Compare commits
166
Commits
| Author | SHA1 | Date | |
|---|---|---|---|
|
|
84779d25b8 | ||
|
|
026d4834d2 | ||
|
|
d0caa1c432 | ||
|
|
603e545060 | ||
|
|
8db0c68dd2 | ||
|
|
6e6a3d9751 | ||
|
|
5f1b8614c6 | ||
|
|
a05a93c187 | ||
|
|
7864972fc1 | ||
|
|
ab73eea018 | ||
|
|
81421858f3 | ||
|
|
9f7baced30 | ||
|
|
a45d86ff2a | ||
|
|
a17c6530ff | ||
|
|
00b1ae1850 | ||
|
|
3c25eeab3d | ||
|
|
ac91aef8f9 | ||
|
|
c86860362a | ||
|
|
f5991995c3 | ||
|
|
b8264a5fb2 | ||
|
|
cf90fad7fb | ||
|
|
f8dae2249f | ||
|
|
fd7b54c41a | ||
|
|
4d8cd51cca | ||
|
|
d25c9ffff0 | ||
|
|
72204b7f01 | ||
|
|
4d037ca87a | ||
|
|
49a47922f2 | ||
|
|
f3b094d8d8 | ||
|
|
46ed6ee7b6 | ||
|
|
22849928ac | ||
|
|
ea0dfd78db | ||
|
|
27d6b2be89 | ||
|
|
7be7168769 | ||
|
|
6f637f57b0 | ||
|
|
bf542247e8 | ||
|
|
0cd6d21e8f | ||
|
|
4557ab17b4 | ||
|
|
844644957d | ||
|
|
a0dfc29c71 | ||
|
|
0446a8fa32 | ||
|
|
4256a61cd7 | ||
|
|
8f647fa4ef | ||
|
|
5da228a103 | ||
|
|
a20ec1f29f | ||
|
|
9066476c67 | ||
|
|
0c639efcd5 | ||
|
|
16cca1a724 | ||
|
|
157e3dc151 | ||
|
|
598a5994bc | ||
|
|
6a088021e7 | ||
|
|
703083a43c | ||
|
|
f45fbc53da | ||
|
|
345abd1a5a | ||
|
|
39e7705da1 | ||
|
|
154e100214 | ||
|
|
04bdfc9409 | ||
|
|
34cf1432e9 | ||
|
|
b723fd39cc | ||
|
|
eb863cff5f | ||
|
|
e9b8ecdff7 | ||
|
|
27c55c2280 | ||
|
|
21f37f99d2 | ||
|
|
63cfc40b41 | ||
|
|
73286c6f4d | ||
|
|
b3c2f05887 | ||
|
|
fd846ee3e1 | ||
|
|
eee909351d | ||
|
|
f79415ed65 | ||
|
|
462b21fe54 | ||
|
|
d3941dfb60 | ||
|
|
2dbc9ca892 | ||
|
|
a631e6909f | ||
|
|
4b1d423210 | ||
|
|
37af117694 | ||
|
|
c1485a49ed | ||
|
|
dcbf4f1850 | ||
|
|
98e9c4cc37 | ||
|
|
cdb92cefd0 | ||
|
|
6c1f52d805 | ||
|
|
eb1493dd80 | ||
|
|
0696c3d662 | ||
|
|
a16f6a2450 | ||
|
|
f01988c0cc | ||
|
|
604cfaa65a | ||
|
|
0c1db68601 | ||
|
|
e1733d5b29 | ||
|
|
7dfaaa6806 | ||
|
|
6d3d44bca6 | ||
|
|
4099390b15 | ||
|
|
69ac786a0c | ||
|
|
bf7a36735e | ||
|
|
72764aef89 | ||
|
|
13ed072d48 | ||
|
|
a472f4a1a3 | ||
|
|
c9a0b7195b | ||
|
|
5162cfb89f | ||
|
|
64494e2b24 | ||
|
|
989866a7aa | ||
|
|
c4a744154e | ||
|
|
21d7b39e0f | ||
|
|
9a98a2359e | ||
|
|
1f6158a726 | ||
|
|
0969273d99 | ||
|
|
37586c597c | ||
|
|
1486c07dad | ||
|
|
d090a456cf | ||
|
|
1017da8f51 | ||
|
|
7789aa9387 | ||
|
|
dde5c2a230 | ||
|
|
ec615d590b | ||
|
|
5d1b2d9261 | ||
|
|
038ee646ee | ||
|
|
5338aa8bb7 | ||
|
|
b6eac61043 | ||
|
|
c869398f60 | ||
|
|
ab6f3a546f | ||
|
|
beeabc53c6 | ||
|
|
f0124d2fc5 | ||
|
|
54dc1405dc | ||
|
|
677aa95e48 | ||
|
|
d0eaf9760a | ||
|
|
f2c9f8e400 | ||
|
|
a8a00821d1 | ||
|
|
67bd515ed7 | ||
|
|
ac8784e122 | ||
|
|
0d98a14863 | ||
|
|
a27ab37d62 | ||
|
|
b553390c8b | ||
|
|
f8a7405fb8 | ||
|
|
76d384d6be | ||
|
|
76382289d0 | ||
|
|
e0d52ab9f7 | ||
|
|
2214130993 | ||
|
|
2b760e8a0a | ||
|
|
ecfd6ff848 | ||
|
|
f96929cc9e | ||
|
|
dd4eb1a7b9 | ||
|
|
b2dce094f7 | ||
|
|
e96797c7db | ||
|
|
157f3e1a43 | ||
|
|
c9c181d25b | ||
|
|
9b5b9ddcbb | ||
|
|
b328746f4a | ||
|
|
da6f519346 | ||
|
|
806595ccc2 | ||
|
|
48183748ba | ||
|
|
06ccc3cc29 | ||
|
|
07be8038a2 | ||
|
|
11801d19a8 | ||
|
|
a7f4a57370 | ||
|
|
323000faee | ||
|
|
bb78d8514f | ||
|
|
fb596116ab | ||
|
|
1cf11888b2 | ||
|
|
7a2bae81d9 | ||
|
|
63889ef70f | ||
|
|
2ba5aeaa93 | ||
|
|
f581eec582 | ||
|
|
defc0b741a | ||
|
|
330764a2e7 | ||
|
|
3e4ae315e7 | ||
|
|
769e9eeba5 | ||
|
|
b0d7d3e909 | ||
|
|
3e6bcc91f8 | ||
|
|
3450a067d3 |
@@ -47,26 +47,36 @@ jobs:
|
||||
strategy:
|
||||
matrix:
|
||||
os: [ubuntu-18.04, macos-10.15]
|
||||
target: [debug, optim]
|
||||
mpi: [sequential, parallel]
|
||||
target: [dbg, opt]
|
||||
mpi: [seq, par]
|
||||
build-system: [make]
|
||||
# 'include' allows us to
|
||||
# - add a variable without creating a new matrix dimension.
|
||||
# - add a new combination ('build-system: cmake' case here)
|
||||
hypre-target: [int32]
|
||||
# 'include' allows us to:
|
||||
# - Add a variable to all jobs without creating a new matrix dimension.
|
||||
# Codecov is defined that way.
|
||||
# - Add a new combination.
|
||||
# 'build-system: cmake' and 'hypre-target: int64'
|
||||
#
|
||||
# note: we will gather coverage info for any non-debug run except the
|
||||
# CMake build.
|
||||
include:
|
||||
- target: debug
|
||||
- target: dbg
|
||||
codecov: NO
|
||||
- target: optim
|
||||
- target: opt
|
||||
codecov: YES
|
||||
- os: ubuntu-18.04
|
||||
target: optim
|
||||
target: opt
|
||||
codecov: NO
|
||||
mpi: parallel
|
||||
mpi: par
|
||||
build-system: cmake
|
||||
name: ${{ matrix.os }}-${{ matrix.target }}-${{ matrix.mpi }}-${{ matrix.build-system }}
|
||||
hypre-target: int32
|
||||
- os: ubuntu-18.04
|
||||
target: opt
|
||||
codecov: NO
|
||||
mpi: par
|
||||
build-system: make
|
||||
hypre-target: int64
|
||||
name: ${{ matrix.os }}-${{ matrix.build-system }}-${{ matrix.target }}-${{ matrix.mpi }}-${{ matrix.hypre-target }}
|
||||
|
||||
runs-on: ${{ matrix.os }}
|
||||
|
||||
@@ -92,7 +102,7 @@ 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 == 'parallel' && matrix.os == 'ubuntu-18.04'
|
||||
if: matrix.mpi == 'par' && matrix.os == 'ubuntu-18.04'
|
||||
run: |
|
||||
sudo apt-get install mpich libmpich-dev
|
||||
export MAKE_CXX_FLAG="MPICXX=mpic++"
|
||||
@@ -103,11 +113,11 @@ jobs:
|
||||
sudo apt-get install lcov
|
||||
|
||||
- name: Set up Homebrew
|
||||
if: ( matrix.mpi == 'parallel' || matrix.codecov == 'YES' ) && matrix.os == 'macos-10.15'
|
||||
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 == 'parallel' && matrix.os == 'macos-10.15'
|
||||
if: matrix.mpi == 'par' && matrix.os == 'macos-10.15'
|
||||
run: |
|
||||
export HOMEBREW_NO_INSTALL_CLEANUP=1
|
||||
brew install openmpi
|
||||
@@ -123,39 +133,40 @@ jobs:
|
||||
# Install will only run on cache miss.
|
||||
- name: cache hypre
|
||||
id: hypre-cache
|
||||
if: matrix.mpi == 'parallel'
|
||||
if: matrix.mpi == 'par'
|
||||
uses: actions/cache@v2
|
||||
with:
|
||||
path: ${{ env.HYPRE_TOP_DIR }}
|
||||
key: ${{ runner.os }}-build-${{ env.HYPRE_TOP_DIR }}-v2
|
||||
key: ${{ runner.os }}-build-${{ env.HYPRE_TOP_DIR }}-${{ matrix.hypre-target }}-v2.0
|
||||
|
||||
- name: get hypre
|
||||
if: matrix.mpi == 'parallel' && steps.hypre-cache.outputs.cache-hit != 'true'
|
||||
uses: mfem/github-actions/build-hypre@v1.0
|
||||
if: matrix.mpi == 'par' && steps.hypre-cache.outputs.cache-hit != 'true'
|
||||
uses: mfem/github-actions/build-hypre@v2.0
|
||||
with:
|
||||
hypre-archive: ${{ env.HYPRE_ARCHIVE }}
|
||||
hypre-dir: ${{ env.HYPRE_TOP_DIR }}
|
||||
archive: ${{ env.HYPRE_ARCHIVE }}
|
||||
dir: ${{ env.HYPRE_TOP_DIR }}
|
||||
target: ${{ matrix.hypre-target }}
|
||||
|
||||
# Get Metis through cache, or build it.
|
||||
# Install will only run on cache miss.
|
||||
- name: cache metis
|
||||
id: metis-cache
|
||||
if: matrix.mpi == 'parallel'
|
||||
if: matrix.mpi == 'par'
|
||||
uses: actions/cache@v2
|
||||
with:
|
||||
path: ${{ env.METIS_TOP_DIR }}
|
||||
key: ${{ runner.os }}-build-${{ env.METIS_TOP_DIR }}-v2
|
||||
key: ${{ runner.os }}-build-${{ env.METIS_TOP_DIR }}-v2.0
|
||||
|
||||
- name: install metis
|
||||
if: matrix.mpi == 'parallel' && steps.metis-cache.outputs.cache-hit != 'true'
|
||||
uses: mfem/github-actions/build-metis@v1.0
|
||||
if: matrix.mpi == 'par' && steps.metis-cache.outputs.cache-hit != 'true'
|
||||
uses: mfem/github-actions/build-metis@v2.0
|
||||
with:
|
||||
metis-archive: ${{ env.METIS_ARCHIVE }}
|
||||
metis-dir: ${{ env.METIS_TOP_DIR }}
|
||||
archive: ${{ env.METIS_ARCHIVE }}
|
||||
dir: ${{ env.METIS_TOP_DIR }}
|
||||
|
||||
# MFEM build and test
|
||||
- name: build
|
||||
uses: mfem/github-actions/build-mfem@v1.0
|
||||
uses: mfem/github-actions/build-mfem@v2.0
|
||||
with:
|
||||
os: ${{ matrix.os }}
|
||||
target: ${{ matrix.target }}
|
||||
@@ -168,17 +179,17 @@ jobs:
|
||||
|
||||
# Run checks (and only checks) on debug targets
|
||||
- name: checks
|
||||
if: matrix.build-system == 'make' && matrix.target == 'debug'
|
||||
if: matrix.build-system == 'make' && matrix.target == 'dbg'
|
||||
run: |
|
||||
cd ${{ env.MFEM_TOP_DIR }} && make check
|
||||
|
||||
- name: unit tests
|
||||
if: matrix.build-system == 'make' && matrix.target == 'optim'
|
||||
if: matrix.build-system == 'make' && matrix.target == 'opt'
|
||||
run: |
|
||||
cd ${{ env.MFEM_TOP_DIR }} && make unittest
|
||||
|
||||
- name: tests
|
||||
if: matrix.build-system == 'make' && matrix.target == 'optim'
|
||||
if: matrix.build-system == 'make' && matrix.target == 'opt'
|
||||
run: |
|
||||
cd ${{ env.MFEM_TOP_DIR }} && make test
|
||||
|
||||
@@ -190,8 +201,8 @@ jobs:
|
||||
# Code coverage (process and upload reports)
|
||||
- name: codecov
|
||||
if: matrix.codecov == 'YES'
|
||||
uses: mfem/github-actions/upload-coverage@v1.0
|
||||
uses: mfem/github-actions/upload-coverage@v2.0
|
||||
with:
|
||||
name: ${{ matrix.os }}-${{ matrix.mpi }}
|
||||
name: ${{ matrix.os }}-${{ matrix.build-system }}-${{ matrix.target }}-${{ matrix.mpi }}-${{ matrix.hypre-target }}
|
||||
project_dir: ${{ env.MFEM_TOP_DIR }}
|
||||
directories: "fem general linalg mesh"
|
||||
|
||||
@@ -53,32 +53,33 @@ jobs:
|
||||
uses: actions/cache@v2
|
||||
with:
|
||||
path: ${{ env.HYPRE_TOP_DIR }}
|
||||
key: ${{ runner.os }}-build-${{ env.HYPRE_TOP_DIR }}-v2
|
||||
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@master
|
||||
uses: mfem/github-actions/build-hypre@v2.0
|
||||
with:
|
||||
hypre-archive: ${{ env.HYPRE_ARCHIVE }}
|
||||
hypre-dir: ${{ env.HYPRE_TOP_DIR }}
|
||||
archive: ${{ env.HYPRE_ARCHIVE }}
|
||||
dir: ${{ env.HYPRE_TOP_DIR }}
|
||||
target: int32
|
||||
|
||||
- name: Cache Metis Install
|
||||
id: metis-cache
|
||||
uses: actions/cache@v2
|
||||
with:
|
||||
path: ${{ env.METIS_TOP_DIR }}
|
||||
key: ${{ runner.os }}-build-${{ env.METIS_TOP_DIR }}-v2
|
||||
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@master
|
||||
uses: mfem/github-actions/build-metis@v2.0
|
||||
with:
|
||||
metis-archive: ${{ env.METIS_ARCHIVE }}
|
||||
metis-dir: ${{ env.METIS_TOP_DIR }}
|
||||
archive: ${{ env.METIS_ARCHIVE }}
|
||||
dir: ${{ env.METIS_TOP_DIR }}
|
||||
|
||||
# MFEM build and test
|
||||
- name: build-mfem
|
||||
uses: mfem/github-actions/build-mfem@master
|
||||
uses: mfem/github-actions/build-mfem@v2.0
|
||||
with:
|
||||
os: ${{ runner.os }}
|
||||
target: optim
|
||||
|
||||
@@ -28,49 +28,24 @@ jobs:
|
||||
access_token: ${{ github.token }}
|
||||
- name: checkout mfem
|
||||
uses: actions/checkout@v2
|
||||
with:
|
||||
path: mfem
|
||||
|
||||
- name: copyright check
|
||||
id: copyright
|
||||
run: |
|
||||
cd mfem
|
||||
if git grep -l "^\(#\|//\).*\(\-2020\|\ 2010,\)" > matches.txt
|
||||
then
|
||||
echo "Please update the following files to Copyright (c) 2010-2021:"
|
||||
cat matches.txt
|
||||
exit 1
|
||||
else
|
||||
echo "No outdated copyright found."
|
||||
fi
|
||||
./config/githooks/pre-push --copyright
|
||||
|
||||
continue-on-error: true
|
||||
|
||||
- name: license check
|
||||
id: license
|
||||
run: |
|
||||
cd mfem
|
||||
if git grep -li "^\(#\|//\).*GNU\ Lesser\ General\ Public\ License" > matches.txt
|
||||
then
|
||||
echo "Please update the following files to the BSD-3 license:"
|
||||
cat matches.txt
|
||||
exit 1
|
||||
else
|
||||
echo "No GNU GPL license found."
|
||||
fi
|
||||
./config/githooks/pre-push --license
|
||||
continue-on-error: true
|
||||
|
||||
- name: release check
|
||||
id: release
|
||||
run: |
|
||||
cd mfem
|
||||
if git grep -l "^\(#\|//\).*LLNL\-CODE\-443211" > matches.txt
|
||||
then
|
||||
echo "Please update the following files to LLNL-CODE-806117:"
|
||||
cat matches.txt
|
||||
exit 1
|
||||
else
|
||||
echo "No outdated release number found."
|
||||
fi
|
||||
./config/githooks/pre-push --release
|
||||
continue-on-error: true
|
||||
|
||||
- name: wrap-up
|
||||
@@ -100,8 +75,7 @@ jobs:
|
||||
|
||||
- name: style check
|
||||
run: |
|
||||
cd tests/scripts
|
||||
./runtest code-style
|
||||
./config/githooks/pre-push --style
|
||||
|
||||
documentation:
|
||||
runs-on: ubuntu-18.04
|
||||
@@ -133,5 +107,4 @@ jobs:
|
||||
run: |
|
||||
git fetch origin master:master
|
||||
git checkout -b gh-actions-branch-history
|
||||
cd tests/scripts
|
||||
./runtest branch-history
|
||||
./config/githooks/pre-push --history
|
||||
|
||||
@@ -26,6 +26,7 @@ CMakeFiles/
|
||||
config/_config.hpp
|
||||
config/config.mk
|
||||
config/sample-runs-build.log
|
||||
config/user.mk
|
||||
doc/CodeDocumentation.conf
|
||||
doc/CodeDocumentation.html
|
||||
doc/CodeDocumentation
|
||||
|
||||
+27
-8
@@ -48,39 +48,58 @@ variables:
|
||||
TPLS_REPO: ssh://git@mybitbucket.llnl.gov:7999/mfem/tpls.git
|
||||
TESTS_REPO: ssh://git@mybitbucket.llnl.gov:7999/mfem/tests.git
|
||||
AUTOTEST_REPO: ssh://git@mybitbucket.llnl.gov:7999/mfem/autotest.git
|
||||
MFEM_DATA_REPO: https://github.com/mfem/data.git
|
||||
ARTIFACTS_DIR: artifacts
|
||||
|
||||
# The pipeline is divided into stages. Usually, these are also synchronization
|
||||
# points, however, we use "needs" keyword to express the DAG of jobs for more
|
||||
# efficiency.
|
||||
# - We use setup phase to download content outside of mfem directory.
|
||||
# - We use setup and setup_baseline phases to download content outside of mfem
|
||||
# directory.
|
||||
# - Allocate/Release is where quartz resources are allocated/released once for all.
|
||||
# - Build and Test is where we build and MFEM for multiple toolchains.
|
||||
# - Baseline_checks gathers baseline-type test suites execution
|
||||
# - Baseline_publish, only available on master, allows to update baseline
|
||||
# results
|
||||
stages:
|
||||
- setup
|
||||
- q_allocate_resources
|
||||
- q_build_and_test
|
||||
- q_release_resources
|
||||
- l_build_and_test
|
||||
- c_build_and_test
|
||||
- setup
|
||||
- setup_baseline
|
||||
- baseline_check
|
||||
- baseline_to_autotest
|
||||
- baseline_publish
|
||||
|
||||
# The setup job in setup stage don't rely on MFEM git repo. It prepares a
|
||||
# pipeline-wide working directory downloading/updating external repos.
|
||||
# TODO: updating tests and tpls is not necessary anymore since pipelines are
|
||||
# now using unique directories so repo are never shared with another pipeline.
|
||||
# This is not memory efficient (we keep a lot of data), hence this reminder.
|
||||
# Setup
|
||||
# setup clones the mfem/data repo in ${BUILD_ROOT}. The build_and_test script
|
||||
# then symlinks the repo to the parent directory of the MFEM source directory.
|
||||
# Unit tests that depend on the mfem/data repo will then detect that this
|
||||
# directory is present and be enabled.
|
||||
setup:
|
||||
tags:
|
||||
- shell
|
||||
- quartz
|
||||
stage: setup
|
||||
variables:
|
||||
GIT_STRATEGY: none
|
||||
script:
|
||||
- mkdir -p ${BUILD_ROOT} && cd ${BUILD_ROOT}
|
||||
- if [ ! -d data ]; then git clone ${MFEM_DATA_REPO}; fi
|
||||
needs: []
|
||||
|
||||
# The setup_baseline job in setup stage_baseline doesn't rely on MFEM git repo.
|
||||
# It prepares a pipeline-wide working directory downloading/updating external
|
||||
# repos. TODO: updating tests and tpls is not necessary anymore since pipelines
|
||||
# are now using unique directories so repo are never shared with another
|
||||
# pipeline. This is not memory efficient (we keep a lot of data), hence this
|
||||
# reminder.
|
||||
setup_baseline:
|
||||
tags:
|
||||
- shell
|
||||
- quartz
|
||||
stage: setup_baseline
|
||||
variables:
|
||||
GIT_STRATEGY: none
|
||||
script:
|
||||
|
||||
+1
-1
@@ -25,7 +25,7 @@
|
||||
.build_and_test_on_lassen:
|
||||
extends: [.build_blueos_3_ppc64le_ib_script, .on_lassen]
|
||||
stage: l_build_and_test
|
||||
needs: []
|
||||
needs: [setup]
|
||||
|
||||
opt_mpi_cuda_xl_16_1_1_8:
|
||||
variables:
|
||||
|
||||
+2
-1
@@ -94,6 +94,7 @@ q_report_failure:
|
||||
.build_and_test_on_quartz:
|
||||
extends: [.build_toss_3_x86_64_ib_script, .on_quartz]
|
||||
stage: q_build_and_test
|
||||
needs: [setup]
|
||||
|
||||
# Build MFEM
|
||||
debug_ser_gcc_4_9_3:
|
||||
@@ -139,7 +140,7 @@ opt_par_gcc_6_1_0_pumi:
|
||||
# Baseline
|
||||
baselinecheck_mfem_intel_quartz:
|
||||
extends: [.baselinecheck_mfem, .on_quartz]
|
||||
needs: [setup]
|
||||
needs: [setup_baseline]
|
||||
|
||||
update_autotest:
|
||||
extends: [.on_quartz]
|
||||
|
||||
-469
@@ -1,469 +0,0 @@
|
||||
# Copyright (c) 2010-2021, 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.
|
||||
|
||||
language: cpp
|
||||
|
||||
os: linux
|
||||
dist: bionic
|
||||
|
||||
stages:
|
||||
- checks
|
||||
- tests
|
||||
- optional
|
||||
|
||||
env:
|
||||
global:
|
||||
- HYPRE_ARCHIVE=v2.19.0.tar.gz
|
||||
HYPRE_URL=https://github.com/hypre-space/hypre/archive/$HYPRE_ARCHIVE
|
||||
HYPRE_TOP_DIR=hypre-2.19.0
|
||||
|
||||
jobs:
|
||||
include:
|
||||
|
||||
# ========================
|
||||
# Checks
|
||||
# ========================
|
||||
# - code-style
|
||||
# - documentation
|
||||
# - gitignore
|
||||
|
||||
- stage: checks
|
||||
os: linux
|
||||
dist: xenial
|
||||
name: "code-style"
|
||||
addons:
|
||||
apt:
|
||||
packages:
|
||||
- astyle=2.05.1-0ubuntu1
|
||||
script:
|
||||
- cd ${TRAVIS_BUILD_DIR}
|
||||
- cd tests/scripts
|
||||
- ./runtest code-style
|
||||
|
||||
- stage: checks
|
||||
os: linux
|
||||
name: "documentation"
|
||||
addons:
|
||||
apt:
|
||||
packages:
|
||||
- doxygen
|
||||
- graphviz
|
||||
script:
|
||||
- cd ${TRAVIS_BUILD_DIR}
|
||||
- cd tests/scripts
|
||||
- ./runtest documentation
|
||||
|
||||
- stage: checks
|
||||
os: linux
|
||||
name: "gitignore"
|
||||
addons:
|
||||
apt:
|
||||
packages:
|
||||
- mpich
|
||||
- libmpich-dev
|
||||
env: MPI=YES
|
||||
before_script:
|
||||
- cd ${TRAVIS_BUILD_DIR}
|
||||
- mpicxx -v
|
||||
- make config MFEM_USE_MPI=YES MFEM_MPI_NP=2
|
||||
- make all -j3
|
||||
- make test-noclean
|
||||
script:
|
||||
- cd tests/scripts
|
||||
- ./runtest gitignore
|
||||
cache:
|
||||
ccache: true
|
||||
directories:
|
||||
- $TRAVIS_BUILD_DIR/../$HYPRE_TOP_DIR/src/hypre
|
||||
- $TRAVIS_BUILD_DIR/../metis-4.0
|
||||
before_cache:
|
||||
- cd $TRAVIS_BUILD_DIR/../metis-4.0;
|
||||
mv libmetis.a Lib ..; rm -rf * ; mv ../libmetis.a ../Lib .;
|
||||
rm -f Lib/*.{c,o}
|
||||
|
||||
# ========================
|
||||
# Optional Checks/Tests
|
||||
# ========================
|
||||
# - branch-history
|
||||
|
||||
- stage: optional
|
||||
name: "branch-history"
|
||||
if: branch != next
|
||||
# need full git history for the binary/big files check
|
||||
git:
|
||||
depth: false
|
||||
script:
|
||||
- cd ${TRAVIS_BUILD_DIR}
|
||||
# update master
|
||||
- git fetch origin master:master
|
||||
# checkout a branch (otherwise Travis works in detached head)
|
||||
- git checkout -b travis_tests
|
||||
- cd tests/scripts
|
||||
- ./runtest branch-history
|
||||
|
||||
# ========================
|
||||
# Linux tests
|
||||
# ========================
|
||||
# - serial + debug
|
||||
# - serial
|
||||
# - parallel + debug
|
||||
# - parallel
|
||||
|
||||
- stage: tests
|
||||
os: linux
|
||||
compiler: gcc
|
||||
name: "Linux: Serial + Debug"
|
||||
env: DEBUG=YES
|
||||
MPI=NO
|
||||
CODECOV=NO
|
||||
MFEM_TEST_TARGET=check
|
||||
cache:
|
||||
ccache: true
|
||||
|
||||
- os: linux
|
||||
compiler: gcc
|
||||
name: "Linux: Serial"
|
||||
env: DEBUG=NO
|
||||
MPI=NO
|
||||
CODECOV=NO
|
||||
MFEM_TEST_TARGET=test
|
||||
cache:
|
||||
ccache: true
|
||||
|
||||
- os: linux
|
||||
compiler: gcc
|
||||
name: "Linux: Parallel + Debug"
|
||||
addons:
|
||||
apt:
|
||||
# sources:
|
||||
# - ubuntu-toolchain-r-test
|
||||
packages:
|
||||
# GCC 4.9
|
||||
# - g++-4.9
|
||||
# MPICH
|
||||
- mpich
|
||||
- libmpich-dev
|
||||
# OpenMPI
|
||||
# - openmpi-bin
|
||||
# - libopenmpi-dev
|
||||
env: DEBUG=YES
|
||||
MPI=YES
|
||||
CODECOV=NO
|
||||
MFEM_TEST_TARGET=check
|
||||
NPROCS=2
|
||||
cache:
|
||||
ccache: true
|
||||
directories:
|
||||
- $TRAVIS_BUILD_DIR/../$HYPRE_TOP_DIR/src/hypre
|
||||
- $TRAVIS_BUILD_DIR/../metis-4.0
|
||||
before_cache:
|
||||
- cd $TRAVIS_BUILD_DIR/../metis-4.0;
|
||||
mv libmetis.a Lib ..; rm -rf * ; mv ../libmetis.a ../Lib .;
|
||||
rm -f Lib/*.{c,o}
|
||||
|
||||
- os: linux
|
||||
compiler: gcc
|
||||
name: "Linux: Parallel"
|
||||
addons:
|
||||
apt:
|
||||
# sources:
|
||||
# - ubuntu-toolchain-r-test
|
||||
packages:
|
||||
# GCC 4.9
|
||||
# - g++-4.9
|
||||
# MPICH
|
||||
- mpich
|
||||
- libmpich-dev
|
||||
# OpenMPI
|
||||
# - openmpi-bin
|
||||
# - libopenmpi-dev
|
||||
env: DEBUG=NO
|
||||
MPI=YES
|
||||
CODECOV=YES
|
||||
MFEM_TEST_TARGET=test
|
||||
NPROCS=2
|
||||
cache:
|
||||
ccache: true
|
||||
directories:
|
||||
- $TRAVIS_BUILD_DIR/../$HYPRE_TOP_DIR/src/hypre
|
||||
- $TRAVIS_BUILD_DIR/../metis-4.0
|
||||
before_cache:
|
||||
- cd $TRAVIS_BUILD_DIR/../metis-4.0;
|
||||
mv libmetis.a Lib ..; rm -rf * ; mv ../libmetis.a ../Lib .;
|
||||
rm -f Lib/*.{c,o}
|
||||
|
||||
- os: linux
|
||||
compiler: gcc
|
||||
name: "Linux: Parallel (cmake)"
|
||||
addons:
|
||||
apt:
|
||||
packages:
|
||||
- mpich
|
||||
- libmpich-dev
|
||||
env: MPI=YES
|
||||
NPROCS=2
|
||||
script:
|
||||
- cd ${TRAVIS_BUILD_DIR}
|
||||
- mkdir ${TRAVIS_BUILD_DIR}/build
|
||||
- cd ${TRAVIS_BUILD_DIR}/build
|
||||
- cmake ..
|
||||
-DMFEM_USE_MPI=ON
|
||||
-DHYPRE_DIR=${TRAVIS_BUILD_DIR}/../$HYPRE_TOP_DIR/src/hypre
|
||||
-DMFEM_MPI_NP=$NPROCS
|
||||
- make -j3 mfem examples
|
||||
- cd ${TRAVIS_BUILD_DIR}/build/tests/unit
|
||||
- make -j3
|
||||
- ctest --output-on-failure
|
||||
cache:
|
||||
ccache: true
|
||||
directories:
|
||||
- $TRAVIS_BUILD_DIR/../$HYPRE_TOP_DIR/src/hypre
|
||||
- $TRAVIS_BUILD_DIR/../metis-4.0
|
||||
before_cache:
|
||||
- cd $TRAVIS_BUILD_DIR/../metis-4.0;
|
||||
mv libmetis.a Lib ..; rm -rf * ; mv ../libmetis.a ../Lib .;
|
||||
rm -f Lib/*.{c,o}
|
||||
|
||||
# ========================
|
||||
# Mac OS X tests
|
||||
# ========================
|
||||
# - serial + debug
|
||||
# - serial
|
||||
# - parallel + debug
|
||||
# - parallel
|
||||
|
||||
- os: osx
|
||||
osx_image: xcode11.2
|
||||
compiler: clang
|
||||
name: "Mac: Serial + Debug"
|
||||
addons:
|
||||
homebrew:
|
||||
packages:
|
||||
- ccache
|
||||
env: DEBUG=YES
|
||||
MPI=NO
|
||||
CODECOV=NO
|
||||
MFEM_TEST_TARGET=check
|
||||
cache:
|
||||
ccache: true
|
||||
|
||||
- os: osx
|
||||
osx_image: xcode11.2
|
||||
compiler: clang
|
||||
name: "Mac: Serial"
|
||||
addons:
|
||||
homebrew:
|
||||
packages:
|
||||
- ccache
|
||||
env: DEBUG=NO
|
||||
MPI=NO
|
||||
CODECOV=NO
|
||||
MFEM_TEST_TARGET=test
|
||||
cache:
|
||||
ccache: true
|
||||
|
||||
- os: osx
|
||||
osx_image: xcode11.2
|
||||
compiler: clang
|
||||
name: "Mac: Parallel + Debug"
|
||||
addons:
|
||||
homebrew:
|
||||
packages:
|
||||
- ccache
|
||||
env: DEBUG=YES
|
||||
MPI=YES
|
||||
CODECOV=NO
|
||||
MFEM_TEST_TARGET=check
|
||||
NPROCS=4
|
||||
TMPDIR=/tmp
|
||||
cache:
|
||||
ccache: true
|
||||
directories:
|
||||
- $TRAVIS_BUILD_DIR/../$HYPRE_TOP_DIR/src/hypre
|
||||
- $TRAVIS_BUILD_DIR/../metis-4.0
|
||||
- $HOME/local-cached
|
||||
before_cache:
|
||||
- cd $TRAVIS_BUILD_DIR/../metis-4.0;
|
||||
mv libmetis.a Lib ..; rm -rf * ; mv ../libmetis.a ../Lib .;
|
||||
rm -f Lib/*.{c,o}
|
||||
|
||||
- os: osx
|
||||
osx_image: xcode11.2
|
||||
compiler: clang
|
||||
name: "Mac: Parallel"
|
||||
addons:
|
||||
homebrew:
|
||||
packages:
|
||||
- ccache
|
||||
env: DEBUG=NO
|
||||
MPI=YES
|
||||
CODECOV=YES
|
||||
MFEM_TEST_TARGET=test
|
||||
NPROCS=4
|
||||
TMPDIR=/tmp
|
||||
cache:
|
||||
ccache: true
|
||||
directories:
|
||||
- $TRAVIS_BUILD_DIR/../$HYPRE_TOP_DIR/src/hypre
|
||||
- $TRAVIS_BUILD_DIR/../metis-4.0
|
||||
- $HOME/local-cached
|
||||
before_cache:
|
||||
- cd $TRAVIS_BUILD_DIR/../metis-4.0;
|
||||
mv libmetis.a Lib ..; rm -rf * ; mv ../libmetis.a ../Lib .;
|
||||
rm -f Lib/*.{c,o}
|
||||
|
||||
before_install:
|
||||
# No addon for brew yet, have to install OSX packages this way.
|
||||
# - if [ $TRAVIS_OS_NAME == "osx" ] && [ $MPI == "YES" ]; then
|
||||
# brew install open-mpi;
|
||||
# fi
|
||||
|
||||
# Disable ccache while building dependencies that are cached:
|
||||
- echo "before \$PATH = $PATH";
|
||||
export PATH=${PATH//\/usr\/lib\/ccache:/};
|
||||
echo "after \$PATH = $PATH"
|
||||
|
||||
# On Mac OS X, build and cache OpenMPI 2.1.6:
|
||||
- if [ $TRAVIS_OS_NAME == "osx" ] && [ $MPI == "YES" ]; then
|
||||
if [ ! -e $HOME/local-cached/bin/mpicc ]; then
|
||||
mkdir -p $HOME/builds && cd $HOME/builds &&
|
||||
wget https://download.open-mpi.org/release/open-mpi/v2.1/openmpi-2.1.6.tar.bz2 &&
|
||||
tar jxf openmpi-2.1.6.tar.bz2 &&
|
||||
mkdir openmpi-build && cd openmpi-build &&
|
||||
../openmpi-2.1.6/configure --prefix=$HOME/local-cached &&
|
||||
make -j3 all && make install;
|
||||
fi;
|
||||
PATH=$HOME/local-cached/bin:$PATH;
|
||||
cd $TRAVIS_BUILD_DIR;
|
||||
fi
|
||||
|
||||
# Update environment to find g++ 4.9 installation first.
|
||||
# - if [ $TRAVIS_OS_NAME == "linux" ]; then
|
||||
# mkdir -p latest-gcc-symlinks;
|
||||
# ln -s /usr/bin/g++-4.9 latest-gcc-symlinks/g++;
|
||||
# ln -s /usr/bin/gcc-4.9 latest-gcc-symlinks/gcc;
|
||||
# ln -s /usr/bin/gcov-4.9 latest-gcc-symlinks/gcov;
|
||||
# export PATH=$PWD/latest-gcc-symlinks:$PATH;
|
||||
# fi
|
||||
|
||||
# Install tool to upload code coverage reports to coveralls.io
|
||||
- if [ "$CODECOV" == "YES" ]; then
|
||||
export PYTHONUSERBASE=$HOME/local;
|
||||
pip install --user cpp-coveralls;
|
||||
pip install --user pyyaml;
|
||||
PATH=$HOME/local/bin:$PATH;
|
||||
fi
|
||||
|
||||
install:
|
||||
# Set MPI compilers, print compiler version
|
||||
- if [ $MPI == "YES" ]; then
|
||||
if [ "$TRAVIS_OS_NAME" == "linux" ]; then
|
||||
export MPICH_CC="$CC";
|
||||
export MPICH_CXX="$CXX";
|
||||
else
|
||||
export OMPI_CC="$CC";
|
||||
export OMPI_CXX="$CXX";
|
||||
mpic++ --showme:version;
|
||||
fi;
|
||||
mpic++ -v;
|
||||
else
|
||||
$CXX -v;
|
||||
fi
|
||||
|
||||
# Back out of the mfem directory to install the libraries
|
||||
- cd ..
