op.test_op)
{
case EvalMode::None:
CeedQFunctionAddOutput(apply_qfunc, "v", test_vdim, CEED_EVAL_NONE);
CeedQFunctionAddOutput(apply_qfunc, "v", test_vdim,
CEED_EVAL_NONE);
break;
case EvalMode::Interp:
CeedQFunctionAddOutput(apply_qfunc, "v", test_vdim, CEED_EVAL_INTERP);
CeedQFunctionAddOutput(apply_qfunc, "v", test_vdim,
CEED_EVAL_INTERP);
break;
case EvalMode::Grad:
CeedQFunctionAddOutput(apply_qfunc, "gv", test_vdim*dim, CEED_EVAL_GRAD);
CeedQFunctionAddOutput(apply_qfunc, "gv", test_vdim*dim,
CEED_EVAL_GRAD);
break;
case EvalMode::InterpAndGrad:
CeedQFunctionAddOutput(apply_qfunc, "v", test_vdim, CEED_EVAL_INTERP);
CeedQFunctionAddOutput(apply_qfunc, "gv", test_vdim*dim, CEED_EVAL_GRAD);
CeedQFunctionAddOutput(apply_qfunc, "v", test_vdim,
CEED_EVAL_INTERP);
CeedQFunctionAddOutput(apply_qfunc, "gv", test_vdim*dim,
CEED_EVAL_GRAD);
break;
}
CeedQFunctionSetContext(apply_qfunc, build_ctx);
@@ -358,14 +306,18 @@ public:
CEED_BASIS_COLLOCATED, CEED_VECTOR_ACTIVE);
break;
case EvalMode::Interp:
CeedOperatorSetField(oper, "u", trial_restr, trial_basis, CEED_VECTOR_ACTIVE);
CeedOperatorSetField(oper, "u", trial_restr, trial_basis,
CEED_VECTOR_ACTIVE);
break;
case EvalMode::Grad:
CeedOperatorSetField(oper, "gu", trial_restr, trial_basis, CEED_VECTOR_ACTIVE);
CeedOperatorSetField(oper, "gu", trial_restr, trial_basis,
CEED_VECTOR_ACTIVE);
break;
case EvalMode::InterpAndGrad:
CeedOperatorSetField(oper, "u", trial_restr, trial_basis, CEED_VECTOR_ACTIVE);
CeedOperatorSetField(oper, "gu", trial_restr, trial_basis, CEED_VECTOR_ACTIVE);
CeedOperatorSetField(oper, "u", trial_restr, trial_basis,
CEED_VECTOR_ACTIVE);
CeedOperatorSetField(oper, "gu", trial_restr, trial_basis,
CEED_VECTOR_ACTIVE);
break;
}
// qdata
@@ -379,14 +331,18 @@ public:
CEED_BASIS_COLLOCATED, CEED_VECTOR_ACTIVE);
break;
case EvalMode::Interp:
CeedOperatorSetField(oper, "v", test_restr, test_basis, CEED_VECTOR_ACTIVE);
CeedOperatorSetField(oper, "v", test_restr, test_basis,
CEED_VECTOR_ACTIVE);
break;
case EvalMode::Grad:
CeedOperatorSetField(oper, "gv", test_restr, test_basis, CEED_VECTOR_ACTIVE);
CeedOperatorSetField(oper, "gv", test_restr, test_basis,
CEED_VECTOR_ACTIVE);
break;
case EvalMode::InterpAndGrad:
CeedOperatorSetField(oper, "v", test_restr, test_basis, CEED_VECTOR_ACTIVE);
CeedOperatorSetField(oper, "gv", test_restr, test_basis, CEED_VECTOR_ACTIVE);
CeedOperatorSetField(oper, "v", test_restr, test_basis,
CEED_VECTOR_ACTIVE);
CeedOperatorSetField(oper, "gv", test_restr, test_basis,
CEED_VECTOR_ACTIVE);
break;
}
@@ -444,7 +400,6 @@ protected:
Coefficient *coeff;
CeedQFunctionContext build_ctx;
public:
MFIntegrator()
: Operator(),
trial_basis(nullptr), test_basis(nullptr), mesh_basis(nullptr),
@@ -453,51 +408,23 @@ public:
apply_qfunc(nullptr), node_coords(nullptr),
qdata(nullptr), coeff(nullptr), build_ctx(nullptr) { }
/** @brief This method assembles the `MFIntegrator` with the given
`CeedOperatorInfo` @a info, an `mfem::FiniteElementSpace` @a fes, an
`mfem::IntegrationRule` @a ir, and `mfem::Coefficient` or
`mfem::VectorCoefficient` @a Q.
The `CeedOperatorInfo` type is expected to inherit from `OperatorInfo`,
and contain a `Context` type relevant to the qFunctions.
public:
/** This method assembles the MFIntegrator.
@param[in] info is the structure describing the CeedOperator to assemble.
@param[in] fes is the finite element space.
@param[in] ir is the integration rule for the operator.
@param[in] Q is the coefficient from the `Integrator`. */
@param[in] info the `CeedOperatorInfo` describing the `CeedOperator`,
the `CeedOperatorInfo` type is expected to inherit from
`OperatorInfo` and contain a `Context` type relevant to
the qFunctions.
@param[in] fes the `FiniteElementSpace` for the form,
@param[in] ir the `IntegrationRule` for the numerical integration,
@param[in] Q `Coefficient` or `VectorCoefficient`. */
template <typename CeedOperatorInfo, typename CoeffType>
void Assemble(CeedOperatorInfo &info,
const mfem::FiniteElementSpace &fes,
const mfem::IntegrationRule &ir,
const mfem::IntegrationRule &irm,
CoeffType *Q)
{
Assemble(info, fes, ir, fes.GetNE(), nullptr, Q);
}
/** @brief This method assembles the `MFIntegrator` with the given
`CeedOperatorInfo` @a info, an `mfem::FiniteElementSpace` @a fes, an
`mfem::IntegrationRule` @a ir, and `mfem::Coefficient` or
`mfem::VectorCoefficient` @a Q for the elements given by the indices
@a indices.
The `CeedOperatorInfo` type is expected to inherit from `OperatorInfo`,
and contain a `Context` type relevant to the qFunctions.
@param[in] info is the structure describing the CeedOperator to assemble.
@param[in] fes is the finite element space.
@param[in] ir is the integration rule for the operator.
@param[in] nelem The number of elements.
@param[in] indices The indices of the elements of same type in the
`FiniteElementSpace`. If `indices == nullptr`, assumes
that the `FiniteElementSpace` is not mixed.
@param[in] Q is the coefficient from the `Integrator`. */
template <typename CeedOperatorInfo, typename CoeffType>
void Assemble(CeedOperatorInfo &info,
const mfem::FiniteElementSpace &fes,
const mfem::IntegrationRule &ir,
int nelem,
const int* indices,
CoeffType *Q)
{
Assemble(info, fes, fes, ir, nelem, indices, Q);
Assemble(info, fes, fes, irm, Q);
}
/** This method assembles the MFIntegrator for mixed forms.
@@ -514,40 +441,12 @@ public:
void Assemble(CeedOperatorInfo &info,
const mfem::FiniteElementSpace &trial_fes,
const mfem::FiniteElementSpace &test_fes,
const mfem::IntegrationRule &ir,
CoeffType *Q)
{
Assemble(info, trial_fes, test_fes, ir, trial_fes.GetNE(), nullptr, Q);
}
/** This method assembles the MFIntegrator for mixed forms.
@param[in] info the `CeedOperatorInfo` describing the `CeedOperator`,
the `CeedOperatorInfo` type is expected to inherit from
`OperatorInfo` and contain a `Context` type relevant to
the qFunctions.
@param[in] trial_fes the trial `FiniteElementSpace` for the form,
@param[in] test_fes the test `FiniteElementSpace` for the form,
@param[in] ir the `IntegrationRule` for the numerical integration,
@param[in] nelem The number of elements,
@param[in] indices The indices of the elements of same type in the
`FiniteElementSpace`. If `indices == nullptr`, assumes
that the `FiniteElementSpace` is not mixed,
@param[in] Q `Coefficient` or `VectorCoefficient`. */
template <typename CeedOperatorInfo, typename CoeffType>
void Assemble(CeedOperatorInfo &info,
const mfem::FiniteElementSpace &trial_fes,
const mfem::FiniteElementSpace &test_fes,
const mfem::IntegrationRule &ir,
int nelem,
const int* indices,
const mfem::IntegrationRule &irm,
CoeffType *Q)
{
Ceed ceed(internal::ceed);
Mesh &mesh = *trial_fes.GetMesh();
MFEM_VERIFY(!(!indices && mesh.GetNumGeometries(mesh.Dimension()) > 1),
"Use ceed::MixedIntegrator on mixed meshes.");
InitCoefficient(Q, mesh, ir, nelem, indices, coeff, info.ctx);
InitCoefficient(Q, mesh, irm, coeff, info.ctx);
bool const_coeff = coeff->IsConstant();
std::string apply_func = const_coeff ? info.apply_func_mf_const
: info.apply_func_mf_quad;
@@ -558,7 +457,7 @@ public:
info.trial_op,
info.test_op
};
CeedInt nqpts, nelem = mesh.GetNE();
CeedInt dim = mesh.SpaceDimension();
CeedInt trial_vdim = trial_fes.GetVDim();
CeedInt test_vdim = test_fes.GetVDim();
@@ -566,22 +465,22 @@ public:
mesh.EnsureNodes();
if ( &trial_fes == &test_fes )
{
InitBasisAndRestriction(trial_fes, ir, nelem, indices, ceed,
InitBasisAndRestriction(trial_fes, irm, ceed,
&trial_basis, &trial_restr);
test_basis = trial_basis;
test_restr = trial_restr;
}
else
{
InitBasisAndRestriction(trial_fes, ir, nelem, indices, ceed,
InitBasisAndRestriction(trial_fes, irm, ceed,
&trial_basis, &trial_restr);
InitBasisAndRestriction(test_fes, ir, nelem, indices, ceed,
InitBasisAndRestriction(test_fes, irm, ceed,
&test_basis, &test_restr);
}
const mfem::FiniteElementSpace *mesh_fes = mesh.GetNodalFESpace();
MFEM_VERIFY(mesh_fes, "the Mesh has no nodal FE space");
InitBasisAndRestriction(*mesh_fes, ir, nelem, indices, ceed, &mesh_basis,
InitBasisAndRestriction(*mesh_fes, irm, ceed, &mesh_basis,
&mesh_restr);
CeedInt trial_nqpts, test_nqpts;
@@ -590,7 +489,7 @@ public:
MFEM_VERIFY(trial_nqpts == test_nqpts,
"Trial and test basis must have the same number of quadrature"
" points.");
CeedInt nqpts = trial_nqpts;
nqpts = trial_nqpts;
InitVector(*mesh.GetNodes(), node_coords);
@@ -671,8 +570,8 @@ public:
// coefficient
if (GridCoefficient *gridCoeff = dynamic_cast<GridCoefficient*>(coeff))
{
InitBasisAndRestriction(*gridCoeff->gf.FESpace(), ir, nelem, indices,
ceed, &gridCoeff->basis, &gridCoeff->restr);
InitBasisAndRestriction(*gridCoeff->gf.FESpace(), irm, ceed,
&gridCoeff->basis, &gridCoeff->restr);
CeedOperatorSetField(oper, "coeff", gridCoeff->restr,
gridCoeff->basis, gridCoeff->coeffVector);
}
@@ -774,4 +673,4 @@ private:
} // namespace mfem
#endif // MFEM_LIBCEED_INTEG
#endif // MFEM_LIBCEED_PAINTEG
-156
View File
@@ -1,156 +0,0 @@
// Copyright (c) 2010-2022, Lawrence Livermore National Security, LLC. Produced
// at the Lawrence Livermore National Laboratory. All Rights reserved. See files
// LICENSE and NOTICE for details. LLNL-CODE-806117.
//
// This file is part of the MFEM library. For more information and source code
// availability visit https://mfem.org.
//
// MFEM is free software; you can redistribute it and/or modify it under the
// terms of the BSD-3 license. We welcome feedback and contributions, see file
// CONTRIBUTING.md for details.
#include "../../gridfunc.hpp"
#include "util.hpp"
namespace mfem
{
namespace ceed
{
#ifdef MFEM_USE_CEED
static CeedElemTopology GetCeedTopology(Geometry::Type geom)
{
switch (geom)
{
case Geometry::SEGMENT:
return CEED_TOPOLOGY_LINE;
case Geometry::TRIANGLE:
return CEED_TOPOLOGY_TRIANGLE;
case Geometry::SQUARE:
return CEED_TOPOLOGY_QUAD;
case Geometry::TETRAHEDRON:
return CEED_TOPOLOGY_TET;
case Geometry::CUBE:
return CEED_TOPOLOGY_HEX;
case Geometry::PRISM:
return CEED_TOPOLOGY_PRISM;
case Geometry::PYRAMID:
return CEED_TOPOLOGY_PYRAMID;
default:
MFEM_ABORT("This type of element is not supported");
return CEED_TOPOLOGY_PRISM; // Silence warning
}
}
static void InitNonTensorBasis(const mfem::FiniteElementSpace &fes,
const mfem::FiniteElement &fe,
const mfem::IntegrationRule &ir,
Ceed ceed, CeedBasis *basis)
{
const mfem::DofToQuad &maps = fe.GetDofToQuad(ir, mfem::DofToQuad::FULL);
mfem::Mesh *mesh = fes.GetMesh();
const int dim = mesh->Dimension();
const int ndofs = maps.ndof;
const int nqpts = maps.nqpt;
mfem::DenseMatrix qX(dim,nqpts);
mfem::Vector qW(nqpts);
for (int i = 0; i < nqpts; i++)
{
const mfem::IntegrationPoint &ip = ir.IntPoint(i);
qX(0,i) = ip.x;
if (dim>1) { qX(1,i) = ip.y; }
if (dim>2) { qX(2,i) = ip.z; }
qW(i) = ip.weight;
}
CeedBasisCreateH1(ceed, GetCeedTopology(fe.GetGeomType()),
fes.GetVDim(), ndofs, nqpts,
maps.Bt.GetData(), maps.Gt.GetData(),
qX.GetData(), qW.GetData(), basis);
}
static void InitTensorBasis(const mfem::FiniteElementSpace &fes,
const mfem::FiniteElement &fe,
const mfem::IntegrationRule &ir,
Ceed ceed, CeedBasis *basis)
{
const mfem::DofToQuad &maps = fe.GetDofToQuad(ir, mfem::DofToQuad::TENSOR);
mfem::Mesh *mesh = fes.GetMesh();
const int ndofs = maps.ndof;
const int nqpts = maps.nqpt;
mfem::Vector qX(nqpts), qW(nqpts);
// The x-coordinates of the first `nqpts` points of the integration rule are
// the points of the corresponding 1D rule. We also scale the weights
// accordingly.
double w_sum = 0.0;
for (int i = 0; i < nqpts; i++)
{
const mfem::IntegrationPoint &ip = ir.IntPoint(i);
qX(i) = ip.x;
qW(i) = ip.weight;
w_sum += ip.weight;
}
qW *= 1.0/w_sum;
CeedBasisCreateTensorH1(ceed, mesh->Dimension(), fes.GetVDim(), ndofs,
nqpts, maps.Bt.GetData(),
maps.Gt.GetData(), qX.GetData(),
qW.GetData(), basis);
}
static void InitBasisImpl(const FiniteElementSpace &fes,
const FiniteElement &fe,
const IntegrationRule &ir,
Ceed ceed, CeedBasis *basis)
{
// Check for FES -> basis, restriction in hash tables
const int P = fe.GetDof();
const int Q = ir.GetNPoints();
const int ncomp = fes.GetVDim();
BasisKey basis_key(&fes, &ir, ncomp, P, Q);
auto basis_itr = mfem::internal::ceed_basis_map.find(basis_key);
const bool tensor = dynamic_cast<const mfem::TensorBasisElement *>
(&fe) != nullptr;
// Init or retreive key values
if (basis_itr == mfem::internal::ceed_basis_map.end())
{
if ( tensor )
{
InitTensorBasis(fes, fe, ir, ceed, basis);
}
else
{
InitNonTensorBasis(fes, fe, ir, ceed, basis);
}
mfem::internal::ceed_basis_map[basis_key] = *basis;
}
else
{
*basis = basis_itr->second;
}
}
void InitBasis(const FiniteElementSpace &fes,
const IntegrationRule &ir,
Ceed ceed, CeedBasis *basis)
{
const mfem::FiniteElement &fe = *fes.GetFE(0);
InitBasisImpl(fes, fe, ir, ceed, basis);
}
void InitBasisWithIndices(const FiniteElementSpace &fes,
const IntegrationRule &ir,
int nelem,
const int* indices,
Ceed ceed, CeedBasis *basis)
{
const mfem::FiniteElement &fe = *fes.GetFE(indices[0]);
InitBasisImpl(fes, fe, ir, ceed, basis);
}
#endif
} // namespace ceed
} // namespace mfem
-57
View File
@@ -1,57 +0,0 @@
// Copyright (c) 2010-2022, Lawrence Livermore National Security, LLC. Produced
// at the Lawrence Livermore National Laboratory. All Rights reserved. See files
// LICENSE and NOTICE for details. LLNL-CODE-806117.
//
// This file is part of the MFEM library. For more information and source code
// availability visit https://mfem.org.
//
// MFEM is free software; you can redistribute it and/or modify it under the
// terms of the BSD-3 license. We welcome feedback and contributions, see file
// CONTRIBUTING.md for details.
#ifndef MFEM_LIBCEED_BASIS
#define MFEM_LIBCEED_BASIS
#include "ceed.hpp"
namespace mfem
{
namespace ceed
{
#ifdef MFEM_USE_CEED
/** @brief Initialize a CeedBasis for non-mixed meshes.
@param[in] fes Input finite element space.
@param[in] ir Input integration rule.
@param[in] ceed Input Ceed object.
@param[out] basis The address of the initialized CeedBasis object.
*/
void InitBasis(const FiniteElementSpace &fes,
const IntegrationRule &ir,
Ceed ceed, CeedBasis *basis);
/** @brief Initialize a CeedBasis for mixed meshes.
@param[in] fes The finite element space.
@param[in] ir is the integration rule for the operator.
@param[in] nelem The number of elements.
@param[in] indices The indices of the elements of same type in the
`FiniteElementSpace`.
@param[in] ceed The Ceed object.
@param[out] basis The `CeedBasis` to initialize. */
void InitBasisWithIndices(const FiniteElementSpace &fes,
const IntegrationRule &ir,
int nelem,
const int* indices,
Ceed ceed, CeedBasis *basis);
#endif
} // namespace ceed
} // namespace mfem
#endif // MFEM_LIBCEED_BASIS
-35
View File
@@ -1,35 +0,0 @@
// Copyright (c) 2010-2022, Lawrence Livermore National Security, LLC. Produced
// at the Lawrence Livermore National Laboratory. All Rights reserved. See files
// LICENSE and NOTICE for details. LLNL-CODE-806117.