|
||||
|
||||
# hypre
|
||||
- if [ $MPI == "YES" ]; then
|
||||
if [ ! -e $HYPRE_TOP_DIR/src/hypre/lib/libHYPRE.a ]; then
|
||||
wget $HYPRE_URL;
|
||||
rm -rf $HYPRE_TOP_DIR;
|
||||
tar xvzf $HYPRE_ARCHIVE;
|
||||
cd $HYPRE_TOP_DIR/src;
|
||||
./configure --disable-fortran CC=mpicc CXX=mpic++;
|
||||
make -j3;
|
||||
cd ../..;
|
||||
else
|
||||
echo "Reusing cached $HYPRE_TOP_DIR/";
|
||||
fi;
|
||||
ln -s $HYPRE_TOP_DIR hypre;
|
||||
else
|
||||
echo "Serial build, not using hypre";
|
||||
fi
|
||||
|
||||
# METIS, use a mirror because the original source server is not always up.
|
||||
# Original url:
|
||||
# http://glaros.dtc.umn.edu/gkhome/fetch/sw/metis/OLD/metis-4.0.3.tar.gz
|
||||
- if [ $MPI == "YES" ]; then
|
||||
if [ ! -e metis-4.0/libmetis.a ]; then
|
||||
wget https://mfem.github.io/tpls/metis-4.0.3.tar.gz;
|
||||
tar xvzf metis-4.0.3.tar.gz;
|
||||
make -j3 -C metis-4.0.3/Lib CC="$CC" OPTFLAGS="-O2";
|
||||
rm -rf metis-4.0;
|
||||
mv metis-4.0.3 metis-4.0;
|
||||
else
|
||||
echo "Reusing cached metis-4.0/";
|
||||
fi;
|
||||
fi
|
||||
|
||||
# Re-enable ccache on linux; enable ccache on mac os:
|
||||
- if [ $TRAVIS_OS_NAME == "linux" ]; then
|
||||
export PATH="/usr/lib/ccache:$PATH";
|
||||
else
|
||||
if [ $TRAVIS_OS_NAME == "osx" ]; then
|
||||
export PATH="/usr/local/opt/ccache/libexec:$PATH";
|
||||
fi;
|
||||
fi
|
||||
|
||||
- printf "which \$CC = "; which $CC;
|
||||
printf "which \$CXX = "; which $CXX
|
||||
|
||||
script:
|
||||
# Compiler
|
||||
- if [ $MPI == "YES" ]; then
|
||||
export MYCXX=mpic++;
|
||||
export MAKE_CXX_FLAG=MPICXX=$MYCXX;
|
||||
else
|
||||
export MYCXX="$CXX";
|
||||
export MAKE_CXX_FLAG=CXX=$MYCXX;
|
||||
fi
|
||||
|
||||
# Print the compiler version
|
||||
- $MYCXX -v
|
||||
|
||||
# Set some variables
|
||||
- cd $TRAVIS_BUILD_DIR;
|
||||
CPPFLAGS="";
|
||||
SKIP_TEST_DIRS="";
|
||||
if [ "$CODECOV" == "YES" ]; then
|
||||
CPPFLAGS="--coverage -g";
|
||||
fi;
|
||||
if [ "$TRAVIS_OS_NAME" != "linux" ] || [ "$DEBUG" == "YES" ]; then
|
||||
CPPFLAGS+=" -pedantic -Wall -Werror";
|
||||
fi
|
||||
|
||||
# Configure the library
|
||||
- make config MFEM_USE_MPI=$MPI MFEM_DEBUG=$DEBUG $MAKE_CXX_FLAG
|
||||
MFEM_MPI_NP=$NPROCS CPPFLAGS="$CPPFLAGS"
|
||||
# Show the configuration
|
||||
- make info
|
||||
# Build the library
|
||||
- make -j3
|
||||
# Build the examples and the miniapps
|
||||
- make -j3 all
|
||||
# Run tests
|
||||
- make $MFEM_TEST_TARGET SKIP_TEST_DIRS="$SKIP_TEST_DIRS"
|
||||
|
||||
after_success:
|
||||
- if [ "$CODECOV" == "YES" ]; then
|
||||
coveralls --include fem --include general --include linalg --include
|
||||
mesh --exclude /usr --gcov-options '\-lp' --root $TRAVIS_BUILD_DIR;
|
||||
fi
|
||||
@@ -173,7 +173,11 @@ Version 4.2.1 (development)
|
||||
mixed meshes. The LOR Transfer miniapp (miniapps/tools/lor-transfer.cpp) now
|
||||
supports meshes with any element geometry.
|
||||
|
||||
- Gitlab CI: use Spack (and Uberenv) to automate the build of TPLs.
|
||||
- Testing improvements:
|
||||
* Transitioned from Travis to GitHub Action for testing/CI on GitHub.
|
||||
* Effectively remove Travis from CI.
|
||||
* Use Spack (and Uberenv) to automate TPL building in LLNL GitLab tests.
|
||||
* Added a set of suggested git hooks for developers in config/githooks.
|
||||
|
||||
- Added new miniapps demonstrating: 1) the use of GSLIB for overlapping grids,
|
||||
see gslib/schwarz_ex1, and 2) coupling different physics in different domains,
|
||||
@@ -231,6 +235,15 @@ Version 4.2.1 (development)
|
||||
|
||||
- Added makefile rule to generate TAGS table for vi or Emacs users.
|
||||
|
||||
API changes
|
||||
-----------
|
||||
- Added an abstract interface `mfem::FaceRestriction` for `H1FaceRestriction`
|
||||
and `L2FaceRestriction`.
|
||||
In order to conform with the semantic of `MultTranspose` in `mfem::Operator`,
|
||||
`mfem::FaceRestriction::MultTranspose` now sets instead of adding values, and
|
||||
`mfem::FaceRestriction::AddMultTranspose` should replace previous calls to
|
||||
`mfem::FaceRestriction::MultTranspose`.
|
||||
|
||||
libCEED integration improvements
|
||||
--------------------------------
|
||||
- Refactor the libCEED integration
|
||||
|
||||
+34
-17
@@ -4,7 +4,9 @@
|
||||
|
||||
<p align="center">
|
||||
<a href="https://github.com/mfem/mfem/blob/master/LICENSE"><img alt="License" src="https://img.shields.io/badge/License-BSD-brightgreen.svg"></a>
|
||||
<a href="https://travis-ci.org/mfem/mfem"><img alt="Build Status" src="https://travis-ci.org/mfem/mfem.svg?branch=master"></a>
|
||||
<a href="https://github.com/mfem/mfem/actions?query=workflow%3Arepo-check+branch%3Amaster"><img alt="Repo check" src="https://github.com/mfem/mfem/actions/workflows/repo-check.yml/badge.svg?branch=master"></a>
|
||||
<a href="https://github.com/mfem/mfem/actions?query=workflow%3Abuild-analysis+branch%3Amaster"><img alt="Build Analysis" src="https://github.com/mfem/mfem/actions/workflows/mfem-analysis.yml/badge.svg?branch=master"></a>
|
||||
<a href="https://github.com/mfem/mfem/actions?query=workflow%3Abuilds-and-tests+branch%3Amaster"><img alt="Builds and Tests" src="https://github.com/mfem/mfem/actions/workflows/builds-and-tests.yml/badge.svg?branch=master"></a>
|
||||
<a href="https://ci.appveyor.com/project/mfem/mfem"><img alt="Build Status" src="https://ci.appveyor.com/api/projects/status/19non9sqm6msi2wy?svg=true"></a>
|
||||
<a href="https://mfem.github.io/doxygen/html/index.html"><img alt="Doxygen" src="https://img.shields.io/badge/code-documented-brightgreen.svg"></a>
|
||||
</p>
|
||||
@@ -63,6 +65,8 @@ Origin](#developers-certificate-of-origin-11) at the end of this file.*
|
||||
development branches off `mfem:master`.
|
||||
- Please follow the [developer guidelines](#developer-guidelines), in particular
|
||||
with regards to documentation and code styling.
|
||||
- Please do not commit large/binary files to the central repository (use a fork
|
||||
instead).
|
||||
- Pull requests should be issued toward `mfem:master`. Make sure
|
||||
to check the items off the [Pull Request Checklist](#pull-request-checklist).
|
||||
- When your contribution is fully working and ready to be reviewed, add
|
||||
@@ -71,6 +75,7 @@ Origin](#developers-certificate-of-origin-11) at the end of this file.*
|
||||
reviewers to evaluate the changes.
|
||||
- The reviewers have 3 weeks to evaluate the PR and work with the author to
|
||||
fix issues and implement improvements.
|
||||
- During review there should be no force pushes/rewriting history in the branch.
|
||||
- After approval, MFEM developers merge the PR manually in the [mfem:next branch](#masternext-workflow).
|
||||
- After a week of testing in `mfem:next`, the original PR is merged in `mfem:master`.
|
||||
- We use [milestones](https://github.com/mfem/mfem/milestones) to coordinate the
|
||||
@@ -91,8 +96,9 @@ The MFEM source code has the following structure:
|
||||
```
|
||||
.
|
||||
├── config
|
||||
│ └── cmake
|
||||
│ └── ...
|
||||
│ ├── cmake
|
||||
│ │ └── ...
|
||||
│ └── githooks
|
||||
├── data
|
||||
├── doc
|
||||
├── examples
|
||||
@@ -363,6 +369,10 @@ Before you can start, you need a GitHub account, here are a few suggestions:
|
||||
two reviewers to evaluate the changes. The reviewers have 3 weeks to evaluate
|
||||
the PR and work with the author to implement improvements and fix issues.
|
||||
|
||||
- Once the `ready-for-review` label has been applied and reviewers have been
|
||||
assigned, the PR is considered under review. To help with the review process
|
||||
there should be no force pushes/rewriting history in the branch.
|
||||
|
||||
- After approval, the PR is [tested](#masternext-workflow) for a week with
|
||||
other approved PRs in the `mfem:next` branch.
|
||||
|
||||
@@ -370,16 +380,20 @@ Before you can start, you need a GitHub account, here are a few suggestions:
|
||||
`mfem:next`, see the [README](tests/scripts/README) file in that directory
|
||||
for more details.
|
||||
|
||||
- Track the Travis CI, Github Actions and Appveyor [continuous integration](#automated-testing)
|
||||
- Track the GitHub Actions and Appveyor [continuous integration](#automated-testing)
|
||||
builds at the end of the PR. These should generally run clean, so address any
|
||||
errors as soon as possible. Please ask if you are unsure how to do that.
|
||||
|
||||
- Note that some tests, such as the `branch-history` check in Travis and Github
|
||||
Actions are safeguards that are allowed to fail in certain cases.
|
||||
- Note that some tests, such as the `branch-history` check in GitHub Actions
|
||||
are safeguards that are allowed to fail in certain cases.
|
||||
|
||||
- Other tests, such as the `code-style`, `documentation` and `gitignore`
|
||||
checks in Travis and Github Actions enforce MFEM-specific rules which are
|
||||
explained in the error messages and the `tests/scripts` directory.
|
||||
checks in GitHub Actions enforce MFEM-specific rules which are explained in
|
||||
the error messages and the `tests/scripts` directory.
|
||||
|
||||
- Also note that the tests `branch-history` and `repos-checks` found in GitHub
|
||||
Actions can be triggered automatically before each push using git hooks. See
|
||||
the [git hooks README](config/githooks/README.md) for a detailed explanation.
|
||||
|
||||
- If triggered, track the status of the LLNL GitLab tests. If failing, ask
|
||||
one of the _LLNL developers_ for details.
|
||||
@@ -399,7 +413,7 @@ Before a PR can be merged, it should satisfy the following:
|
||||
- [ ] Does `make` or `cmake` have a new target?
|
||||
- [ ] Did the requirements or the installation process change? *(rare)*
|
||||
- [ ] Update continuous integration server configurations if necessary (e.g. with new version requirements for each of MFEM's dependencies)
|
||||
- [ ] `.travis.yml`
|
||||
- [ ] `.github`
|
||||
- [ ] `.appveyor.yml`
|
||||
- [ ] Update `.gitignore`:
|
||||
- [ ] Check if `make distclean; git status` shows any files that were generated from the source by the project (not an IDE) but we don't want to track in the repository.
|
||||
@@ -516,7 +530,7 @@ MFEM uses a `master`/`next`-branch workflow as described below:
|
||||
- [ ] `doc/CodeDocumentation.conf.in`
|
||||
- [ ] Check that version requirements for each of MFEM's dependencies are documented in `INSTALL` and up-to-date
|
||||
- [ ] Check that continuous integration server configurations reflect the dependency version requirements of the new release
|
||||
- [ ] `.travis.yml`
|
||||
- [ ] `.github`
|
||||
- [ ] `.appveyor.yml`
|
||||
- [ ] Update the `CHANGELOG` to organize all release contributions
|
||||
- [ ] Review the whole source code once over
|
||||
@@ -578,14 +592,17 @@ MFEM uses a `master`/`next`-branch workflow as described below:
|
||||
MFEM has several levels of automated testing running on GitHub, as well as on
|
||||
local Mac and Linux workstations, and Livermore Computing clusters at LLNL.
|
||||
|
||||
### Linux and Mac smoke tests
|
||||
We use Travis CI and Github Actions to drive the default tests on the `master`
|
||||
and `next` branches. See the `.travis` file and the logs at
|
||||
[https://travis-ci.org/mfem/mfem](https://travis-ci.org/mfem/mfem).
|
||||
In addition, developers can set local git hooks to run some quick checks on
|
||||
commit or push, see the [README](config/githooks/README.md) in the `config/githooks`
|
||||
directory.
|
||||
|
||||
Testing using Travis CI and Github Actions should be kept lightweight, as there
|
||||
is a time constraint on jobs. Two virtual machines are configured - Mac (OS X)
|
||||
and Linux.
|
||||
### Linux and Mac smoke tests
|
||||
We use GitHub Actions to drive the default tests on the `master` and `next`
|
||||
branches. See the `.github/workflows` files and the logs at
|
||||
[https://github.com/mfem/mfem/actions](https://github.com/mfem/mfem/actions).
|
||||
|
||||
Testing using GitHub Actions should be kept lightweight, as there is a time
|
||||
constraint on jobs. Two virtual machines are configured - Mac (OS X) and Linux.
|
||||
|
||||
- Tests on the `master` branch are triggered whenever a PR is issued on this branch.
|
||||
- Tests on the `next` branch are currently scheduled to run each night.
|
||||
|
||||
@@ -0,0 +1,41 @@
|
||||
Finite Element Discretization Library
|
||||
__
|
||||
_ __ ___ / _| ___ _ __ ___
|
||||
| '_ ` _ \ | |_ / _ \| '_ ` _ \
|
||||
| | | | | || _|| __/| | | | | |
|
||||
|_| |_| |_||_| \___||_| |_| |_|
|
||||
|
||||
https://mfem.org
|
||||
|
||||
|
||||
This directory contains recommended git hooks, which are scripts that can be
|
||||
used to improve your development experience with MFEM:
|
||||
|
||||
### The hooks
|
||||
|
||||
* `pre-commit` is a hook that will be applied before each commit and run
|
||||
`astyle` on the code. This will ensure that your changes comply with the MFEM
|
||||
code styling guidelines.
|
||||
|
||||
* `pre-push` is a hook that will be applied before each push to run a quick set
|
||||
of tests that verify that your files headers are in compliance, and that you did
|
||||
not add any large files to the repo.
|
||||
|
||||
### Setup
|
||||
|
||||
To setup the git hooks, run `make hooks`, which creates symlinks to the hooks in
|
||||
the `.git/hooks` directory. Individual hooks can be enabled by manually creating
|
||||
symlinks.
|
||||
|
||||
(You may also copy the scripts directly and customize them further, but this way
|
||||
you may miss additional updates in the future.)
|
||||
|
||||
### Failures
|
||||
|
||||
The `branch-history` check can fail in some cases when the history is OK. For
|
||||
example, when a large number of files were modified for a legitimate reason, or
|
||||
when a picture was added for documentation.
|
||||
|
||||
If that is the case, make sure the failure is indeed justified, and rerun the
|
||||
push command with the `--no-verify` option. This will skip the hooks, allowing
|
||||
you to push those changes.
|
||||
Executable
+4
@@ -0,0 +1,4 @@
|
||||
#!/bin/sh
|
||||
|
||||
# Apply automated code formatting
|
||||
make -C $(git rev-parse --show-toplevel) style
|
||||
Executable
+107
@@ -0,0 +1,107 @@
|
||||
#!/bin/bash
|
||||
|
||||
# Copyright (c) 2010-2021, Lawrence Livermore National Security, LLC. Produced
|
||||
# at the Lawrence Livermore National Laboratory. All Rights reserved. See files
|
||||
# LICENSE and NOTICE for details. LLNL-CODE-806117.
|
||||
#
|
||||
# This file is part of the MFEM library. For more information and source code
|
||||
# availability visit https://mfem.org.
|
||||
#
|
||||
# MFEM is free software; you can redistribute it and/or modify it under the
|
||||
# terms of the BSD-3 license. We welcome feedback and contributions, see file
|
||||
# CONTRIBUTING.md for details.
|
||||
|
||||
option=${1:-""}
|
||||
|
||||
if [[ "${option}" == "--help" ]]; then
|
||||
echo "This script runs checks on the repository."
|
||||
echo "It has 2 modes: with and without an option."
|
||||
echo ""
|
||||
echo "Options are used in GitHub Actions and can be:"
|
||||
echo " --copyright"
|
||||
echo " --license"
|
||||
echo " --release"
|
||||
echo " --style"
|
||||
echo " --history"
|
||||
echo ""
|
||||
echo "As a githook, the script is used without options."
|
||||
echo "In that case, it will run all the checks except style."
|
||||
echo ""
|
||||
echo "Use --help to print this help message."
|
||||
fi
|
||||
|
||||
cd $(git rev-parse --show-toplevel)
|
||||
|
||||
# copyright check
|
||||
copyright=true
|
||||
if [[ "${option}" == "--copyright" || "${option}" == "" ]]; then
|
||||
if git grep -l "^\(#\|//\).*\(\-2020\|\ 2010,\)" > matches.txt; then
|
||||
echo "Please update the following files to Copyright (c) 2010-2021:"
|
||||
cat matches.txt
|
||||
copyright=false
|
||||
fi
|
||||
fi
|
||||
|
||||
# license check
|
||||
license=true
|
||||
if [[ "${option}" == "--license" || "${option}" == "" ]]; then
|
||||
if git grep -li "^\(#\|//\).*GNU\ Lesser\ General\ Public\ License" > matches.txt; then
|
||||
echo "Please update the following files to the BSD-3 license:"
|
||||
cat matches.txt
|
||||
license=false
|
||||
fi
|
||||
fi
|
||||
|
||||
# release check
|
||||
release=true
|
||||
if [[ "${option}" == "--release" || "${option}" == "" ]]; then
|
||||
if git grep -l "^\(#\|//\).*LLNL\-CODE\-443211" > matches.txt
|
||||
then
|
||||
echo "Please update the following files to LLNL-CODE-806117:"
|
||||
cat matches.txt
|
||||
release=false
|
||||
fi
|
||||
fi
|
||||
|
||||
# wrap-up
|
||||
code=0
|
||||
if ! $copyright ; then
|
||||
echo "copyright check failed, unroll log for details"
|
||||
code=1
|
||||
fi
|
||||
if ! $license ; then
|
||||
echo "license check failed, unroll log for details"
|
||||
code=1
|
||||
fi
|
||||
if ! $release ; then
|
||||
echo "release check failed, unroll log for details"
|
||||
code=1
|
||||
fi
|
||||
|
||||
# `code-style` is not just a check, it will actually reformat the code if
|
||||
# necessary. This means that if one pushes while the repo is in dirty state
|
||||
# (changes not staged), those changes may be mixed with format changes.
|
||||
# To activate this, you will need to hard-copy this hook script in the hook
|
||||
# directory and uncomment only then. (See README.md)
|
||||
#
|
||||
## style check
|
||||
#if [[ "${option}" == "--style" || "${option}" == "" ]]; then
|
||||
if [[ "${option}" == "--style" ]]; then
|
||||
if which astyle && [[ "$(astyle --version)" == "Artistic Style Version 2.05.1" ]]; then
|
||||
cd tests/scripts
|
||||
if ! ./runtest code-style; then code=1; fi
|
||||
cd -
|
||||
else
|
||||
echo "Warning: astyle not found or version is not 2.05.1"
|
||||
fi
|
||||
fi
|
||||
|
||||
# branch-history
|
||||
if [[ "${option}" == "--history" || "${option}" == "" ]]; then
|
||||
git fetch origin master:master
|
||||
cd tests/scripts
|
||||
if ! ./runtest branch-history; then code=1; fi
|
||||
cd -
|
||||
fi
|
||||
|
||||
exit $code
|
||||
+1
-1
@@ -105,7 +105,7 @@ private:
|
||||
Vector diag(fespace.GetTrueVSize());
|
||||
bfs.Last()->AssembleDiagonal(diag);
|
||||
|
||||
Solver* smoother = new OperatorChebyshevSmoother(opr.Ptr(), diag,
|
||||
Solver* smoother = new OperatorChebyshevSmoother(*opr, diag,
|
||||
*essentialTrueDofs.Last(), 2);
|
||||
AddLevel(opr.Ptr(), smoother, true, true);
|
||||
}
|
||||
|
||||
+1
-1
@@ -115,7 +115,7 @@ private:
|
||||
Vector diag(fespace.GetTrueVSize());
|
||||
bfs.Last()->AssembleDiagonal(diag);
|
||||
|
||||
Solver* smoother = new OperatorChebyshevSmoother(opr.Ptr(), diag,
|
||||
Solver* smoother = new OperatorChebyshevSmoother(*opr, diag,
|
||||
*essentialTrueDofs.Last(), 2, fespace.GetParMesh()->GetComm());
|
||||
|
||||
AddLevel(opr.Ptr(), smoother, true, true);
|
||||
|
||||
@@ -1,960 +0,0 @@
|
||||
#include "DofMapsDST.hpp"
|
||||
|
||||
double testcoeff(const Vector & x)
|
||||
{
|
||||
return sin(3*M_PI*(x.Sum()));
|
||||
}
|
||||
|
||||
int get_rank(int tdof, std::vector<int> & tdof_offsets)
|
||||
{
|
||||
int size = tdof_offsets.size();
|
||||
if (size == 1) { return 0; }
|
||||
std::vector<int>::iterator up;
|
||||
up=std::upper_bound(tdof_offsets.begin(), tdof_offsets.end(),tdof); //
|
||||
return std::distance(tdof_offsets.begin(),up)-1;
|
||||
}
|
||||
|
||||
void ComputeTdofOffsets(const MPI_Comm & comm, const ParFiniteElementSpace * pfes,
|
||||
std::vector<int> & tdof_offsets)
|
||||
{
|
||||
int num_procs;
|
||||
MPI_Comm_size(comm, &num_procs);
|
||||
tdof_offsets.resize(num_procs);
|
||||
int mytoffset = pfes->GetMyTDofOffset();
|
||||
MPI_Allgather(&mytoffset,1,MPI_INT,&tdof_offsets[0],1,MPI_INT,comm);
|
||||
}
|
||||
|
||||
void GetSubdomainijk(int ip, const Array<int> nxyz, Array<int> & ijk)
|
||||
{
|
||||
ijk.SetSize(3);
|
||||
ijk[2] = ip/(nxyz[0]*nxyz[1]);
|
||||
ijk[1] = (ip-ijk[2]*nxyz[0]*nxyz[1])/nxyz[0];
|
||||
ijk[0] = (ip-ijk[2]*nxyz[0]*nxyz[1])%nxyz[0];
|
||||
}
|
||||
void GetDirectionijk(int id, Array<int> & ijk)
|
||||
{
|
||||
ijk.SetSize(3);
|
||||
int n = 3;
|
||||
ijk[2] = id/(n*n) - 1;
|
||||
ijk[1] = (id-(ijk[2]+1)*n*n)/n - 1;
|
||||
ijk[0] = (id-(ijk[2]+1)*n*n)%n - 1;
|
||||
}
|
||||
|
||||
int GetSubdomainId(const Array<int> nxyz, Array<int> & ijk)
|
||||
{
|
||||
int dim=ijk.Size();
|
||||
int k = (dim==2)? 0 : ijk[2];
|
||||
return k*nxyz[1]*nxyz[0] + ijk[1]*nxyz[0] + ijk[0];
|
||||
}
|
||||
|
||||
int GetDirectionId(const Array<int> & ijk)
|
||||
{
|
||||
int n = 3;
|
||||
int dim = ijk.Size();
|
||||
int k = (dim == 2) ? -1 : ijk[2];
|
||||
return (k+1)*n*n + (ijk[1]+1)*n + ijk[0]+1;
|
||||
}
|
||||
|
||||
void DofMaps::Init()
|
||||
{
|
||||
comm = pfes->GetComm();
|
||||
MPI_Comm_size(comm, &num_procs);
|
||||
MPI_Comm_rank(comm, &myid);
|
||||
|
||||
dim = pfes->GetParMesh()->Dimension();
|
||||
ComputeTdofOffsets(comm, pfes, tdof_offsets);
|
||||
myelemoffset = part->myelem_offset;
|
||||
mytoffset = pfes->GetMyTDofOffset();
|
||||
subdomain_rank = part->subdomain_rank;
|
||||
nrsubdomains = part->nrsubdomains;
|
||||
nxyz.SetSize(3);
|
||||
for (int i = 0; i<3; i++) { nxyz[i] = part->nxyz[i]; }
|
||||
|
||||
//compute sign factors for tdofs
|
||||
int lsize = pfes->GetVSize();
|
||||
int tsize = pfes->GetTrueVSize();
|
||||
tdof_sign.SetSize(tsize);
|
||||
for (int i = 0; i<lsize; i++)
|
||||
{
|
||||
int j = pfes->GetGlobalTDofNumber(i);
|
||||
if (j<mytoffset || j>=mytoffset+tsize) continue;
|
||||
tdof_sign[j-mytoffset] = pfes->GetDofSign(i);
|
||||
}
|
||||
}
|
||||
|
||||
DofMaps::DofMaps(ParFiniteElementSpace *pfes_, ParMeshPartition * part_, bool CompFlag_)
|
||||
: pfes(pfes_), part(part_), CompFlag(CompFlag_)
|
||||
{
|
||||
Init();
|
||||
Setup();
|
||||
}
|
||||
|
||||
void DofMaps::Setup()
|
||||
{
|
||||
// Setup the local FiniteElementSpaces
|
||||
const FiniteElementCollection * fec = pfes->FEColl();
|
||||
fes.SetSize(nrsubdomains);
|
||||
for (int i = 0; i<nrsubdomains; i++)
|
||||
{
|
||||
fes[i] = nullptr; // initialize with null on all procs
|
||||
if (myid == subdomain_rank[i])
|
||||
{
|
||||
fes[i] = new FiniteElementSpace(part->subdomain_mesh[i],fec);
|
||||
}
|
||||
}
|
||||
// cout << "Computing Overlap Tdofs" << endl;
|
||||
SubdomainToSubdomainMapsSetup();
|
||||
// TestSubdomainToSubdomainMaps();
|
||||
|
||||
SubdomainToGlobalMapsSetup();
|
||||
// TestSubdomainToGlobalMaps();
|
||||
}
|
||||
|
||||
void DofMaps::SubdomainToSubdomainMapsSetup()
|
||||
{
|
||||
ComputeOvlpElems();
|
||||
ComputeOvlpTdofs();
|
||||
}
|
||||
|
||||
void DofMaps::AddElementToOvlpLists(int l, int iel,
|
||||
const Array<bool> & neg, const Array<bool> & pos)
|
||||
{
|
||||
int kbeg = (dim == 2) ? 0 : -1;
|
||||
int kend = (dim == 2) ? 0 : 1;
|
||||
Array<int> dijk(3);
|
||||
for (int k = kbeg; k<=kend; k++)
|
||||
{
|
||||
if (dim == 3)
|
||||
{
|
||||
if (k == -1 && !neg[2]) continue;
|
||||
if (k == 1 && !pos[2]) continue;
|
||||
}
|
||||
|
||||
for (int j = -1; j<=1; j++)
|
||||
{
|
||||
if (j== -1 && !neg[1]) continue;
|
||||
if (j== 1 && !pos[1]) continue;
|
||||
for (int i = -1; i<=1; i++)
|
||||
{
|
||||
// cases to skip
|
||||
if (i==-1 && !neg[0]) continue;
|
||||
if (i== 1 && !pos[0]) continue;
|
||||
|
||||
if (i==0 && j==0 && k == 0) continue;
|
||||
dijk[0] = i; dijk[1] = j; dijk[2] = (dim==2)?-1 : k;
|
||||
int DirId = GetDirectionId(dijk);
|
||||
OvlpElems[l][DirId].Append(iel);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
void DofMaps::ComputeOvlpElems()
|
||||
{
|
||||
// first compute the element in the overlaps
|
||||
OvlpElems.resize(nrsubdomains);
|
||||
int nlayers = 2*part->OvlpNlayers;
|
||||
// loop through subdomains
|
||||
for (int l = 0; l<nrsubdomains; l++)
|
||||
{
|
||||
if (myid == subdomain_rank[l])
|
||||
{
|
||||
Array<int> ijk;
|
||||
GetSubdomainijk(l,nxyz,ijk);
|
||||
Mesh * mesh = part->subdomain_mesh[l];
|
||||
OvlpElems[l].resize(pow(3,dim));
|
||||
Vector pmin, pmax;
|
||||
mesh->GetBoundingBox(pmin,pmax);
|
||||
double h = part->MeshSize;
|
||||
// loop through the elements in the mesh and assign them to the
|
||||
// appropriate lists of overlaps
|
||||
for (int iel=0; iel< mesh->GetNE(); iel++)
|
||||
{
|
||||
// Get element center
|
||||
Vector center(dim);
|
||||
int geom = mesh->GetElementBaseGeometry(iel);
|
||||
ElementTransformation * tr = mesh->GetElementTransformation(iel);
|
||||
tr->Transform(Geometries.GetCenter(geom),center);
|
||||
|
||||
Array<bool> pos(dim); pos = false;
|
||||
Array<bool> neg(dim); neg = false;
|
||||
// loop through dimensions
|
||||
for (int d=0;d<dim; d++)
|
||||
{
|
||||
if (ijk[d]>0 && center[d] < pmin[d]+h*nlayers)
|
||||
{
|
||||
neg[d] = true;
|
||||
}
|
||||
|
||||
if (ijk[d]<nxyz[d]-1 && center[d] > pmax[d]-h*nlayers)
|
||||
{
|
||||
pos[d] = true;
|
||||
}
|
||||
}
|
||||
// Add the element to the appropriate lists
|
||||
AddElementToOvlpLists(l,iel,neg,pos);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
void DofMaps::ComputeOvlpTdofs()
|
||||
{
|
||||
OvlpTDofs.resize(nrsubdomains);
|
||||
int nrneighbors = pow(3,dim); // including its self
|
||||
|
||||
// loop through subdomains
|
||||
for (int l = 0; l<nrsubdomains; l++)
|
||||
{
|
||||
if (myid != subdomain_rank[l]) continue;
|
||||
int ntdofs = fes[l]->GetTrueVSize();
|
||||
Array<int> tdof_marker(ntdofs);
|
||||
OvlpTDofs[l].resize(nrneighbors);
|
||||
// loop through neighboring directions/neighbors
|
||||
for (int d=0; d<nrneighbors; d++)
|
||||
{
|
||||
tdof_marker = 0;
|
||||
Array<int> tdoflist;
|
||||
// Get the direction
|
||||
Array<int> dijk;
|
||||
GetDirectionijk(l,dijk);
|
||||
int nel = OvlpElems[l][d].Size();
|
||||
Array<int>Elems = OvlpElems[l][d];
|
||||
for (int iel = 0; iel<nel; ++iel)
|
||||
{
|
||||
int jel = Elems[iel];
|
||||
Array<int> ElemDofs;
|
||||
|
||||
fes[l]->GetElementDofs(jel,ElemDofs);
|
||||
int ndof = ElemDofs.Size();
|
||||
for (int i = 0; i<ndof; ++i)
|
||||
{
|
||||
int dof_ = ElemDofs[i];
|
||||
int dof = (dof_ >= 0) ? dof_ : abs(dof_) - 1;
|
||||
if (!tdof_marker[dof])
|
||||
{
|
||||
tdoflist.Append(dof); // dofs of ip0 in ovlp
|
||||
tdof_marker[dof] = 1;
|
||||
}
|
||||
}
|
||||
}
|
||||
OvlpTDofs[l][d] = tdoflist;
|
||||
if (CompFlag)
|
||||
{
|
||||
for (int i=0; i<tdoflist.Size(); i++)
|
||||
{
|
||||
tdoflist[i] += fes[l]->GetTrueVSize();
|
||||
}
|
||||
OvlpTDofs[l][d].Append(tdoflist);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
void DofMaps::PrintOvlpTdofs()
|
||||
{
|
||||
int nrneighbors = pow(3,dim); // including its self
|
||||
if (myid == 0)
|
||||
{
|
||||
for (int i = 0; i<nrsubdomains; i++)
|
||||
{
|
||||
if (myid != subdomain_rank[i]) continue;
|
||||
Array<int> ijk;
|
||||
GetSubdomainijk(i,nxyz,ijk);
|
||||
cout << "subdomain = " ; ijk.Print();
|
||||
cout << "myid = " << myid << endl;
|
||||
cout << "ip = " << i << endl;
|
||||
for (int d = 0; d<nrneighbors; d++)
|
||||
{
|
||||
Array<int> dijk;
|
||||
GetDirectionijk(d,dijk);
|
||||
cout << "direction = " ; dijk.Print();
|
||||
|
||||
if (OvlpTDofs[i][d].Size())
|
||||
{
|
||||
cout << "OvlpTdofs = " ;
|
||||
OvlpTDofs[i][d].Print(cout,OvlpTDofs[i][d].Size() );
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
void DofMaps::TransferToNeighbors(const Array<int> & SubdomainIds, const Array<Vector *> & x,
|
||||
std::vector<std::vector<Vector * >> & OvlpSol)
|
||||
{
|
||||
// 2D for now....
|
||||
MFEM_VERIFY(SubdomainIds.Size() == x.Size(), "TransferToNeighbors: Size inconsistency");
|
||||
int nrsendIds = SubdomainIds.Size();
|
||||
int nrneighbors = pow(3,dim);
|
||||
MPI_Request *recv_requests = new MPI_Request[nrsendIds*nrneighbors];
|
||||
MPI_Request *send_requests = new MPI_Request[nrsendIds*nrneighbors];
|
||||
MPI_Status *recv_statuses = new MPI_Status[nrsendIds*nrneighbors];
|
||||
MPI_Status *send_statuses = new MPI_Status[nrsendIds*nrneighbors];
|
||||
Array<Vector * > send_buffer(nrsendIds*nrneighbors);
|
||||
Array<Vector * > recv_buffer(nrsendIds*nrneighbors);
|
||||
int send_counter = 0;
|
||||
int recv_counter = 0;
|
||||
for (int is = 0; is<nrsendIds; is++)
|
||||
{
|
||||
int i0 = SubdomainIds[is];
|
||||
Array<int> ijk;
|
||||
GetSubdomainijk(i0,nxyz,ijk);
|
||||
for (int d=0;d<nrneighbors; d++)
|
||||
{
|
||||
Array<int>directions;
|
||||
GetDirectionijk(d,directions);
|
||||
|
||||
if (dim == 2 && directions[0] == 0 && directions[1] == 0) continue;
|
||||
if (dim == 3 && directions[0] == 0
|
||||
&& directions[1] == 0
|
||||
&& directions[2] == 0) continue;
|
||||
int i = ijk[0] + directions[0];
|
||||
if (i<0 || i>=nxyz[0]) continue;
|
||||
int j = ijk[1] + directions[1];
|
||||
if (j<0 || j>=nxyz[1]) continue;
|
||||
int k = (dim ==3 ) ? ijk[2] + directions[2] : 0;
|
||||
if (k<0 || k>=nxyz[2]) continue;
|
||||
Array<int>ijk1(3);
|
||||
ijk1[0] = i;
|
||||
ijk1[1] = j;
|
||||
ijk1[2] = k;
|
||||
int i1 = GetSubdomainId(nxyz,ijk1);
|
||||
if (myid == subdomain_rank[i0])
|
||||
{
|
||||
Array<int> tdofs0 = OvlpTDofs[i0][d]; // map of dofs in the overlap
|
||||
send_buffer[send_counter] = new Vector(tdofs0.Size());
|
||||
x[is]->GetSubVector(tdofs0,*send_buffer[send_counter]);
|
||||
// Destination rank
|
||||
int dest = subdomain_rank[i1];
|
||||
int tag = i0 * nrneighbors + d;
|
||||
|
||||
int count = tdofs0.Size();
|
||||
MPI_Isend(send_buffer[send_counter]->GetData(),count,MPI_DOUBLE,dest,
|
||||
tag,comm,&send_requests[send_counter]);
|
||||
send_counter++;
|
||||
|
||||
}
|
||||
if (myid == subdomain_rank[i1])
|
||||
{
|
||||
Array<int> direction1(3); direction1 = -1;
|
||||
for (int dd=0;dd<dim;dd++)
|
||||
{
|
||||
direction1[dd] = -directions[dd];
|
||||
}
|
||||
int d1 = GetDirectionId(direction1);
|
||||
|
||||
int count = OvlpTDofs[i1][d1].Size();
|
||||
recv_buffer[recv_counter] = new Vector(count);
|
||||
int src = subdomain_rank[i0];
|
||||
int tag = i0 * nrneighbors + d;
|
||||
MPI_Irecv(recv_buffer[recv_counter]->GetData(), count,MPI_DOUBLE,src,
|
||||
tag,comm, &recv_requests[recv_counter]);
|
||||
recv_counter++;
|
||||
}
|
||||
}
|
||||
}
|
||||
MPI_Waitall(send_counter, send_requests, send_statuses);
|
||||
MPI_Waitall(recv_counter, recv_requests, recv_statuses);
|
||||
|
||||
delete [] send_statuses;
|
||||
delete [] send_requests;
|
||||
delete [] recv_statuses;
|
||||
delete [] recv_requests;
|
||||
|
||||
for (int i = 0; i<send_counter; i++)
|
||||
{
|
||||
delete send_buffer[i];
|
||||
}
|
||||
send_buffer.DeleteAll();
|
||||
|
||||
|
||||
// Extract the transfered solutions
|
||||
recv_counter = 0;
|
||||
for (int is = 0; is<nrsendIds; is++)
|
||||
{
|
||||
int i0 = SubdomainIds[is];
|
||||
Array<int> ijk;
|
||||
GetSubdomainijk(i0,nxyz,ijk);
|
||||
for (int d=0;d<nrneighbors; d++)
|
||||
{
|
||||
Array<int>directions;
|
||||
GetDirectionijk(d,directions);
|
||||
if (dim == 2 && directions[0] == 0 && directions[1] == 0) continue;
|
||||
if (dim == 3 && directions[0] == 0
|
||||
&& directions[1] == 0
|
||||
&& directions[2] == 0) continue;
|
||||
int i = ijk[0] + directions[0];
|
||||
if (i<0 || i>=nxyz[0]) continue;
|
||||
int j = ijk[1] + directions[1];
|
||||
if (j<0 || j>=nxyz[1]) continue;
|
||||
int k = (dim ==3 ) ? ijk[2] + directions[2] : 0;
|
||||
if (k<0 || k>=nxyz[2]) continue;
|
||||
|
||||
Array<int>ijk1(3);
|
||||
ijk1[0] = i;
|
||||
ijk1[1] = j;
|
||||
ijk1[2] = k;
|
||||
int i1 = GetSubdomainId(nxyz,ijk1);
|
||||
if (myid == subdomain_rank[i1])
|
||||
{
|
||||
Array<int> direction1(3); direction1 = -1;
|
||||
for (int d=0;d<dim;d++)
|
||||
{
|
||||
direction1[d] = -directions[d];
|
||||
}
|
||||
int d1 = GetDirectionId(direction1);
|
||||
Array<int> tdofs1 = OvlpTDofs[i1][d1];
|
||||
if (!OvlpSol[i1][d1])
|
||||
{
|
||||
OvlpSol[i1][d1] = new Vector(2*fes[i1]->GetTrueVSize());
|
||||
}
|
||||
*OvlpSol[i1][d1] = 0.0;
|
||||
OvlpSol[i1][d1]->SetSubVector(tdofs1,*recv_buffer[recv_counter]);
|
||||
recv_counter++;
|
||||
}
|
||||
}
|
||||
}
|
||||
for (int i = 0; i<recv_counter; i++)
|
||||
{
|
||||
delete recv_buffer[i];
|
||||
}
|
||||
recv_buffer.DeleteAll();
|
||||
}
|
||||
|
||||
void DofMaps::TestSubdomainToSubdomainMaps()
|
||||
{
|
||||
// testing inter-subdomain communication
|
||||
FunctionCoefficient c1(testcoeff);
|
||||
int nrsub = nrsubdomains;
|
||||
Array<int> subdomain_ids(nrsub);
|
||||
Array<Vector*> x(nrsub);
|
||||
for (int i = 0; i<nrsub; i++)
|
||||
{
|
||||
x[i] = nullptr;
|
||||
subdomain_ids[i] = i;
|
||||
if (fes[i])
|
||||
{
|
||||
ComplexGridFunction gf(fes[i]);
|
||||
gf = 0.0;
|
||||
gf.ProjectCoefficient(c1,c1);
|
||||
x[i] = new Vector(2*fes[i]->GetTrueVSize());
|
||||
*x[i] = gf;
|
||||
}
|
||||
}
|
||||
|
||||
std::vector<std::vector<Vector * >> OvlpSol;
|
||||
|
||||
OvlpSol.resize(nrsubdomains);
|
||||
int nrneighbors = pow(3,dim);
|
||||
for (int ip = 0; ip<nrsubdomains; ip++)
|
||||
{
|
||||
if (myid == subdomain_rank[ip])
|
||||
{
|
||||
OvlpSol[ip].resize(nrneighbors);
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
TransferToNeighbors(subdomain_ids,x,OvlpSol);
|
||||
|
||||
string keys = "keys amrRljc\n";
|
||||
for (int i0 = 0 ; i0< nrsubdomains; i0++)
|
||||
{
|
||||
if (fes[i0])
|
||||
{
|
||||
ComplexGridFunction gf0(fes[i0]);
|
||||
for (int d = 0; d<nrneighbors; d++)
|
||||
{
|
||||
if(OvlpSol[i0][d])
|
||||
{
|
||||
Array<int>dijk;
|
||||
GetDirectionijk(d,dijk);
|
||||
Array<int>ijk;
|
||||
GetSubdomainijk(i0,nxyz,ijk);
|
||||
ostringstream oss;
|
||||
oss << "myid: " << myid
|
||||
<< ", subdomain: (" << ijk[0] << "," << ijk[1] <<")"
|
||||
<< ", direction: (" << dijk[0] << "," << dijk[1] <<")";
|
||||
|
||||
gf0 = 0.0;
|
||||
gf0.real().SetVector(*OvlpSol[i0][d],0);
|
||||
gf0.imag().SetVector(*OvlpSol[i0][d],fes[i0]->GetTrueVSize());
|
||||
char vishost[] = "localhost";
|
||||
int visport = 19916;
|
||||
socketstream sol_sock(vishost, visport);
|
||||
sol_sock.precision(8);
|
||||
sol_sock << "solution\n" << *(part->subdomain_mesh[i0]) << gf0.real()
|
||||
<< keys
|
||||
<< "window_title '" << oss.str() << "'" << flush;
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
for (int i = 0; i<nrsub; i++)
|
||||
{
|
||||
delete x[i];
|
||||
}
|
||||
}
|
||||
|
||||
void DofMaps::SubdomainToGlobalMapsSetup()
|
||||
{
|
||||
// workspace for MPI_AlltoAll
|
||||
send_count.SetSize(num_procs); send_count = 0;
|
||||
send_displ.SetSize(num_procs); send_displ = 0;
|
||||
recv_count.SetSize(num_procs); recv_count = 0;
|
||||
recv_displ.SetSize(num_procs); recv_displ = 0;
|
||||
|
||||
// 1. Communicate to the subdomain rank the list of tdofs
|
||||
// a. Compute send count
|
||||
for (int ip = 0; ip<nrsubdomains; ++ip)
|
||||
{
|
||||
// avoid any communication if on subdomain rank
|
||||
int nel = part->local_element_map[ip].Size();
|
||||
|
||||
for (int iel = 0; iel<nel; iel++)
|
||||
{
|
||||
int elem_idx = part->local_element_map[ip][iel] - myelemoffset;
|
||||
// int ndofs = local_tdofs[ip].Size();
|
||||
int ndofs = pfes->GetFE(elem_idx)->GetDof();
|
||||
|
||||
send_count[subdomain_rank[ip]] += 2 + ndofs;
|
||||
}
|
||||
}
|
||||
// b. Compute receive count
|
||||
MPI_Alltoall(send_count,1,MPI_INT,recv_count,1,MPI_INT,comm);
|
||||
for (int k=0; k<num_procs-1; k++)
|
||||
{
|
||||
send_displ[k+1] = send_displ[k] + send_count[k];
|
||||
recv_displ[k+1] = recv_displ[k] + recv_count[k];
|
||||
}
|
||||
sbuff_size = send_count.Sum();
|
||||
rbuff_size = recv_count.Sum();
|
||||
// c. Allocate and fill the send buffer
|
||||
Array<int> sendbuf(sbuff_size); sendbuf = 0;
|
||||
Array<int> soffs(num_procs); soffs = 0;
|
||||
for (int ip = 0; ip<nrsubdomains; ++ip)
|
||||
{
|
||||
int nel = part->local_element_map[ip].Size();
|
||||
for (int iel = 0; iel<nel; iel++)
|
||||
{
|
||||
int elem_idx = part->local_element_map[ip][iel] - myelemoffset;
|
||||
Array<int>ElemDofs;
|
||||
pfes->GetElementDofs(elem_idx,ElemDofs);
|
||||
int ndofs = ElemDofs.Size();
|
||||
|
||||
int j = send_displ[subdomain_rank[ip]] + soffs[subdomain_rank[ip]];
|
||||
sendbuf[j] = ip;
|
||||
sendbuf[j+1] = ndofs;
|
||||
|
||||
for (int k = 0; k < ndofs ; ++k)
|
||||
{
|
||||
int edof_ = ElemDofs[k];
|
||||
int edof = (edof_ >= 0) ? edof_ : abs(edof_) - 1;
|
||||
sendbuf[j+2+k] = pfes->GetGlobalTDofNumber(edof);
|
||||
}
|
||||
soffs[subdomain_rank[ip]] += 2 + ndofs;
|
||||
}
|
||||
}
|
||||
|
||||
// d. Communication
|
||||
Array<int> recvbuf(rbuff_size);
|
||||
MPI_Alltoallv(sendbuf, send_count, send_displ, MPI_INT, recvbuf,
|
||||
recv_count, recv_displ, MPI_INT, comm);
|
||||
|
||||
// 3. Extract from recv_buffer
|
||||
std::vector<Array<int>> global_tdofs(nrsubdomains);
|
||||
int k=0;
|
||||
while (k<rbuff_size)
|
||||
{
|
||||
int ip = recvbuf[k++];
|
||||
int ndofs = recvbuf[k++];
|
||||
for (int i = 0; i < ndofs; ++i)
|
||||
{
|
||||
global_tdofs[ip].Append(recvbuf[i+k]);
|
||||
}
|
||||
k += ndofs;
|
||||
}
|
||||
|
||||
SubdomainGTrueDofs.resize(nrsubdomains);
|
||||
// 4. Construct SubdomainTdof to Global mesh tdof maps
|
||||
for (int ip=0; ip<nrsubdomains; ++ip)
|
||||
{
|
||||
if (myid != subdomain_rank[ip]) continue;
|
||||
int nrdof = fes[ip]->GetTrueVSize();
|
||||
|
||||
SubdomainGTrueDofs[ip].SetSize(nrdof);
|
||||
int nel = part->element_map[ip].Size();
|
||||
int k = 0;
|
||||
for (int iel = 0; iel<nel; ++iel)
|
||||
{
|
||||
Array<int> elem_dofs;
|
||||
fes[ip]->GetElementDofs(iel,elem_dofs);
|
||||
int ndof = elem_dofs.Size();
|
||||
for (int i = 0; i<ndof; ++i)
|
||||
{
|
||||
int edof_ = elem_dofs[i];
|
||||
int edof = (edof_ >= 0) ? edof_ : abs(edof_) - 1;
|
||||
// rearranging dofs from serial fespace to pfes ordering
|
||||
SubdomainGTrueDofs[ip][edof] = global_tdofs[ip][k++];
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// 5. Communicate SubdomainGTrueDofs to participating ranks
|
||||
send_count = 0; send_displ = 0;
|
||||
recv_count = 0; recv_displ = 0;
|
||||
|
||||
for (int ip = 0; ip < nrsubdomains; ++ip)
|
||||
{
|
||||
if (myid != subdomain_rank[ip]) continue;
|
||||
int ndofs = SubdomainGTrueDofs[ip].Size();
|
||||
for (int i = 0; i<ndofs; ++i)
|
||||
{
|
||||
int tdof = SubdomainGTrueDofs[ip][i];
|
||||
int rank = get_rank(tdof,tdof_offsets);
|
||||
if (rank == subdomain_rank[ip]) continue; // <--------------
|
||||
send_count[rank] += 2; // 1 for the dof and 1 for the ip that goes to
|
||||
}
|
||||
}
|
||||
|
||||
// communicate so that recv_count is constructed
|
||||
MPI_Alltoall(send_count,1,MPI_INT,recv_count,1,MPI_INT,comm);
|
||||
//
|
||||
for (int k=0; k<num_procs-1; k++)
|
||||
{
|
||||
send_displ[k+1] = send_displ[k] + send_count[k];
|
||||
recv_displ[k+1] = recv_displ[k] + recv_count[k];
|
||||
}
|
||||
sbuff_size = send_count.Sum();
|
||||
rbuff_size = recv_count.Sum();
|
||||
|
||||
sendbuf.SetSize(sbuff_size);
|
||||
sendbuf = 0; soffs = 0;
|
||||
|
||||
for (int ip = 0; ip < nrsubdomains; ip++)
|
||||
{
|
||||
if (myid != subdomain_rank[ip]) continue;
|
||||
int ndofs = SubdomainGTrueDofs[ip].Size();
|
||||
// loop through dofs
|
||||
for (int i = 0; i<ndofs; ++i)
|
||||
{
|
||||
int tdof = SubdomainGTrueDofs[ip][i];
|
||||
int irank = get_rank(tdof,tdof_offsets);
|
||||
if (irank == subdomain_rank[ip]) continue; // <--------------
|
||||
int j = send_displ[irank] + soffs[irank];
|
||||
sendbuf[j] = ip;
|
||||
sendbuf[j+1] = SubdomainGTrueDofs[ip][i];
|
||||
soffs[irank] += 2 ;
|
||||
}
|
||||
}
|
||||
|
||||
recvbuf.SetSize(rbuff_size);
|
||||
MPI_Alltoallv(sendbuf, send_count, send_displ, MPI_INT, recvbuf,
|
||||
recv_count, recv_displ, MPI_INT, comm);
|
||||
|
||||
// List of tdofs owned by the processor for subdomains not owned
|
||||
SubdomainLTrueDofs.resize(nrsubdomains);
|
||||
for (int k=0; k<rbuff_size/2; k++)
|
||||
{
|
||||
int ip = recvbuf[2*k];
|
||||
int tdof = recvbuf[2*k+1];
|
||||
SubdomainLTrueDofs[ip].Append(tdof);
|
||||
}
|
||||
}
|
||||
|
||||
// Restriction of global residual to subdomain residuals
|
||||
void DofMaps::GlobalToSubdomains(const Vector & y, Array<Vector*> & x)
|
||||
{
|
||||
send_count = 0; send_displ = 0;
|
||||
recv_count = 0; recv_displ = 0;
|
||||
|
||||
// Compute send_counts
|
||||
int m = (CompFlag) ? 2 : 1 ;
|
||||
for (int ip = 0; ip < nrsubdomains; ip++)
|
||||
{
|
||||
if (myid == subdomain_rank[ip]) continue; // <---------------
|
||||
int ndofs = SubdomainLTrueDofs[ip].Size();
|
||||
send_count[subdomain_rank[ip]] += m * ndofs;
|
||||
}
|
||||
|
||||
// communicate so that recv_count is constructed
|
||||
MPI_Alltoall(send_count,1,MPI_INT,recv_count,1,MPI_INT,comm);
|
||||
|
||||
for (int k=0; k<num_procs-1; k++)
|
||||
{
|
||||
send_displ[k+1] = send_displ[k] + send_count[k];
|
||||
recv_displ[k+1] = recv_displ[k] + recv_count[k];
|
||||
}
|
||||
sbuff_size = send_count.Sum();
|
||||
rbuff_size = recv_count.Sum();
|
||||
|
||||
Array<double> sendbuf(sbuff_size); sendbuf = 0;
|
||||
Array<int> soffs(num_procs); soffs = 0;
|
||||
|
||||
for (int ip = 0; ip < nrsubdomains; ip++)
|
||||
{
|
||||
if (myid == subdomain_rank[ip]) continue; // <---------------
|
||||
int ndofs = SubdomainLTrueDofs[ip].Size();
|
||||
for (int i = 0; i<ndofs; i++)
|
||||
{
|
||||
int tdof = SubdomainLTrueDofs[ip][i];
|
||||
int j = send_displ[subdomain_rank[ip]] + soffs[subdomain_rank[ip]];
|
||||
soffs[subdomain_rank[ip]] +=m;
|
||||
int k = tdof - mytoffset;
|
||||
// sendbuf[j] = y[k];
|
||||
sendbuf[j] = tdof_sign[k]*y[k];
|
||||
if (CompFlag)
|
||||
{ // if complex valued
|
||||
int tsize = pfes->GetTrueVSize();
|
||||
// sendbuf[j+1] = y[k+tsize];
|
||||
sendbuf[j+1] = tdof_sign[k]*y[k+tsize];
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// communication
|
||||
Array<double> recvbuf(rbuff_size);
|
||||
MPI_Alltoallv(sendbuf, send_count, send_displ, MPI_DOUBLE, recvbuf,
|
||||
recv_count, recv_displ, MPI_DOUBLE, comm);
|
||||
Array<int> roffs(num_procs);
|
||||
roffs = 0;
|
||||
// Now each process will construct the res vector
|
||||
x.SetSize(nrsubdomains);
|
||||
for (int ip = 0; ip < nrsubdomains; ip++)
|
||||
{
|
||||
if (myid != subdomain_rank[ip]) continue;
|
||||
int ndof = SubdomainGTrueDofs[ip].Size();
|
||||
if (!x[ip]) x[ip] = new Vector(m*ndof);
|
||||
*x[ip] = 0.0;
|
||||
// extract the data from receiv buffer
|
||||
for (int i=0; i<ndof; i++)
|
||||
{
|
||||
// pick up the tdof and find its rank
|
||||
int tdof = SubdomainGTrueDofs[ip][i];
|
||||
int tdof_rank = get_rank(tdof,tdof_offsets);
|
||||
if (tdof_rank != subdomain_rank[ip]) // <---------------
|
||||
{
|
||||
int k = recv_displ[tdof_rank] + roffs[tdof_rank];
|
||||
roffs[tdof_rank] += m;
|
||||
(*x[ip])[i] = recvbuf[k];
|
||||
if (CompFlag)
|
||||
{
|
||||
(*x[ip])[i+ndof] = recvbuf[k+1];
|
||||
}
|
||||
}
|
||||
else
|
||||
{
|
||||
int k = tdof - mytoffset;
|
||||
// (*x[ip])[i] = y[k];
|
||||
(*x[ip])[i] = tdof_sign[k]*y[k];
|
||||
if (CompFlag)
|
||||
{
|
||||
int gtsize = pfes->GetTrueVSize();
|
||||
(*x[ip])[i+ndof] = tdof_sign[k]*y[k+gtsize];
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// Prolongation of subdomain solutions to the global solution
|
||||
void DofMaps::SubdomainsToGlobal(const Array<Vector*> & x, Vector & y)
|
||||
{
|
||||
send_count = 0; send_displ = 0;
|
||||
recv_count = 0; recv_displ = 0;
|
||||
|
||||
// Compute send_counts
|
||||
int m = (CompFlag) ? 2 : 1 ;
|
||||
for (int ip = 0; ip < nrsubdomains; ip++)
|
||||
{
|
||||
if (myid != subdomain_rank[ip]) continue;
|
||||
int ndofs = SubdomainGTrueDofs[ip].Size();
|
||||
for (int i=0; i<ndofs; i++)
|
||||
{
|
||||
// pick up the tdof and find its rank
|
||||
int tdof = SubdomainGTrueDofs[ip][i];
|
||||
int tdof_rank = get_rank(tdof,tdof_offsets);
|
||||
if (tdof_rank == subdomain_rank[ip]) continue;
|
||||
send_count[tdof_rank] +=m;
|
||||
}
|
||||
}
|
||||
|
||||
MPI_Alltoall(send_count,1,MPI_INT,recv_count,1,MPI_INT,comm);
|
||||
|
||||
for (int k=0; k<num_procs-1; k++)
|
||||
{
|
||||
send_displ[k+1] = send_displ[k] + send_count[k];
|
||||
recv_displ[k+1] = recv_displ[k] + recv_count[k];
|
||||
}
|
||||
sbuff_size = send_count.Sum();
|
||||
rbuff_size = recv_count.Sum();
|
||||
|
||||
Array<double> sendbuf(sbuff_size); sendbuf = 0;
|
||||
Array<int> soffs(num_procs); soffs = 0;
|
||||
|
||||
for (int ip = 0; ip < nrsubdomains; ip++)
|
||||
{
|
||||
if (myid != subdomain_rank[ip]) continue;
|
||||
int ndofs = SubdomainGTrueDofs[ip].Size();
|
||||
// loop through dofs
|
||||
for (int i=0; i<ndofs; i++)
|
||||
{
|
||||
// pick up the dof and find its tdof_rank
|
||||
int tdof = SubdomainGTrueDofs[ip][i];
|
||||
int tdof_rank = get_rank(tdof,tdof_offsets);
|
||||
// offset
|
||||
if (tdof_rank == subdomain_rank[ip]) continue;
|
||||
int k = send_displ[tdof_rank] + soffs[tdof_rank];
|
||||
soffs[tdof_rank] +=m;
|
||||
sendbuf[k] = (*x[ip])[i];
|
||||
if (CompFlag)
|
||||
{
|
||||
sendbuf[k+1] = (*x[ip])[i+ndofs];
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
Array<double> recvbuf(rbuff_size);
|
||||
Array<int> roffs(num_procs); roffs = 0;
|
||||
MPI_Alltoallv(sendbuf, send_count, send_displ, MPI_DOUBLE, recvbuf,
|
||||
recv_count, recv_displ, MPI_DOUBLE, comm);
|
||||
|
||||
for (int ip = 0; ip < nrsubdomains; ip++)
|
||||
{
|
||||
if (myid == subdomain_rank[ip])
|
||||
{
|
||||
int ndofs = SubdomainGTrueDofs[ip].Size();
|
||||
for (int i = 0; i<ndofs; i++)
|
||||
{
|
||||
int tdof = SubdomainGTrueDofs[ip][i];
|
||||
int k = tdof - mytoffset;
|
||||
if (k<0 || k>=pfes->GetTrueVSize()) continue;
|
||||
y[k] += tdof_sign[k] * (*x[ip])[i];
|
||||
if (CompFlag)
|
||||
{
|
||||
int gtsize = pfes->GetTrueVSize();
|
||||
y[k+gtsize] += tdof_sign[k]*(*x[ip])[i+ndofs];
|
||||
}
|
||||
}
|
||||
}
|
||||
else
|
||||
{
|
||||
int ndofs = SubdomainLTrueDofs[ip].Size();
|
||||
for (int i = 0; i<ndofs; i++)
|
||||
{
|
||||
int tdof = SubdomainLTrueDofs[ip][i];
|
||||
int k = tdof - mytoffset;
|
||||
int j = recv_displ[subdomain_rank[ip]] + roffs[subdomain_rank[ip]];
|
||||
roffs[subdomain_rank[ip]] +=m;
|
||||
y[k] += tdof_sign[k] * recvbuf[j];
|
||||
if (CompFlag)
|
||||
{
|
||||
int tsize = pfes->GetTrueVSize();
|
||||
y[k+tsize] += tdof_sign[k]*recvbuf[j+1];
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
void DofMaps::TestSubdomainToGlobalMaps()
|
||||
{
|
||||
cout << "Testing Subdomain To Global Maps" << endl;
|
||||
FunctionCoefficient c1(testcoeff);
|
||||
Array<Vector*> x(nrsubdomains);
|
||||
Vector y(pfes->GetTrueVSize()); y = 0.0;
|
||||
for (int i = 0 ; i<nrsubdomains; i++)
|
||||
{
|
||||
if (myid != subdomain_rank[i]) continue;
|
||||
x[i] = new Vector(fes[i]->GetTrueVSize());
|
||||
GridFunction gf(fes[i]);
|
||||
gf = 0.0;
|
||||
|
||||
if (i==3) gf.ProjectCoefficient(c1);
|
||||
*x[i] = gf;
|
||||
}
|
||||
|
||||
SubdomainsToGlobal(x,y);
|
||||
|
||||
// cout << "1: myid = " << myid << ", y = "; y.Print();
|
||||
|
||||
string keys = (dim==2) ? "keys amrRljc\n": "keys m\n";
|
||||
ParGridFunction pgf(pfes);
|
||||
|
||||
const Operator &P = *pfes->GetProlongationMatrix();
|
||||
P.Mult(y, pgf);
|
||||
|
||||
char vishost[] = "localhost";
|
||||
int visport = 19916;
|
||||
socketstream sol_sock(vishost, visport);
|
||||
sol_sock.precision(8);
|
||||
sol_sock << "parallel " << num_procs << " " << myid << "\n"
|
||||
<< "solution\n" << *pfes->GetParMesh() << pgf
|
||||
<< keys << flush;
|
||||
|
||||
ParGridFunction pgf1(pfes);
|
||||
pgf1.ProjectCoefficient(c1);
|
||||
Vector y1(pfes->GetTrueVSize());
|
||||
const SparseMatrix * R = pfes->GetRestrictionMatrix();
|
||||
|
||||
R->Mult(pgf1,y1);
|
||||
// P.MultTranspose(pgf1,y1);
|
||||
Array<Vector*> x1;
|
||||
GlobalToSubdomains(y1,x1);
|
||||
|
||||
|
||||
// for (int i = 0 ; i<nrsubdomains; i++)
|
||||
// {
|
||||
// if (myid != subdomain_rank[i]) continue;
|
||||
// ostringstream mesh_name;
|
||||
// mesh_name << "output/mesh." << setfill('0') << setw(6) << i;
|
||||
// ofstream mesh_ofs(mesh_name.str().c_str());
|
||||
// mesh_ofs.precision(8);
|
||||
// fes[i]->GetMesh()->Print(mesh_ofs);
|
||||
// GridFunction gf(fes[i]);
|
||||
// gf = x1[i];
|
||||
// ostringstream gf_name;
|
||||
// gf_name << "output/gf." << setfill('0') << setw(6) << i;
|
||||
// ofstream gf_ofs(gf_name.str().c_str());
|
||||
// gf_ofs.precision(8);
|
||||
// gf.Save(gf_ofs);
|
||||
// }
|
||||
|
||||
|
||||
|
||||
int nrsub = nrsubdomains;
|
||||
for (int i = 0 ; i<nrsub; i++)
|
||||
{
|
||||
if (myid == subdomain_rank[i])
|
||||
{
|
||||
socketstream sol_sock1(vishost, visport);
|
||||
sol_sock1.precision(8);
|
||||
sol_sock1 << "parallel " << nrsub << " " << i << "\n";
|
||||
GridFunction gf(fes[i]);
|
||||
GridFunction gf1(fes[i]);
|
||||
gf1.ProjectCoefficient(c1);
|
||||
gf = *x1[i];
|
||||
gf1-=gf;
|
||||
cout << "ip, Diff norm = " <<i<<", " << gf1.Norml2() << endl;
|
||||
sol_sock1 << "solution\n" << *fes[i]->GetMesh() << gf
|
||||
<< keys << flush;
|
||||
}
|
||||
MPI_Barrier(MPI_COMM_WORLD);
|
||||
}
|
||||
|
||||
socketstream gf_sock(vishost, visport);
|
||||
gf_sock.precision(8);
|
||||
gf_sock << "parallel " << num_procs << " " << myid << "\n"
|
||||
<< "solution\n" << *pfes->GetParMesh() << pgf1
|
||||
<< keys << flush;
|
||||
}
|
||||
|
||||
|
||||
DofMaps::~DofMaps()
|
||||
{
|
||||
for (int i = 0; i<nrsubdomains; i++)
|
||||
{
|
||||
delete fes[i];
|
||||
}
|
||||
}
|
||||
@@ -1,109 +0,0 @@
|
||||
#pragma once
|
||||
#include "../common/Utilities.hpp"
|
||||
#include "../common/PML.hpp"
|
||||
using namespace std;
|
||||
using namespace mfem;
|
||||
|
||||
double testcoeff(const Vector & x);
|
||||
int get_rank(int tdof, std::vector<int> & tdof_offsets);
|
||||
|
||||
|
||||
void ComputeTdofOffsets(const MPI_Comm & comm, const ParFiniteElementSpace * pfes,
|
||||
std::vector<int> & tdof_offsets);
|
||||
|
||||
void GetSubdomainijk(int ip, const Array<int> nxyz, Array<int> & ijk);
|
||||
void GetDirectionijk(int id, Array<int> & ijk);
|
||||
int GetSubdomainId(const Array<int> nxyz, Array<int> & ijk);
|
||||
int GetDirectionId(const Array<int> & ijk);
|
||||
|
||||
|
||||
// class handling two types of dof maps
|
||||
// 1. Subdomain truedofs ---> Global truedofs
|
||||
// 2. Subdomain truedofs ---> Neighbor truedofs
|
||||
class DofMaps
|
||||
{
|
||||
private:
|
||||
// The FE space of the problem (H1/Hcurl)
|
||||
ParFiniteElementSpace *pfes = nullptr;
|
||||
|
||||
// The given partition of the parmesh
|
||||
ParMeshPartition *part = nullptr;
|
||||
// partition in x-y-z
|
||||
Array<int> nxyz;
|
||||
|
||||
// MPI parameters
|
||||
MPI_Comm comm = MPI_COMM_WORLD;
|
||||
int num_procs, myid;
|
||||
|
||||
// true dof offset and element offset of the processor
|
||||
vector<int> tdof_offsets;
|
||||
int mytoffset;
|
||||
int myelemoffset;
|
||||
|
||||
int dim;
|
||||
// Total number of subdomains
|
||||
int nrsubdomains;
|
||||
|
||||
// Array specifying the subdomain rank
|
||||
Array<int> subdomain_rank;
|
||||
|
||||
// Complex flag
|
||||
bool CompFlag;
|
||||
|
||||
// sign factors
|
||||
Array<int> tdof_sign;
|
||||
// Initializing mpi and helper parameters
|
||||
void Init();
|
||||
|
||||
// 1. Setting up the subdomains FE spaces
|
||||
// 2. Setting up the subdomains-to-subdomains maps
|
||||
// 3. Setting up the subdomain-to-global maps
|
||||
void Setup();
|
||||
|
||||
// -----------------------------------------------
|
||||
// Subdomain to Subdomain maps
|
||||
// -----------------------------------------------
|
||||
std::vector<std::vector<Array<int>>> OvlpElems;
|
||||
void AddElementToOvlpLists(int l, int iel,
|
||||
const Array<bool> & neg,
|
||||
const Array<bool> & pos);
|
||||
std::vector<std::vector<Array<int>>> OvlpTDofs;
|
||||
void SubdomainToSubdomainMapsSetup();
|
||||
void ComputeOvlpElems();
|
||||
void ComputeOvlpTdofs();
|
||||
void PrintOvlpTdofs();
|
||||
|
||||
// -----------------------------------------------
|
||||
// Subdomain to Global maps
|
||||
// -----------------------------------------------
|
||||
std::vector<Array<int>> SubdomainGTrueDofs; // Subdomain Tdofs to Global Tdofs
|
||||
std::vector<Array<int>> SubdomainLTrueDofs; // Subdomain Tdofs to Local (on rank) Tdofs
|
||||
|
||||
Array<int> send_count, send_displ;
|
||||
Array<int> recv_count, recv_displ;
|
||||
int sbuff_size = 0;
|
||||
int rbuff_size = 0;
|
||||
void SubdomainToGlobalMapsSetup();
|
||||
|
||||
// Testing
|
||||
void TestSubdomainToGlobalMaps();
|
||||
void TestSubdomainToSubdomainMaps();
|
||||
|
||||
public:
|
||||
// constructor
|
||||
|
||||
// FiniteElementSpaces of the subdomains
|
||||
Array<FiniteElementSpace *> fes;
|
||||
|
||||
DofMaps(ParFiniteElementSpace *fespace_, ParMeshPartition * part_, bool CompFlag_ = false);
|
||||
~DofMaps();
|
||||
// Transfering from subdomains SubdomainIds to all their neighbors
|
||||
void TransferToNeighbors(const Array<int> & SubdomainIds, const Array<Vector *> & x,
|
||||
std::vector<std::vector<Vector * >> & OvlpSol);
|
||||
|
||||
// Prolongation of subdomain solutions to the global solution
|
||||
void SubdomainsToGlobal(const Array<Vector*> & x, Vector & y);
|
||||
// Restriction of global residual to subdomain residuals
|
||||
// bool comp: true for complex valued problems
|
||||
void GlobalToSubdomains(const Vector & y, Array<Vector*> & x);
|
||||
};
|
||||
@@ -1,849 +0,0 @@
|
||||
//Parallel Diagonal Source Transfer Preconditioner
|
||||
|
||||
#include "ParDST.hpp"
|
||||
|
||||
ParDST::ParDST(ParSesquilinearForm * bf_, Array2D<double> & Pmllength_,
|
||||
double omega_, Coefficient * Q_, int nrlayers_ , int nx_, int ny_, int nz_)
|
||||
: Solver(2*bf_->ParFESpace()->GetTrueVSize(), 2*bf_->ParFESpace()->GetTrueVSize()),
|
||||
bf(bf_), Pmllength(Pmllength_), omega(omega_),
|
||||
Q(Q_), nrlayers(nrlayers_)
|
||||
{
|
||||
nx = nx_; ny = ny_; nz = nz_;
|
||||
Init();
|
||||
}
|
||||
ParDST::ParDST(ParSesquilinearForm * bf_, Array2D<double> & Pmllength_,
|
||||