//
// This file is part of the MFEM library. For more information and source code
// availability visit https://mfem.org.
//
// MFEM is free software; you can redistribute it and/or modify it under the
// terms of the BSD-3 license. We welcome feedback and contributions, see file
// CONTRIBUTING.md for details.
#ifndef MFEM_LIBCEED_CEED
#define MFEM_LIBCEED_CEED
#ifdef MFEM_USE_CEED
#include <ceed.h>
#if !CEED_VERSION_GE(0,10,0)
#error MFEM requires a libCEED version >= 0.10.0
#endif
namespace mfem
{
namespace internal
{
extern Ceed ceed;
} // namespace internal
} // namespace mfem
#endif // MFEM_USE_CEED
#endif // MFEM_LIBCEED_CEED
-441
View File
@@ -1,441 +0,0 @@
// Copyright (c) 2010-2022, Lawrence Livermore National Security, LLC. Produced
// at the Lawrence Livermore National Laboratory. All Rights reserved. See files
// LICENSE and NOTICE for details. LLNL-CODE-806117.
//
// This file is part of the MFEM library. For more information and source code
// availability visit https://mfem.org.
//
// MFEM is free software; you can redistribute it and/or modify it under the
// terms of the BSD-3 license. We welcome feedback and contributions, see file
// CONTRIBUTING.md for details.
#ifndef MFEM_LIBCEED_COEFF
#define MFEM_LIBCEED_COEFF
#ifdef MFEM_USE_CEED
#include "../../../general/forall.hpp"
#include "../../../config/config.hpp"
#include "../../../linalg/vector.hpp"
#include "../../../linalg/dtensor.hpp"
#include "../../../mesh/mesh.hpp"
#include "../../gridfunc.hpp"
#include "util.hpp"
#include "ceed.hpp"
namespace mfem
{
class Mesh;
class IntegrationRule;
class Coefficient;
class VectorCoefficient;
class GridFunction;
namespace ceed
{
struct Coefficient
{
const int ncomp;
Coefficient(int ncomp_) : ncomp(ncomp_) { }
virtual bool IsConstant() const { return true; }
virtual ~Coefficient() { }
};
struct VariableCoefficient : Coefficient
{
CeedVector coeffVector = nullptr;
const CeedEvalMode emode;
VariableCoefficient(int ncomp_, CeedEvalMode emode_)
: Coefficient(ncomp_), emode(emode_) { }
virtual bool IsConstant() const override { return false; }
~VariableCoefficient()
{
CeedVectorDestroy(&coeffVector);
}
};
struct GridCoefficient : VariableCoefficient
{
const mfem::GridFunction &gf;
CeedBasis basis;
CeedElemRestriction restr;
GridCoefficient(const mfem::GridFunction &gf_)
: VariableCoefficient(gf_.VectorDim(), CEED_EVAL_INTERP),
gf(gf_)
{
InitVector(gf, coeffVector);
}
};
struct QuadCoefficient : VariableCoefficient
{
mfem::Vector coeff;
CeedElemRestriction restr;
QuadCoefficient(int ncomp_) : VariableCoefficient(ncomp_, CEED_EVAL_NONE) { }
};
/** @brief Initializes an mfem::ceed::Coefficient @a coeff_ptr from an
mfem::Coefficient @a Q, an mfem::Mesh @a mesh, and an mfem::IntegrationRule
@a ir.
@param[in] Q is the coefficient from the `Integrator`.
@param[in] mesh is the mesh.
@param[in] ir is the integration rule.
@param[out] coeff_ptr is the structure to store the coefficient for the
`CeedOperator`.
@param[out] ctx is the Context associated to the QFunction. */
template <typename Context>
void InitCoefficient(mfem::Coefficient *Q, mfem::Mesh &mesh,
const mfem::IntegrationRule &ir,
Coefficient*& coeff_ptr, Context &ctx)
{
if ( Q == nullptr )
{
Coefficient *ceedCoeff = new Coefficient(1);
ctx.coeff = 1.0;
coeff_ptr = ceedCoeff;
}
else if (ConstantCoefficient *const_coeff =
dynamic_cast<ConstantCoefficient*>(Q))
{
Coefficient *ceedCoeff = new Coefficient(1);
ctx.coeff = const_coeff->constant;
coeff_ptr = ceedCoeff;
}
else if (GridFunctionCoefficient* gf_coeff =
dynamic_cast<GridFunctionCoefficient*>(Q))
{
GridCoefficient *ceedCoeff =
new GridCoefficient(*gf_coeff->GetGridFunction());
coeff_ptr = ceedCoeff;
}
else if (QuadratureFunctionCoefficient *cQ =
dynamic_cast<QuadratureFunctionCoefficient*>(Q))
{
QuadCoefficient *ceedCoeff = new QuadCoefficient(1);
const int ne = mesh.GetNE();
const int nq = ir.GetNPoints();
const mfem::QuadratureFunction &qFun = cQ->GetQuadFunction();
MFEM_VERIFY(qFun.Size() == nq * ne,
"Incompatible QuadratureFunction dimension \n");
MFEM_VERIFY(&ir == &qFun.GetSpace()->GetElementIntRule(0),
"IntegrationRule used within integrator and in"
" QuadratureFunction appear to be different");
qFun.Read();
ceedCoeff->coeff.MakeRef(const_cast<mfem::QuadratureFunction &>(qFun),0);
InitVector(ceedCoeff->coeff, ceedCoeff->coeffVector);
coeff_ptr = ceedCoeff;
}
else
{
QuadCoefficient *ceedCoeff = new QuadCoefficient(1);
const int ne = mesh.GetNE();
const int nq = ir.GetNPoints();
ceedCoeff->coeff.SetSize(nq * ne);
auto C = Reshape(ceedCoeff->coeff.HostWrite(), nq, ne);
for (int e = 0; e < ne; ++e)
{
mfem::ElementTransformation &T = *mesh.GetElementTransformation(e);
for (int q = 0; q < nq; ++q)
{
C(q,e) = Q->Eval(T, ir.IntPoint(q));
}
}
InitVector(ceedCoeff->coeff, ceedCoeff->coeffVector);
coeff_ptr = ceedCoeff;
}
}
/** @brief Initializes an mfem::ceed::Coefficient @a coeff_ptr from an
mfem::VectorCoefficient @a VQ, an mfem::Mesh @a mesh, and an
mfem::IntegrationRule @a ir.
@param[in] VQ is the vector coefficient from the `Integrator`.
@param[in] mesh is the mesh.
@param[in] ir is the integration rule.
@param[out] coeff_ptr is the structure to store the coefficient for the
`CeedOperator`.
@param[out] ctx is the Context associated to the QFunction. */
template <typename Context>
void InitCoefficient(mfem::VectorCoefficient *VQ, mfem::Mesh &mesh,
const mfem::IntegrationRule &ir,
Coefficient *&coeff_ptr, Context &ctx)
{
if (VectorConstantCoefficient *const_coeff =
dynamic_cast<VectorConstantCoefficient*>(VQ))
{
const int vdim = const_coeff->GetVDim();
const mfem::Vector &val = const_coeff->GetVec();
Coefficient *ceedCoeff = new Coefficient(vdim);
for (int i = 0; i < vdim; i++)
{
ctx.coeff[i] = val[i];
}
coeff_ptr = ceedCoeff;
}
else if (VectorGridFunctionCoefficient* vgf_coeff =
dynamic_cast<VectorGridFunctionCoefficient*>(VQ))
{
GridCoefficient *ceedCoeff =
new GridCoefficient(*vgf_coeff->GetGridFunction());
coeff_ptr = ceedCoeff;
}
else if (VectorQuadratureFunctionCoefficient *cQ =
dynamic_cast<VectorQuadratureFunctionCoefficient*>(VQ))
{
QuadCoefficient *ceedCoeff = new QuadCoefficient(cQ->GetVDim());
const int dim = mesh.Dimension();
const int ne = mesh.GetNE();
const int nq = ir.GetNPoints();
const mfem::QuadratureFunction &qFun = cQ->GetQuadFunction();
MFEM_VERIFY(qFun.Size() == dim * nq * ne,
"Incompatible QuadratureFunction dimension \n");
MFEM_VERIFY(&ir == &qFun.GetSpace()->GetElementIntRule(0),
"IntegrationRule used within integrator and in"
" QuadratureFunction appear to be different");
qFun.Read();
ceedCoeff->coeff.MakeRef(const_cast<mfem::QuadratureFunction &>(qFun),0);
InitVector(ceedCoeff->coeff, ceedCoeff->coeffVector);
coeff_ptr = ceedCoeff;
}
else
{
const int dim = mesh.Dimension();
QuadCoefficient *ceedCoeff = new QuadCoefficient(dim);
const int ne = mesh.GetNE();
const int nq = ir.GetNPoints();
ceedCoeff->coeff.SetSize(dim * nq * ne);
auto C = Reshape(ceedCoeff->coeff.HostWrite(), dim, nq, ne);
mfem::DenseMatrix Q_ir;
for (int e = 0; e < ne; ++e)
{
mfem::ElementTransformation &T = *mesh.GetElementTransformation(e);
VQ->Eval(Q_ir, T, ir);
for (int q = 0; q < nq; ++q)
{
for (int i = 0; i < dim; ++i)
{
C(i,q,e) = Q_ir(i,q);
}
}
}
InitVector(ceedCoeff->coeff, ceedCoeff->coeffVector);
coeff_ptr = ceedCoeff;
}
}
/** @brief Initializes an mfem::ceed::Coefficient @a coeff_ptr from an
mfem::Coefficient @a Q, an mfem::Mesh @a mesh, and an mfem::IntegrationRule
@a ir for the elements given by the indices @a indices.