double omega_, VectorCoefficient * VQ_, int nrlayers_ , int nx_, int ny_, int nz_)
|
||||
: Solver(2*bf_->ParFESpace()->GetTrueVSize(), 2*bf_->ParFESpace()->GetTrueVSize()),
|
||||
bf(bf_), Pmllength(Pmllength_), omega(omega_),
|
||||
VQ(VQ_), nrlayers(nrlayers_)
|
||||
{
|
||||
nx = nx_; ny = ny_; nz = nz_;
|
||||
Init();
|
||||
}
|
||||
ParDST::ParDST(ParSesquilinearForm * bf_, Array2D<double> & Pmllength_,
|
||||
double omega_, MatrixCoefficient * MQ_, int nrlayers_ , int nx_, int ny_, int nz_)
|
||||
: Solver(2*bf_->ParFESpace()->GetTrueVSize(), 2*bf_->ParFESpace()->GetTrueVSize()),
|
||||
bf(bf_), Pmllength(Pmllength_), omega(omega_),
|
||||
MQ(MQ_), nrlayers(nrlayers_)
|
||||
{
|
||||
nx = nx_; ny = ny_; nz = nz_;
|
||||
Init();
|
||||
}
|
||||
|
||||
void ParDST::Init()
|
||||
{
|
||||
pfes = bf->ParFESpace();
|
||||
fec = pfes->FEColl();
|
||||
|
||||
comm = pfes->GetComm();
|
||||
MPI_Comm_size(comm, &num_procs);
|
||||
MPI_Comm_rank(comm, &myid);
|
||||
|
||||
//1. Indentify problem ... Helmholtz or Maxwell
|
||||
prob_kind = fec->GetContType();
|
||||
if (myid == 0)
|
||||
{
|
||||
cout << " 1. Indentify problem to be solved ... " << endl;
|
||||
if (prob_kind == 0) cout << " Helmholtz" << endl;
|
||||
if (prob_kind == 1) cout << " Maxwell" << endl;
|
||||
}
|
||||
|
||||
//2. Create the parallel mesh partition
|
||||
pmesh = pfes->GetParMesh();
|
||||
dim = pmesh->Dimension();
|
||||
if (myid == 0)
|
||||
{
|
||||
cout << "\n 2. Generating ParMesh partitioning ... " << endl;
|
||||
}
|
||||
ovlpnrlayers = nrlayers+1;
|
||||
part = new ParMeshPartition(pmesh,nx,ny,nz,ovlpnrlayers);
|
||||
nxyz.SetSize(3);
|
||||
nxyz[0] = nx = part->nxyz[0];
|
||||
nxyz[1] = ny = part->nxyz[1];
|
||||
nxyz[2] = nz = part->nxyz[2];
|
||||
|
||||
nrsubdomains = part->nrsubdomains;
|
||||
SubdomainRank = part->subdomain_rank;
|
||||
|
||||
for (int ip = 0; ip<nrsubdomains; ip++)
|
||||
{
|
||||
if (myid == SubdomainRank[ip])
|
||||
{
|
||||
RankSubdomains.Append(ip);
|
||||
}
|
||||
}
|
||||
|
||||
cout << " myid: " << myid
|
||||
<< ", nrsubdomains: " << RankSubdomains.Size() << endl;
|
||||
|
||||
MPI_Barrier(comm);
|
||||
|
||||
if (myid == 0)
|
||||
{
|
||||
cout << " Done ! " << endl;
|
||||
}
|
||||
//3. Setup info for sweeps
|
||||
if (myid == 0)
|
||||
{
|
||||
cout << "\n 3. Computing sweeps info ..." << endl;
|
||||
}
|
||||
sweeps = new Sweep(dim);
|
||||
if (myid == 0)
|
||||
{
|
||||
cout << " Done ! " << endl;
|
||||
}
|
||||
//4. Create LocalToGlobal maps
|
||||
// (local GridFunctions/Vector to Global ParGridFunction/Vector)
|
||||
if (myid == 0)
|
||||
{
|
||||
cout << "\n 4. Computing true dofs maps ..." << endl;
|
||||
}
|
||||
|
||||
// if (myid == SubdomainRank[0])
|
||||
// {
|
||||
// cout << "myid = " << myid << endl;
|
||||
// char vishost[] = "localhost";
|
||||
// int visport = 19916;
|
||||
// socketstream mesh_sock1(vishost, visport);
|
||||
// mesh_sock1.precision(8);
|
||||
// mesh_sock1 << "mesh\n"
|
||||
// << *part->subdomain_mesh[0] << "window_title 'Subdomain'" << flush;
|
||||
// part->subdomain_mesh[0]->Print();
|
||||
|
||||
// }
|
||||
bool comp = true;
|
||||
|
||||
dmaps = new DofMaps(pfes,part, comp);
|
||||
|
||||
if (myid == 0)
|
||||
{
|
||||
cout << " Done ! " << endl;
|
||||
}
|
||||
// 4. Setting up the local problems
|
||||
if (myid == 0)
|
||||
{
|
||||
cout << "\n 5. Setting up the subdomain problems ..." << endl;
|
||||
}
|
||||
|
||||
SetupSubdomainProblems();
|
||||
|
||||
if (myid == 0)
|
||||
{
|
||||
cout << " Done ! " << endl;
|
||||
}
|
||||
if (myid == 0)
|
||||
{
|
||||
cout << "\n 6. Mark subdomain overlap truedofs ..." << endl;
|
||||
}
|
||||
MarkSubdomainOverlapDofs(comp);
|
||||
if (myid == 0)
|
||||
{
|
||||
cout << " Done ! " << endl;
|
||||
}
|
||||
}
|
||||
|
||||
void ParDST::Mult(const Vector &r, Vector &z) const
|
||||
{
|
||||
// Initialize transfered residuals to 0.0;
|
||||
for (int ip=0; ip<nrsubdomains; ip++)
|
||||
{
|
||||
if (myid != SubdomainRank[ip]) continue;
|
||||
for (int i=0;i<sweeps->nsweeps; i++)
|
||||
{
|
||||
*f_transf[ip][i] = 0.0;
|
||||
}
|
||||
}
|
||||
|
||||
// restrict given residual to subdomains
|
||||
dmaps->GlobalToSubdomains(r,f_orig);
|
||||
|
||||
for (int ip=0; ip<nrsubdomains; ip++)
|
||||
{
|
||||
if (myid != SubdomainRank[ip]) continue;
|
||||
Array<int> ijk(3);
|
||||
GetSubdomainijk(ip,nxyz,ijk);
|
||||
Array2D<int> direct(dim,2); direct = 0;
|
||||
for (int d=0;d<dim; d++)
|
||||
{
|
||||
if (ijk[d] > 0) direct[d][0] = 1;
|
||||
if (ijk[d] < part->nxyz[d]-1) direct[d][1] = 1;
|
||||
}
|
||||
GetChiRes(*f_orig[ip],ip,direct);
|
||||
}
|
||||
|
||||
z = 0.0;
|
||||
int nsteps;
|
||||
switch(dim)
|
||||
{
|
||||
case 1: nsteps = nx; break;
|
||||
case 2: nsteps = nx+ny-1; break;
|
||||
default: nsteps = nx+ny+nz-2; break;
|
||||
}
|
||||
int nsweeps = sweeps->nsweeps;
|
||||
// 1. Loop through sweeps
|
||||
if (dim == 3 && nz == 1) { nsweeps = 4; } // x-y partition only;
|
||||
for (int l=0; l<nsweeps; l++)
|
||||
{
|
||||
// 2. loop through diagonals/steps of each sweep
|
||||
for (int s = 0; s<nsteps; s++)
|
||||
{
|
||||
Array2D<int> subdomains;
|
||||
GetStepSubdomains(l,s,subdomains);
|
||||
int nsubdomains = subdomains.NumRows();
|
||||
|
||||
// 3. Loop through the subdomains on the diagonal
|
||||
Array<int> subdomain_ids;
|
||||
for (int sb=0; sb < nsubdomains; sb++)
|
||||
{
|
||||
Array<int> ijk(dim); ijk = 0;
|
||||
for (int d=0; d<dim; d++) ijk[d] = subdomains[sb][d];
|
||||
int ip = GetSubdomainId(nxyz,ijk);
|
||||
subdomain_ids.Append(ip);
|
||||
if (myid != SubdomainRank[ip]) continue;
|
||||
|
||||
int n = dmaps->fes[ip]->GetTrueVSize();
|
||||
Vector res_local(2*n); res_local = 0.0;
|
||||
|
||||
if (l==0) { res_local += *f_orig[ip]; }
|
||||
res_local += *f_transf[ip][l];
|
||||
if (res_local.Norml2() < 1e-12)
|
||||
{
|
||||
*subdomain_sol[ip] = 0.0;
|
||||
continue;
|
||||
}
|
||||
PmlMatInv[ip]->Mult(res_local, *subdomain_sol[ip]);
|
||||
}
|
||||
// 4. Transfer solutions to neighbors so that the subdomain
|
||||
// residuals are updated
|
||||
TransferSources(l,subdomain_ids);
|
||||
}
|
||||
// 5. Update the global solution
|
||||
dmaps->SubdomainsToGlobal(subdomain_sol,z);
|
||||
}
|
||||
}
|
||||
|
||||
void ParDST::SetupSubdomainProblems()
|
||||
{
|
||||
sqf.SetSize(nrsubdomains);
|
||||
Optr.SetSize(nrsubdomains);
|
||||
PmlMat.SetSize(nrsubdomains);
|
||||
PmlMatInv.SetSize(nrsubdomains);
|
||||
f_orig.SetSize(nrsubdomains);
|
||||
f_transf.SetSize(nrsubdomains);
|
||||
subdomain_sol.SetSize(nrsubdomains);
|
||||
for (int ip=0; ip<nrsubdomains; ip++)
|
||||
{
|
||||
sqf[ip] = nullptr;
|
||||
f_orig[ip] = nullptr;
|
||||
subdomain_sol[ip] = nullptr;
|
||||
PmlMat[ip] = nullptr;
|
||||
PmlMatInv[ip] = nullptr;
|
||||
Optr[ip] = nullptr;
|
||||
|
||||
if (myid != SubdomainRank[ip]) continue;
|
||||
subdomain_sol[ip] = new Vector(2*dmaps->fes[ip]->GetTrueVSize());
|
||||
if (prob_kind == 0)
|
||||
{
|
||||
SetHelmholtzPmlSystemMatrix(ip);
|
||||
}
|
||||
else if (prob_kind == 1)
|
||||
{
|
||||
SetMaxwellPmlSystemMatrix(ip);
|
||||
}
|
||||
PmlMat[ip] = Optr[ip]->As<ComplexSparseMatrix>();
|
||||
|
||||
PmlMatInv[ip] = new ComplexUMFPackSolver;
|
||||
PmlMatInv[ip]->Control[UMFPACK_ORDERING] = UMFPACK_ORDERING_METIS;
|
||||
PmlMatInv[ip]->SetOperator(*PmlMat[ip]);
|
||||
|
||||
// HYPRE_Int rowstarts[2]; rowstarts[0] = 0;
|
||||
// rowstarts[1] = dmaps->fes[ip]->GetTrueVSize();
|
||||
// HypreParMatrix * HypreMat_r =
|
||||
// new HypreParMatrix(MPI_COMM_SELF,rowstarts[1],rowstarts,
|
||||
// &(PmlMat[ip]->real()));
|
||||
// HypreParMatrix * HypreMat_i =
|
||||
// new HypreParMatrix(MPI_COMM_SELF,rowstarts[1],rowstarts,
|
||||
// &(PmlMat[ip]->imag()));
|
||||
// ComplexHypreParMatrix * HypreMat =
|
||||
// new ComplexHypreParMatrix(HypreMat_r,HypreMat_i,true,true);
|
||||
// PmlMatInv[ip] = new ComplexMUMPSSolver;
|
||||
// PmlMatInv[ip]->SetOperator(*HypreMat);
|
||||
// delete HypreMat;
|
||||
int ndofs = dmaps->fes[ip]->GetTrueVSize();
|
||||
f_transf[ip].SetSize(sweeps->nsweeps);
|
||||
for (int i=0;i<sweeps->nsweeps; i++)
|
||||
{
|
||||
f_transf[ip][i] = new Vector(2*ndofs);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
|
||||
void ParDST::SetHelmholtzPmlSystemMatrix(int ip)
|
||||
{
|
||||
MFEM_VERIFY(part->subdomain_mesh[ip], "Null mesh pointer");
|
||||
Mesh * mesh = part->subdomain_mesh[ip];
|
||||
double h = part->MeshSize;
|
||||
Array2D<double> length(dim,2);
|
||||
length = h*(nrlayers);
|
||||
|
||||
Array<int> ijk;
|
||||
GetSubdomainijk(ip,nxyz,ijk);
|
||||
int i = ijk[0];
|
||||
int j = ijk[1];
|
||||
int k = ijk[2];
|
||||
|
||||
if (i == 0 ) length[0][0] = Pmllength[0][0];
|
||||
if (i == nx-1 ) length[0][1] = Pmllength[0][1];
|
||||
if (dim > 1)
|
||||
{
|
||||
if (j == 0 ) length[1][0] = Pmllength[1][0];
|
||||
if (j == ny-1 ) length[1][1] = Pmllength[1][1];
|
||||
}
|
||||
if (dim == 3)
|
||||
{
|
||||
if (k == 0 ) length[2][0] = Pmllength[2][0];
|
||||
if (k == nz-1 ) length[2][1] = Pmllength[2][1];
|
||||
}
|
||||
|
||||
CartesianPML pml(mesh, length);
|
||||
pml.SetOmega(omega);
|
||||
|
||||
Array <int> ess_tdof_list;
|
||||
if (mesh->bdr_attributes.Size())
|
||||
{
|
||||
Array<int> ess_bdr(mesh->bdr_attributes.Max());
|
||||
ess_bdr = 1;
|
||||
dmaps->fes[ip]->GetEssentialTrueDofs(ess_bdr, ess_tdof_list);
|
||||
}
|
||||
|
||||
ConstantCoefficient one(1.0);
|
||||
ConstantCoefficient sigma(-pow(omega, 2));
|
||||
PmlMatrixCoefficient c1_re(dim,pml_detJ_JT_J_inv_Re,&pml);
|
||||
PmlMatrixCoefficient c1_im(dim,pml_detJ_JT_J_inv_Im,&pml);
|
||||
PmlCoefficient detJ_re(pml_detJ_Re,&pml);
|
||||
PmlCoefficient detJ_im(pml_detJ_Im,&pml);
|
||||
ProductCoefficient c2_re0(sigma, detJ_re);
|
||||
ProductCoefficient c2_im0(sigma, detJ_im);
|
||||
ProductCoefficient c2_re(c2_re0, *Q);
|
||||
ProductCoefficient c2_im(c2_im0, *Q);
|
||||
sqf[ip] = new SesquilinearForm (dmaps->fes[ip],bf->GetConvention());
|
||||
|
||||
sqf[ip]->AddDomainIntegrator(new DiffusionIntegrator(c1_re),
|
||||
new DiffusionIntegrator(c1_im));
|
||||
sqf[ip]->AddDomainIntegrator(new MassIntegrator(c2_re),
|
||||
new MassIntegrator(c2_im));
|
||||
sqf[ip]->Assemble(0);
|
||||
|
||||
Optr[ip] = new OperatorPtr;
|
||||
sqf[ip]->FormSystemMatrix(ess_tdof_list,*Optr[ip]);
|
||||
}
|
||||
|
||||
void ParDST::SetMaxwellPmlSystemMatrix(int ip)
|
||||
{
|
||||
MFEM_VERIFY(part->subdomain_mesh[ip], "Null mesh pointer");
|
||||
Mesh * mesh = part->subdomain_mesh[ip];
|
||||
double h = part->MeshSize;
|
||||
Array2D<double> length(dim,2);
|
||||
length = h*(nrlayers);
|
||||
|
||||
Array<int> ijk;
|
||||
GetSubdomainijk(ip,nxyz,ijk);
|
||||
int i = ijk[0];
|
||||
int j = ijk[1];
|
||||
int k = ijk[2];
|
||||
|
||||
if (i == 0 ) length[0][0] = Pmllength[0][0];
|
||||
if (i == nx-1 ) length[0][1] = Pmllength[0][1];
|
||||
if (dim > 1)
|
||||
{
|
||||
if (j == 0 ) length[1][0] = Pmllength[1][0];
|
||||
if (j == ny-1 ) length[1][1] = Pmllength[1][1];
|
||||
}
|
||||
if (dim == 3)
|
||||
{
|
||||
if (k == 0 ) length[2][0] = Pmllength[2][0];
|
||||
if (k == nz-1 ) length[2][1] = Pmllength[2][1];
|
||||
}
|
||||
|
||||
CartesianPML pml(mesh, length);
|
||||
pml.SetOmega(omega);
|
||||
Array <int> ess_tdof_list;
|
||||
if (mesh->bdr_attributes.Size())
|
||||
{
|
||||
Array<int> ess_bdr(mesh->bdr_attributes.Max());
|
||||
ess_bdr = 1;
|
||||
dmaps->fes[ip]->GetEssentialTrueDofs(ess_bdr, ess_tdof_list);
|
||||
}
|
||||
|
||||
ConstantCoefficient omeg(-pow(omega, 2));
|
||||
int cdim = (dim == 2) ? 1 : dim;
|
||||
|
||||
PmlMatrixCoefficient pml_c1_Re(cdim,detJ_inv_JT_J_Re, &pml);
|
||||
PmlMatrixCoefficient pml_c1_Im(cdim,detJ_inv_JT_J_Im, &pml);
|
||||
|
||||
PmlMatrixCoefficient pml_c2_Re(dim, detJ_JT_J_inv_Re,&pml);
|
||||
PmlMatrixCoefficient pml_c2_Im(dim, detJ_JT_J_inv_Im,&pml);
|
||||
ScalarMatrixProductCoefficient c2_Re0(omeg,pml_c2_Re);
|
||||
ScalarMatrixProductCoefficient c2_Im0(omeg,pml_c2_Im);
|
||||
|
||||
MatrixCoefficient * c2_Re=nullptr;
|
||||
MatrixCoefficient * c2_Im=nullptr;
|
||||
|
||||
if (Q)
|
||||
{
|
||||
c2_Re = new ScalarMatrixProductCoefficient(*Q,c2_Re0);
|
||||
c2_Im = new ScalarMatrixProductCoefficient(*Q,c2_Im0);
|
||||
}
|
||||
else if (VQ)
|
||||
{
|
||||
MFEM_ABORT("Vector Coeffiecient not supported ");
|
||||
}
|
||||
else if (MQ)
|
||||
{
|
||||
c2_Re = new MatrixMatrixProductCoefficient(c2_Re0,*MQ);
|
||||
c2_Im = new MatrixMatrixProductCoefficient(c2_Im0,*MQ);
|
||||
}
|
||||
|
||||
sqf[ip] = new SesquilinearForm(dmaps->fes[ip],bf->GetConvention());
|
||||
|
||||
sqf[ip]->AddDomainIntegrator(new CurlCurlIntegrator(pml_c1_Re),
|
||||
new CurlCurlIntegrator(pml_c1_Im));
|
||||
sqf[ip]->AddDomainIntegrator(new VectorFEMassIntegrator(*c2_Re),
|
||||
new VectorFEMassIntegrator(*c2_Im));
|
||||
sqf[ip]->Assemble(0);
|
||||
|
||||
Optr[ip] = new OperatorPtr;
|
||||
sqf[ip]->FormSystemMatrix(ess_tdof_list,*Optr[ip]);
|
||||
delete c2_Re;
|
||||
delete c2_Im;
|
||||
}
|
||||
|
||||
|
||||
void ParDST::MarkSubdomainOverlapDofs(const bool comp)
|
||||
{
|
||||
// First mark the elements
|
||||
// cout<< "Compute Overlap Elements (in each possible direction) " << endl;
|
||||
// Lists of elements
|
||||
// x,y,z = +/- 1 ovlp
|
||||
NovlpElems.resize(nrsubdomains);
|
||||
|
||||
for (int ip = 0; ip<nrsubdomains; ip++)
|
||||
{
|
||||
if (myid != SubdomainRank[ip]) continue;
|
||||
Array<int> ijk;
|
||||
GetSubdomainijk(ip,nxyz,ijk);
|
||||
|
||||
Mesh * mesh = dmaps->fes[ip]->GetMesh();
|
||||
NovlpElems[ip].resize(2*dim);
|
||||
|
||||
Vector pmin, pmax;
|
||||
mesh->GetBoundingBox(pmin,pmax);
|
||||
double h = part->MeshSize;
|
||||
// Loop through elements
|
||||
for (int iel=0; iel<mesh->GetNE(); iel++)
|
||||
{
|
||||
// Get element center
|
||||
Vector center(dim);
|
||||
int geom = mesh->GetElementBaseGeometry(iel);
|
||||
ElementTransformation * tr = mesh->GetElementTransformation(iel);
|
||||
tr->Transform(Geometries.GetCenter(geom),center);
|
||||
|
||||
// Assign elements to the appropriate lists
|
||||
for (int d=0;d<dim; d++)
|
||||
{
|
||||
if (ijk[d]>0)
|
||||
{
|
||||
if (center[d] >= pmin[d]+h*ovlpnrlayers)
|
||||
{
|
||||
NovlpElems[ip][d].Append(iel);
|
||||
}
|
||||
}
|
||||
else
|
||||
{
|
||||
NovlpElems[ip][d].Append(iel);
|
||||
}
|
||||
|
||||
if (ijk[d]<nxyz[d]-1)
|
||||
{
|
||||
if (center[d] <= pmax[d]-h*ovlpnrlayers)
|
||||
{
|
||||
NovlpElems[ip][dim+d].Append(iel);
|
||||
}
|
||||
}
|
||||
else
|
||||
{
|
||||
NovlpElems[ip][dim+d].Append(iel);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// mark dofs
|
||||
NovlpDofs.resize(nrsubdomains);
|
||||
int mm = (comp) ? 2 : 1; // complex or real valued
|
||||
for (int ip = 0; ip<nrsubdomains; ip++)
|
||||
{
|
||||
if (myid != SubdomainRank[ip]) continue;
|
||||
FiniteElementSpace * fes = dmaps->fes[ip];
|
||||
// Loop through the marked elements
|
||||
NovlpDofs[ip].resize(2*dim);
|
||||
int n = fes->GetTrueVSize();
|
||||
Array<int> marker(n);
|
||||
for (int d=0;d<2*dim; d++)
|
||||
{
|
||||
marker = 0;
|
||||
int m = 0;
|
||||
int melems = NovlpElems[ip][d].Size();
|
||||
for (int iel=0; iel<melems; iel++)
|
||||
{
|
||||
Array<int> ElemDofs;
|
||||
int el = NovlpElems[ip][d][iel];
|
||||
fes->GetElementDofs(el,ElemDofs);
|
||||
int ndof = ElemDofs.Size();
|
||||
for (int i = 0; i<ndof; ++i)
|
||||
{
|
||||
int eldof = ElemDofs[i];
|
||||
int tdof = (eldof >= 0) ? eldof : abs(eldof) - 1;
|
||||
if (marker[tdof] == 1) continue;
|
||||
marker[tdof] = 1;
|
||||
m++;
|
||||
}
|
||||
}
|
||||
int k = mm*(n-m);
|
||||
NovlpDofs[ip][d].SetSize(k);
|
||||
int l = 0;
|
||||
for (int i = 0; i<n; i++)
|
||||
{
|
||||
if (marker[i]==0)
|
||||
{
|
||||
NovlpDofs[ip][d][l] = i; // real dofs
|
||||
if (comp)
|
||||
{
|
||||
NovlpDofs[ip][d][l+k/2] = i+fes->GetTrueVSize();
|
||||
}
|
||||
l++;
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
void ParDST::GetChiRes(Vector & res, int ip, Array2D<int> direct) const
|
||||
{
|
||||
for (int d=0; d<dim; d++)
|
||||
{
|
||||
// negative direction
|
||||
if (direct[d][0]==1) res.SetSubVector(NovlpDofs[ip][d],0.0);
|
||||
// possitive direction
|
||||
if (direct[d][1]==1) res.SetSubVector(NovlpDofs[ip][d+dim],0.0);
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
|
||||
void ParDST::PlotLocal(Vector & sol, socketstream & sol_sock, int ip) const
|
||||
{
|
||||
FiniteElementSpace * fes = dmaps->fes[ip];
|
||||
Mesh * mesh = fes->GetMesh();
|
||||
GridFunction gf(fes);
|
||||
double * data = sol.GetData();
|
||||
gf.SetData(data);
|
||||
|
||||
string keys;
|
||||
keys = "keys mrRljc\n";
|
||||
sol_sock << "solution\n" << *mesh << gf << keys << flush;
|
||||
}
|
||||
|
||||
|
||||
void ParDST::GetStepSubdomains(const int sweep, const int step, Array2D<int> & subdomains) const
|
||||
{
|
||||
Array<int> aux;
|
||||
switch(dim)
|
||||
{
|
||||
case 2:
|
||||
for (int i=nx-1;i>=0; i--)
|
||||
{
|
||||
int j;
|
||||
switch (sweep)
|
||||
{
|
||||
case 0: j = step-i; break;
|
||||
case 1: j = step-nx+i+1; break;
|
||||
case 2: j = nx+i-step-1; break;
|
||||
default: j = nx+ny-i-step-2; break;
|
||||
}
|
||||
if (j<0 || j>=ny) continue;
|
||||
aux.Append(i); aux.Append(j);
|
||||
}
|
||||
break;
|
||||
default:
|
||||
for (int i=nx-1;i>=0; i--)
|
||||
{
|
||||
for (int j=ny-1;j>=0; j--)
|
||||
{
|
||||
int k;
|
||||
switch (sweep)
|
||||
{
|
||||
case 0: k = step-i-j; break;
|
||||
case 1: k = step-nx+i+1-j; break;
|
||||
case 2: k = step-ny+j+1-i; break;
|
||||
case 3: k = step-nx-ny+i+j+2; break;
|
||||
case 4: k = i+j+nz-1-step; break;
|
||||
case 5: k = nx+nz-i+j-step-2; break;
|
||||
case 6: k = ny+nz+i-j-step-2; break;
|
||||
default: k = nx+ny+nz-i-j-step-3; break;
|
||||
}
|
||||
if (k<0 || k>=nz) continue;
|
||||
aux.Append(i); aux.Append(j); aux.Append(k);
|
||||
}
|
||||
}
|
||||
break;
|
||||
}
|
||||
|
||||
int nrows = aux.Size()/dim;
|
||||
int ncols = dim;
|
||||
|
||||
subdomains.SetSize(nrows,ncols);
|
||||
for (int r=0;r<nrows; r++)
|
||||
{
|
||||
for (int c=0; c<ncols; c++)
|
||||
{
|
||||
int k = r*ncols + c;
|
||||
subdomains[r][c] = aux[k];
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
void ParDST::TransferSources(int sweep, const Array<int> & subdomain_ids) const
|
||||
{
|
||||
OvlpSol.resize(nrsubdomains);
|
||||
int nrneighbors = pow(3,dim);
|
||||
for (int ip = 0; ip<nrsubdomains; ip++)
|
||||
{
|
||||
if (myid == SubdomainRank[ip])
|
||||
{
|
||||
OvlpSol[ip].resize(nrneighbors);
|
||||
}
|
||||
}
|
||||
int m = subdomain_ids.Size();
|
||||
Array<Vector *> x(m);
|
||||
for (int i = 0; i<m; i++)
|
||||
{
|
||||
x[i] = nullptr;
|
||||
int ip = subdomain_ids[i];
|
||||
if (myid != SubdomainRank[ip]) continue;
|
||||
x[i] = new Vector(subdomain_sol[ip]->GetData(),subdomain_sol[ip]->Size());
|
||||
}
|
||||
dmaps->TransferToNeighbors(subdomain_ids,x,OvlpSol);
|
||||
for (int i = 0; i<m; i++)
|
||||
{
|
||||
delete x[i]; x[i] = nullptr;
|
||||
}
|
||||
// Update residuals
|
||||
// Find all neighbors of patch ip0
|
||||
for (int is = 0; is<m; is++)
|
||||
{
|
||||
int ip0 = subdomain_ids[is];
|
||||
Array<int> ijk;
|
||||
Array<int> ijk1(3);
|
||||
GetSubdomainijk(ip0,nxyz,ijk);
|
||||
Array<int> directions(3);
|
||||
for (int i=-1; i<2; i++)
|
||||
{
|
||||
int i1 = ijk[0] + i;
|
||||
if (i1 <0 || i1>=nx) continue;
|
||||
directions[0] = i;
|
||||
ijk1[0] = i1;
|
||||
for (int j=-1; j<2; j++)
|
||||
{
|
||||
int j1 = ijk[1] + j;
|
||||
if (j1 <0 || j1>=ny) continue;
|
||||
directions[1] = j;
|
||||
ijk1[1] = j1;
|
||||
int kbeg = (dim == 2) ? 0 : -1;
|
||||
int kend = (dim == 2) ? 1 : 2;
|
||||
for (int k=kbeg; k<kend; k++)
|
||||
{
|
||||
int k1 = ijk[2] + k;
|
||||
if (k1 <0 || k1>=nz) continue;
|
||||
directions[2] = (dim == 3) ? k : -1 ;
|
||||
if (i==0 && j==0 && k==0) continue;
|
||||
|
||||
int l = GetSweepToTransfer(sweep,directions);
|
||||
if (l == -1) continue;
|
||||
ijk1[2] = k1;
|
||||
int ip1 = GetSubdomainId(nxyz,ijk1);
|
||||
|
||||
if (myid != SubdomainRank[ip1]) continue;
|
||||
Array<int>directions1(3); directions1 = -1;
|
||||
for (int i = 0; i<dim; i++) directions1[i] = -directions[i];
|
||||
int dir = GetDirectionId(directions1);
|
||||
int n = dmaps->fes[ip1]->GetTrueVSize();
|
||||
Vector res(2*n);
|
||||
PmlMat[ip1]->Mult(*OvlpSol[ip1][dir],res);
|
||||
|
||||
Array2D<int> direct(dim,2); direct = 0;
|
||||
for (int d = 0; d<dim; d++)
|
||||
{
|
||||
if (directions[d]==1) direct[d][0] = 1;
|
||||
if (directions[d]==-1) direct[d][1] = 1;
|
||||
}
|
||||
GetChiRes(res,ip1,direct);
|
||||
*f_transf[ip1][l] -= res;
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
for (int ip = 0; ip<nrsubdomains; ip++)
|
||||
{
|
||||
if (myid == SubdomainRank[ip])
|
||||
{
|
||||
for (int i = 0; i<nrneighbors; i++)
|
||||
{
|
||||
if (OvlpSol[ip][i])
|
||||
{
|
||||
delete OvlpSol[ip][i];
|
||||
}
|
||||
}
|
||||
OvlpSol[ip].clear();
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
int ParDST::GetSweepToTransfer(const int s, Array<int> directions) const
|
||||
{
|
||||
int l1=-1;
|
||||
int nsweeps = sweeps->nsweeps;
|
||||
Array<int> sweep0;
|
||||
sweeps->GetSweep(s,sweep0);
|
||||
switch (dim)
|
||||
{
|
||||
case 2:
|
||||
for (int l=s; l<nsweeps; l++)
|
||||
{
|
||||
// Rule 1: the transfer source direction has to be similar with
|
||||
// the sweep direction
|
||||
Array<int> sweep1;
|
||||
sweeps->GetSweep(l,sweep1);
|
||||
int ddot = 0;
|
||||
for (int d=0; d<dim; d++) ddot+= sweep1[d] * directions[d];
|
||||
if (ddot <= 0) continue;
|
||||
|
||||
// Rule 2: The horizontal or vertical transfer source cannot be used
|
||||
// Case of horizontal or vertical transfer source
|
||||
// (it can't be both 0 cause it's skipped)
|
||||
if (directions[0]==0 || directions[1] == 0)
|
||||
{
|
||||
if (sweep0[0] == -sweep1[0] && sweep0[1] == -sweep1[1]) continue;
|
||||
}
|
||||
l1 = l;
|
||||
break;
|
||||
}
|
||||
break;
|
||||
default:
|
||||
for (int l=s; l<nsweeps; l++)
|
||||
{
|
||||
// Rule 1: (similar directions) the transfer source direction has to be similar with
|
||||
// the sweep direction
|
||||
Array<int> sweep1;
|
||||
sweeps->GetSweep(l,sweep1);
|
||||
int ddot = 0;
|
||||
bool similar = true;
|
||||
for (int d=0; d<dim; d++)
|
||||
{
|
||||
if (sweep1[d] * directions[d] < 0) similar = false;
|
||||
ddot+= sweep1[d] * directions[d];
|
||||
}
|
||||
if (!similar || ddot<=0) continue; // not similar
|
||||
|
||||
// Rule 2: (oposite directions) the transfer source direction has to be similar with
|
||||
// the sweep direction
|
||||
//
|
||||
// check any of the projections onto the planes
|
||||
// (xy, xz, yz)
|
||||
|
||||
if ( (directions[0]==0 && directions[1] != 0) ||
|
||||
(directions[0]!=0 && directions[1] == 0) ||
|
||||
(directions[0]==0 && directions[2] != 0) ||
|
||||
(directions[0]!=0 && directions[2] == 0) ||
|
||||
(directions[2]==0 && directions[1] != 0) ||
|
||||
(directions[2]!=0 && directions[1] == 0) )
|
||||
{
|
||||
if (sweep0[0] == -sweep1[0] &&
|
||||
sweep0[1] == -sweep1[1] &&
|
||||
sweep0[2] == -sweep1[2]) continue;
|
||||
}
|
||||
l1 = l;
|
||||
break;
|
||||
}
|
||||
break;
|
||||
}
|
||||
|
||||
return l1;
|
||||
}
|
||||
|
||||
void ParDST::CorrectOrientation(int ip,Vector &x) const
|
||||
{
|
||||
FiniteElementSpace * fespace = dmaps->fes[ip];
|
||||
Mesh * mesh = fespace->GetMesh();
|
||||
int nrelems = mesh->GetNE();
|
||||
// GridFunction test;
|
||||
// test.SetFromTrueDofs(x)
|
||||
Array<int> signs(fespace->GetTrueVSize()); signs = 0;
|
||||
for (int iel=0; iel<nrelems; iel++)
|
||||
{
|
||||
Array<int> ElemDofs;
|
||||
fespace->GetElementDofs(iel,ElemDofs);
|
||||
int ndofs = ElemDofs.Size();
|
||||
ElemDofs.Print();
|
||||
for (int i = 0; i< ndofs; i++)
|
||||
{
|
||||
int pdof_ = ElemDofs[i];
|
||||
if (pdof_ < 0)
|
||||
{
|
||||
signs[abs(pdof_)-1] += 1.0 ;
|
||||
}
|
||||
else
|
||||
{
|
||||
signs[pdof_] -= 1.0 ;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
cout << "signs = " ; signs.Print();
|
||||
for (int i = 0; i<fespace->GetTrueVSize(); i++)
|
||||
{
|
||||
if (signs[i]<0)
|
||||
{
|
||||
x(i) *= -1.0;
|
||||
x(i+fespace->GetTrueVSize()) *= -1.0;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
ParDST::~ParDST()
|
||||
{
|
||||
|
||||
for (int ip=0; ip<nrsubdomains; ip++)
|
||||
{
|
||||
delete Optr[ip];
|
||||
delete subdomain_sol[ip];
|
||||
delete PmlMatInv[ip];
|
||||
delete sqf[ip];
|
||||
if (myid != SubdomainRank[ip]) continue;
|
||||
for (int i=0;i<sweeps->nsweeps; i++)
|
||||
{
|
||||
delete f_transf[ip][i];
|
||||
}
|
||||
delete f_orig[ip];
|
||||
}
|
||||
f_orig.DeleteAll();
|
||||
delete dmaps;
|
||||
delete sweeps;
|
||||
delete part;
|
||||
|
||||
}
|
||||
@@ -1,70 +0,0 @@
|
||||
#pragma once
|
||||
#include "../common/Utilities.hpp"
|
||||
#include "../common/PML.hpp"
|
||||
#include "DofMapsDST.hpp"
|
||||
using namespace std;
|
||||
using namespace mfem;
|
||||
|
||||
|
||||
class ParDST : public Solver//
|
||||
{
|
||||
private:
|
||||
MPI_Comm comm = MPI_COMM_WORLD;
|
||||
int num_procs, myid;
|
||||
// Constructor inputs
|
||||
int prob_kind;
|
||||
ParSesquilinearForm *bf=nullptr;
|
||||
ParFiniteElementSpace * pfes = nullptr;
|
||||
ParMesh * pmesh = nullptr;
|
||||
ParMeshPartition * part = nullptr;
|
||||
Array<int> SubdomainRank;
|
||||
Array<int> RankSubdomains;
|
||||
const FiniteElementCollection * fec = nullptr;
|
||||
Array2D<double> Pmllength;
|
||||
int dim = 2;
|
||||
double omega = 0.5;
|
||||
Coefficient * Q=nullptr;
|
||||
VectorCoefficient * VQ=nullptr;
|
||||
MatrixCoefficient * MQ=nullptr;
|
||||
int nrlayers;
|
||||
int ovlpnrlayers;
|
||||
int nrsubdomains = 0;
|
||||
int nx,ny,nz;
|
||||
Array<int> nxyz;
|
||||
Sweep * sweeps = nullptr;
|
||||
DofMaps * dmaps = nullptr;
|
||||
Array< SesquilinearForm * > sqf;
|
||||
Array< OperatorPtr * > Optr;
|
||||
Array<ComplexSparseMatrix *> PmlMat;
|
||||
Array<ComplexUMFPackSolver *> PmlMatInv;
|
||||
// Array<ComplexMUMPSSolver *> PmlMatInv;
|
||||
mutable Array<Vector *> f_orig;
|
||||
mutable Array<Array<Vector * >> f_transf;
|
||||
mutable Array<Vector * > subdomain_sol;
|
||||
mutable std::vector<std::vector<Vector * >> OvlpSol;
|
||||
void SetupSubdomainProblems();
|
||||
std::vector<std::vector<Array<int>>> NovlpElems;
|
||||
std::vector<std::vector<Array<int>>> NovlpDofs;
|
||||
void MarkSubdomainOverlapDofs(const bool comp = false);
|
||||
void SetHelmholtzPmlSystemMatrix(int ip);
|
||||
void SetMaxwellPmlSystemMatrix(int ip);
|
||||
void GetChiRes(Vector & res, int ip, Array2D<int> direct) const;
|
||||
void PlotLocal(Vector & sol, socketstream & sol_sock, int ip) const;
|
||||
void GetStepSubdomains(const int sweep, const int step, Array2D<int> & subdomains) const;
|
||||
void TransferSources(int sweep, const Array<int> & subdomain_ids) const;
|
||||
int GetSweepToTransfer(const int s, Array<int> directions) const;
|
||||
void CorrectOrientation(int ip, Vector & x) const;
|
||||
void Init();
|
||||
public:
|
||||
ParDST(ParSesquilinearForm * bf_, Array2D<double> & Pmllength_,
|
||||
double omega_, Coefficient * Q_, int nrlayers_, int nx_=2, int ny_=2, int nz_=2);
|
||||
ParDST(ParSesquilinearForm * bf_, Array2D<double> & Pmllength_,
|
||||
double omega_, VectorCoefficient * VQ_, int nrlayers_, int nx_=2, int ny_=2, int nz_=2);
|
||||
ParDST(ParSesquilinearForm * bf_, Array2D<double> & Pmllength_,
|
||||
double omega_, MatrixCoefficient * MQ_, int nrlayers_, int nx_=2, int ny_=2, int nz_=2);
|
||||
virtual void SetOperator(const Operator &op) {}
|
||||
virtual void Mult(const Vector &r, Vector &z) const;
|
||||
virtual ~ParDST();
|
||||
};
|
||||
|
||||
|
||||
File diff suppressed because it is too large
Load Diff
@@ -1,151 +0,0 @@
|
||||
#pragma once
|
||||
#include "mfem.hpp"
|
||||
#include <fstream>
|
||||
#include <iostream>
|
||||
using namespace std;
|
||||
using namespace mfem;
|
||||
|
||||
|
||||
struct UniqueIndexGenerator
|
||||
{
|
||||
int counter = 0;
|
||||
std::unordered_map<int,int> idx;
|
||||
int Get(int i)
|
||||
{
|
||||
std::unordered_map<int,int>::iterator f = idx.find(i);
|
||||
if (f == idx.end())
|
||||
{
|
||||
idx[i] = counter;
|
||||
return counter++;
|
||||
}
|
||||
else
|
||||
{
|
||||
return (*f).second;
|
||||
}
|
||||
}
|
||||
void Reset()
|
||||
{
|
||||
counter = 0;
|
||||
idx.clear();
|
||||
}
|
||||
};
|
||||
|
||||
|
||||
double GetUniformMeshElementSize(Mesh * mesh);
|
||||
Mesh * ExtendMesh(Mesh * mesh, const Array<int> & directions);
|
||||
|
||||
class CartesianMeshPartition
|
||||
{
|
||||
private:
|
||||
Mesh *mesh=nullptr;
|
||||
public:
|
||||
int nrpatch;
|
||||
int nxyz[3];
|
||||
double MeshSize;
|
||||
std::vector<Array<int>> element_map;
|
||||
Array3D<int>subdomains;
|
||||
// constructor
|
||||
CartesianMeshPartition(Mesh * mesh_,int & nx, int & ny, int & nz);
|
||||
~CartesianMeshPartition() {};
|
||||
};
|
||||
|
||||
class OverlappingCartesianMeshPartition
|
||||
{
|
||||
private:
|
||||
Mesh *mesh=nullptr;
|
||||
public:
|
||||
int nrpatch;
|
||||
double MeshSize;
|
||||
int nxyz[3];
|
||||
std::vector<Array<int>> element_map;
|
||||
Array3D<int> subdomains;
|
||||
// constructor
|
||||
OverlappingCartesianMeshPartition(Mesh * mesh_,int & nx, int & ny, int & nz);
|
||||
OverlappingCartesianMeshPartition(Mesh * mesh_,int & nx, int & ny, int & nz, int ovlp_nlayers);
|
||||
~OverlappingCartesianMeshPartition() {};
|
||||
};
|
||||
|
||||
class STPOverlappingCartesianMeshPartition // Special layered partition for STP
|
||||
{
|
||||
private:
|
||||
Mesh *mesh=nullptr;
|
||||
public:
|
||||
int nrpatch;
|
||||
int nx, ny, nz;
|
||||
std::vector<Array<int>> element_map;
|
||||
// constructor
|
||||
STPOverlappingCartesianMeshPartition(Mesh * mesh_);
|
||||
~STPOverlappingCartesianMeshPartition() {};
|
||||
};
|
||||
|
||||
class MeshPartition
|
||||
{
|
||||
private:
|
||||
Mesh *mesh=nullptr;
|
||||
void AddElementToMesh(Mesh * mesh,mfem::Element::Type elem_type,int * ind);
|
||||
void GetNumVertices(int type, mfem::Element::Type & elem_type, int & nrvert);
|
||||
void PrintElementMap();
|
||||
public:
|
||||
int nrpatch;
|
||||
double MeshSize;
|
||||
std::vector<Array<int>> element_map;
|
||||
Array3D<int> subdomains;
|
||||
Array<Mesh *> patch_mesh;
|
||||
int partition_kind;
|
||||
int nxyz[3];
|
||||
// constructor
|
||||
MeshPartition(Mesh * mesh_, int part, int mx=1, int my=1, int mz=1, int ovl_nlayers=0);
|
||||
~MeshPartition();
|
||||
};
|
||||
|
||||
void SaveMeshPartition(Array<Mesh * > meshes,
|
||||
string mfilename="output/mesh.",
|
||||
string sfilename="output/sol.");
|
||||
|
||||
|
||||
#ifdef MFEM_USE_MPI
|
||||
|
||||
class CartesianParMeshPartition
|
||||
{
|
||||
private:
|
||||
ParMesh *pmesh=nullptr;
|
||||
public:
|
||||
int nrsubdomains;
|
||||
int nxyz[3];
|
||||
double MeshSize;
|
||||
std::vector<Array<int>> local_element_map;
|
||||
Array<int> subdomain_rank;
|
||||
Array3D<int>subdomains;
|
||||
// constructor
|
||||
CartesianParMeshPartition(ParMesh * pmesh_,int & nx, int & ny, int & nz,
|
||||
int ovlp_nlayers);
|
||||
~CartesianParMeshPartition() {};
|
||||
};
|
||||
|
||||
class ParMeshPartition
|
||||
{
|
||||
private:
|
||||
MPI_Comm comm;
|
||||
ParMesh *pmesh=nullptr;
|
||||
void AddElementToMesh(Mesh * mesh,mfem::Element::Type elem_type,int * ind);
|
||||
void GetNumVertices(int type, mfem::Element::Type & elem_type, int & nrvert);
|
||||
void PrintElementMap();
|
||||
public:
|
||||
int nrsubdomains;
|
||||
int OvlpNlayers;
|
||||
int myelem_offset = 0;
|
||||
double MeshSize;
|
||||
std::vector<Array<int>> element_map;
|
||||
std::vector<Array<int>> local_element_map;
|
||||
Array3D<int> subdomains;
|
||||
Array<Mesh *> subdomain_mesh;
|
||||
Array<int> subdomain_rank;
|
||||
int partition_kind;
|
||||
int nxyz[3];
|
||||
// constructor
|
||||
ParMeshPartition(ParMesh * pmesh_, int mx=1, int my=1, int mz=1, int ovl_nlayers=0);
|
||||
void SaveMeshPartition();
|
||||
~ParMeshPartition();
|
||||
};
|
||||
|
||||
#endif
|
||||
@@ -1,570 +0,0 @@
|
||||
#include "PML.hpp"
|
||||
|
||||
CartesianPML::CartesianPML(Mesh *mesh_, Array2D<double> length_)
|
||||
: mesh(mesh_), length(length_)
|
||||
{
|
||||
dim = mesh->Dimension();
|
||||
SetBoundaries();
|
||||
}
|
||||
|
||||
void CartesianPML::SetBoundaries()
|
||||
{
|
||||
comp_dom_bdr.SetSize(dim, 2);
|
||||
dom_bdr.SetSize(dim, 2);
|
||||
// initialize
|
||||
for (int i = 0; i < dim; i++)
|
||||
{
|
||||
dom_bdr(i, 0) = infinity();
|
||||
dom_bdr(i, 1) = -infinity();
|
||||
}
|
||||
|
||||
for (int i = 0; i < mesh->GetNBE(); i++)
|
||||
{
|
||||
Array<int> bdr_vertices;
|
||||
mesh->GetBdrElementVertices(i, bdr_vertices);
|
||||
for (int j = 0; j < bdr_vertices.Size(); j++)
|
||||
{
|
||||
for (int k = 0; k < dim; k++)
|
||||
{
|
||||
dom_bdr(k, 0) = min(dom_bdr(k, 0), mesh->GetVertex(bdr_vertices[j])[k]);
|
||||
dom_bdr(k, 1) = max(dom_bdr(k, 1), mesh->GetVertex(bdr_vertices[j])[k]);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
#ifdef MFEM_USE_MPI
|
||||
ParMesh * pmesh = dynamic_cast<ParMesh *>(mesh);
|
||||
if (pmesh)
|
||||
{
|
||||
for (int d=0; d<dim; d++)
|
||||
{
|
||||
MPI_Allreduce(MPI_IN_PLACE,&dom_bdr(d,0),1,MPI_DOUBLE,MPI_MIN,pmesh->GetComm());
|
||||
MPI_Allreduce(MPI_IN_PLACE,&dom_bdr(d,1),1,MPI_DOUBLE,MPI_MAX,pmesh->GetComm());
|
||||
}
|
||||
}
|
||||
#endif
|
||||
|
||||
for (int i = 0; i < dim; i++)
|
||||
{
|
||||
comp_dom_bdr(i, 0) = dom_bdr(i, 0) + length(i, 0);
|
||||
comp_dom_bdr(i, 1) = dom_bdr(i, 1) - length(i, 1);
|
||||
}
|
||||
}
|
||||
|
||||
void CartesianPML::SetAttributes(Mesh *mesh_)
|
||||
{
|
||||
int nrelem = mesh_->GetNE();
|
||||
elems.SetSize(nrelem);
|
||||
|
||||
for (int i = 0; i < nrelem; ++i)
|
||||
{
|
||||
elems[i] = 1;
|
||||
bool in_pml = false;
|
||||
Element *el = mesh_->GetElement(i);
|
||||
Array<int> vertices;
|
||||
// Initialize Attribute
|
||||
el->SetAttribute(1);
|
||||
el->GetVertices(vertices);
|
||||
int nrvert = vertices.Size();
|
||||
// Check if any vertex is in the pml
|
||||
for (int iv = 0; iv < nrvert; ++iv)
|
||||
{
|
||||
int vert_idx = vertices[iv];
|
||||
double *coords = mesh_->GetVertex(vert_idx);
|
||||
for (int comp = 0; comp < dim; ++comp)
|
||||
{
|
||||
if (coords[comp] > comp_dom_bdr(comp, 1) ||
|
||||
coords[comp] < comp_dom_bdr(comp, 0))
|
||||
{
|
||||
in_pml = true;
|
||||
break;
|
||||
}
|
||||
}
|
||||
}
|
||||
if (in_pml)
|
||||
{
|
||||
elems[i] = 0;
|
||||
el->SetAttribute(2);
|
||||
}
|
||||
}
|
||||
mesh_->SetAttributes();
|
||||
}
|
||||
|
||||
void CartesianPML::StretchFunction(const Vector &x,
|
||||
vector<complex<double>> &dxs, double omega)
|
||||
{
|
||||
complex<double> zi = complex<double>(0., 1.);
|
||||
|
||||
double n = 2.0;
|
||||
double c = 10.0;
|
||||
// double c = log(omega);
|
||||
double coeff;
|
||||
// Stretch in each direction independently
|
||||
for (int i = 0; i < dim; ++i)
|
||||
{
|
||||
dxs[i] = 1.0;
|
||||
if (x(i) >= comp_dom_bdr(i, 1))
|
||||
{
|
||||
coeff = n * c / omega / pow(length(i, 1), n);
|
||||
dxs[i] = 1.0 + zi * coeff * abs(pow(x(i) - comp_dom_bdr(i, 1), n - 1.0));
|
||||
}
|
||||
if (x(i) <= comp_dom_bdr(i, 0))
|
||||
{
|
||||
coeff = n * c / omega / pow(length(i, 0), n);
|
||||
dxs[i] = 1.0 + zi * coeff * abs(pow(x(i) - comp_dom_bdr(i, 0), n - 1.0));
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
ToroidPML::ToroidPML(Mesh *mesh_)
|
||||
: mesh(mesh_)
|
||||
{
|
||||
dim = mesh->Dimension();
|
||||
zlim.SetSize(2);
|
||||
rlim.SetSize(2);
|
||||
alim.SetSize(2);
|
||||
zpml_thickness.SetSize(2);
|
||||
rpml_thickness.SetSize(2);
|
||||
apml_thickness.SetSize(2);
|
||||
SetBoundaries();
|
||||
}
|
||||
|
||||
void ToroidPML::SetBoundaries()
|
||||
{
|
||||
mesh->EnsureNodes();
|
||||
int nrnodes = mesh->GetNodalFESpace()->GetTrueVSize()/dim;
|
||||
double zmin = infinity();
|
||||
double zmax = -infinity();
|
||||
double rmin = infinity();
|
||||
double rmax = -infinity();
|
||||
double amin = infinity(); // in degrees
|
||||
double amax = -infinity(); // in degrees
|
||||
for (int i = 0; i<nrnodes; i++)
|
||||
{
|
||||
Vector coord(dim);
|
||||
mesh->GetNode(i,coord);
|
||||
for (int d = 0; d<dim; d++)
|
||||
{
|
||||
if (abs(coord[d])<1e-13) coord[d] = 0.0;
|
||||
}
|
||||
// Find r and a for this point
|
||||
double x = coord[0];
|
||||
double y = coord[1];
|
||||
double z = 0.0;
|
||||
if (dim == 3) z = coord[2];
|
||||
double a = GetAngle(x,y);
|
||||
double r = sqrt(x*x + y*y);
|
||||
|
||||
zmin = min(zmin,z);
|
||||
zmax = max(zmax,z);
|
||||
rmin = min(rmin,r);
|
||||
rmax = max(rmax,r);
|
||||
amin = min(amin,a);
|
||||
amax = max(amax,a);
|
||||
}
|
||||
|
||||
zlim[0] = zmin;
|
||||
zlim[1] = zmax;
|
||||
rlim[0] = rmin;
|
||||
rlim[1] = rmax;
|
||||
alim[0] = amin;
|
||||
alim[1] = amax;
|
||||
|
||||
|
||||
#ifdef MFEM_USE_MPI
|
||||
ParMesh * pmesh = dynamic_cast<ParMesh *>(mesh);
|
||||
if (pmesh)
|
||||
{
|
||||
MPI_Allreduce(MPI_IN_PLACE,&zlim[0],1,MPI_DOUBLE,MPI_MIN,pmesh->GetComm());
|
||||
MPI_Allreduce(MPI_IN_PLACE,&zlim[1],1,MPI_DOUBLE,MPI_MAX,pmesh->GetComm());
|
||||
MPI_Allreduce(MPI_IN_PLACE,&rlim[0],1,MPI_DOUBLE,MPI_MIN,pmesh->GetComm());
|
||||
MPI_Allreduce(MPI_IN_PLACE,&rlim[1],1,MPI_DOUBLE,MPI_MAX,pmesh->GetComm());
|
||||
MPI_Allreduce(MPI_IN_PLACE,&alim[0],1,MPI_DOUBLE,MPI_MIN,pmesh->GetComm());
|
||||
MPI_Allreduce(MPI_IN_PLACE,&alim[1],1,MPI_DOUBLE,MPI_MAX,pmesh->GetComm());
|
||||
}
|
||||
#endif
|
||||
}
|
||||
|
||||
void ToroidPML::SetAttributes(Mesh *mesh_)
|
||||
{
|
||||
int nrelem = mesh_->GetNE();
|
||||
elems.SetSize(nrelem);
|
||||
|
||||
// Loop through the elements and identify which of them are in the PML
|
||||
for (int i = 0; i < nrelem; ++i)
|
||||
{
|
||||
// initialize with 1
|
||||
elems[i] = 1;
|
||||
Element *el = mesh_->GetElement(i);
|
||||
// Initialize attribute
|
||||
el->SetAttribute(1);
|
||||
|
||||
Array<int> vertices;
|
||||
el->GetVertices(vertices);
|
||||
int nrvert = vertices.Size();
|
||||
// Check if any vertex is in the pml
|
||||
bool in_pml = false;
|
||||
for (int iv = 0; iv < nrvert; ++iv)
|
||||
{
|
||||
int vert_idx = vertices[iv];
|
||||
double *coords = mesh_->GetVertex(vert_idx);
|
||||
double x = coords[0];
|
||||
double y = coords[1];
|
||||
double a = GetAngle(x,y);
|
||||
double r = sqrt(x*x + y*y);
|
||||
|
||||
if (astretch)
|
||||
{
|
||||
if ( (a <= alim[0]+apml_thickness[0]) ||
|
||||
(a >= alim[1]-apml_thickness[1]) )
|
||||
{
|
||||
in_pml = true;
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
if (rstretch)
|
||||
{
|
||||
if ( (r <= rlim[0]+rpml_thickness[0]) ||
|
||||
(r >= rlim[1]-rpml_thickness[1]) )
|
||||
{
|
||||
in_pml = true;
|
||||
break;
|
||||
}
|
||||
}
|
||||
}
|
||||
if (in_pml)
|
||||
{
|
||||
elems[i] = 0;
|
||||
el->SetAttribute(2);
|
||||
}
|
||||
|
||||
// Vector center;
|
||||
// mesh_->GetElementCenter(i,center);
|
||||
// double x = center[0];
|
||||
// double y = center[1];
|
||||
// double a = GetAngle(x,y);
|
||||
// double r = sqrt(x*x + y*y);
|
||||
// // check upper and lower bound
|
||||
// if (astretch)
|
||||
// {
|
||||
// if ( (a <= alim[0]+apml_thickness[0]) ||
|
||||
// (a >= alim[1]-apml_thickness[1]) )
|
||||
// {
|
||||
// elems[i] = 0;
|
||||
// el->SetAttribute(2);
|
||||
// }
|
||||
// }
|
||||
// if (rstretch)
|
||||
// {
|
||||
// if ( (r <= rlim[0]+rpml_thickness[0]) ||
|
||||
// (r >= rlim[1]-rpml_thickness[1]) )
|
||||
// {
|
||||
// elems[i] = 0;
|
||||
// el->SetAttribute(2);
|
||||
// }
|
||||
// }
|
||||
}
|
||||
mesh_->SetAttributes();
|
||||
}
|
||||
|
||||
|
||||
double ToroidPML::GetAngle(const double x, const double y)
|
||||
{
|
||||
// Find r and a for this point
|
||||
double arad;
|
||||
if (x == 0.0)
|
||||
{
|
||||
arad = (y > 0.0)? M_PI/2.0 : 3.0 * M_PI/2.0;
|
||||
}
|
||||
else
|
||||
{
|
||||
arad = atan(y/x);
|
||||
int k = 0;
|
||||
if (x<0)
|
||||
{
|
||||
k = 1;
|
||||
}
|
||||
else if (y<0)
|
||||
{
|
||||
k = 2;
|
||||
}
|
||||
arad += k*M_PI;
|
||||
}
|
||||
return arad * 180.0/M_PI;
|
||||
}
|
||||
|
||||
// void ToroidPML::StretchFunction(const Vector &X,
|
||||
// vector<complex<double>> &dxs, double omega)
|
||||
void ToroidPML::StretchFunction(const Vector &X, ComplexDenseMatrix & J, double omega)
|
||||
|
||||
{
|
||||
complex<double> zi = complex<double>(0., 1.);
|
||||
|
||||
double n = 2.0;
|
||||
double c = 10.0;
|
||||
// double c = log(omega);
|
||||
// Stretch in the azimuthal direction
|
||||
double x = X[0];
|
||||
double y = X[1];
|
||||
if (abs(x) < 1e-12) x = 0.0;
|
||||
if (abs(y) < 1e-12) y = 0.0;
|
||||
double a = GetAngle(x,y);
|
||||
double r = sqrt(x*x + y*y);
|
||||
// dxs[0] = 1.0;
|
||||
// dxs[1] = 1.0;
|
||||
J = 0.0;
|
||||
J(0,0) = 1.0;
|
||||
J(1,1) = 1.0;
|
||||
if (dim == 3) J(2,2) = 1.0;
|
||||
|
||||
if (astretch)
|
||||
{
|
||||
double th = a * M_PI/180.0;
|
||||
double thl, thL, thH;
|
||||
bool in_pml = false;
|
||||
// negative direction
|
||||
if (a <= alim[0]+apml_thickness[0])
|
||||
{
|
||||
in_pml = true;
|
||||
thL = alim[1] * M_PI/180.0;
|
||||
thH = apml_thickness[1] * M_PI/180.0;
|
||||
thl = thL + thH;
|
||||
}
|
||||
// positive direction
|
||||
if (a >= alim[1]-apml_thickness[1])
|
||||
{
|
||||
in_pml = true;
|
||||
thL = alim[1] * M_PI/180.0;
|
||||
thH = apml_thickness[1] * M_PI/180.0;
|
||||
thl = thL - thH;
|
||||
}
|
||||
// double c1 = min(20.0*M_PI/180.0,thH);
|
||||
if (in_pml)
|
||||
{
|
||||
double c1 = thH;
|
||||
double coeff = n * c / omega / pow(c1,n);
|
||||
double f_th = pow(th - thl,n-1);
|
||||
double th_x = - y / (r * r);
|
||||
double th_y = x / (r * r);
|
||||
|
||||
J(0,0) = 1.0 + zi * coeff * abs(f_th * th_x);
|
||||
J(0,1) = zi * f_th * th_y;
|
||||
J(1,0) = zi * f_th * th_x;
|
||||
J(1,1) = 1.0 + zi * coeff * abs(f_th * th_y);
|
||||
}
|
||||
}
|
||||
// Stretch in the radial direction
|
||||
if (rstretch)
|
||||
{ // negative
|
||||
double rl, rL, rH;
|
||||
bool in_pml = false;
|
||||
if (r <= rlim[0]+rpml_thickness[0])
|
||||
{
|
||||
in_pml = true;
|
||||
rL = rlim[0];
|
||||
rH = rpml_thickness[0];
|
||||
rl = rL + rH;
|
||||
}
|
||||
// positive direction
|
||||
if (r >= rlim[1]-rpml_thickness[1])
|
||||
{
|
||||
in_pml = true;
|
||||
rL = rlim[1];
|
||||
rH = rpml_thickness[1];
|
||||
rl = rL - rH;
|
||||
}
|
||||
|
||||
if (in_pml)
|
||||
{
|
||||
double coeff = n * c / omega / pow (rH,n);
|
||||
double f_r = pow(r-rl,n-1.0);
|
||||
double r_x = x / r;
|
||||
double r_y = y / r;
|
||||
|
||||
J(0,0) = 1.0 + zi * coeff * abs(f_r*r_x);
|
||||
J(0,1) = zi * f_r * r_y;
|
||||
J(1,0) = zi * f_r * r_x;
|
||||
J(1,1) = 1.0 + zi * coeff * abs(f_r*r_y);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
double pml_detJ_Re(const Vector & x, CartesianPML * pml)
|
||||
{
|
||||
int dim = pml->dim;
|
||||
double omega = pml->omega;
|
||||
std::vector<std::complex<double>> dxs(dim);
|
||||
complex<double> det(1.0,0.0);
|
||||
pml->StretchFunction(x, dxs, omega);
|
||||
for (int i=0; i<dim; ++i) det *= dxs[i];
|
||||
return det.real();
|
||||
}
|
||||
|
||||
double pml_detJ_Im(const Vector & x, CartesianPML * pml)
|
||||
{
|
||||
int dim = pml->dim;
|
||||
double omega = pml->omega;
|
||||
std::vector<std::complex<double>> dxs(dim);
|
||||
complex<double> det(1.0,0.0);
|
||||
pml->StretchFunction(x, dxs, omega);
|
||||
for (int i=0; i<dim; ++i) det *= dxs[i];
|
||||
return det.imag();
|
||||
}
|
||||
|
||||
void pml_detJ_JT_J_inv_Re(const Vector & x, CartesianPML * pml , DenseMatrix & M)
|
||||
{
|
||||
int dim = pml->dim;
|
||||
double omega = pml->omega;
|
||||
std::vector<std::complex<double>> dxs(dim);
|
||||
complex<double> det(1.0,0.0);
|
||||
pml->StretchFunction(x, dxs, omega);
|
||||
|
||||
for (int i = 0; i<dim; ++i)
|
||||
{
|
||||
det *= dxs[i];
|
||||
}
|
||||
|
||||
M=0.0;
|
||||
for (int i = 0; i<dim; ++i)
|
||||
{
|
||||
M(i,i) = (det / pow(dxs[i],2)).real();
|
||||
}
|
||||
}
|
||||
|
||||
void pml_detJ_JT_J_inv_Im(const Vector & x, CartesianPML * pml , DenseMatrix & M)
|
||||
{
|
||||
int dim = pml->dim;
|
||||
double omega = pml->omega;
|
||||
|
||||
std::vector<std::complex<double>> dxs(dim);
|
||||
complex<double> det = 1.0;
|
||||
pml->StretchFunction(x, dxs, omega);
|
||||
|
||||
for (int i = 0; i<dim; ++i)
|
||||
{
|
||||
det *= dxs[i];
|
||||
}
|
||||
|
||||
M=0.0;
|
||||
for (int i = 0; i<dim; ++i)
|
||||
{
|
||||
M(i,i) = (det / pow(dxs[i],2)).imag();
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
void detJ_JT_J_inv_Re(const Vector &x, CartesianPML * pml, DenseMatrix &M)
|
||||
{
|
||||
int dim = pml->dim;
|
||||
double omega = pml->omega;
|
||||
vector<complex<double>> dxs(dim);
|
||||
complex<double> det(1.0, 0.0);
|
||||
pml->StretchFunction(x, dxs, omega);
|
||||
|
||||
for (int i = 0; i < dim; ++i)
|
||||
{
|
||||
det *= dxs[i];
|
||||
}
|
||||
|
||||
M = 0.0;
|
||||
for (int i = 0; i < dim; ++i)
|
||||
{
|
||||
M(i, i) = (det / pow(dxs[i], 2)).real();
|
||||
}
|
||||
}
|
||||
|
||||
void detJ_JT_J_inv_Im(const Vector &x, CartesianPML * pml, DenseMatrix &M)
|
||||
{
|
||||
int dim = pml->dim;
|
||||
double omega = pml->omega;
|
||||
vector<complex<double>> dxs(dim);
|
||||
complex<double> det = 1.0;
|
||||
pml->StretchFunction(x, dxs, omega);
|
||||
|
||||
for (int i = 0; i < dim; ++i)
|
||||
{
|
||||
det *= dxs[i];
|
||||
}
|
||||
|
||||
M = 0.0;
|
||||
for (int i = 0; i < dim; ++i)
|
||||
{
|
||||
M(i, i) = (det / pow(dxs[i], 2)).imag();
|
||||
}
|
||||
}
|
||||
|
||||
void detJ_JT_J_inv_abs(const Vector &x, CartesianPML * pml, DenseMatrix &M)
|
||||
{
|
||||
int dim = pml->dim;
|
||||
double omega = pml->omega;
|
||||
vector<complex<double>> dxs(dim);
|
||||
complex<double> det = 1.0;
|
||||
pml->StretchFunction(x, dxs, omega);
|
||||
|
||||
for (int i = 0; i < dim; ++i)
|
||||
{
|
||||
det *= dxs[i];
|
||||
}
|
||||
|
||||
M = 0.0;
|
||||
for (int i = 0; i < dim; ++i)
|
||||
{
|
||||
M(i, i) = abs(det / pow(dxs[i], 2));
|
||||
}
|
||||
}
|
||||
|
||||
void detJ_inv_JT_J_Re(const Vector &x, CartesianPML * pml, DenseMatrix &M)
|
||||
{
|
||||
int dim = pml->dim;
|
||||
double omega = pml->omega;
|
||||
vector<complex<double>> dxs(dim);
|
||||
complex<double> det(1.0, 0.0);
|
||||
pml->StretchFunction(x, dxs, omega);
|
||||
|
||||
for (int i = 0; i < dim; ++i)
|
||||
{
|
||||
det *= dxs[i];
|
||||
}
|
||||
|
||||
// in the 2D case the coefficient is scalar 1/det(J)
|
||||
if (dim == 2)
|
||||
{
|
||||
M = (1.0 / det).real();
|
||||
}
|
||||
else
|
||||
{
|
||||
M = 0.0;
|
||||
for (int i = 0; i < dim; ++i)
|
||||
{
|
||||
M(i, i) = (pow(dxs[i], 2) / det).real();
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
void detJ_inv_JT_J_Im(const Vector &x, CartesianPML * pml, DenseMatrix &M)
|
||||
{
|
||||
int dim = pml->dim;
|
||||
double omega = pml->omega;
|
||||
vector<complex<double>> dxs(dim);
|
||||
complex<double> det = 1.0;
|
||||
pml->StretchFunction(x, dxs, omega);
|
||||
|
||||
for (int i = 0; i < dim; ++i)
|
||||
{
|
||||
det *= dxs[i];
|
||||
}
|
||||
|
||||
if (dim == 2)
|
||||
{
|
||||
M = (1.0 / det).imag();
|
||||
}
|
||||
else
|
||||
{
|
||||
M = 0.0;
|
||||
for (int i = 0; i < dim; ++i)
|
||||
{
|
||||
M(i, i) = (pow(dxs[i], 2) / det).imag();
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -1,179 +0,0 @@
|
||||
#pragma once
|
||||
#include "mfem.hpp"
|
||||
#include "complex_linalg.hpp"
|
||||
#include <fstream>
|
||||
#include <iostream>
|
||||
using namespace std;
|
||||
using namespace mfem;
|
||||
|
||||
|
||||
// Class for setting up a simple Cartesian PML region
|
||||
class CartesianPML
|
||||
{
|
||||
private:
|
||||
Mesh *mesh;
|
||||
|
||||
// Length of the PML Region in each direction
|
||||
Array2D<double> length;
|
||||
|
||||
// Computational Domain Boundary
|
||||
Array2D<double> comp_dom_bdr;
|
||||
|
||||
// Domain Boundary
|
||||
Array2D<double> dom_bdr;
|
||||
|
||||
// Integer Array identifying elements in the pml
|
||||
// 0: in the pml, 1: not in the pml
|
||||
Array<int> elems;
|
||||
|
||||
// Compute Domain and Computational Domain Boundaries
|
||||
void SetBoundaries();
|
||||
|
||||
public:
|
||||
// Constructor
|
||||
CartesianPML(Mesh *mesh_,Array2D<double> length_);
|
||||
|
||||
int dim;
|
||||
double omega;
|
||||
// Return Computational Domain Boundary
|
||||
Array2D<double> GetCompDomainBdr() {return comp_dom_bdr;}
|
||||
|
||||
// Return Domain Boundary
|
||||
Array2D<double> GetDomainBdr() {return dom_bdr;}
|
||||
|
||||
// Return Marker list for elements
|
||||
Array<int> * GetMarkedPMLElements() {return &elems;}
|
||||
|
||||
// Mark element in the PML region
|
||||
void SetAttributes(Mesh *mesh_);
|
||||
|
||||
void SetOmega(double omega_) {omega = omega_;}
|
||||
|
||||
// PML complex stretching function
|
||||
void StretchFunction(const Vector &x, vector<complex<double>> &dxs, double omega);
|
||||
};
|
||||
|
||||
class ToroidPML
|
||||
{
|
||||
private:
|
||||
Mesh *mesh;
|
||||
|
||||
Vector zlim, zpml_thickness; // range in axial direction
|
||||
Vector rlim, rpml_thickness; // range in radial direction
|
||||
Vector alim, apml_thickness; // range in azimuthal direction
|
||||
|
||||
// Integer Array identifying elements in the pml
|
||||
// 0: in the pml, 1: not in the pml
|
||||
Array<int> elems;
|
||||
|
||||
double GetAngle(const double x, const double y);
|
||||
|
||||
// Compute Domain and Computational Domain Boundaries
|
||||
void SetBoundaries();
|
||||
|
||||
bool zstretch = false;
|
||||
bool rstretch = false;
|
||||
bool astretch = false;
|
||||
|
||||
public:
|
||||
// Constructor
|
||||
ToroidPML(Mesh *mesh_);
|
||||
|
||||
int dim;
|
||||
double omega;
|
||||
// Return Computational Domain Boundary
|
||||
|
||||
// Return Domain Boundary
|
||||
void GetDomainBdrs(Vector & zlim_, Vector & rlim_, Vector & alim_)
|
||||
{
|
||||
zlim_.SetSize(2); zlim_ = zlim;
|
||||
rlim_.SetSize(2); rlim_ = rlim;
|
||||
alim_.SetSize(2); alim_ = alim;
|
||||
}
|
||||
|
||||
void SetPmlWidth(const Vector & zpml, const Vector & rpml, const Vector & apml)
|
||||
{
|
||||
MFEM_VERIFY(zpml.Size() == 2 , "Check zpml size");
|
||||
MFEM_VERIFY(rpml.Size() == 2 , "Check rpml size");
|
||||
MFEM_VERIFY(apml.Size() == 2 , "Check apml size");
|
||||
zpml_thickness = zpml;
|
||||
rpml_thickness = rpml;
|
||||
apml_thickness = apml;
|
||||
}
|
||||
|
||||
void SetPmlAxes(const bool zstretch_,
|
||||
const bool rstretch_,
|
||||
const bool astretch_ )
|
||||
{
|
||||
zstretch = zstretch_;
|
||||
rstretch = rstretch_;
|
||||
astretch = astretch_;
|
||||
}
|
||||
|
||||
// // Return Marker list for elements
|
||||
Array<int> * GetMarkedPMLElements() {return &elems;}
|
||||
|
||||
// Mark element in the PML region
|
||||
void SetAttributes(Mesh *mesh_);
|
||||
|
||||
void SetOmega(double omega_) {omega = omega_;}
|
||||
|
||||
// PML complex stretching function
|
||||
// void StretchFunction(const Vector &X, vector<complex<double>> &dxs, double omega);
|
||||
void StretchFunction(const Vector &X, ComplexDenseMatrix & J, double omega);
|
||||
};
|
||||
|
||||
class PmlCoefficient : public Coefficient
|
||||
{
|
||||
private:
|
||||
CartesianPML * pml = nullptr;
|
||||
double (*Function)(const Vector &, CartesianPML * );
|
||||
public:
|
||||
PmlCoefficient(double (*F)(const Vector &, CartesianPML *), CartesianPML * pml_)
|
||||
: pml(pml_), Function(F)
|
||||
{}
|
||||
virtual double Eval(ElementTransformation &T, const IntegrationPoint &ip)
|
||||
{
|
||||
double x[3];
|
||||
Vector transip(x, 3);
|
||||
T.Transform(ip, transip);
|
||||
return ((*Function)(transip, pml));
|
||||
}
|
||||
};
|
||||
|
||||
|
||||
// This includes scalar coefficients
|
||||
class PmlMatrixCoefficient : public MatrixCoefficient
|
||||
{
|
||||
private:
|
||||
CartesianPML * pml = nullptr;
|
||||
void (*Function)(const Vector &, CartesianPML * , DenseMatrix &);
|
||||
public:
|
||||
PmlMatrixCoefficient(int dim, void(*F)(const Vector &, CartesianPML *,
|
||||
DenseMatrix &),
|
||||
CartesianPML * pml_)
|
||||
: MatrixCoefficient(dim), pml(pml_), Function(F)
|
||||
{}
|
||||
virtual void Eval(DenseMatrix &K, ElementTransformation &T,
|
||||
const IntegrationPoint &ip)
|
||||
{
|
||||
double x[3];
|
||||
Vector transip(x, 3);
|
||||
T.Transform(ip, transip);
|
||||
K.SetSize(height, width);
|
||||
(*Function)(transip, pml, K);
|
||||
}
|
||||
};
|
||||
|
||||
// Helmholtz pml Functions
|
||||
double pml_detJ_Re(const Vector & x, CartesianPML * pml);
|
||||
double pml_detJ_Im(const Vector & x, CartesianPML * pml);
|
||||
void pml_detJ_JT_J_inv_Re(const Vector & x, CartesianPML * pml , DenseMatrix & M);
|
||||
void pml_detJ_JT_J_inv_Im(const Vector & x, CartesianPML * pml , DenseMatrix & M);
|
||||
|
||||
// Maxwell Pml functions
|
||||
void detJ_JT_J_inv_Re(const Vector &x, CartesianPML * pml, DenseMatrix &M);