@param[in] Q is the coefficient from the `Integrator`.
@param[in] mesh is the mesh.
@param[in] ir is the integration rule.
@param[in] nelem The number of elements.
@param[in] indices The indices of the elements of same type in the
`FiniteElementSpace`.
@param[out] coeff_ptr is the structure to store the coefficient for the
`CeedOperator`.
@param[out] ctx is the Context associated to the QFunction. */
template <typename Context>
void InitCoefficientWithIndices(mfem::Coefficient *Q, mfem::Mesh &mesh,
const mfem::IntegrationRule &ir,
int nelem,
const int* indices,
Coefficient*& coeff_ptr, Context &ctx)
{
if ( Q == nullptr )
{
Coefficient *ceedCoeff = new Coefficient(1);
ctx.coeff = 1.0;
coeff_ptr = ceedCoeff;
}
else if (ConstantCoefficient *const_coeff =
dynamic_cast<ConstantCoefficient*>(Q))
{
Coefficient *ceedCoeff = new Coefficient(1);
ctx.coeff = const_coeff->constant;
coeff_ptr = ceedCoeff;
}
else if (GridFunctionCoefficient* gf_coeff =
dynamic_cast<GridFunctionCoefficient*>(Q))
{
GridCoefficient *ceedCoeff =
new GridCoefficient(*gf_coeff->GetGridFunction());
coeff_ptr = ceedCoeff;
}
else if (QuadratureFunctionCoefficient *cQ =
dynamic_cast<QuadratureFunctionCoefficient*>(Q))
{
QuadCoefficient *ceedCoeff = new QuadCoefficient(1);
const int ne = mesh.GetNE();
const int nq = ir.GetNPoints();
const mfem::QuadratureFunction &qFun = cQ->GetQuadFunction();
MFEM_VERIFY(qFun.Size() == nq * ne,
"Incompatible QuadratureFunction dimension \n");
MFEM_VERIFY(&ir == &qFun.GetSpace()->GetElementIntRule(0),
"IntegrationRule used within integrator and in"
" QuadratureFunction appear to be different");
ceedCoeff->coeff.SetSize(nq * nelem);
Memory<int> m_indices((int*)indices, nelem, false);
auto in = Reshape(qFun.Read(), nq, ne);
auto d_indices = Read(m_indices, nelem);
auto out = Reshape(ceedCoeff->coeff.Write(), nq, nelem);
MFEM_FORALL(i, nelem * nq,
{
const int q = i%nq;
const int sub_e = i/nq;
const int e = d_indices[sub_e];
out(q, sub_e) = in(q, e);
});
m_indices.DeleteDevice();
InitVector(ceedCoeff->coeff, ceedCoeff->coeffVector);
coeff_ptr = ceedCoeff;
}
else
{
QuadCoefficient *ceedCoeff = new QuadCoefficient(1);
const int nq = ir.GetNPoints();
ceedCoeff->coeff.SetSize(nq * nelem);
auto C = Reshape(ceedCoeff->coeff.HostWrite(), nq, nelem);
for (int i = 0; i < nelem; ++i)
{
const int e = indices[i];
mfem::ElementTransformation &T = *mesh.GetElementTransformation(e);
for (int q = 0; q < nq; ++q)
{
C(q, i) = Q->Eval(T, ir.IntPoint(q));
}
}
InitVector(ceedCoeff->coeff, ceedCoeff->coeffVector);
coeff_ptr = ceedCoeff;
}
}
/** @brief Initializes an mfem::ceed::Coefficient @a coeff_ptr from an
mfem::VectorCoefficient @a Q, an mfem::Mesh @a mesh, and an
mfem::IntegrationRule @a ir for the elements given by the indices @a indices.
@param[in] VQ is the vector coefficient from the `Integrator`.
@param[in] mesh is the mesh.
@param[in] ir is the integration rule.
@param[in] nelem The number of elements.
@param[in] indices The indices of the elements of same type in the
`FiniteElementSpace`.
@param[out] coeff_ptr is the structure to store the coefficient for the
`CeedOperator`.
@param[out] ctx is the Context associated to the QFunction. */
template <typename Context>
void InitCoefficientWithIndices(mfem::VectorCoefficient *VQ, mfem::Mesh &mesh,
const mfem::IntegrationRule &ir,
int nelem,
const int* indices,
Coefficient *&coeff_ptr, Context &ctx)
{
if (VectorConstantCoefficient *const_coeff =
dynamic_cast<VectorConstantCoefficient*>(VQ))
{
const int vdim = const_coeff->GetVDim();
const mfem::Vector &val = const_coeff->GetVec();
Coefficient *ceedCoeff = new Coefficient(vdim);
for (int i = 0; i < vdim; i++)
{
ctx.coeff[i] = val[i];
}
coeff_ptr = ceedCoeff;
}
else if (VectorGridFunctionCoefficient* vgf_coeff =
dynamic_cast<VectorGridFunctionCoefficient*>(VQ))
{
GridCoefficient *ceedCoeff =
new GridCoefficient(*vgf_coeff->GetGridFunction());
coeff_ptr = ceedCoeff;
}
else if (VectorQuadratureFunctionCoefficient *cQ =
dynamic_cast<VectorQuadratureFunctionCoefficient*>(VQ))
{
QuadCoefficient *ceedCoeff = new QuadCoefficient(cQ->GetVDim());
const int dim = mesh.Dimension();
const int ne = mesh.GetNE();
const int nq = ir.GetNPoints();
const mfem::QuadratureFunction &qFun = cQ->GetQuadFunction();
MFEM_VERIFY(qFun.Size() == dim * nq * ne,
"Incompatible QuadratureFunction dimension \n");
MFEM_VERIFY(&ir == &qFun.GetSpace()->GetElementIntRule(0),
"IntegrationRule used within integrator and in"
" QuadratureFunction appear to be different");
ceedCoeff->coeff.SetSize(dim * nq * nelem);
Memory<int> m_indices((int*)indices, nelem, false);
auto in = Reshape(qFun.Read(), dim, nq, ne);
auto d_indices = Read(m_indices, nelem);
auto out = Reshape(ceedCoeff->coeff.Write(), dim, nq, nelem);
MFEM_FORALL(i, nelem * nq,
{
const int q = i%nq;
const int sub_e = i/nq;
const int e = d_indices[sub_e];
for (int d = 0; d < dim; d++)
{
out(d, q, sub_e) = in(d, q, e);
}
});
m_indices.DeleteDevice();
InitVector(ceedCoeff->coeff, ceedCoeff->coeffVector);
coeff_ptr = ceedCoeff;
}
else
{
const int dim = mesh.Dimension();
QuadCoefficient *ceedCoeff = new QuadCoefficient(dim);
const int nq = ir.GetNPoints();
ceedCoeff->coeff.SetSize(dim * nq * nelem);
auto C = Reshape(ceedCoeff->coeff.HostWrite(), dim, nq, nelem);
mfem::DenseMatrix Q_ir;
for (int i = 0; i < nelem; ++i)
{
const int e = indices[i];
mfem::ElementTransformation &T = *mesh.GetElementTransformation(e);
VQ->Eval(Q_ir, T, ir);
for (int q = 0; q < nq; ++q)
{
for (int d = 0; d < dim; ++d)
{
C(d, q, i) = Q_ir(d, q);
}
}
}
InitVector(ceedCoeff->coeff, ceedCoeff->coeffVector);
coeff_ptr = ceedCoeff;
}
}
template <typename Coeff, typename Context>
void InitCoefficient(Coeff *Q, mfem::Mesh &mesh,
const mfem::IntegrationRule &ir, int nelem,
const int* indices, Coefficient *&coeff_ptr, Context &ctx)
{
if (indices)
{
InitCoefficientWithIndices(Q, mesh, ir, nelem, indices, coeff_ptr, ctx);
}
else
{
InitCoefficient(Q, mesh, ir, coeff_ptr, ctx);
}
}
} // namespace ceed
} // namespace mfem
#endif
#endif // MFEM_LIBCEED_COEFF
-32
View File
@@ -1,32 +0,0 @@
// Copyright (c) 2010-2022, Lawrence Livermore National Security, LLC. Produced
// at the Lawrence Livermore National Laboratory. All Rights reserved. See files
// LICENSE and NOTICE for details. LLNL-CODE-806117.
//
// This file is part of the MFEM library. For more information and source code
// availability visit https://mfem.org.
//
// MFEM is free software; you can redistribute it and/or modify it under the
// terms of the BSD-3 license. We welcome feedback and contributions, see file
// CONTRIBUTING.md for details.
#ifndef MFEM_LIBCEED_INTERFACE
#define MFEM_LIBCEED_INTERFACE
// Object wrapping a CeedOperator in a mfem::Operator.
#include "operator.hpp"
// Functions to initialize CeedBasis objects.
#include "basis.hpp"
// Functions to initialize CeedRestriction objects.
#include "restriction.hpp"
// Functions to initialize coefficients.
#include "coefficient.hpp"
// PA or MF Operator using libCEED.
#include "integrator.hpp"
// PA Operator supporting mixed finite element spaces.