|
||||
void detJ_JT_J_inv_Im(const Vector &x, CartesianPML * pml, DenseMatrix &M);
|
||||
void detJ_JT_J_inv_abs(const Vector &x, CartesianPML * pml, DenseMatrix &M);
|
||||
void detJ_inv_JT_J_Re(const Vector &x, CartesianPML * pml, DenseMatrix &M);
|
||||
void detJ_inv_JT_J_Im(const Vector &x, CartesianPML * pml, DenseMatrix &M);
|
||||
@@ -1,619 +0,0 @@
|
||||
#include "Utilities.hpp"
|
||||
|
||||
Sweep::Sweep(int dim_) : dim(dim_)
|
||||
{
|
||||
nsweeps = pow(2,dim);
|
||||
sweeps.resize(nsweeps);
|
||||
|
||||
for (int is = 0; is<nsweeps; is++)
|
||||
{
|
||||
sweeps[is].SetSize(dim);
|
||||
}
|
||||
|
||||
switch(dim)
|
||||
{
|
||||
case 1:
|
||||
sweeps[0][0] = 1;
|
||||
sweeps[1][0] = -1;
|
||||
break;
|
||||
case 2:
|
||||
sweeps[0][0] = 1; sweeps[0][1] = 1;
|
||||
sweeps[1][0] = -1; sweeps[1][1] = 1;
|
||||
sweeps[2][0] = 1; sweeps[2][1] = -1;
|
||||
sweeps[3][0] = -1; sweeps[3][1] = -1;
|
||||
break;
|
||||
default:
|
||||
sweeps[0][0] = 1; sweeps[0][1] = 1; sweeps[0][2] = 1;
|
||||
sweeps[1][0] = -1; sweeps[1][1] = 1; sweeps[1][2] = 1;
|
||||
sweeps[2][0] = 1; sweeps[2][1] = -1; sweeps[2][2] = 1;
|
||||
sweeps[3][0] = -1; sweeps[3][1] = -1; sweeps[3][2] = 1;
|
||||
sweeps[4][0] = 1; sweeps[4][1] = 1; sweeps[4][2] = -1;
|
||||
sweeps[5][0] = -1; sweeps[5][1] = 1; sweeps[5][2] = -1;
|
||||
sweeps[6][0] = 1; sweeps[6][1] = -1; sweeps[6][2] = -1;
|
||||
sweeps[7][0] = -1; sweeps[7][1] = -1; sweeps[7][2] = -1;
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
Sweep::~Sweep()
|
||||
{
|
||||
for (int i = 0; i<nsweeps; i++)
|
||||
{
|
||||
sweeps[i].DeleteAll();
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
double CutOffFncn(const Vector &x, const Vector & pmin, const Vector & pmax, const Array2D<double> & h_)
|
||||
{
|
||||
int dim = pmin.Size();
|
||||
Vector h0(dim);
|
||||
Vector h1(dim);
|
||||
for (int i=0; i<dim; i++)
|
||||
{
|
||||
h0(i) = h_[i][0];
|
||||
h1(i) = h_[i][1];
|
||||
}
|
||||
Vector x0(dim);
|
||||
Vector x1(dim);
|
||||
x0 = pmin; x0+=h0;
|
||||
x1 = pmax; x1-=h1;
|
||||
|
||||
double f = 1.0;
|
||||
for (int i = 0; i<dim; i++)
|
||||
{
|
||||
double val = 1.0;
|
||||
if( x(i) >= pmax(i) || x(i) <= pmin(i))
|
||||
{
|
||||
val = 0.0;
|
||||
}
|
||||
else if (x(i) < pmax(i) && x(i) >= x1(i))
|
||||
{
|
||||
if(h1(i) != 0.0)
|
||||
// val = (x(i)-pmax(i))/(x1(i)-pmax(i));
|
||||
val = pow((x(i)-pmax(i))/(x1(i)-pmax(i)),1.0);
|
||||
}
|
||||
else if (x(i) > pmin(i) && x(i) <= x0(i))
|
||||
{
|
||||
if (h0(i) != 0.0)
|
||||
// val = (x(i)-pmin(i))/(x0(i)-pmin(i));
|
||||
val = pow((x(i)-pmin(i))/(x0(i)-pmin(i)),1.0);
|
||||
}
|
||||
|
||||
if (h0(i) == 0 && x(i) <= x1(i))
|
||||
{
|
||||
val = 1.0;
|
||||
}
|
||||
if (h1(i) == 0 && x(i) >= x0(i))
|
||||
{
|
||||
val = 1.0;
|
||||
}
|
||||
f *= val;
|
||||
}
|
||||
return f;
|
||||
}
|
||||
|
||||
double ChiFncn(const Vector &x, const Vector & pmin, const Vector & pmax, const Array2D<double> & h_)
|
||||
{
|
||||
int dim = pmin.Size();
|
||||
Vector h0(dim);
|
||||
Vector h1(dim);
|
||||
for (int i=0; i<dim; i++)
|
||||
{
|
||||
h0(i) = h_[i][0];
|
||||
h1(i) = h_[i][1];
|
||||
}
|
||||
Vector x0(dim);
|
||||
Vector x1(dim);
|
||||
x0 = pmin; x0+=h0;
|
||||
x1 = pmax; x1-=h1;
|
||||
|
||||
double f = 1.0;
|
||||
for (int i = 0; i<dim; i++)
|
||||
{
|
||||
double val = 1.0;
|
||||
if( x(i) >= pmax(i) || x(i) <= pmin(i))
|
||||
{
|
||||
val = 0.0;
|
||||
}
|
||||
else if (x(i) < pmax(i) && x(i) >= x1(i))
|
||||
{
|
||||
if(h1(i) != 0.0)
|
||||
val = (x(i)-pmax(i))/(x1(i)-pmax(i));
|
||||
// This function has to be changed to smth more reasonable
|
||||
// val = pow((x(i)-pmax(i))/(x1(i)-pmax(i)),100.0);
|
||||
}
|
||||
else if (x(i) > pmin(i) && x(i) <= x0(i))
|
||||
{
|
||||
if (h0(i) != 0.0)
|
||||
val = (x(i)-pmin(i))/(x0(i)-pmin(i));
|
||||
// val = pow((x(i)-pmin(i))/(x0(i)-pmin(i)),100.0);
|
||||
}
|
||||
|
||||
if (h0(i) == 0 && x(i) <= x1(i))
|
||||
{
|
||||
val = 1.0;
|
||||
}
|
||||
if (h1(i) == 0 && x(i) >= x0(i))
|
||||
{
|
||||
val = 1.0;
|
||||
}
|
||||
f *= val;
|
||||
}
|
||||
return f;
|
||||
}
|
||||
|
||||
|
||||
DofMap::DofMap(FiniteElementSpace * fes , MeshPartition * partition)
|
||||
{
|
||||
const FiniteElementCollection * fec = fes->FEColl();
|
||||
nrpatch = partition->nrpatch;
|
||||
|
||||
fespaces.SetSize(nrpatch);
|
||||
|
||||
Dof2GlobalDof.resize(nrpatch);
|
||||
|
||||
for (int ip=0; ip<nrpatch; ++ip)
|
||||
{
|
||||
// create finite element spaces for each patch
|
||||
fespaces[ip] = new FiniteElementSpace(partition->patch_mesh[ip],fec);
|
||||
|
||||
// construct the patch tdof to global tdof map
|
||||
int nrdof = fespaces[ip]->GetTrueVSize();
|
||||
Dof2GlobalDof[ip].SetSize(2*nrdof);
|
||||
|
||||
// loop through the elements in the patch
|
||||
for (int iel = 0; iel<partition->element_map[ip].Size(); ++iel)
|
||||
{
|
||||
// index in the global mesh
|
||||
int iel_idx = partition->element_map[ip][iel];
|
||||
// get the dofs of this element
|
||||
Array<int> ElemDofs;
|
||||
Array<int> GlobalElemDofs;
|
||||
fespaces[ip]->GetElementDofs(iel,ElemDofs);
|
||||
fes->GetElementDofs(iel_idx,GlobalElemDofs);
|
||||
// the sizes have to match
|
||||
MFEM_VERIFY(ElemDofs.Size() == GlobalElemDofs.Size(),
|
||||
"Size inconsistency");
|
||||
// loop through the dofs and take into account the signs;
|
||||
int ndof = ElemDofs.Size();
|
||||
for (int i = 0; i<ndof; ++i)
|
||||
{
|
||||
int pdof_ = ElemDofs[i];
|
||||
int gdof_ = GlobalElemDofs[i];
|
||||
int pdof = (pdof_ >= 0) ? pdof_ : abs(pdof_) - 1;
|
||||
int gdof = (gdof_ >= 0) ? gdof_ : abs(gdof_) - 1;
|
||||
Dof2GlobalDof[ip][pdof] = gdof;
|
||||
Dof2GlobalDof[ip][pdof+nrdof] = gdof+fes->GetTrueVSize();
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
DofMap::DofMap(FiniteElementSpace * fes , MeshPartition * partition, int nrlayers)
|
||||
{
|
||||
|
||||
nx = partition->nxyz[0];
|
||||
ny = partition->nxyz[1];
|
||||
nz = partition->nxyz[2];
|
||||
|
||||
int partition_kind = partition->partition_kind;
|
||||
// Mesh * mesh = fespace->GetMesh();
|
||||
const FiniteElementCollection * fec = fes->FEColl();
|
||||
nrpatch = partition->nrpatch;
|
||||
|
||||
fespaces.SetSize(nrpatch);
|
||||
PmlMeshes.SetSize(nrpatch);
|
||||
// Extend patch meshes to include pml
|
||||
|
||||
for (int ip = 0; ip<nrpatch; ip++)
|
||||
{
|
||||
int k = ip/(nx*ny);
|
||||
int j = (ip-k*nx*ny)/nx;
|
||||
int i = (ip-k*nx*ny)%nx;
|
||||
|
||||
Array<int> directions;
|
||||
if (i > 0)
|
||||
{
|
||||
for (int i=0; i<nrlayers; i++)
|
||||
{
|
||||
directions.Append(-1);
|
||||
}
|
||||
}
|
||||
if (j > 0)
|
||||
{
|
||||
for (int i=0; i<nrlayers; i++)
|
||||
{
|
||||
directions.Append(-2);
|
||||
}
|
||||
}
|
||||
if (k > 0)
|
||||
{
|
||||
for (int i=0; i<nrlayers; i++)
|
||||
{
|
||||
directions.Append(-3);
|
||||
}
|
||||
}
|
||||
if (i < nx-1)
|
||||
{
|
||||
for (int i=0; i<nrlayers; i++)
|
||||
{
|
||||
if (partition_kind == 3 || partition_kind == 2) directions.Append(1);
|
||||
}
|
||||
}
|
||||
if (j < ny-1)
|
||||
{
|
||||
for (int i=0; i<nrlayers; i++)
|
||||
{
|
||||
if (partition_kind == 3 || partition_kind == 2) directions.Append(2);
|
||||
}
|
||||
}
|
||||
if (k < nz-1)
|
||||
{
|
||||
for (int i=0; i<nrlayers; i++)
|
||||
{
|
||||
if (partition_kind == 3 || partition_kind == 2) directions.Append(1);
|
||||
}
|
||||
}
|
||||
PmlMeshes[ip] = ExtendMesh(partition->patch_mesh[ip],directions);
|
||||
}
|
||||
|
||||
// Save PML_meshes
|
||||
string meshpath;
|
||||
string solpath;
|
||||
if (partition_kind == 3 || partition_kind == 2)
|
||||
{
|
||||
meshpath = "output/mesh_ovlp_pml.";
|
||||
solpath = "output/sol_ovlp_pml.";
|
||||
}
|
||||
else if (partition_kind == 4)
|
||||
{
|
||||
meshpath = "output/mesh_novlp_pml.";
|
||||
solpath = "output/sol_novlp_pml.";
|
||||
}
|
||||
else
|
||||
{
|
||||
MFEM_ABORT("This partition kind not supported yet");
|
||||
}
|
||||
|
||||
// SaveMeshPartition(PmlMeshes, meshpath, solpath);
|
||||
|
||||
PmlFespaces.SetSize(nrpatch);
|
||||
Dof2GlobalDof.resize(nrpatch);
|
||||
Dof2PmlDof.resize(nrpatch);
|
||||
|
||||
for (int ip=0; ip<nrpatch; ++ip)
|
||||
{
|
||||
// create finite element spaces for each patch
|
||||
fespaces[ip] = new FiniteElementSpace(partition->patch_mesh[ip],fec);
|
||||
PmlFespaces[ip] = new FiniteElementSpace(PmlMeshes[ip],fec);
|
||||
|
||||
// construct the patch tdof to global tdof map
|
||||
int nrdof = fespaces[ip]->GetTrueVSize();
|
||||
Dof2GlobalDof[ip].SetSize(2*nrdof);
|
||||
Dof2PmlDof[ip].SetSize(2*nrdof);
|
||||
|
||||
// build dof maps between patch and extended patch
|
||||
// loop through the patch elements and constract the dof map
|
||||
// The same elements in the extended mesh have the same ordering (but not the dofs)
|
||||
|
||||
// loop through the elements in the patch
|
||||
for (int iel = 0; iel<partition->element_map[ip].Size(); ++iel)
|
||||
{
|
||||
// index in the global mesh
|
||||
int iel_idx = partition->element_map[ip][iel];
|
||||
// get the dofs of this element
|
||||
Array<int> ElemDofs;
|
||||
Array<int> PmlElemDofs;
|
||||
Array<int> GlobalElemDofs;
|
||||
fespaces[ip]->GetElementDofs(iel,ElemDofs);
|
||||
PmlFespaces[ip]->GetElementDofs(iel,PmlElemDofs);
|
||||
fes->GetElementDofs(iel_idx,GlobalElemDofs);
|
||||
// the sizes have to match
|
||||
MFEM_VERIFY(ElemDofs.Size() == GlobalElemDofs.Size(),
|
||||
"Size inconsistency");
|
||||
MFEM_VERIFY(ElemDofs.Size() == PmlElemDofs.Size(),
|
||||
"Size inconsistency");
|
||||
// loop through the dofs and take into account the signs;
|
||||
int ndof = ElemDofs.Size();
|
||||
for (int i = 0; i<ndof; ++i)
|
||||
{
|
||||
int pdof_ = ElemDofs[i];
|
||||
int gdof_ = GlobalElemDofs[i];
|
||||
int pmldof_ = PmlElemDofs[i];
|
||||
int pdof = (pdof_ >= 0) ? pdof_ : abs(pdof_) - 1;
|
||||
int gdof = (gdof_ >= 0) ? gdof_ : abs(gdof_) - 1;
|
||||
int pmldof = (pmldof_ >= 0) ? pmldof_ : abs(pmldof_) - 1;
|
||||
|
||||
Dof2GlobalDof[ip][pdof] = gdof;
|
||||
Dof2GlobalDof[ip][pdof+nrdof] = gdof+fes->GetTrueVSize();
|
||||
Dof2PmlDof[ip][pdof] = pmldof;
|
||||
Dof2PmlDof[ip][pdof+nrdof] = pmldof+PmlFespaces[ip]->GetTrueVSize();
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
LocalDofMap::LocalDofMap(const FiniteElementCollection * fec_, MeshPartition * part1_,
|
||||
MeshPartition * part2_):fec(fec_), part1(part1_), part2(part2_)
|
||||
{
|
||||
// Each overlapping patch has 2 non-overlapping subdomains
|
||||
// Thre are n non-overlapping and and n-1 overlapping subdomains
|
||||
int nrpatch = part2->nrpatch;
|
||||
MFEM_VERIFY(part1->nrpatch-1 == part2->nrpatch, "Check number of subdomains");
|
||||
|
||||
cout << "Constructing local dof maps" << endl;
|
||||
map1.resize(nrpatch);
|
||||
map2.resize(nrpatch);
|
||||
for (int ip=0; ip<nrpatch; ip++)
|
||||
{
|
||||
// Get the 3 meshes involved
|
||||
Mesh * mesh = part2->patch_mesh[ip];
|
||||
Mesh * mesh1 = part1->patch_mesh[ip];
|
||||
Mesh * mesh2 = part1->patch_mesh[ip+1];
|
||||
|
||||
// Define the fespaces
|
||||
FiniteElementSpace fespace(mesh, fec);
|
||||
FiniteElementSpace fespace1(mesh1, fec);
|
||||
FiniteElementSpace fespace2(mesh2, fec);
|
||||
|
||||
int ndof1 = fespace1.GetTrueVSize();
|
||||
int ndof2 = fespace2.GetTrueVSize();
|
||||
|
||||
map1[ip].SetSize(2*ndof1); // times 2 because it's complex
|
||||
map2[ip].SetSize(2*ndof2); // times 2 because it's complex
|
||||
|
||||
// loop through the elements in the patches
|
||||
// map 1 is constructed by the first half of elements
|
||||
// map 2 is constructed by the second half of elements
|
||||
|
||||
for (int iel = 0; iel<part1->element_map[ip].Size(); ++iel)
|
||||
{
|
||||
// index in the overlapping mesh
|
||||
int iel_idx = iel;
|
||||
Array<int> ElemDofs;
|
||||
Array<int> GlobalElemDofs;
|
||||
fespace1.GetElementDofs(iel,ElemDofs);
|
||||
fespace.GetElementDofs(iel_idx,GlobalElemDofs);
|
||||
// the sizes have to match
|
||||
MFEM_VERIFY(ElemDofs.Size() == GlobalElemDofs.Size(),
|
||||
"Size inconsistency");
|
||||
// loop through the dofs and take into account the signs;
|
||||
int ndof = ElemDofs.Size();
|
||||
for (int i = 0; i<ndof; ++i)
|
||||
{
|
||||
int pdof_ = ElemDofs[i];
|
||||
int gdof_ = GlobalElemDofs[i];
|
||||
int pdof = (pdof_ >= 0) ? pdof_ : abs(pdof_) - 1;
|
||||
int gdof = (gdof_ >= 0) ? gdof_ : abs(gdof_) - 1;
|
||||
map1[ip][pdof] = gdof;
|
||||
map1[ip][pdof+ndof1] = gdof+fespace.GetTrueVSize();
|
||||
}
|
||||
}
|
||||
for (int iel = 0; iel<part1->element_map[ip+1].Size(); ++iel)
|
||||
{
|
||||
// index in the overlapping mesh
|
||||
int k = part1->element_map[ip].Size();
|
||||
int iel_idx = iel+k;
|
||||
Array<int> ElemDofs;
|
||||
Array<int> GlobalElemDofs;
|
||||
fespace2.GetElementDofs(iel,ElemDofs);
|
||||
fespace.GetElementDofs(iel_idx,GlobalElemDofs);
|
||||
// the sizes have to match
|
||||
MFEM_VERIFY(ElemDofs.Size() == GlobalElemDofs.Size(),
|
||||
"Size inconsistency");
|
||||
// loop through the dofs and take into account the signs;
|
||||
int ndof = ElemDofs.Size();
|
||||
for (int i = 0; i<ndof; ++i)
|
||||
{
|
||||
int pdof_ = ElemDofs[i];
|
||||
int gdof_ = GlobalElemDofs[i];
|
||||
int pdof = (pdof_ >= 0) ? pdof_ : abs(pdof_) - 1;
|
||||
int gdof = (gdof_ >= 0) ? gdof_ : abs(gdof_) - 1;
|
||||
map2[ip][pdof] = gdof;
|
||||
map2[ip][pdof+ndof2] = gdof+fespace.GetTrueVSize();
|
||||
}
|
||||
}
|
||||
}
|
||||
};
|
||||
|
||||
|
||||
NeighborDofMaps::NeighborDofMaps(MeshPartition * part_, FiniteElementSpace * fes_,
|
||||
DofMap * dmap_,
|
||||
int ovlp_layers_) : part(part_), fes(fes_),
|
||||
dmap(dmap_),
|
||||
ovlp_layers(ovlp_layers_)
|
||||
{
|
||||
|
||||
nrsubdomains = part->nrpatch;
|
||||
nxyz.SetSize(3);
|
||||
mesh = fes->GetMesh();
|
||||
dim = mesh->Dimension();
|
||||
for (int d=0; d<3; d++) nxyz[d] = part->nxyz[d];
|
||||
MarkOvlpElements();
|
||||
ComputeNeighborDofMaps();
|
||||
}
|
||||
|
||||
void NeighborDofMaps::MarkOvlpElements()
|
||||
{
|
||||
// Lists of elements
|
||||
// x,y,z = +/- 1 ovlp
|
||||
OvlpElems.resize(nrsubdomains);
|
||||
|
||||
for (int ip = 0; ip<nrsubdomains; ip++)
|
||||
{
|
||||
int i0,j0,k0;
|
||||
Getijk(ip,i0,j0,k0);
|
||||
int ijk[dim]; ijk[0] = i0; ijk[1]=j0;
|
||||
if (dim==3) ijk[2] = k0;
|
||||
|
||||
FiniteElementSpace * sub_fes = dmap->fespaces[ip];
|
||||
Mesh * sub_mesh = sub_fes->GetMesh();
|
||||
// OvlpElems[ip].resize(2*dim);
|
||||
OvlpElems[ip].resize(pow(3,dim));
|
||||
|
||||
Vector pmin, pmax;
|
||||
sub_mesh->GetBoundingBox(pmin,pmax);
|
||||
double h = part->MeshSize;
|
||||
// Loop through elements
|
||||
for (int iel=0; iel<sub_mesh->GetNE(); iel++)
|
||||
{
|
||||
// Get element center
|
||||
Vector center(dim);
|
||||
int geom = sub_mesh->GetElementBaseGeometry(iel);
|
||||
ElementTransformation * tr = sub_mesh->GetElementTransformation(iel);
|
||||
tr->Transform(Geometries.GetCenter(geom),center);
|
||||
|
||||
// loop through dimensions
|
||||
Array<bool> pos(dim); pos = 0;
|
||||
Array<bool> neg(dim); neg = 0;
|
||||
|
||||
for (int d=0;d<dim; d++)
|
||||
{
|
||||
if (ijk[d]>0 && center[d] < pmin[d]+2.0*h*ovlp_layers)
|
||||
{
|
||||
neg[d] = true;
|
||||
}
|
||||
|
||||
if (ijk[d]<nxyz[d]-1 && center[d] > pmax[d]-2.0*h*ovlp_layers)
|
||||
{
|
||||
pos[d] = true;
|
||||
}
|
||||
}
|
||||
SetElementToOverlap(ip,iel,neg,pos);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
void NeighborDofMaps::ComputeNeighborDofMaps()
|
||||
{
|
||||
OvlpDofMaps.resize(nrsubdomains);
|
||||
|
||||
// Array<UniqueIndexGen * > Gen(nrsubdomains);
|
||||
// // construct unique number generator for the elements of a patch
|
||||
// for (int ip = 0; ip<nrsubdomains; ip++)
|
||||
// {
|
||||
// Gen[ip] = new UniqueIndexGen;
|
||||
// // register the elements
|
||||
// int nel = part->element_map[ip].Size();
|
||||
// for (int iel=0; iel<nel; iel++)
|
||||
// {
|
||||
// int iel_idx = part->element_map[ip][iel];
|
||||
// Gen[ip]->Set(iel_idx);
|
||||
// }
|
||||
// }
|
||||
|
||||
// construct dof maps
|
||||
int nrneighbors = pow(3,dim); // including its self
|
||||
|
||||
for (int ip0 = 0; ip0<nrsubdomains; ip0++)
|
||||
{
|
||||
OvlpDofMaps[ip0].resize(nrneighbors);
|
||||
|
||||
FiniteElementSpace * fes0 = dmap->fespaces[ip0];
|
||||
int tdofs0 = fes0->GetTrueVSize();
|
||||
Array<int> marker0(tdofs0); marker0 = 0;
|
||||
int i0, j0, k0;
|
||||
Array<int> ijk(dim);
|
||||
Getijk(ip0, i0,j0,k0);
|
||||
|
||||
int kbeg = (dim == 2) ? 0 : -1;
|
||||
int kend = (dim == 2) ? 1 : 2;
|
||||
for (int k=kbeg; k<kend; k++)
|
||||
{
|
||||
int k1 = k0 + k;
|
||||
if (k1 <0 || k1>=nxyz[2]) continue;
|
||||
int kk = (dim == 2) ? -1 : k;
|
||||
for (int j=-1; j<2; j++)
|
||||
{
|
||||
int j1 = j0 + j;
|
||||
if (j1 <0 || j1>=nxyz[1]) continue;
|
||||
for (int i=-1; i<2; i++)
|
||||
{
|
||||
int i1 = i0 + i;
|
||||
if (i1 <0 || i1>=nxyz[0]) continue;
|
||||
|
||||
Array<int> ip0list; marker0 = 0;
|
||||
int directionId = GetDirectionId(i,j,kk);
|
||||
|
||||
Array<int> Elems = OvlpElems[ip0][directionId];
|
||||
int nel = Elems.Size();
|
||||
|
||||
for (int iel = 0; iel<nel; ++iel)
|
||||
{
|
||||
int iel0 = Elems[iel];
|
||||
Array<int> ElemDofs0;
|
||||
|
||||
fes0->GetElementDofs(iel0,ElemDofs0);
|
||||
int ndof = ElemDofs0.Size();
|
||||
// since the elements are added to the subdomain meshes
|
||||
// in the same ordered fashion (as they come from the
|
||||
// original mesh) then the ordering of elements in each
|
||||
// subdomain is the same. Hence the dof ovlp lists
|
||||
// can be computed for each subdomain independendly
|
||||
for (int l = 0; l<ndof; ++l)
|
||||
{
|
||||
int dof0_ = ElemDofs0[l];
|
||||
int dof0 = (dof0_ >= 0) ? dof0_ : abs(dof0_) - 1;
|
||||
if (!marker0[dof0])
|
||||
{
|
||||
ip0list.Append(dof0); // dofs of ip0 in ovlp
|
||||
marker0[dof0] = 1;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
OvlpDofMaps[ip0][directionId].Append(ip0list);
|
||||
int tsize = fes0->GetTrueVSize();
|
||||
// Imaginary part
|
||||
for (int l=0;l<ip0list.Size(); l++) { ip0list[l] += tsize; }
|
||||
OvlpDofMaps[ip0][directionId].Append(ip0list);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
void NeighborDofMaps::GetNeighborDofMap(const int ip,
|
||||
const Array<int> & directions,
|
||||
Array<int> & dofmap)
|
||||
{
|
||||
int k = (dim == 2) ? -1 : directions[2];
|
||||
int directionid = GetDirectionId(directions[0],directions[1],k);
|
||||
dofmap = OvlpDofMaps[ip][directionid];
|
||||
}
|
||||
|
||||
|
||||
void NeighborDofMaps::SetElementToOverlap(int ip, int iel,
|
||||
const Array<bool> & neg,
|
||||
const Array<bool> & pos)
|
||||
{
|
||||
int kbeg = (dim == 2) ? 0 : -1;
|
||||
int kend = (dim == 2) ? 0 : 1;
|
||||
for (int k = kbeg; k<=kend; k++)
|
||||
{
|
||||
if (dim == 3)
|
||||
{
|
||||
if (k == -1 && !neg[2]) continue;
|
||||
if (k == 1 && !pos[2]) continue;
|
||||
}
|
||||
for (int j = -1; j<=1; j++)
|
||||
{
|
||||
if (j== -1 && !neg[1]) continue;
|
||||
if (j== 1 && !pos[1]) continue;
|
||||
for (int i = -1; i<=1; i++)
|
||||
{
|
||||
// cases to skip
|
||||
if (i==-1 && !neg[0]) continue;
|
||||
if (i== 1 && !pos[0]) continue;
|
||||
|
||||
if (i==0 && j==0 && k == 0) continue;
|
||||
int kk = (dim==2)?-1 : k;
|
||||
int DirId = GetDirectionId(i,j,kk);
|
||||
OvlpElems[ip][DirId].Append(iel);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -1,179 +0,0 @@
|
||||
#pragma once
|
||||
#include "MeshPartition.hpp"
|
||||
#include "complex_linalg.hpp"
|
||||
|
||||
struct UniqueIndexGen
|
||||
{
|
||||
int counter = 0;
|
||||
std::unordered_map<int,int> idx;
|
||||
|
||||
void Set(int i)
|
||||
{
|
||||
std::unordered_map<int,int>::iterator f = idx.find(i);
|
||||
if (f == idx.end())
|
||||
{
|
||||
idx[i] = counter;
|
||||
counter++;
|
||||
}
|
||||
}
|
||||
|
||||
int Get(int i)
|
||||
{
|
||||
std::unordered_map<int,int>::iterator f = idx.find(i);
|
||||
if (f == idx.end())
|
||||
{
|
||||
return -1;
|
||||
}
|
||||
else
|
||||
{
|
||||
return (*f).second;
|
||||
}
|
||||
}
|
||||
void Reset()
|
||||
{
|
||||
counter = 0;
|
||||
idx.clear();
|
||||
}
|
||||
};
|
||||
|
||||
|
||||
struct Sweep
|
||||
{
|
||||
private:
|
||||
int dim;
|
||||
std::vector<Array<int>> sweeps;
|
||||
public:
|
||||
int nsweeps;
|
||||
Sweep(int dim_);
|
||||
~Sweep();
|
||||
void GetSweep(const int i, Array<int> & sweep)
|
||||
{
|
||||
MFEM_VERIFY(i<nsweeps, "Sweep number out of bounds");
|
||||
sweep.SetSize(dim);
|
||||
sweep = sweeps[i];
|
||||
}
|
||||
|
||||
|
||||
};
|
||||
|
||||
|
||||
|
||||
// Function coefficient that takes the bounding box of the mesh as an input
|
||||
class CutOffFnCoefficient : public Coefficient
|
||||
{
|
||||
private:
|
||||
double (*Function)(const Vector &, const Vector &, const Vector &, const Array2D<double> &);
|
||||
Vector pmin, pmax;
|
||||
Array2D<double> h; // specify the with of the cutoff function (h in each direction)
|
||||
|
||||
|
||||
public:
|
||||
CutOffFnCoefficient(double (*F)(const Vector &, const Vector &, const Vector &, const Array2D<double> &),
|
||||
const Vector & pmin_, const Vector & pmax_, Array2D<double> & h_)
|
||||
: Function(F), pmin(pmin_), pmax(pmax_), h(h_)
|
||||
{}
|
||||
virtual double Eval(ElementTransformation &T, const IntegrationPoint &ip)
|
||||
{
|
||||
double x[3];
|
||||
Vector transip(x, 3);
|
||||
T.Transform(ip, transip);
|
||||
return ((*Function)(transip, pmin, pmax, h));
|
||||
}
|
||||
};
|
||||
|
||||
double CutOffFncn(const Vector &x, const Vector & pmin,
|
||||
const Vector & pmax, const Array2D<double> & h_);
|
||||
double ChiFncn(const Vector &x, const Vector & pmin,
|
||||
const Vector & pmax, const Array2D<double> & h_);
|
||||
|
||||
class DofMap // Constructs dof maps for a given partition
|
||||
{
|
||||
public:
|
||||
int nrpatch, nx, ny, nz;
|
||||
vector<Array<int>> Dof2GlobalDof;
|
||||
vector<Array<int>> Dof2PmlDof;
|
||||
Array<Mesh *> PmlMeshes;
|
||||
Array<FiniteElementSpace *> fespaces;
|
||||
Array<FiniteElementSpace *> PmlFespaces;
|
||||
// constructor
|
||||
// Non PML constructor dof map
|
||||
DofMap(FiniteElementSpace * fes, MeshPartition * partition);
|
||||
// PML
|
||||
DofMap(FiniteElementSpace * fes , MeshPartition * partition, int nrlayers);
|
||||
~DofMap(){};
|
||||
};
|
||||
|
||||
|
||||
|
||||
class LocalDofMap // Constructs dof mapbetween two partitions
|
||||
{
|
||||
const FiniteElementCollection *fec=nullptr;
|
||||
MeshPartition * part1=nullptr;
|
||||
MeshPartition * part2=nullptr;
|
||||
public:
|
||||
int nrpatch, nx, ny, nz;
|
||||
vector<Array<int>> map1;
|
||||
vector<Array<int>> map2;
|
||||
// constructor
|
||||
LocalDofMap(const FiniteElementCollection * fec_, MeshPartition * part1_,
|
||||
MeshPartition * part2_);
|
||||
~LocalDofMap();
|
||||
};
|
||||
|
||||
|
||||
struct NeighborDofMaps
|
||||
{
|
||||
private:
|
||||
int dim;
|
||||
MeshPartition * part = nullptr;
|
||||
FiniteElementSpace * fes = nullptr;
|
||||
Mesh * mesh = nullptr;
|
||||
std::vector<std::vector<Array<int>>> OvlpElems;
|
||||
std::vector<std::vector<Array<int>>> OvlpDofMaps;
|
||||
|
||||
DofMap * dmap = nullptr;
|
||||
int nrsubdomains = 0;
|
||||
int ovlp_layers = 0;
|
||||
Array<int> nxyz;
|
||||
void SetElementToOverlap(int ip, int iel,
|
||||
const Array<bool> & neg,
|
||||
const Array<bool> & pos);
|
||||
|
||||
void MarkOvlpElements();
|
||||
void ComputeNeighborDofMaps();
|
||||
|
||||
void Getijk(int ip, int & i, int & j, int & k) const
|
||||
{
|
||||
k = ip/(nxyz[0]*nxyz[1]);
|
||||
j = (ip-k*nxyz[0]*nxyz[1])/nxyz[0];
|
||||
i = (ip-k*nxyz[0]*nxyz[1])%nxyz[0];
|
||||
}
|
||||
|
||||
int GetPatchId(const Array<int> & ijk) const
|
||||
{
|
||||
int d=ijk.Size();
|
||||
int z = (d==2)? 0 : ijk[2];
|
||||
return part->subdomains(ijk[0],ijk[1],z);
|
||||
}
|
||||
int GetDirectionId(int i, int j, int k=-1)
|
||||
{
|
||||
int n = 3;
|
||||
return (k+1)*n*n + (j+1)*n + i+1;
|
||||
}
|
||||
void GetDirections(const int id, int & i, int & j, int & k)
|
||||
{
|
||||
int n = 3;
|
||||
k = id/(n*n) - 1;
|
||||
j = (id-(k+1)*n*n)/n - 1;
|
||||
i = (id-(k+1)*n*n)%n - 1;
|
||||
}
|
||||
|
||||
public:
|
||||
NeighborDofMaps(MeshPartition * part_,
|
||||
FiniteElementSpace * fes_,
|
||||
DofMap * dmap_,
|
||||
int ovlp_layers_);
|
||||
|
||||
void GetNeighborDofMap(const int ip, const Array<int> & directions,
|
||||
Array<int> & dofmap);
|
||||
};
|
||||
@@ -1,358 +0,0 @@
|
||||
|
||||
#include "../../../linalg/kernels.hpp"
|
||||
#include "complex_linalg.hpp"
|
||||
|
||||
|
||||
ComplexDenseMatrix::ComplexDenseMatrix(){}
|
||||
|
||||
ComplexDenseMatrix::ComplexDenseMatrix(int s)
|
||||
{
|
||||
MFEM_ASSERT(s >= 0, "invalid ComplexDenseMatrix size: " << s);
|
||||
height = s;
|
||||
width = s;
|
||||
if (s > 0)
|
||||
{
|
||||
data = new complex<double>[s*s];
|
||||
*this = 0.0; // init with zeroes
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
ComplexDenseMatrix::ComplexDenseMatrix(int m, int n)
|
||||
{
|
||||
MFEM_VERIFY(m >= 0 && n >= 0,
|
||||
"invalid DenseMatrix size: " << m << " x " << n);
|
||||
const int s = m*n;
|
||||
height = m;
|
||||
width = n;
|
||||
if (s > 0)
|
||||
{
|
||||
data = new complex<double>[s];
|
||||
*this = 0.0; // init with zeroes
|
||||
}
|
||||
}
|
||||
|
||||
void ComplexDenseMatrix::SetSize(int h, int w)
|
||||
{
|
||||
MFEM_VERIFY(h >= 0 && w >= 0,
|
||||
"invalid ComplexDenseMatrix size: " << h << " x " << w);
|
||||
if (Height() == h && Width() == w)
|
||||
{
|
||||
return;
|
||||
}
|
||||
height = h;
|
||||
width = w;
|
||||
const int hw = h*w;
|
||||
delete data;
|
||||
data = new complex<double>[hw];
|
||||
*this = 0.0; // init with zeroes
|
||||
}
|
||||
|
||||
ComplexDenseMatrix &ComplexDenseMatrix::operator=(double c)
|
||||
{
|
||||
const int s = Height()*Width();
|
||||
for (int i = 0; i < s; i++)
|
||||
{
|
||||
data[i] = c;
|
||||
}
|
||||
return *this;
|
||||
}
|
||||
|
||||
ComplexDenseMatrix &ComplexDenseMatrix::operator=(complex<double> c)
|
||||
{
|
||||
const int s = Height()*Width();
|
||||
for (int i = 0; i < s; i++)
|
||||
{
|
||||
data[i] = c;
|
||||
}
|
||||
return *this;
|
||||
}
|
||||
|
||||
|
||||
std::complex<double> ComplexDenseMatrix::Det() const
|
||||
{
|
||||
MFEM_ASSERT(Height() == Width() && Height() > 0,
|
||||
"The matrix must be square and "
|
||||
<< "sized larger than zero to compute the determinant."
|
||||
<< " Height() = " << Height()
|
||||
<< ", Width() = " << Width());
|
||||
|
||||
switch (Height())
|
||||
{
|
||||
case 1:
|
||||
return data[0];
|
||||
|
||||
case 2:
|
||||
return data[0] * data[3] - data[1] * data[2];
|
||||
|
||||
case 3:
|
||||
{
|
||||
const complex<double> *d = data;
|
||||
return
|
||||
d[0] * (d[4] * d[8] - d[5] * d[7]) +
|
||||
d[3] * (d[2] * d[7] - d[1] * d[8]) +
|
||||
d[6] * (d[1] * d[5] - d[2] * d[4]);
|
||||
}
|
||||
default:
|
||||
{
|
||||
MFEM_ABORT("dim>3 not supported yet");
|
||||
return 0;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
DenseMatrix * ComplexDenseMatrix::real() const
|
||||
{
|
||||
DenseMatrix * Ar = new DenseMatrix(height,width);
|
||||
double * data = Ar->Data();
|
||||
complex<double> * zdata = this->data;
|
||||
for (int s = 0; s<height*width; s++)
|
||||
{
|
||||
data[s] = zdata[s].real();
|
||||
}
|
||||
return Ar;
|
||||
}
|
||||
DenseMatrix * ComplexDenseMatrix::imag() const
|
||||
{
|
||||
DenseMatrix * Ai = new DenseMatrix(height,width);
|
||||
double * data = Ai->Data();
|
||||
complex<double> * zdata = this->data;
|
||||
for (int s = 0; s<height*width; s++)
|
||||
{
|
||||
data[s] = zdata[s].imag();
|
||||
}
|
||||
return Ai;
|
||||
}
|
||||
|
||||
void ComplexDenseMatrix::GetReal(DenseMatrix & Ar)
|
||||
{
|
||||
MFEM_ASSERT(Ar.Height() == height && Ar.Width() == width, "Incompatible dimensions");
|
||||
double * data = Ar.Data();
|
||||
complex<double> * zdata = this->data;
|
||||
for (int s = 0; s<height*width; s++)
|
||||
{
|
||||
data[s] = zdata[s].real();
|
||||
}
|
||||
}
|
||||
|
||||
void ComplexDenseMatrix::GetImag(DenseMatrix & Ai)
|
||||
{
|
||||
double * data = Ai.Data();
|
||||
complex<double> * zdata = this->data;
|
||||
for (int s = 0; s<height*width; s++)
|
||||
{
|
||||
data[s] = zdata[s].imag();
|
||||
}
|
||||
}
|
||||
|
||||
ComplexDenseMatrix &ComplexDenseMatrix::operator=(const ComplexDenseMatrix &m)
|
||||
{
|
||||
SetSize(m.height, m.width);
|
||||
|
||||
const int hw = height * width;
|
||||
for (int i = 0; i < hw; i++)
|
||||
{
|
||||
data[i] = m.data[i];
|
||||
}
|
||||
return *this;
|
||||
}
|
||||
|
||||
ComplexDenseMatrix &ComplexDenseMatrix::operator+=(const complex<double> *m)
|
||||
{
|
||||
const int hw = Height()*Width();
|
||||
for (int i = 0; i < hw; i++)
|
||||
{
|
||||
data[i] += m[i];
|
||||
}
|
||||
return *this;
|
||||
}
|
||||
|
||||
ComplexDenseMatrix &ComplexDenseMatrix::operator+=(const ComplexDenseMatrix &m)
|
||||
{
|
||||
MFEM_ASSERT(Height() == m.Height() && Width() == m.Width(),
|
||||
"incompatible matrix sizes.");
|
||||
return *this += m.GetData();
|
||||
}
|
||||
|
||||
ComplexDenseMatrix &ComplexDenseMatrix::operator-=(const ComplexDenseMatrix &m)
|
||||
{
|
||||
int s = Height()*Width();
|
||||
complex<double> * mdata = m.GetData();
|
||||
for (int i = 0; i < s; i++)
|
||||
{
|
||||
data[i] -= mdata[s];
|
||||
}
|
||||
return *this;
|
||||
}
|
||||
|
||||
ComplexDenseMatrix &ComplexDenseMatrix::operator*=(complex<double> c)
|
||||
{
|
||||
int s = Height()*Width();
|
||||
for (int i = 0; i < s; i++)
|
||||
{
|
||||
data[i] *= c;
|
||||
}
|
||||
return *this;
|
||||
}
|
||||
|
||||
|
||||
void ComplexDenseMatrix::Print(std::ostream &out, int width_) const
|
||||
{
|
||||
// save current output flags
|
||||
ios::fmtflags old_flags = out.flags();
|
||||
// output flags = scientific + show sign
|
||||
out << setiosflags(ios::scientific | ios::showpos);
|
||||
for (int i = 0; i < height; i++)
|
||||
{
|
||||
out << "[row " << i << "]\n";
|
||||
for (int j = 0; j < width; j++)
|
||||
{
|
||||
out << (*this)(i,j);
|
||||
if (j+1 == width || (j+1) % width_ == 0)
|
||||
{
|
||||
out << '\n';
|
||||
}
|
||||
else
|
||||
{
|
||||
out << ' ';
|
||||
}
|
||||
}
|
||||
}
|
||||
// reset output flags to original values
|
||||
out.flags(old_flags);
|
||||
}
|
||||
|
||||
void ComplexDenseMatrix::PrintMatlab(std::ostream &out) const
|
||||
{
|
||||
// save current output flags
|
||||
// ios::fmtflags old_flags = out.flags();
|
||||
// output flags = scientific + show sign
|
||||
// out << setiosflags(ios::scientific | ios::showpos);
|
||||
for (int i = 0; i < height; i++)
|
||||
{
|
||||
for (int j = 0; j < width; j++)
|
||||
{
|
||||
out << (*this)(i,j);
|
||||
out << ' ';
|
||||
}
|
||||
out << "\n";
|
||||
}
|
||||
// reset output flags to original values
|
||||
// out.flags(old_flags);
|
||||
}
|
||||
|
||||
ComplexDenseMatrixInverse::ComplexDenseMatrixInverse(const ComplexDenseMatrix & A) : ComplexDenseMatrix(A.Height())
|
||||
{
|
||||
MFEM_VERIFY(A.Height() == A.Width(), "The matrix is not square");
|
||||
MFEM_VERIFY(A.Height() < 4, "dim > 3 is not supported yet");
|
||||
|
||||
std::complex<double> detA = A.Det();
|
||||
MFEM_VERIFY(abs(A.Det())>1e-14, "The given matrix is singular");
|
||||
|
||||
std::complex<double> * d = this->Data();
|
||||
std::complex<double> *dA = A.GetData();
|
||||
switch (A.Height())
|
||||
{
|
||||
case 1:
|
||||
d[0] = 1.0/dA[0];
|
||||
break;
|
||||
case 2:
|
||||
d[0] = 1.0/detA * dA[3];
|
||||
d[1] = -1.0/detA * dA[1];
|
||||
d[2] = -1.0/detA * dA[2];
|
||||
d[3] = 1.0/detA * dA[0];
|
||||
break;
|
||||
case 3:
|
||||
d[0] = 1.0/detA*(dA[4]*dA[8] - dA[5]*dA[7]);
|
||||
d[1] = -1.0/detA*(dA[1]*dA[8] - dA[2]*dA[7]);
|
||||
d[2] = 1.0/detA*(dA[1]*dA[5] - dA[2]*dA[4]);
|
||||
d[3] = -1.0/detA*(dA[3]*dA[8] - dA[5]*dA[6]);
|
||||
d[4] = 1.0/detA*(dA[0]*dA[8] - dA[2]*dA[6]);
|
||||
d[5] = -1.0/detA*(dA[0]*dA[5] - dA[2]*dA[3]);
|
||||
d[6] = 1.0/detA*(dA[3]*dA[7] - dA[4]*dA[6]);
|
||||
d[7] = -1.0/detA*(dA[0]*dA[7] - dA[1]*dA[6]);
|
||||
d[8] = 1.0/detA*(dA[0]*dA[4] - dA[1]*dA[3]);
|
||||
break;
|
||||
default:
|
||||
// Should be unreachable
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
|
||||
/// Matrix matrix multiplication. A = B * C.
|
||||
void Mult(const ComplexDenseMatrix &b, const ComplexDenseMatrix &c, ComplexDenseMatrix &a)
|
||||
{
|
||||
MFEM_ASSERT(a.Height() == b.Height() && a.Width() == c.Width() &&
|
||||
b.Width() == c.Height(), "incompatible dimensions");
|
||||
|
||||
const int ah = a.Height();
|
||||
const int aw = a.Width();
|
||||
const int bw = b.Width();
|
||||
complex<double> *ad = a.Data();
|
||||
const complex<double> *bd = b.Data();
|
||||
const complex<double> *cd = c.Data();
|
||||
kernels::Mult(ah,aw,bw,bd,cd,ad);
|
||||
}
|
||||
|
||||
/// Multiply the transpose of a matrix A with a matrix B: At*B
|
||||
void MultAtB(const ComplexDenseMatrix &A, const ComplexDenseMatrix &B, ComplexDenseMatrix &AtB)
|
||||
{
|
||||
MFEM_ASSERT(A.Width() == AtB.Height() && B.Width() == AtB.Width() &&
|
||||
A.Height() == B.Height(), "incompatible dimensions");
|
||||
const int ah = A.Height();
|
||||
const int aw = A.Width();
|
||||
const int bw = B.Width();
|
||||
const complex<double> *ad = A.Data();
|
||||
const complex<double> *bd = B.Data();
|
||||
complex<double> *cd = AtB.Data();
|
||||
|
||||
for (int j = 0; j < bw; j++)
|
||||
{
|
||||
const complex<double> *ap = ad;
|
||||
for (int i = 0; i < aw; i++)
|
||||
{
|
||||
complex<double> d = 0.0;
|
||||
for (int k = 0; k < ah; k++)
|
||||
{
|
||||
d += ap[k] * bd[k];
|
||||
}
|
||||
*(cd++) = d;
|
||||
ap += ah;
|
||||
}
|
||||
bd += ah;
|
||||
}
|
||||
}
|
||||
|
||||
/// Multiply the conjugate transpose of a matrix A with a matrix B: At*B
|
||||
void MultAhB(const ComplexDenseMatrix &A, const ComplexDenseMatrix &B, ComplexDenseMatrix &AtB)
|
||||
{
|
||||
MFEM_ASSERT(A.Width() == AtB.Height() && B.Width() == AtB.Width() &&
|
||||
A.Height() == B.Height(), "incompatible dimensions");
|
||||
MFEM_ASSERT(A.Width() == AtB.Height() && B.Width() == AtB.Width() &&
|
||||
A.Height() == B.Height(), "incompatible dimensions");
|
||||
const int ah = A.Height();
|
||||
const int aw = A.Width();
|
||||
const int bw = B.Width();
|
||||
const complex<double> *ad = A.Data();
|
||||
const complex<double> *bd = B.Data();
|
||||
complex<double> *cd = AtB.Data();
|
||||
|
||||
for (int j = 0; j < bw; j++)
|
||||
{
|
||||
const complex<double> *ap = ad;
|
||||
for (int i = 0; i < aw; i++)
|
||||
{
|
||||
complex<double> d = 0.0;
|
||||
for (int k = 0; k < ah; k++)
|
||||
{
|
||||
d += conj(ap[k]) * bd[k];
|
||||
}
|
||||
*(cd++) = d;
|
||||
ap += ah;
|
||||
}
|
||||
bd += ah;
|
||||
}
|
||||
|
||||
}
|
||||
@@ -1,98 +0,0 @@
|
||||
#pragma once
|
||||
#include "mfem.hpp"
|
||||
#include <fstream>
|
||||
#include <iostream>
|
||||
|
||||
using namespace std;
|
||||
using namespace mfem;
|
||||
|
||||
class ComplexDenseMatrix
|
||||
{
|
||||
private:
|
||||
std::complex<double> * data = nullptr;
|
||||
int height = 0;
|
||||
int width = 0;
|
||||
public:
|
||||
ComplexDenseMatrix();
|
||||
|
||||
/// Creates square matrix of size s.
|
||||
explicit ComplexDenseMatrix(int s);
|
||||
|
||||
/// Creates rectangular matrix of size m x n.
|
||||
ComplexDenseMatrix(int m, int n);
|
||||
|
||||
/// Change the size of the DenseMatrix to s x s.
|
||||
void SetSize(int s) { SetSize(s, s); }
|
||||
|
||||
/// Change the size of the DenseMatrix to h x w.
|
||||
void SetSize(int h, int w);
|
||||
|
||||
/// Returns the matrix data array.
|
||||
inline complex<double> *Data() const
|
||||
{ return const_cast<complex<double>*>((const complex<double>*)data);}
|
||||
|
||||
/// Returns the matrix data array.
|
||||
inline complex<double> *GetData() const { return Data(); }
|
||||
|
||||
/// Returns reference to a_{ij}.
|
||||
inline complex<double> &operator()(int i, int j);
|
||||
inline const complex<double> &operator()(int i, int j) const;
|
||||
|
||||
inline int Height() const { return height; }
|
||||
inline int Width() const { return width; }
|
||||
|
||||
/// Sets the matrix elements equal to constant c
|
||||
ComplexDenseMatrix &operator=(std::complex<double> c);
|
||||
ComplexDenseMatrix &operator=(double c);
|
||||
|
||||
/// Sets the matrix size and elements equal to those of m
|
||||
ComplexDenseMatrix &operator=(const ComplexDenseMatrix &m);
|
||||
ComplexDenseMatrix &operator+=(const complex<double> *m);
|
||||
ComplexDenseMatrix &operator+=(const ComplexDenseMatrix &m);
|
||||
ComplexDenseMatrix &operator-=(const ComplexDenseMatrix &m);
|
||||
ComplexDenseMatrix &operator*=(complex<double> c);
|
||||
|
||||
/// Calculates the determinant of the matrix
|
||||
/// (for 2x2, 3x3)
|
||||
std::complex<double> Det() const;
|
||||
|
||||
virtual void Print(std::ostream &out = mfem::out, int width_ = 4) const;
|
||||
virtual void PrintMatlab(std::ostream &out = mfem::out) const;
|
||||
|
||||
DenseMatrix * real() const;
|
||||
DenseMatrix * imag() const;
|
||||
|
||||
void GetReal(DenseMatrix & Ar);
|
||||
void GetImag(DenseMatrix & Ai);
|
||||
};
|
||||
|
||||
inline complex<double> &ComplexDenseMatrix::operator()(int i, int j)
|
||||
{
|
||||
MFEM_VERIFY(data && i >= 0 && i < height && j >= 0 && j < width, "");
|
||||
// return data[i*width+j];
|
||||
return data[j*height+i];
|
||||
}
|
||||
|
||||
inline const complex<double> &ComplexDenseMatrix::operator()(int i, int j) const
|
||||
{
|
||||
MFEM_VERIFY(data && i >= 0 && i < height && j >= 0 && j < width, "");
|
||||
// return data[i*width+j];
|
||||
return data[j*height+i];
|
||||
}
|
||||
|
||||
|
||||
class ComplexDenseMatrixInverse : public ComplexDenseMatrix
|
||||
{
|
||||
private:
|
||||
public:
|
||||
ComplexDenseMatrixInverse(const ComplexDenseMatrix & );
|
||||
};
|
||||
|
||||
/// Matrix matrix multiplication. A = B * C.
|
||||
void Mult(const ComplexDenseMatrix &b, const ComplexDenseMatrix &c, ComplexDenseMatrix &a);
|
||||
|
||||
/// Multiply the transpose of a matrix A with a matrix B: At*B
|
||||
void MultAtB(const ComplexDenseMatrix &A, const ComplexDenseMatrix &B, ComplexDenseMatrix &AtB);
|
||||
|
||||
/// Multiply the conjugate transpose of a matrix A with a matrix B: At*B
|
||||
void MultAhB(const ComplexDenseMatrix &A, const ComplexDenseMatrix &B, ComplexDenseMatrix &AtB);
|
||||
@@ -1,404 +0,0 @@
|
||||
//
|
||||
// Compile with: make helmholtzp
|
||||
//
|
||||
// Sample runs: mpirun -np 4 ./helmholtzp -nd 2 -nx 4 -ny 4 -sr 3 -pr 3 -k 16.0 -o 2
|
||||
// mpirun -np 4 ./helmholtzp -nd 3 -nx 2 -ny 2 -nz 2 -sr 3 -pr 1 -k 2.0 -o 2
|
||||
//
|
||||
#include "mfem.hpp"
|
||||
#include <fstream>
|
||||
#include <iostream>
|
||||
#include "ParDST/ParDST.hpp"
|
||||
|
||||
using namespace std;
|
||||
using namespace mfem;
|
||||
|
||||
// Exact solution and r.h.s., see below for implementation.
|
||||
double f_exact_Re(const Vector &x);
|
||||
double f_exact_Im(const Vector &x);
|
||||
|
||||
double wavespeed(const Vector &x);
|
||||
|
||||
double funccoeff_re(const Vector & x);
|
||||
double funccoeff_im(const Vector & x);
|
||||
|
||||
|
||||
int dim;
|
||||
double omega;
|
||||
int sol = 1;
|
||||
double length = 1.0;
|
||||
double pml_length = 0.25;
|
||||
Array2D<double>comp_bdr;
|
||||
|
||||
int main(int argc, char *argv[])
|
||||
{
|
||||
// 1. Initialize MPI.
|
||||
int num_procs, myid;
|
||||
MPI_Init(&argc, &argv);
|
||||
MPI_Comm_size(MPI_COMM_WORLD, &num_procs);
|
||||
MPI_Comm_rank(MPI_COMM_WORLD, &myid);
|
||||
// 2. Parse command-line options.
|
||||
// finite element order of approximation
|
||||
int order = 1;
|
||||
bool visualization = 1;
|
||||
// number of wavelengths
|
||||
double k = 0.5;
|
||||
// number of serial refinements
|
||||
int ser_ref_levels = 1;
|
||||
// number of parallel refinements
|
||||
int par_ref_levels = 2;
|
||||
// dimension
|
||||
int nd = 2;
|
||||
int nx=2;
|
||||
int ny=2;
|
||||
int nz=2;
|
||||
bool herm_conv = true;
|
||||
|
||||
// optional command line inputs
|
||||
OptionsParser args(argc, argv);
|
||||
args.AddOption(&order, "-o", "--order",
|
||||
"Finite element order (polynomial degree) or -1 for"
|
||||
" isoparametric space.");
|
||||
args.AddOption(&nd, "-nd", "--dim","Problem space dimension");
|
||||
args.AddOption(&nx, "-nx", "--nx","Number of subdomains in x direction");
|
||||
args.AddOption(&ny, "-ny", "--ny","Number of subdomains in y direction");
|
||||
args.AddOption(&nz, "-nz", "--nz","Number of subdomains in z direction");
|
||||
args.AddOption(&sol, "-sol", "--exact",
|
||||
"Exact solution flag - 0:polynomial, 1: plane wave, -1: unknown exact");
|
||||
args.AddOption(&k, "-k", "--wavelengths",
|
||||
"Number of wavelengths.");
|
||||
args.AddOption(&pml_length, "-pml_length", "--pml_length",
|
||||
"Length of the PML region in each direction");
|
||||
args.AddOption(&length, "-length", "--length",
|
||||
"length of the domain in each direction.");
|
||||
args.AddOption(&ser_ref_levels, "-sr", "--ser_ref_levels",
|
||||
"Number of Serial Refinements.");
|
||||
args.AddOption(&par_ref_levels, "-pr", "--par_ref_levels",
|
||||
"Number of Parallel Refinements.");
|
||||
args.AddOption(&herm_conv, "-herm", "--hermitian", "-no-herm",
|
||||
"--no-hermitian", "Use convention for Hermitian operators.");
|
||||
args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
|
||||
"--no-visualization",
|
||||
"Enable or disable GLVis visualization.");
|
||||
args.Parse();
|
||||
// check if the inputs are correct
|
||||
if (!args.Good())
|
||||
{
|
||||
if (myid == 0)
|
||||
{
|
||||
args.PrintUsage(cout);
|
||||
}
|
||||
MPI_Finalize();
|
||||
return 1;
|
||||
}
|
||||
if (myid == 0)
|
||||
{
|
||||
args.PrintOptions(cout);
|
||||
}
|
||||
// Angular frequency
|
||||
omega = 2.0 * M_PI * k;
|
||||
|
||||
// 3. Read the mesh from the given mesh file.
|
||||
Mesh *mesh;
|
||||
|
||||
if (nd == 2)
|
||||
{
|
||||
mesh = new Mesh(1, 1, Element::QUADRILATERAL, true, length, length, false);
|
||||
}
|
||||
else
|
||||
{
|
||||
mesh = new Mesh(1, 1, 1, Element::HEXAHEDRON, true, length, length, length,false);
|
||||
}
|
||||
|
||||
// 3. Executing uniform h-refinement
|
||||
dim = mesh->Dimension();
|
||||
for (int i = 0; i < ser_ref_levels; i++ )
|
||||
{
|
||||
mesh->UniformRefinement();
|
||||
}
|
||||
|
||||
// 4. Define a parallel mesh by a partitioning of the serial mesh.
|
||||
// ParMesh *pmesh = new ParMesh(MPI_COMM_WORLD, *mesh);
|
||||
int nprocs;
|
||||
int nprocsx;
|
||||
int nprocsy;
|
||||
int nprocsz;
|
||||
if (dim == 2)
|
||||
{
|
||||
nprocs = sqrt(num_procs);
|
||||
// nprocsx = nprocs;
|
||||
// nprocsy = nprocs;
|
||||
nprocsx = 1;
|
||||
nprocsy = num_procs;
|
||||
nprocsz = 1;
|
||||
}
|
||||
else
|
||||
{
|
||||
nprocs = cbrt(num_procs);
|
||||
// nprocsx = nprocs;
|
||||
// nprocsy = nprocs;
|
||||
// nprocsz = nprocs;
|
||||
nprocsx = 1;
|
||||
if (nz != 1)
|
||||
{
|
||||
nprocsy = sqrt(num_procs);
|
||||
nprocsz = nprocsy;
|
||||
}
|
||||
else
|
||||
{
|
||||
nprocsy = num_procs;
|
||||
nprocsz = 1;
|
||||
}
|
||||
}
|
||||
// MFEM_VERIFY(nprocs*nprocs == num_procs, "Check MPI partitioning");
|
||||
// int nxyz[3] = {num_procs,1,1};
|
||||
// int nxyz[3] = {nprocs,nprocs,1};
|
||||
// int nxyz[3] = {1,num_procs,1};
|
||||
|
||||
int nxyz[3] = {nprocsx,nprocsy,nprocsz};
|
||||
// int nxyz[3] = {num_procs,1,1};
|
||||
int * part = mesh->CartesianPartitioning(nxyz);
|
||||
ParMesh *pmesh = new ParMesh(MPI_COMM_WORLD,*mesh,part);
|
||||
// ParMesh *pmesh = new ParMesh(MPI_COMM_WORLD,*mesh);
|
||||
delete [] part;
|
||||
delete mesh;
|
||||
|
||||
for (int l = 0; l < par_ref_levels; l++)
|
||||
{
|
||||
pmesh->UniformRefinement();
|
||||
}
|
||||
|
||||
|
||||
double hl = GetUniformMeshElementSize(pmesh);
|
||||
int nrlayers = 3;
|
||||
|
||||
Array2D<double> lengths(dim,2);
|
||||
lengths = hl*nrlayers;
|
||||
// lengths[0][1] = 0.0;
|
||||
// lengths[1][1] = 0.0;
|
||||
// lengths[1][0] = 0.0;
|
||||
// lengths[0][0] = 0.0;
|
||||
CartesianPML pml(pmesh,lengths);
|
||||
pml.SetOmega(omega);
|
||||
comp_bdr.SetSize(dim,2);
|
||||
comp_bdr = pml.GetCompDomainBdr();
|
||||
|
||||
// 6. Define a finite element space on the mesh.
|
||||
FiniteElementCollection *fec = new H1_FECollection(order, dim);
|
||||
ParFiniteElementSpace *fespace = new ParFiniteElementSpace(pmesh, fec);
|
||||
HYPRE_Int size = fespace->GlobalTrueVSize();
|
||||
|
||||
if (myid == 0)
|
||||
{
|
||||
cout << "Number of finite element unknowns: " << size << endl;
|
||||
}
|
||||
// 6. Set up the linear form (Real and Imaginary part)
|
||||
FunctionCoefficient f_Re(f_exact_Re);
|
||||
FunctionCoefficient f_Im(f_exact_Im);
|
||||
|
||||
// 8. Setup Complex Operator convention
|
||||
ComplexOperator::Convention conv =
|
||||
herm_conv ? ComplexOperator::HERMITIAN : ComplexOperator::BLOCK_SYMMETRIC;
|
||||
|
||||
// ParLinearForm *b_Re(new ParLinearForm);
|
||||
ParComplexLinearForm b(fespace, conv);
|
||||
b.AddDomainIntegrator(new DomainLFIntegrator(f_Re),
|
||||
new DomainLFIntegrator(f_Im));
|
||||
b.real().Vector::operator=(0.0);
|
||||
b.imag().Vector::operator=(0.0);
|
||||
b.Assemble();
|
||||
|
||||
// 7. Set up the bilinear form (Real and Imaginary part)
|
||||
ConstantCoefficient one(1.0);
|
||||
ConstantCoefficient sigma(-pow(omega, 2));
|
||||
|
||||
FunctionCoefficient ws(wavespeed);
|
||||
|
||||
PmlMatrixCoefficient c1_re(dim,pml_detJ_JT_J_inv_Re,&pml);
|
||||
PmlMatrixCoefficient c1_im(dim,pml_detJ_JT_J_inv_Im,&pml);
|
||||
|
||||
PmlCoefficient detJ_re(pml_detJ_Re,&pml);
|
||||
PmlCoefficient detJ_im(pml_detJ_Im,&pml);
|
||||
|
||||
ProductCoefficient c2_re0(sigma, detJ_re);
|
||||
ProductCoefficient c2_im0(sigma, detJ_im);
|
||||
|
||||
ProductCoefficient c2_re(c2_re0, ws);
|
||||
ProductCoefficient c2_im(c2_im0, ws);
|
||||
|
||||
ParSesquilinearForm a(fespace,conv);
|
||||
a.AddDomainIntegrator(new DiffusionIntegrator(c1_re),
|
||||
new DiffusionIntegrator(c1_im));
|
||||
a.AddDomainIntegrator(new MassIntegrator(c2_re),
|
||||
new MassIntegrator(c2_im));
|
||||
a.Assemble();
|
||||
a.Finalize();
|
||||
|
||||
Array<int> ess_tdof_list;
|
||||
Array<int> ess_bdr(pmesh->bdr_attributes.Max());
|
||||
ess_bdr = 1;
|
||||
fespace->GetEssentialTrueDofs(ess_bdr, ess_tdof_list);
|
||||
|
||||
// Solution grid function
|
||||
ParComplexGridFunction p_gf(fespace); p_gf = 0.0;
|
||||
OperatorHandle Ah;
|
||||
Vector X, B;
|
||||
|
||||
a.FormLinearSystem(ess_tdof_list, p_gf, b, Ah, X, B);
|
||||
{
|
||||
StopWatch chrono;
|
||||
chrono.Clear();
|
||||
chrono.Start();
|
||||
ParDST S(&a,lengths,omega, &ws,nrlayers,nx,ny,nz);
|
||||
chrono.Stop();
|
||||
double t1 = chrono.RealTime();
|
||||
|
||||
chrono.Clear();
|
||||
chrono.Start();
|
||||
// X = 0.0;
|
||||
GMRESSolver gmres(MPI_COMM_WORLD);
|
||||
gmres.SetPreconditioner(S);
|
||||
gmres.SetOperator(*Ah);
|
||||
gmres.SetRelTol(1e-6);
|
||||
gmres.SetMaxIter(20);
|
||||
gmres.SetPrintLevel(1);
|
||||
gmres.Mult(B, X);
|
||||
chrono.Stop();
|
||||
|
||||
double t2 = chrono.RealTime();
|
||||
|
||||
MPI_Barrier(MPI_COMM_WORLD);
|
||||
|
||||
|
||||
cout << " myid: " << myid
|
||||
<< ", setup time: " << t1
|
||||
<< ", solution time: " << t2 << endl;
|
||||
|
||||
|
||||
a.RecoverFEMSolution(X,B,p_gf);
|
||||
if (visualization)
|
||||
{
|
||||
char vishost[] = "localhost";
|
||||
int visport = 19916;
|
||||
string keys;
|
||||
if (dim ==2 )
|
||||
{
|
||||
keys = "keys mrRljc\n";
|
||||
}
|
||||
else
|
||||
{
|
||||
keys = "keys mc\n";
|
||||
}
|
||||
socketstream sol_sock_re(vishost, visport);
|
||||
sol_sock_re.precision(8);
|
||||
sol_sock_re << "parallel " << num_procs << " " << myid << "\n"
|
||||
<< "solution\n" << *pmesh << p_gf.real() << keys
|
||||
<< "window_title 'Numerical Pressure: Real Part' " << flush;
|
||||
|
||||
socketstream sol_sock_im(vishost, visport);
|
||||
sol_sock_im.precision(8);
|
||||
sol_sock_im << "parallel " << num_procs << " " << myid << "\n"
|
||||
<< "solution\n" << *pmesh << p_gf.imag() << keys
|
||||
<< "window_title 'Numerical Pressure: Imag Part' " << flush;
|
||||
}
|
||||
}
|
||||
|
||||
delete fespace;
|
||||
delete fec;
|
||||
delete pmesh;
|
||||
MPI_Finalize();
|
||||
return 0;
|
||||
}
|
||||
|
||||
|
||||
double f_exact_Re(const Vector &x)
|
||||
{
|
||||
|
||||
int nrsources = (dim == 2) ? 4 : 8;
|
||||
Vector x0(nrsources);
|
||||
Vector y0(nrsources);
|
||||
Vector z0(nrsources);
|
||||
x0(0) = 0.25; y0(0) = 0.25; z0(0) = 0.25;
|
||||
x0(1) = 0.75; y0(1) = 0.25; z0(1) = 0.25;
|
||||
x0(2) = 0.25; y0(2) = 0.75; z0(2) = 0.25;
|
||||
x0(3) = 0.75; y0(3) = 0.75; z0(3) = 0.25;
|
||||
if (dim == 3)
|
||||
{
|
||||
x0(4) = 0.25; y0(4) = 0.25; z0(4) = 0.75;
|
||||
x0(5) = 0.75; y0(5) = 0.25; z0(5) = 0.75;
|
||||
x0(6) = 0.25; y0(6) = 0.75; z0(6) = 0.75;
|
||||
x0(7) = 0.75; y0(7) = 0.75; z0(7) = 0.75;
|
||||
}
|
||||
|
||||
double n = 4.0*omega/M_PI;
|
||||
double coeff = 16.0*omega*omega/M_PI/M_PI/M_PI;
|
||||
|
||||
double f_re = 0.0;
|
||||
// for (int i = 0; i<1; i++)
|
||||
for (int i = 0; i<nrsources; i++)
|
||||
{
|
||||
double beta = pow(x0(i)-x(0),2) + pow(y0(i)-x(1),2);
|
||||
if (dim == 3) { beta += pow(z0(i)-x(2),2); }
|
||||
double alpha = -pow(n,2) * beta;
|
||||
f_re += coeff*exp(alpha);
|
||||
}
|
||||
|
||||
bool in_pml = false;
|
||||
for (int i = 0; i<dim; i++)
|
||||
{
|
||||
if (x(i)<=comp_bdr(i,0) || x(i)>=comp_bdr(i,1))
|
||||
{
|
||||
in_pml = true;
|
||||
break;
|
||||
}
|
||||
}
|
||||
if (in_pml) f_re = 0.0;
|
||||
|
||||
return f_re;
|
||||
|
||||
}
|
||||
double f_exact_Im(const Vector &x)
|
||||
{
|
||||
double f_im;
|
||||
f_im = 0.0;
|
||||
return f_im;
|
||||
}
|
||||
|
||||
double wavespeed(const Vector &x)
|
||||
{
|
||||
double ws;
|
||||
ws = 1.0;
|
||||
return ws;
|
||||
}
|
||||
|
||||
double funccoeff_re(const Vector & x)
|
||||
{
|
||||
return sin(3*M_PI*(x.Sum()));
|
||||
}
|
||||
|
||||
double funccoeff_im(const Vector & x)
|
||||
{
|
||||
return cos(10*M_PI*(x.Sum()));
|
||||
}
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
@@ -1,67 +0,0 @@
|
||||
# Copyright (c) 2010, Lawrence Livermore National Security, LLC. Produced at the
|
||||
# Lawrence Livermore National Laboratory. LLNL-CODE-443211. All Rights reserved.
|
||||
# See file COPYRIGHT for details.
|
||||
#
|
||||
# This file is part of the MFEM library. For more information and source code
|
||||
# availability see http://mfem.org.
|
||||
#
|
||||
# MFEM is free software; you can redistribute it and/or modify it under the
|
||||
# terms of the GNU Lesser General Public License (as published by the Free
|
||||
# Software Foundation) version 2.1 dated February 1999.
|
||||
|
||||
# Use the MFEM build directory
|
||||
MFEM_DIR ?= ../..