#include "mixed_integrator.hpp"
// Utility functions
#include "util.hpp"
// Wrapper to include <ceed.h>
#include "ceed.hpp"
#endif // MFEM_LIBCEED_INTERFACE
-126
View File
@@ -1,126 +0,0 @@
// Copyright (c) 2010-2022, Lawrence Livermore National Security, LLC. Produced
// at the Lawrence Livermore National Laboratory. All Rights reserved. See files
// LICENSE and NOTICE for details. LLNL-CODE-806117.
//
// This file is part of the MFEM library. For more information and source code
// availability visit https://mfem.org.
//
// MFEM is free software; you can redistribute it and/or modify it under the
// terms of the BSD-3 license. We welcome feedback and contributions, see file
// CONTRIBUTING.md for details.
#ifndef MFEM_LIBCEED_MIXED_INTEGRATOR
#define MFEM_LIBCEED_MIXED_INTEGRATOR
#include "ceed.hpp"
#include "integrator.hpp"
#include <unordered_map>
namespace mfem
{
namespace ceed
{
/** @brief This class wraps a `ceed::PAIntegrator` or `ceed::MFIntegrator` to
support mixed finite element spaces. */
template <typename CeedInteg>
class MixedIntegrator : public ceed::Operator
{
#ifdef MFEM_USE_CEED
using ElementKey = std::pair<int, int>; //< Element::Type, Order >
struct key_hash
{
std::size_t operator()(const ElementKey& k) const
{
return k.first + 2 * k.second;
}
};
using ElementsMap = std::unordered_map<const ElementKey, int*, key_hash>;
std::vector<CeedInteg*> sub_ops;
public:
template <typename Integrator, typename CeedOperatorInfo, typename CoeffType>
void Assemble(const Integrator &integ,
CeedOperatorInfo &info,
const mfem::FiniteElementSpace &fes,
CoeffType *Q)
{
ElementsMap count;
ElementsMap element_indices;
ElementsMap offsets;
// Count the number of elements of each type
for (int i = 0; i < fes.GetNE(); i++)
{
ElementKey key(fes.GetElementType(i), fes.GetElementOrder(i));
auto value = count.find(key);
if (value == count.end())
{
count[key] = new int(1);
}
else
{
(*value->second)++;
}
}
// Initialization of the arrays
for ( const auto& value : count )
{
element_indices[value.first] = new int[*value.second];
offsets[value.first] = new int(0);
}
// Populates the indices arrays for each element type
for (int i = 0; i < fes.GetNE(); i++)
{
ElementKey key(fes.GetElementType(i), fes.GetElementOrder(i));
int &offset = *(offsets[key]);
int* indices_array = element_indices[key];
indices_array[offset] = i;
offset++;
}
// Create composite CeedOperator
CeedCompositeOperatorCreate(internal::ceed, &oper);
// Create each sub-CeedOperator
sub_ops.reserve(element_indices.size());
for (const auto& value : element_indices)
{
const int* indices = value.second;
const int first_index = indices[0];
const mfem::FiniteElement &el = *fes.GetFE(first_index);
auto &T = *fes.GetMesh()->GetElementTransformation(first_index);
MFEM_ASSERT(!integ.GetIntegrationRule(),
"Mixed mesh integrators should not have an"
" IntegrationRule.");
const IntegrationRule &ir = GetRule(integ, el, el, T);
auto sub_op = new CeedInteg();
int nelem = *count[value.first];
sub_op->Assemble(info, fes, ir, nelem, indices, Q);
sub_ops.push_back(sub_op);
CeedCompositeOperatorAddSub(oper, sub_op->GetCeedOperator());
}
const int ndofs = fes.GetVDim() * fes.GetNDofs();
CeedVectorCreate(internal::ceed, ndofs, &u);
CeedVectorCreate(internal::ceed, ndofs, &v);
}
virtual ~MixedIntegrator()
{
for (auto sub_op : sub_ops)
{
delete sub_op;
}
}
#endif
};
} // namespace ceed
} // namespace mfem
#endif // MFEM_LIBCEED_MIXED_INTEGRATOR
-301
View File
@@ -1,301 +0,0 @@
// Copyright (c) 2010-2022, Lawrence Livermore National Security, LLC. Produced
// at the Lawrence Livermore National Laboratory. All Rights reserved. See files
// LICENSE and NOTICE for details. LLNL-CODE-806117.
//
// This file is part of the MFEM library. For more information and source code
// availability visit https://mfem.org.
//
// MFEM is free software; you can redistribute it and/or modify it under the
// terms of the BSD-3 license. We welcome feedback and contributions, see file
// CONTRIBUTING.md for details.
#include "../../../fem/gridfunc.hpp"
#include "ceed.hpp"
namespace mfem
{
namespace ceed
{
#ifdef MFEM_USE_CEED
static void InitNativeRestr(const mfem::FiniteElementSpace &fes,
Ceed ceed, CeedElemRestriction *restr)
{
const mfem::FiniteElement *fe = fes.GetFE(0);
const int P = fe->GetDof();
CeedInt compstride = fes.GetOrdering()==Ordering::byVDIM ? 1 : fes.GetNDofs();
const mfem::Table &el_dof = fes.GetElementToDofTable();
mfem::Array<int> tp_el_dof(el_dof.Size_of_connections());
const mfem::TensorBasisElement * tfe =
dynamic_cast<const mfem::TensorBasisElement *>(fe);
const int stride = compstride == 1 ? fes.GetVDim() : 1;
const mfem::Array<int>& dof_map = tfe->GetDofMap();
for (int i = 0; i < fes.GetNE(); i++)
{
const int el_offset = P * i;
for (int j = 0; j < P; j++)
{
tp_el_dof[j+el_offset] = stride*el_dof.GetJ()[dof_map[j]+el_offset];
}
}
CeedElemRestrictionCreate(ceed, fes.GetNE(), P, fes.GetVDim(),
compstride, (fes.GetVDim())*(fes.GetNDofs()),
CEED_MEM_HOST, CEED_COPY_VALUES,
tp_el_dof.GetData(), restr);
}
static void InitLexicoRestr(const mfem::FiniteElementSpace &fes,
Ceed ceed, CeedElemRestriction *restr)
{
const mfem::FiniteElement *fe = fes.GetFE(0);
const int P = fe->GetDof();
CeedInt compstride = fes.GetOrdering()==Ordering::byVDIM ? 1 : fes.GetNDofs();
const mfem::Table &el_dof = fes.GetElementToDofTable();
mfem::Array<int> tp_el_dof(el_dof.Size_of_connections());
const int stride = compstride == 1 ? fes.GetVDim() : 1;
for (int e = 0; e < fes.GetNE(); e++)
{
for (int i = 0; i < P; i++)
{
tp_el_dof[i + e*P] = stride*el_dof.GetJ()[i + e*P];
}
}
CeedElemRestrictionCreate(ceed, fes.GetNE(), P, fes.GetVDim(),
compstride, (fes.GetVDim())*(fes.GetNDofs()),
CEED_MEM_HOST, CEED_COPY_VALUES,
tp_el_dof.GetData(), restr);
}
static void InitRestrictionImpl(const mfem::FiniteElementSpace &fes,
Ceed ceed, CeedElemRestriction *restr)
{
const mfem::FiniteElement *fe = fes.GetFE(0);
const mfem::TensorBasisElement * tfe =
dynamic_cast<const mfem::TensorBasisElement *>(fe);
if ( tfe && tfe->GetDofMap().Size()>0 ) // Native ordering using dof_map
{
InitNativeRestr(fes, ceed, restr);
}
else // Lexicographic ordering
{
InitLexicoRestr(fes, ceed, restr);
}
}
static void InitNativeRestrWithIndices(
const mfem::FiniteElementSpace &fes,
int nelem,
const int* indices,
Ceed ceed, CeedElemRestriction *restr)
{
const mfem::FiniteElement *fe = fes.GetFE(indices[0]);
const int P = fe->GetDof();
CeedInt compstride = fes.GetOrdering()==Ordering::byVDIM ? 1 : fes.GetNDofs();
mfem::Array<int> tp_el_dof(nelem*P);
const mfem::TensorBasisElement * tfe =
dynamic_cast<const mfem::TensorBasisElement *>(fe);
Array<int> dofs;
const int stride = compstride == 1 ? fes.GetVDim() : 1;
const mfem::Array<int>& dof_map = tfe->GetDofMap();
for (int i = 0; i < nelem; i++)
{
const int elem_index = indices[i];
fes.GetElementDofs(elem_index, dofs);
const int el_offset = P * i;
for (int j = 0; j < P; j++)
{
tp_el_dof[j + el_offset] = stride*dofs[dof_map[j]];
}
}
CeedElemRestrictionCreate(ceed, nelem, P, fes.GetVDim(),
compstride, (fes.GetVDim())*(fes.GetNDofs()),
CEED_MEM_HOST, CEED_COPY_VALUES,
tp_el_dof.GetData(), restr);
}
static void InitLexicoRestrWithIndices(
const mfem::FiniteElementSpace &fes,
int nelem,
const int* indices,
Ceed ceed, CeedElemRestriction *restr)
{
const mfem::FiniteElement *fe = fes.GetFE(indices[0]);
const int P = fe->GetDof();
CeedInt compstride = fes.GetOrdering()==Ordering::byVDIM ? 1 : fes.GetNDofs();
mfem::Array<int> tp_el_dof(nelem*P);
Array<int> dofs;
const int stride = compstride == 1 ? fes.GetVDim() : 1;
for (int i = 0; i < nelem; i++)
{
const int elem_index = indices[i];
fes.GetElementDofs(elem_index, dofs);
const int el_offset = P * i;
for (int j = 0; j < P; j++)
{
tp_el_dof[j + el_offset] = stride*dofs[j];
}
}
CeedElemRestrictionCreate(ceed, nelem, P, fes.GetVDim(),
compstride, (fes.GetVDim())*(fes.GetNDofs()),
CEED_MEM_HOST, CEED_COPY_VALUES,
tp_el_dof.GetData(), restr);
}
static void InitRestrictionWithIndicesImpl(
const mfem::FiniteElementSpace &fes,
int nelem,
const int* indices,
Ceed ceed, CeedElemRestriction *restr)
{
const mfem::FiniteElement *fe = fes.GetFE(indices[0]);
const mfem::TensorBasisElement * tfe =
dynamic_cast<const mfem::TensorBasisElement *>(fe);
if ( tfe && tfe->GetDofMap().Size()>0 ) // Native ordering using dof_map
{
InitNativeRestrWithIndices(fes, nelem, indices, ceed, restr);
}
else // Lexicographic ordering
{
InitLexicoRestrWithIndices(fes, nelem, indices, ceed, restr);
}
}
static void InitCoeffRestrictionWithIndicesImpl(
const mfem::FiniteElementSpace &fes,
int nelem,
const int* indices,
int nquads,
int ncomp,
Ceed ceed,
CeedElemRestriction *restr)
{
mfem::Array<int> tp_el_dof(nelem*nquads);
const int stride_quad = ncomp;
const int stride_elem = ncomp*nquads;
// TODO generalize to support different #quads
for (int i = 0; i < nelem; i++)
{
const int elem_index = indices[i];
const int el_offset = elem_index * stride_elem;
for (int j = 0; j < nquads; j++)
{
tp_el_dof[j + nquads * i] = j * stride_quad + el_offset;
}
}
CeedElemRestrictionCreate(ceed, nelem, nquads, ncomp, 1,
ncomp*fes.GetNE()*nquads,
CEED_MEM_HOST, CEED_COPY_VALUES,
tp_el_dof.GetData(), restr);
}
void InitStridedRestriction(const mfem::FiniteElementSpace &fes,
CeedInt nelem, CeedInt nqpts, CeedInt qdatasize,
const CeedInt *strides,
CeedElemRestriction *restr)
{
RestrKey restr_key(&fes, nelem, nqpts, qdatasize, restr_type::Strided);
auto restr_itr = mfem::internal::ceed_restr_map.find(restr_key);
if (restr_itr == mfem::internal::ceed_restr_map.end())
{
CeedElemRestrictionCreateStrided(mfem::internal::ceed, nelem, nqpts, qdatasize,
nelem*nqpts*qdatasize,
strides,
restr);
// Will be automatically destroyed when @a fes gets destroyed.
mfem::internal::ceed_restr_map[restr_key] = *restr;
}
else
{
*restr = restr_itr->second;
}
}
void InitRestriction(const FiniteElementSpace &fes,
Ceed ceed,
CeedElemRestriction *restr)
{
// Check for FES -> basis, restriction in hash tables
const mfem::FiniteElement *fe = fes.GetFE(0);
const int P = fe->GetDof();
const int nelem = fes.GetNE();
const int ncomp = fes.GetVDim();
RestrKey restr_key(&fes, nelem, P, ncomp, restr_type::Standard);
auto restr_itr = mfem::internal::ceed_restr_map.find(restr_key);
// Init or retreive key values
if (restr_itr == mfem::internal::ceed_restr_map.end())
{
InitRestrictionImpl(fes, ceed, restr);
mfem::internal::ceed_restr_map[restr_key] = *restr;
}
else
{
*restr = restr_itr->second;
}
}
void InitRestrictionWithIndices(const FiniteElementSpace &fes,
int nelem,
const int* indices,
Ceed ceed,
CeedElemRestriction *restr)
{
// Check for FES -> basis, restriction in hash tables
const mfem::FiniteElement *fe = fes.GetFE(indices[0]);
const int P = fe->GetDof();
const int ncomp = fes.GetVDim();
RestrKey restr_key(&fes, nelem, P, ncomp, restr_type::Standard);
auto restr_itr = mfem::internal::ceed_restr_map.find(restr_key);
// Init or retreive key values
if (restr_itr == mfem::internal::ceed_restr_map.end())
{
InitRestrictionWithIndicesImpl(fes, nelem, indices, ceed, restr);
mfem::internal::ceed_restr_map[restr_key] = *restr;
}
else
{
*restr = restr_itr->second;
}
}
void InitCoeffRestrictionWithIndices(const FiniteElementSpace &fes,
int nelem,
const int* indices,
int nquads,
int ncomp,
Ceed ceed,
CeedElemRestriction *restr)
{
// Check for FES -> basis, restriction in hash tables
RestrKey restr_key(&fes, nelem, nquads, ncomp, restr_type::Coeff);
auto restr_itr = mfem::internal::ceed_restr_map.find(restr_key);
// Init or retreive key values
if (restr_itr == mfem::internal::ceed_restr_map.end())
{
InitCoeffRestrictionWithIndicesImpl(fes, nelem, indices, nquads, ncomp,
ceed, restr);
mfem::internal::ceed_restr_map[restr_key] = *restr;
}
else
{
*restr = restr_itr->second;
}
}
#endif
} // namespace ceed
} // namespace mfem
-88
View File
@@ -1,88 +0,0 @@
// Copyright (c) 2010-2022, Lawrence Livermore National Security, LLC. Produced
// at the Lawrence Livermore National Laboratory. All Rights reserved. See files
// LICENSE and NOTICE for details. LLNL-CODE-806117.
//
// This file is part of the MFEM library. For more information and source code
// availability visit https://mfem.org.
//
// MFEM is free software; you can redistribute it and/or modify it under the
// terms of the BSD-3 license. We welcome feedback and contributions, see file
// CONTRIBUTING.md for details.
#ifndef MFEM_LIBCEED_RESTR
#define MFEM_LIBCEED_RESTR
#include "ceed.hpp"
namespace mfem
{
namespace ceed
{
#ifdef MFEM_USE_CEED
/** @brief Initialize a CeedElemRestriction for non-mixed meshes.
@param[in] fes Input finite element space.
@param[in] ceed Input Ceed object.
@param[out] restr The address of the initialized CeedElemRestriction object.
*/
void InitRestriction(const FiniteElementSpace &fes,
Ceed ceed,
CeedElemRestriction *restr);
/** @brief Initialize a CeedElemRestriction for mixed meshes.
@param[in] fes The finite element space.
@param[in] ceed The Ceed object.
@param[in] nelem The number of elements.
@param[in] indices The indices of the elements of same type in the
`FiniteElementSpace`.
@param[out] restr The `CeedElemRestriction` to initialize. */
void InitRestrictionWithIndices(const FiniteElementSpace &fes,
int nelem,
const int* indices,
Ceed ceed,
CeedElemRestriction *restr);
/** @brief Initialize a strided CeedElemRestriction
@param[in] nelem is the number of elements.
@param[in] nqpts is the total number of quadrature points.
@param[in] qdatasize is the number of data per quadrature point.
@param[in] strides Array for strides between [nodes, components, elements].
Data for node i, component j, element k can be found in the L-vector at
index i*strides[0] + j*strides[1] + k*strides[2]. CEED_STRIDES_BACKEND may
be used with vectors created by a Ceed backend.
@param[out] restr The `CeedElemRestriction` to initialize. */
void InitStridedRestriction(const mfem::FiniteElementSpace &fes,
CeedInt nelem, CeedInt nqpts, CeedInt qdatasize,
const CeedInt *strides,
CeedElemRestriction *restr);
/** @brief Initialize a CeedElemRestriction for a mfem::Coefficient on a mixed
mesh.
@param[in] fes The finite element space.
@param[in] nelem is the number of elements.
@param[in] indices The indices of the elements of same type in the
`FiniteElementSpace`.
@param[in] nquads is the total number of quadrature points
@param[in] ncomp is the number of data per quadrature point
@param[in] ceed The Ceed object.
@param[out] restr The `CeedElemRestriction` to initialize. */
void InitCoeffRestrictionWithIndices(const FiniteElementSpace &fes,
int nelem,
const int* indices,
int nquads,
int ncomp,
Ceed ceed,
CeedElemRestriction *restr);
#endif
} // namespace ceed
} // namespace mfem
#endif // MFEM_LIBCEED_RESTR
-281
View File
@@ -1,281 +0,0 @@
// Copyright (c) 2010-2022, Lawrence Livermore National Security, LLC. Produced
// at the Lawrence Livermore National Laboratory. All Rights reserved. See files
// LICENSE and NOTICE for details. LLNL-CODE-806117.
//
// This file is part of the MFEM library. For more information and source code
// availability visit https://mfem.org.
//
// MFEM is free software; you can redistribute it and/or modify it under the
// terms of the BSD-3 license. We welcome feedback and contributions, see file
// CONTRIBUTING.md for details.
#include "../../../general/device.hpp"
#include "../../../fem/gridfunc.hpp"
#include "../../../linalg/dtensor.hpp"
#include "basis.hpp"
#include "restriction.hpp"
#include "ceed.hpp"
#include <sys/types.h>
#include <sys/stat.h>
#ifndef _WIN32
typedef struct stat struct_stat;
#else
#define stat(dir, buf) _stat(dir, buf)
#define S_ISDIR(mode) _S_IFDIR(mode)
typedef struct _stat struct_stat;
#endif
namespace mfem
{
bool DeviceCanUseCeed()
{
return Device::Allows(Backend::CEED_MASK);
}
namespace ceed
{
void RemoveBasisAndRestriction(const mfem::FiniteElementSpace *fes)
{
#ifdef MFEM_USE_CEED
auto itb = mfem::internal::ceed_basis_map.begin();
while (itb != mfem::internal::ceed_basis_map.end())
{
if (std::get<0>(itb->first)==fes)
{
CeedBasisDestroy(&itb->second);
itb = mfem::internal::ceed_basis_map.erase(itb);
}
else
{
itb++;
}
}
auto itr = mfem::internal::ceed_restr_map.begin();
while (itr != mfem::internal::ceed_restr_map.end())
{
if (std::get<0>(itr->first)==fes)
{
CeedElemRestrictionDestroy(&itr->second);
itr = mfem::internal::ceed_restr_map.erase(itr);
}
else
{
itr++;
}
}
#endif
}
#ifdef MFEM_USE_CEED
void InitVector(const mfem::Vector &v, CeedVector &cv)
{
CeedVectorCreate(mfem::internal::ceed, v.Size(), &cv);
CeedScalar *cv_ptr;
CeedMemType mem;
CeedGetPreferredMemType(mfem::internal::ceed, &mem);
if ( Device::Allows(Backend::DEVICE_MASK) && mem==CEED_MEM_DEVICE )
{
cv_ptr = const_cast<CeedScalar*>(v.Read());
}
else
{
cv_ptr = const_cast<CeedScalar*>(v.HostRead());
mem = CEED_MEM_HOST;
}
CeedVectorSetArray(cv, mem, CEED_USE_POINTER, cv_ptr);
}
void InitBasisAndRestriction(const FiniteElementSpace &fes,
const IntegrationRule &irm,
Ceed ceed, CeedBasis *basis,
CeedElemRestriction *restr)
{
InitBasis(fes, irm, ceed, basis);
InitRestriction(fes, ceed, restr);
}
void InitBasisAndRestrictionWithIndices(const FiniteElementSpace &fes,
const IntegrationRule &irm,
int nelem,
const int* indices,
Ceed ceed, CeedBasis *basis,
CeedElemRestriction *restr)
{
InitBasisWithIndices(fes, irm, nelem, indices, ceed, basis);
InitRestrictionWithIndices(fes, nelem, indices, ceed, restr);
}
void InitBasisAndRestriction(const FiniteElementSpace &fes,
const IntegrationRule &irm,
int nelem,
const int* indices,
Ceed ceed, CeedBasis *basis,
CeedElemRestriction *restr)
{
if (indices)
{
InitBasisAndRestrictionWithIndices(fes,irm,nelem,indices,ceed,basis,restr);
}
else
{
InitBasisAndRestriction(fes,irm,ceed,basis,restr);
}
}
// Assumes a tensor-product operator with one active field
int CeedOperatorGetActiveField(CeedOperator oper, CeedOperatorField *field)
{
int ierr;
Ceed ceed;
ierr = CeedOperatorGetCeed(oper, &ceed); CeedChk(ierr);
CeedQFunction qf;
bool isComposite;
ierr = CeedOperatorIsComposite(oper, &isComposite); CeedChk(ierr);
CeedOperator *subops;
if (isComposite)
{
ierr = CeedOperatorGetSubList(oper, &subops); CeedChk(ierr);
ierr = CeedOperatorGetQFunction(subops[0], &qf); CeedChk(ierr);
}
else
{
ierr = CeedOperatorGetQFunction(oper, &qf); CeedChk(ierr);
}
CeedInt numinputfields, numoutputfields;
ierr = CeedQFunctionGetNumArgs(qf, &numinputfields, &numoutputfields);
CeedOperatorField *inputfields;
if (isComposite)
{
ierr = CeedOperatorGetFields(subops[0], &numinputfields, &inputfields,
&numoutputfields, NULL); CeedChk(ierr);
}
else
{
ierr = CeedOperatorGetFields(oper, &numinputfields, &inputfields,
&numoutputfields, NULL); CeedChk(ierr);
}
CeedVector if_vector;
bool found = false;
int found_index = -1;
for (int i = 0; i < numinputfields; ++i)
{
ierr = CeedOperatorFieldGetVector(inputfields[i], &if_vector); CeedChk(ierr);
if (if_vector == CEED_VECTOR_ACTIVE)
{
if (found)
{
return CeedError(ceed, 1, "Multiple active vectors in CeedOperator!");
}
found = true;
found_index = i;
}
}
if (!found)
{
return CeedError(ceed, 1, "No active vector in CeedOperator!");
}
*field = inputfields[found_index];
return 0;
}
template <>
const IntegrationRule & GetRule<MassIntegrator>(
const MassIntegrator &integ,
const FiniteElement &trial_fe,
const FiniteElement &test_fe,
ElementTransformation &trans)
{
return MassIntegrator::GetRule(trial_fe, test_fe, trans);
}
template <>
const IntegrationRule & GetRule<VectorMassIntegrator>(
const VectorMassIntegrator &integ,
const FiniteElement &trial_fe,
const FiniteElement &test_fe,
ElementTransformation &trans)
{
return MassIntegrator::GetRule(trial_fe, test_fe, trans);
}
template <>
const IntegrationRule & GetRule<ConvectionIntegrator>(
const ConvectionIntegrator &integ,
const FiniteElement &trial_fe,
const FiniteElement &test_fe,
ElementTransformation &trans)
{
return ConvectionIntegrator::GetRule(trial_fe, test_fe, trans);
}
template <>
const IntegrationRule & GetRule<VectorConvectionNLFIntegrator>(
const VectorConvectionNLFIntegrator &integ,
const FiniteElement &trial_fe,
const FiniteElement &test_fe,
ElementTransformation &trans)
{
return VectorConvectionNLFIntegrator::GetRule(trial_fe, trans);
}
template <>
const IntegrationRule & GetRule<DiffusionIntegrator>(
const DiffusionIntegrator &integ,
const FiniteElement &trial_fe,
const FiniteElement &test_fe,
ElementTransformation &trans)
{
return DiffusionIntegrator::GetRule(trial_fe, test_fe);
}
template <>
const IntegrationRule & GetRule<VectorDiffusionIntegrator>(
const VectorDiffusionIntegrator &integ,
const FiniteElement &trial_fe,
const FiniteElement &test_fe,
ElementTransformation &trans)
{
return DiffusionIntegrator::GetRule(trial_fe, test_fe);
}
std::string ceed_path;
const std::string &GetCeedPath()
{
if (ceed_path.empty())
{
const char *install_dir = MFEM_INSTALL_DIR "/include/mfem/fem/ceed";
const char *source_dir = MFEM_SOURCE_DIR "/fem/ceed";
struct_stat m_stat;
if (stat(install_dir, &m_stat) == 0 && S_ISDIR(m_stat.st_mode))
{
ceed_path = install_dir;
}
else if (stat(source_dir, &m_stat) == 0 && S_ISDIR(m_stat.st_mode))
{
ceed_path = source_dir;
}
else
{
MFEM_ABORT("Cannot find libCEED kernels in MFEM_INSTALL_DIR or "
"MFEM_SOURCE_DIR");
}
// Could be useful for debugging:
// out << "Using libCEED dir: " << ceed_path << std::endl;
}
return ceed_path;
}
#endif
} // namespace ceed
} // namespace mfem
@@ -11,7 +11,7 @@
#include "mass.hpp"
#include "../../../../config/config.hpp"
#include "../../config/config.hpp"
#ifdef MFEM_USE_CEED
#include "mass_qf.h"
#endif
@@ -28,7 +28,7 @@ struct MassOperatorInfo : public OperatorInfo
MassContext ctx;
MassOperatorInfo()
{
header = "/integrators/mass/mass_qf.h";
header = "/mass_qf.h";
build_func_const = ":f_build_mass_const";
build_qf_const = &f_build_mass_const;
build_func_quad = ":f_build_mass_quad";
@@ -59,30 +59,6 @@ PAMassIntegrator::PAMassIntegrator(const mfem::FiniteElementSpace &fes,
#endif
}
MixedPAMassIntegrator::MixedPAMassIntegrator(const MassIntegrator &integ,
const mfem::FiniteElementSpace &fes,
mfem::Coefficient *Q)
{
#ifdef MFEM_USE_CEED
MassOperatorInfo info;
Assemble(integ, info, fes, Q);
#else
MFEM_ABORT("MFEM must be built with MFEM_USE_CEED=YES to use libCEED.");
#endif
}
MixedPAMassIntegrator::MixedPAMassIntegrator(const VectorMassIntegrator &integ,
const mfem::FiniteElementSpace &fes,
mfem::Coefficient *Q)
{
#ifdef MFEM_USE_CEED
MassOperatorInfo info;
Assemble(integ, info, fes, Q);
#else
MFEM_ABORT("MFEM must be built with MFEM_USE_CEED=YES to use libCEED.");
#endif
}
MFMassIntegrator::MFMassIntegrator(const mfem::FiniteElementSpace &fes,
const mfem::IntegrationRule &irm,
mfem::Coefficient *Q)
@@ -96,30 +72,6 @@ MFMassIntegrator::MFMassIntegrator(const mfem::FiniteElementSpace &fes,
#endif
}
MixedMFMassIntegrator::MixedMFMassIntegrator(const MassIntegrator &integ,
const mfem::FiniteElementSpace &fes,
mfem::Coefficient *Q)
{
#ifdef MFEM_USE_CEED
MassOperatorInfo info;
Assemble(integ, info, fes, Q);
#else
MFEM_ABORT("MFEM must be built with MFEM_USE_CEED=YES to use libCEED.");
#endif
}
MixedMFMassIntegrator::MixedMFMassIntegrator(const VectorMassIntegrator &integ,
const mfem::FiniteElementSpace &fes,
mfem::Coefficient *Q)
{
#ifdef MFEM_USE_CEED
MassOperatorInfo info;
Assemble(integ, info, fes, Q);
#else
MFEM_ABORT("MFEM must be built with MFEM_USE_CEED=YES to use libCEED.");
#endif
}
} // namespace ceed
} // namespace mfem
@@ -12,9 +12,8 @@
#ifndef MFEM_LIBCEED_MASS_HPP
#define MFEM_LIBCEED_MASS_HPP
#include "../../interface/integrator.hpp"
#include "../../interface/mixed_integrator.hpp"
#include "../../../fespace.hpp"
#include "integrator.hpp"
#include "../fespace.hpp"
namespace mfem
{
@@ -27,43 +26,19 @@ class PAMassIntegrator : public PAIntegrator
{
public:
PAMassIntegrator(const mfem::FiniteElementSpace &fes,
const mfem::IntegrationRule &ir,
const mfem::IntegrationRule &irm,
mfem::Coefficient *Q);
};
class MixedPAMassIntegrator : public MixedIntegrator<PAIntegrator>
{
public:
MixedPAMassIntegrator(const MassIntegrator &integ,
const mfem::FiniteElementSpace &fes,
mfem::Coefficient *Q);
MixedPAMassIntegrator(const VectorMassIntegrator &integ,
const mfem::FiniteElementSpace &fes,
mfem::Coefficient *Q);
};
/// Represent a MassIntegrator with AssemblyLevel::None using libCEED.
class MFMassIntegrator : public MFIntegrator
{
public:
MFMassIntegrator(const mfem::FiniteElementSpace &fes,
const mfem::IntegrationRule &ir,
const mfem::IntegrationRule &irm,
mfem::Coefficient *Q);
};
class MixedMFMassIntegrator : public MixedIntegrator<MFIntegrator>
{
public:
MixedMFMassIntegrator(const MassIntegrator &integ,
const mfem::FiniteElementSpace &fes,
mfem::Coefficient *Q);
MixedMFMassIntegrator(const VectorMassIntegrator &integ,
const mfem::FiniteElementSpace &fes,
mfem::Coefficient *Q);
};
}
}
@@ -11,7 +11,7 @@
#include "nlconvection.hpp"
#include "../../../../config/config.hpp"
#include "../../config/config.hpp"
#ifdef MFEM_USE_CEED
#include "nlconvection_qf.h"
#endif
@@ -28,7 +28,7 @@ struct NLConvectionOperatorInfo : public OperatorInfo
NLConvectionContext ctx;
NLConvectionOperatorInfo(int dim)
{
header = "/integrators/nlconvection/nlconvection_qf.h";
header = "/nlconvection_qf.h";
build_func_const = ":f_build_conv_const";
build_qf_const = &f_build_conv_const;
build_func_quad = ":f_build_conv_quad";
@@ -60,19 +60,6 @@ PAVectorConvectionNLFIntegrator::PAVectorConvectionNLFIntegrator(
#endif
}
MixedPAVectorConvectionNLIntegrator::MixedPAVectorConvectionNLIntegrator(
const VectorConvectionNLFIntegrator &integ,
const mfem::FiniteElementSpace &fes,
mfem::Coefficient *Q)
{
#ifdef MFEM_USE_CEED
NLConvectionOperatorInfo info(fes.GetMesh()->Dimension());
Assemble(integ, info, fes, Q);
#else
MFEM_ABORT("MFEM must be built with MFEM_USE_CEED=YES to use libCEED.");
#endif
}
MFVectorConvectionNLFIntegrator::MFVectorConvectionNLFIntegrator(
const mfem::FiniteElementSpace &fes,
const mfem::IntegrationRule &irm,
@@ -87,19 +74,6 @@ MFVectorConvectionNLFIntegrator::MFVectorConvectionNLFIntegrator(
#endif
}
MixedMFVectorConvectionNLIntegrator::MixedMFVectorConvectionNLIntegrator(
const VectorConvectionNLFIntegrator &integ,
const mfem::FiniteElementSpace &fes,
mfem::Coefficient *Q)
{
#ifdef MFEM_USE_CEED
NLConvectionOperatorInfo info(fes.GetMesh()->Dimension());
Assemble(integ, info, fes, Q);
#else
MFEM_ABORT("MFEM must be built with MFEM_USE_CEED=YES to use libCEED.");
#endif
}
} // namespace ceed
} // namespace mfem
@@ -12,9 +12,8 @@
#ifndef MFEM_LIBCEED_NLCONV_HPP
#define MFEM_LIBCEED_NLCONV_HPP
#include "../../interface/integrator.hpp"
#include "../../interface/mixed_integrator.hpp"
#include "../../../fespace.hpp"
#include "integrator.hpp"
#include "../fespace.hpp"
namespace mfem
{
@@ -32,15 +31,6 @@ public:
mfem::Coefficient *coeff);
};
class MixedPAVectorConvectionNLIntegrator : public MixedIntegrator<PAIntegrator>
{
public:
MixedPAVectorConvectionNLIntegrator(
const VectorConvectionNLFIntegrator &integ,
const mfem::FiniteElementSpace &fes,
mfem::Coefficient *Q);
};
/** Represent a VectorConvectionNLFIntegrator with AssemblyLevel::None
using libCEED. */
class MFVectorConvectionNLFIntegrator : public MFIntegrator
@@ -51,15 +41,6 @@ public:
mfem::Coefficient *coeff);
};
class MixedMFVectorConvectionNLIntegrator : public MixedIntegrator<MFIntegrator>
{
public:
MixedMFVectorConvectionNLIntegrator(
const VectorConvectionNLFIntegrator &integ,
const mfem::FiniteElementSpace &fes,
mfem::Coefficient *Q);
};
}
}
@@ -327,13 +327,13 @@ CEED_QFUNCTION(f_apply_conv_mf_const)(void *ctx, CeedInt Q,
const CeedScalar A33 = J11 * J22 - J12 * J21;
const CeedScalar w = qw[i] * coeff;
const CeedScalar qd00 = w * A11;
const CeedScalar qd10 = w * A21;
const CeedScalar qd20 = w * A31;
const CeedScalar qd01 = w * A12;
const CeedScalar qd01 = w * A21;
const CeedScalar qd02 = w * A31;
const CeedScalar qd10 = w * A12;
const CeedScalar qd11 = w * A22;
const CeedScalar qd21 = w * A32;
const CeedScalar qd02 = w * A13;
const CeedScalar qd12 = w * A23;
const CeedScalar qd12 = w * A32;
const CeedScalar qd20 = w * A13;
const CeedScalar qd21 = w * A23;
const CeedScalar qd22 = w * A33;
const CeedScalar u0 = u[i + Q * 0];
const CeedScalar u1 = u[i + Q * 1];
@@ -440,13 +440,13 @@ CEED_QFUNCTION(f_apply_conv_mf_quad)(void *ctx, CeedInt Q,
const CeedScalar A33 = J11 * J22 - J12 * J21;
const CeedScalar w = qw[i] * c[i];
const CeedScalar qd00 = w * A11;
const CeedScalar qd10 = w * A21;
const CeedScalar qd20 = w * A31;
const CeedScalar qd01 = w * A12;
const CeedScalar qd01 = w * A21;
const CeedScalar qd02 = w * A31;
const CeedScalar qd10 = w * A12;
const CeedScalar qd11 = w * A22;
const CeedScalar qd21 = w * A32;
const CeedScalar qd02 = w * A13;
const CeedScalar qd12 = w * A23;
const CeedScalar qd12 = w * A32;
const CeedScalar qd20 = w * A13;
const CeedScalar qd21 = w * A23;
const CeedScalar qd22 = w * A33;
const CeedScalar u0 = u[i + Q * 0];
const CeedScalar u1 = u[i + Q * 1];
@@ -11,11 +11,9 @@
#include "operator.hpp"
#include "../../../config/config.hpp"
#include "../../../linalg/vector.hpp"
#include "../../fespace.hpp"
#include "util.hpp"
#include "ceed.hpp"
#include "../../config/config.hpp"
#include "../../linalg/vector.hpp"
#include "../fespace.hpp"
namespace mfem
{
@@ -12,8 +12,8 @@
#ifndef MFEM_LIBCEED_OPERATOR
#define MFEM_LIBCEED_OPERATOR
#include "../../../linalg/operator.hpp"
#include "ceed.hpp"
#include "util.hpp"
#include "../../linalg/operator.hpp"
namespace mfem
{
@@ -41,6 +41,7 @@ public:
void AddMult(const mfem::Vector &x, mfem::Vector &y) const;
void GetDiagonal(mfem::Vector &diag) const;
using mfem::Operator::SetupRAP;
virtual ~Operator()
{
#ifdef MFEM_USE_CEED
@@ -10,9 +10,7 @@
// CONTRIBUTING.md for details.
#include "solvers-atpmg.hpp"
#include "../interface/ceed.hpp"
#include "../interface/util.hpp"
#include "util.hpp"
#ifdef MFEM_USE_CEED
#include <ceed/backend.h>
@@ -12,10 +12,12 @@
#ifndef MFEM_CEEDSOLVERS_ATPMG_H
#define MFEM_CEEDSOLVERS_ATPMG_H
#include "../interface/ceed.hpp"
#include "../../config/config.hpp"
#ifdef MFEM_USE_CEED
#include <ceed.h>
namespace mfem
{

Some files were not shown because too many files have changed in this diff Show More