|
||||
MFEM_BUILD_DIR ?= ../..
|
||||
SRC = $(if $(MFEM_DIR:../../..=),$(MFEM_DIR)/examples/maxwell-solver-dev/,)
|
||||
CONFIG_MK = $(MFEM_BUILD_DIR)/config/config.mk
|
||||
|
||||
MFEM_LIB_FILE = mfem_is_not_built
|
||||
-include $(CONFIG_MK)
|
||||
|
||||
SEQ_EXAMPLES =
|
||||
PAR_EXAMPLES = helmholtzp maxwellp
|
||||
ifeq ($(MFEM_USE_MPI),NO)
|
||||
EXAMPLES = $(SEQ_EXAMPLES)
|
||||
else
|
||||
EXAMPLES = $(PAR_EXAMPLES) $(SEQ_EXAMPLES)
|
||||
endif
|
||||
|
||||
.SUFFIXES:
|
||||
.SUFFIXES: .o .cpp .mk
|
||||
.PHONY: all clean
|
||||
.PRECIOUS: %.o
|
||||
|
||||
COMMON_O= common/PML.o common/MeshPartition.o \
|
||||
common/Utilities.o common/complex_linalg.o\
|
||||
ParDST/ParDST.o ParDST/DofMapsDST.o
|
||||
|
||||
# Remove built-in rules
|
||||
%: %.cpp
|
||||
%.o: %.cpp
|
||||
|
||||
all: $(EXAMPLES)
|
||||
|
||||
# Rules for building the EXAMPLES
|
||||
|
||||
%: $(SRC)%.cpp $(COMMON_O) $(MFEM_LIB_FILE) $(CONFIG_MK)
|
||||
$(MFEM_CXX) $(MFEM_FLAGS) $< -o $@ $(COMMON_O) $(MFEM_LIBS)
|
||||
|
||||
# Rules for compiling miniapp dependencies
|
||||
$(COMMON_O) $($(EXAMPLES)): \
|
||||
%.o: $(SRC)%.cpp $(SRC)%.hpp $(CONFIG_MK)
|
||||
$(MFEM_CXX) $(MFEM_FLAGS) -c $(<) -o $(@)
|
||||
|
||||
# Generate an error message if the MFEM library is not built and exit
|
||||
$(MFEM_LIB_FILE):
|
||||
$(error The MFEM library is not built)
|
||||
|
||||
clean:
|
||||
rm -f *.o *~ $(SEQ_EXAMPLES) $(PAR_EXAMPLES)
|
||||
rm -f DST/*.o
|
||||
rm -f ParDST/*.o
|
||||
rm -f common/*.o
|
||||
rm -f DST2D/*.o
|
||||
rm -rf *.dSYM *.TVD.*breakpoints
|
||||
rm output/*
|
||||
|
||||
|
||||
@@ -1,535 +0,0 @@
|
||||
//
|
||||
// Compile with: make maxwellp
|
||||
//
|
||||
// Sample runs: mpirun -np 4 ./maxwellp -nd 2 -nx 4 -ny 4 -sr 3 -pr 3 -k 16.0 -o 2
|
||||
// mpirun -np 4 ./maxwellp -nd 3 -nx 2 -ny 2 -nz 2 -sr 3 -pr 1 -k 2.0 -o 2
|
||||
//
|
||||
|
||||
#include "mfem.hpp"
|
||||
#include <fstream>
|
||||
#include <iostream>
|
||||
#include "ParDST/ParDST.hpp"
|
||||
|
||||
using namespace std;
|
||||
using namespace mfem;
|
||||
|
||||
void source_re(const Vector &x, Vector & f);
|
||||
void source_im(const Vector &x, Vector & f);
|
||||
void exact_re(const Vector & x, Vector & E);
|
||||
void exact_im(const Vector & x, Vector & E);
|
||||
void maxwell_solution(const Vector & x, double E[], double curl2E[]);
|
||||
double wavespeed(const Vector &x);
|
||||
void Mwavespeed(const Vector & x, DenseMatrix & M);
|
||||
|
||||
void ess_data_func(const Vector & x, Vector & E);
|
||||
|
||||
|
||||
double mu = 1.0;
|
||||
double epsilon = 1.0;
|
||||
double omega;
|
||||
int dim;
|
||||
double length = 1.0;
|
||||
Array2D<double> comp_bdr;
|
||||
Array2D<double> domain_bdr;
|
||||
bool exact_known = false;
|
||||
|
||||
int main(int argc, char *argv[])
|
||||
{
|
||||
// 1. Parse command-line options.
|
||||
int num_procs, myid;
|
||||
MPI_Init(&argc, &argv);
|
||||
MPI_Comm_size(MPI_COMM_WORLD, &num_procs);
|
||||
MPI_Comm_rank(MPI_COMM_WORLD, &myid);
|
||||
int order = 1;
|
||||
// number of serial refinements
|
||||
int ser_ref_levels = 1;
|
||||
// number of parallel refinements
|
||||
int par_ref_levels = 2;
|
||||
// number of wavelengths
|
||||
double k = 5.0; //
|
||||
bool herm_conv = true;
|
||||
bool visualization = 1;
|
||||
int nd=2;
|
||||
int nx=2;
|
||||
int ny=2;
|
||||
int nz=2;
|
||||
OptionsParser args(argc, argv);
|
||||
args.AddOption(&order, "-o", "--order",
|
||||
"Finite element order (polynomial degree).");
|
||||
args.AddOption(&nd, "-nd", "--dim","Problem space dimension");
|
||||
args.AddOption(&nx, "-nx", "--nx","Number of subdomains in x direction");
|
||||
args.AddOption(&ny, "-ny", "--ny","Number of subdomains in y direction");
|
||||
args.AddOption(&nz, "-nz", "--nz","Number of subdomains in z direction");
|
||||
args.AddOption(&ser_ref_levels, "-sr", "--ser_ref_levels",
|
||||
"Number of Serial Refinements.");
|
||||
args.AddOption(&par_ref_levels, "-pr", "--par_ref_levels",
|
||||
"Number of Parallel Refinements.");
|
||||
args.AddOption(&mu, "-mu", "--permeability",
|
||||
"Permeability of free space (or 1/(spring constant)).");
|
||||
args.AddOption(&epsilon, "-eps", "--permittivity",
|
||||
"Permittivity of free space (or mass constant).");
|
||||
args.AddOption(&k, "-k", "--wavelengths",
|
||||
"Number of wavelengths.");
|
||||
args.AddOption(&herm_conv, "-herm", "--hermitian", "-no-herm",
|
||||
"--no-hermitian", "Use convention for Hermitian operators.");
|
||||
args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
|
||||
"--no-visualization",
|
||||
"Enable or disable GLVis visualization.");
|
||||
args.Parse();
|
||||
|
||||
// check if the inputs are correct
|
||||
if (!args.Good())
|
||||
{
|
||||
if (myid == 0)
|
||||
{
|
||||
args.PrintUsage(cout);
|
||||
}
|
||||
MPI_Finalize();
|
||||
return 1;
|
||||
}
|
||||
if (myid == 0)
|
||||
{
|
||||
args.PrintOptions(cout);
|
||||
}
|
||||
|
||||
// Angular frequency
|
||||
omega = 2.0 * M_PI * k;
|
||||
|
||||
Mesh *mesh;
|
||||
|
||||
|
||||
int nel = 1;
|
||||
if (nd == 2)
|
||||
{
|
||||
mesh = new Mesh(nel, nel, Element::QUADRILATERAL, true, length, length, false);
|
||||
}
|
||||
else
|
||||
{
|
||||
mesh = new Mesh(nel, nel, nel, Element::HEXAHEDRON, true, length, length, length,false);
|
||||
}
|
||||
|
||||
dim = mesh->Dimension();
|
||||
|
||||
// 4. Refine the mesh to increase the resolution.
|
||||
for (int l = 0; l < ser_ref_levels; l++)
|
||||
{
|
||||
mesh->UniformRefinement();
|
||||
}
|
||||
|
||||
// 4. Define a parallel mesh by a partitioning of the serial mesh.
|
||||
// ParMesh *pmesh = new ParMesh(MPI_COMM_WORLD, *mesh);
|
||||
int nprocs;
|
||||
int nprocsx;
|
||||
int nprocsy;
|
||||
int nprocsz;
|
||||
if (dim == 2)
|
||||
{
|
||||
nprocs = sqrt(num_procs);
|
||||
nprocsx = nprocs;
|
||||
nprocsy = nprocs;
|
||||
nprocsz = 1;
|
||||
}
|
||||
else
|
||||
{
|
||||
nprocs = cbrt(num_procs);
|
||||
nprocsx = nprocs;
|
||||
nprocsy = nprocs;
|
||||
nprocsz = nprocs;
|
||||
}
|
||||
int nxyz[3] = {nprocsx,nprocsy,nprocsz};
|
||||
int * part = mesh->CartesianPartitioning(nxyz);
|
||||
// ParMesh *pmesh = new ParMesh(MPI_COMM_WORLD,*mesh,part);
|
||||
ParMesh *pmesh = new ParMesh(MPI_COMM_WORLD,*mesh);
|
||||
delete [] part;
|
||||
|
||||
|
||||
delete mesh;
|
||||
|
||||
for (int l = 0; l < par_ref_levels; l++)
|
||||
{
|
||||
pmesh->UniformRefinement();
|
||||
}
|
||||
|
||||
|
||||
double hl = GetUniformMeshElementSize(pmesh);
|
||||
int nrlayers = 3;
|
||||
Array2D<double> lengths(dim,2);
|
||||
lengths = hl*nrlayers;
|
||||
// lengths[0][1] = 0.0;
|
||||
// lengths[1][1] = 0.0;
|
||||
// lengths[1][0] = 0.0;
|
||||
// lengths[0][0] = 0.0;
|
||||
if (exact_known) lengths = 0.0;
|
||||
// CartesianPML pml(mesh,lengths);
|
||||
CartesianPML pml(pmesh,lengths);
|
||||
pml.SetOmega(omega);
|
||||
comp_bdr.SetSize(dim,2);
|
||||
comp_bdr = pml.GetCompDomainBdr();
|
||||
|
||||
|
||||
// 6. Define a finite element space on the mesh. Here we use the Nedelec
|
||||
// finite elements of the specified order.
|
||||
FiniteElementCollection *fec = new ND_FECollection(order, dim);
|
||||
ParFiniteElementSpace *fespace = new ParFiniteElementSpace(pmesh, fec);
|
||||
HYPRE_Int size = fespace->GlobalTrueVSize();
|
||||
|
||||
if (myid == 0)
|
||||
{
|
||||
cout << "Number of finite element unknowns: " << size << endl;
|
||||
}
|
||||
|
||||
// 7. Determine the list of true essential boundary dofs. In this example,
|
||||
// the boundary conditions are defined based on the specific mesh and the
|
||||
// problem type.
|
||||
Array<int> ess_tdof_list;
|
||||
if (pmesh->bdr_attributes.Size())
|
||||
{
|
||||
Array<int> ess_bdr(pmesh->bdr_attributes.Max());
|
||||
ess_bdr = 1;
|
||||
fespace->GetEssentialTrueDofs(ess_bdr, ess_tdof_list);
|
||||
}
|
||||
|
||||
// 8. Setup Complex Operator convention
|
||||
ComplexOperator::Convention conv =
|
||||
herm_conv ? ComplexOperator::HERMITIAN : ComplexOperator::BLOCK_SYMMETRIC;
|
||||
|
||||
// 9. Set up the linear form b(.) which corresponds to the right-hand side of
|
||||
// the FEM linear system.
|
||||
VectorFunctionCoefficient f_re(dim, source_re);
|
||||
VectorFunctionCoefficient f_im(dim, source_re);
|
||||
ParComplexLinearForm b(fespace, conv);
|
||||
b.AddDomainIntegrator(new VectorFEDomainLFIntegrator(f_re),
|
||||
new VectorFEDomainLFIntegrator(f_im));
|
||||
b.Vector::operator=(0.0);
|
||||
b.Assemble();
|
||||
|
||||
// 10. Define the solution vector x as a complex finite element grid function
|
||||
// corresponding to fespace.
|
||||
ParComplexGridFunction x(fespace);
|
||||
x = 0.0;
|
||||
// VectorFunctionCoefficient done(dim,ess_data_func);
|
||||
// x.ProjectCoefficient(done,done);
|
||||
VectorFunctionCoefficient E_re(dim,exact_re);
|
||||
VectorFunctionCoefficient E_im(dim,exact_re);
|
||||
if (exact_known)
|
||||
{
|
||||
x.ProjectCoefficient(E_re,E_re);
|
||||
}
|
||||
// 11. Set up the sesquilinear form a(.,.)
|
||||
//
|
||||
// 1/mu (1/det(J) J^T J Curl E, Curl F)
|
||||
// - omega^2 * epsilon (det(J) * (J^T J)^-1 * E, F)
|
||||
//
|
||||
FunctionCoefficient ws(wavespeed);
|
||||
|
||||
// MatrixFunctionCoefficient Mws(dim,Mwavespeed);
|
||||
|
||||
// DenseMatrix M(dim); M = 0.0;
|
||||
// M(0,0) = -pow(omega, 2);
|
||||
// M(1,1) = -pow(omega, 2);
|
||||
// M(2,2) = -pow(omega, 2);
|
||||
// MatrixConstantCoefficient Momeg(M);
|
||||
MatrixFunctionCoefficient eps_func(dim,Mwavespeed);
|
||||
|
||||
ConstantCoefficient omeg(-pow(omega, 2));
|
||||
int cdim = (dim == 2) ? 1 : dim;
|
||||
PmlMatrixCoefficient pml_c1_Re(cdim,detJ_inv_JT_J_Re, &pml);
|
||||
PmlMatrixCoefficient pml_c1_Im(cdim,detJ_inv_JT_J_Im, &pml);
|
||||
|
||||
PmlMatrixCoefficient pml_c2_Re(dim, detJ_JT_J_inv_Re,&pml);
|
||||
PmlMatrixCoefficient pml_c2_Im(dim, detJ_JT_J_inv_Im,&pml);
|
||||
ScalarMatrixProductCoefficient c2_Re0(omeg,pml_c2_Re);
|
||||
ScalarMatrixProductCoefficient c2_Im0(omeg,pml_c2_Im);
|
||||
// ScalarMatrixProductCoefficient c2_Re(ws,c2_Re0);
|
||||
// ScalarMatrixProductCoefficient c2_Im(ws,c2_Im0);
|
||||
|
||||
MatrixMatrixProductCoefficient c2_Re(c2_Re0,eps_func);
|
||||
MatrixMatrixProductCoefficient c2_Im(c2_Im0,eps_func);
|
||||
|
||||
// MatrixMatrixProductCoefficient c2_Re0(Momeg,pml_c2_Re);
|
||||
// MatrixMatrixProductCoefficient c2_Im0(Momeg,pml_c2_Im);
|
||||
// MatrixMatrixProductCoefficient c2_Re(Mws,c2_Re0);
|
||||
// MatrixMatrixProductCoefficient c2_Im(Mws,c2_Im0);
|
||||
|
||||
|
||||
ParSesquilinearForm a(fespace, conv);
|
||||
a.AddDomainIntegrator(new CurlCurlIntegrator(pml_c1_Re),
|
||||
new CurlCurlIntegrator(pml_c1_Im));
|
||||
a.AddDomainIntegrator(new VectorFEMassIntegrator(c2_Re),
|
||||
new VectorFEMassIntegrator(c2_Im));
|
||||
|
||||
a.Assemble(0);
|
||||
|
||||
OperatorHandle Ah;
|
||||
Vector B, X;
|
||||
a.FormLinearSystem(ess_tdof_list, x, b, Ah, X, B);
|
||||
|
||||
ComplexSparseMatrix * Ac = Ah.As<ComplexSparseMatrix>();
|
||||
StopWatch chrono;
|
||||
|
||||
|
||||
chrono.Clear();
|
||||
chrono.Start();
|
||||
ParDST * S = new ParDST(&a,lengths, omega, &ws, nrlayers, nx, ny, nz);
|
||||
chrono.Stop();
|
||||
double t1 = chrono.RealTime();
|
||||
|
||||
chrono.Clear();
|
||||
chrono.Start();
|
||||
// X = 0.0;
|
||||
GMRESSolver gmres(MPI_COMM_WORLD);
|
||||
// gmres.iterative_mode = true;
|
||||
gmres.SetPreconditioner(*S);
|
||||
gmres.SetOperator(*Ac);
|
||||
gmres.SetRelTol(1e-8);
|
||||
gmres.SetMaxIter(100);
|
||||
gmres.SetPrintLevel(1);
|
||||
gmres.Mult(B, X);
|
||||
delete S;
|
||||
chrono.Stop();
|
||||
double t2 = chrono.RealTime();
|
||||
|
||||
MPI_Barrier(MPI_COMM_WORLD);
|
||||
|
||||
|
||||
cout << " myid: " << myid
|
||||
<< ", setup time: " << t1
|
||||
<< ", solution time: " << t2 << endl;
|
||||
|
||||
a.RecoverFEMSolution(X, b, x);
|
||||
|
||||
|
||||
if (visualization)
|
||||
{
|
||||
char vishost[] = "localhost";
|
||||
int visport = 19916;
|
||||
string keys;
|
||||
if (dim ==2 )
|
||||
{
|
||||
keys = "keys mrRljc\n";
|
||||
}
|
||||
else
|
||||
{
|
||||
keys = "keys mc\n";
|
||||
}
|
||||
socketstream sol_sock_re(vishost, visport);
|
||||
sol_sock_re.precision(8);
|
||||
sol_sock_re << "parallel " << num_procs << " " << myid << "\n"
|
||||
<< "solution\n" << *pmesh << x.real() << keys
|
||||
<< "window_title 'E: Real Part' " << flush;
|
||||
|
||||
socketstream sol_sock_im(vishost, visport);
|
||||
sol_sock_im.precision(8);
|
||||
sol_sock_im << "parallel " << num_procs << " " << myid << "\n"
|
||||
<< "solution\n" << *pmesh << x.imag() << keys
|
||||
<< "window_title 'E: Imag Part' " << flush;
|
||||
|
||||
{
|
||||
ParGridFunction x_t(fespace);
|
||||
x_t = x.real();
|
||||
|
||||
socketstream sol_sock(vishost, visport);
|
||||
sol_sock.precision(8);
|
||||
sol_sock << "parallel " << num_procs << " " << myid << "\n"
|
||||
<< "solution\n" << *pmesh << x_t << keys << "autoscale off\n"
|
||||
<< "window_title 'Harmonic Solution (t = 0.0 T)'"
|
||||
<< "pause\n" << flush;
|
||||
|
||||
if (myid == 0)
|
||||
{
|
||||
cout << "GLVis visualization paused."
|
||||
<< " Press space (in the GLVis window) to resume it.\n";
|
||||
}
|
||||
|
||||
int num_frames = 32;
|
||||
int i = 0;
|
||||
while (sol_sock)
|
||||
{
|
||||
double t = (double)(i % num_frames) / num_frames;
|
||||
ostringstream oss;
|
||||
oss << "Harmonic Solution (t = " << t << " T)";
|
||||
|
||||
add(cos(2.0*M_PI*t), x.real(), sin(2.0*M_PI*t), x.imag(), x_t);
|
||||
sol_sock << "parallel " << num_procs << " " << myid << "\n";
|
||||
sol_sock << "solution\n" << *pmesh << x_t
|
||||
<< "window_title '" << oss.str() << "'" << flush;
|
||||
i++;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// 18. Free the used memory.
|
||||
delete fespace;
|
||||
delete fec;
|
||||
delete pmesh;
|
||||
MPI_Finalize();
|
||||
return 0;
|
||||
}
|
||||
|
||||
void source_re(const Vector &x, Vector &f)
|
||||
{
|
||||
f = 0.0;
|
||||
if (exact_known)
|
||||
{
|
||||
double E[3], curl2E[3];
|
||||
maxwell_solution(x, E, curl2E);
|
||||
// curl ( curl E) +/- omega^2 E = f
|
||||
double coeff = -omega * omega;
|
||||
f(0) = curl2E[0] + coeff * E[0];
|
||||
f(1) = curl2E[1] + coeff * E[1];
|
||||
if (dim == 2)
|
||||
{
|
||||
if (x.Size() == 3) {f(2)=0.0;}
|
||||
}
|
||||
else
|
||||
{
|
||||
f(2) = curl2E[2] + coeff * E[2];
|
||||
}
|
||||
}
|
||||
else
|
||||
{
|
||||
int nrsources = (dim == 2) ? 4 : 8;
|
||||
Vector x0(nrsources);
|
||||
Vector y0(nrsources);
|
||||
Vector z0(nrsources);
|
||||
x0(0) = 0.25; y0(0) = 0.25; z0(0) = 0.25;
|
||||
x0(1) = 0.75; y0(1) = 0.25; z0(1) = 0.25;
|
||||
x0(2) = 0.25; y0(2) = 0.75; z0(2) = 0.25;
|
||||
x0(3) = 0.75; y0(3) = 0.75; z0(3) = 0.25;
|
||||
if (dim == 3)
|
||||
{
|
||||
x0(4) = 0.25; y0(4) = 0.25; z0(4) = 0.75;
|
||||
x0(5) = 0.75; y0(5) = 0.25; z0(5) = 0.75;
|
||||
x0(6) = 0.25; y0(6) = 0.75; z0(6) = 0.75;
|
||||
x0(7) = 0.75; y0(7) = 0.75; z0(7) = 0.75;
|
||||
}
|
||||
|
||||
double n = 4.0*omega/M_PI;
|
||||
double coeff = 16.0*omega*omega/M_PI/M_PI/M_PI;
|
||||
|
||||
for (int i = 0; i<nrsources; i++)
|
||||
{
|
||||
double beta = pow(x0(i)-x(0),2) + pow(y0(i)-x(1),2);
|
||||
if (dim == 3) { beta += pow(z0(i)-x(2),2); }
|
||||
double alpha = -pow(n,2) * beta;
|
||||
f[0] += coeff*exp(alpha);
|
||||
}
|
||||
|
||||
bool in_pml = false;
|
||||
for (int i = 0; i<dim; i++)
|
||||
{
|
||||
if (x(i)<=comp_bdr(i,0) || x(i)>=comp_bdr(i,1))
|
||||
{
|
||||
in_pml = true;
|
||||
break;
|
||||
}
|
||||
}
|
||||
if (in_pml) f = 0.0;
|
||||
}
|
||||
}
|
||||
|
||||
void source_im(const Vector &x, Vector &f)
|
||||
{
|
||||
f = 0.0;
|
||||
}
|
||||
|
||||
double wavespeed(const Vector &x)
|
||||
{
|
||||
double ws;
|
||||
ws = 1.0;
|
||||
return ws;
|
||||
}
|
||||
|
||||
void Mwavespeed(const Vector & x, DenseMatrix & M)
|
||||
{
|
||||
M = 0.0;
|
||||
M(0,0) = 1.0;
|
||||
M(1,1) = 1.0;
|
||||
// M(2,2) = 4.0*x(0)-1.0;
|
||||
if (dim == 3) M(2,2) = 1.0;
|
||||
}
|
||||
|
||||
|
||||
void exact_re(const Vector & x, Vector & E)
|
||||
{
|
||||
double curl2E[3];
|
||||
maxwell_solution(x, E, curl2E);
|
||||
}
|
||||
void exact_im(const Vector & x, Vector & E)
|
||||
{
|
||||
// double curl2E[3];
|
||||
// maxwell_solution(x, E, curl2E);
|
||||
E = 0.0;
|
||||
}
|
||||
void maxwell_solution(const Vector & x, double E[], double curl2E[])
|
||||
{
|
||||
// point source
|
||||
if (dim == 2)
|
||||
{
|
||||
// shift to avoid singularity
|
||||
double x0 = x(0) + 0.1;
|
||||
double x1 = x(1) + 0.1;
|
||||
//
|
||||
double r = sqrt(x0 * x0 + x1 * x1);
|
||||
|
||||
E[0] = cos(omega * r);
|
||||
E[1] = 0.0;
|
||||
|
||||
double r_x = x0 / r;
|
||||
double r_y = x1 / r;
|
||||
double r_xy = -(r_x / r) * r_y;
|
||||
double r_yx = r_xy;
|
||||
double r_yy = (1.0 / r) * (1.0 - r_y * r_y);
|
||||
|
||||
curl2E[0] = omega * ((r_yy ) * sin(omega * r) + (omega * r_y * r_y) * cos(omega * r));
|
||||
curl2E[1] = -omega * (r_yx * sin(omega * r) + omega * r_y * r_x * cos(omega * r));
|
||||
curl2E[2] = 0.0;
|
||||
}
|
||||
else
|
||||
{
|
||||
// shift to avoid singularity
|
||||
double x0 = x(0) + 0.1;
|
||||
double x1 = x(1) + 0.1;
|
||||
double x2 = x(2) + 0.1;
|
||||
//
|
||||
double r = sqrt(x0 * x0 + x1 * x1 + x2 * x2);
|
||||
|
||||
E[0] = cos(omega * r);
|
||||
E[1] = 0.0;
|
||||
E[2] = 0.0;
|
||||
|
||||
double r_x = x0 / r;
|
||||
double r_y = x1 / r;
|
||||
double r_z = x2 / r;
|
||||
double r_xy = -(r_x / r) * r_y;
|
||||
double r_xz = -(r_x / r) * r_z;
|
||||
double r_yx = r_xy;
|
||||
double r_yy = (1.0 / r) * (1.0 - r_y * r_y);
|
||||
double r_zx = r_xz;
|
||||
double r_zz = (1.0 / r) * (1.0 - r_z * r_z);
|
||||
|
||||
curl2E[0] = omega * ((r_yy + r_zz) * sin(omega * r) +
|
||||
(omega * r_y * r_y + omega * r_z * r_z) * cos(omega * r));
|
||||
curl2E[1] = -omega * (r_yx * sin(omega * r) + omega * r_y * r_x * cos(omega * r));
|
||||
curl2E[2] = -omega * (r_zx * sin(omega * r) + omega * r_z * r_x * cos(omega * r));
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
void ess_data_func(const Vector & x, Vector & E)
|
||||
{
|
||||
E = 0.0;
|
||||
if (x(1)==0.0) E[0] = sin(x(0)+x(1));
|
||||
|
||||
bool in_pml = false;
|
||||
for (int i = 0; i<dim; i++)
|
||||
{
|
||||
if (x(i)<comp_bdr(i,0) || x(i)>comp_bdr(i,1))
|
||||
{
|
||||
in_pml = true;
|
||||
break;
|
||||
}
|
||||
}
|
||||
if (in_pml) E = 0.0;
|
||||
|
||||
}
|
||||
+171
-148
@@ -31,7 +31,7 @@ void BilinearForm::AllocMat()
|
||||
const Table &elem_dof = fes->GetElementToDofTable();
|
||||
Table dof_dof;
|
||||
|
||||
if (fbfi.Size() > 0)
|
||||
if (interior_face_integs.Size() > 0)
|
||||
{
|
||||
// the sparsity pattern is defined from the map: face->element->dof
|
||||
Table face_dof, dof_face;
|
||||
@@ -99,15 +99,15 @@ BilinearForm::BilinearForm (FiniteElementSpace * f, BilinearForm * bf, int ps)
|
||||
ext = NULL;
|
||||
|
||||
// Copy the pointers to the integrators
|
||||
dbfi = bf->dbfi;
|
||||
domain_integs = bf->domain_integs;
|
||||
|
||||
bbfi = bf->bbfi;
|
||||
bbfi_marker = bf->bbfi_marker;
|
||||
boundary_integs = bf->boundary_integs;
|
||||
boundary_integs_marker = bf->boundary_integs_marker;
|
||||
|
||||
fbfi = bf->fbfi;
|
||||
interior_face_integs = bf->interior_face_integs;
|
||||
|
||||
bfbfi = bf->bfbfi;
|
||||
bfbfi_marker = bf->bfbfi_marker;
|
||||
boundary_face_integs = bf->boundary_face_integs;
|
||||
boundary_face_integs_marker = bf->boundary_face_integs_marker;
|
||||
|
||||
AllocMat();
|
||||
}
|
||||
@@ -234,46 +234,47 @@ void BilinearForm::Finalize (int skip_zeros)
|
||||
|
||||
void BilinearForm::AddDomainIntegrator(BilinearFormIntegrator *bfi)
|
||||
{
|
||||
dbfi.Append(bfi);
|
||||
dbfi_marker.Append(NULL); // NULL marker means apply everywhere
|
||||
domain_integs.Append(bfi);
|
||||
domain_integs_marker.Append(NULL); // NULL marker means apply everywhere
|
||||
}
|
||||
|
||||
void BilinearForm::AddDomainIntegrator(BilinearFormIntegrator *bfi,
|
||||
Array<int> &elem_marker)
|
||||
{
|
||||
dbfi.Append(bfi);
|
||||
dbfi_marker.Append(&elem_marker);
|
||||
domain_integs.Append(bfi);
|
||||
domain_integs_marker.Append(&elem_marker);
|
||||
}
|
||||
|
||||
void BilinearForm::AddBoundaryIntegrator (BilinearFormIntegrator * bfi)
|
||||
{
|
||||
bbfi.Append (bfi);
|
||||
bbfi_marker.Append(NULL); // NULL marker means apply everywhere
|
||||
boundary_integs.Append (bfi);
|
||||
boundary_integs_marker.Append(NULL); // NULL marker means apply everywhere
|
||||
}
|
||||
|
||||
void BilinearForm::AddBoundaryIntegrator (BilinearFormIntegrator * bfi,
|
||||
Array<int> &bdr_marker)
|
||||
{
|
||||
bbfi.Append (bfi);
|
||||
bbfi_marker.Append(&bdr_marker);
|
||||
boundary_integs.Append (bfi);
|
||||
boundary_integs_marker.Append(&bdr_marker);
|
||||
}
|
||||
|
||||
void BilinearForm::AddInteriorFaceIntegrator(BilinearFormIntegrator * bfi)
|
||||
{
|
||||
fbfi.Append (bfi);
|
||||
interior_face_integs.Append (bfi);
|
||||
}
|
||||
|
||||
void BilinearForm::AddBdrFaceIntegrator(BilinearFormIntegrator *bfi)
|
||||
{
|
||||
bfbfi.Append(bfi);
|
||||
bfbfi_marker.Append(NULL); // NULL marker means apply everywhere
|
||||
boundary_face_integs.Append(bfi);
|
||||
// NULL marker means apply everywhere
|
||||
boundary_face_integs_marker.Append(NULL);
|
||||
}
|
||||
|
||||
void BilinearForm::AddBdrFaceIntegrator(BilinearFormIntegrator *bfi,
|
||||
Array<int> &bdr_marker)
|
||||
{
|
||||
bfbfi.Append(bfi);
|
||||
bfbfi_marker.Append(&bdr_marker);
|
||||
boundary_face_integs.Append(bfi);
|
||||
boundary_face_integs_marker.Append(&bdr_marker);
|
||||
}
|
||||
|
||||
void BilinearForm::ComputeElementMatrix(int i, DenseMatrix &elmat)
|
||||
@@ -285,14 +286,14 @@ void BilinearForm::ComputeElementMatrix(int i, DenseMatrix &elmat)
|
||||
return;
|
||||
}
|
||||
|
||||
if (dbfi.Size())
|
||||
if (domain_integs.Size())
|
||||
{
|
||||
const FiniteElement &fe = *fes->GetFE(i);
|
||||
ElementTransformation *eltrans = fes->GetElementTransformation(i);
|
||||
dbfi[0]->AssembleElementMatrix(fe, *eltrans, elmat);
|
||||
for (int k = 1; k < dbfi.Size(); k++)
|
||||
domain_integs[0]->AssembleElementMatrix(fe, *eltrans, elmat);
|
||||
for (int k = 1; k < domain_integs.Size(); k++)
|
||||
{
|
||||
dbfi[k]->AssembleElementMatrix(fe, *eltrans, elemmat);
|
||||
domain_integs[k]->AssembleElementMatrix(fe, *eltrans, elemmat);
|
||||
elmat += elemmat;
|
||||
}
|
||||
}
|
||||
@@ -306,14 +307,14 @@ void BilinearForm::ComputeElementMatrix(int i, DenseMatrix &elmat)
|
||||
|
||||
void BilinearForm::ComputeBdrElementMatrix(int i, DenseMatrix &elmat)
|
||||
{
|
||||
if (bbfi.Size())
|
||||
if (boundary_integs.Size())
|
||||
{
|
||||
const FiniteElement &be = *fes->GetBE(i);
|
||||
ElementTransformation *eltrans = fes->GetBdrElementTransformation(i);
|
||||
bbfi[0]->AssembleElementMatrix(be, *eltrans, elmat);
|
||||
for (int k = 1; k < bbfi.Size(); k++)
|
||||
boundary_integs[0]->AssembleElementMatrix(be, *eltrans, elmat);
|
||||
for (int k = 1; k < boundary_integs.Size(); k++)
|
||||
{
|
||||
bbfi[k]->AssembleElementMatrix(be, *eltrans, elemmat);
|
||||
boundary_integs[k]->AssembleElementMatrix(be, *eltrans, elemmat);
|
||||
elmat += elemmat;
|
||||
}
|
||||
}
|
||||
@@ -407,13 +408,14 @@ void BilinearForm::Assemble(int skip_zeros)
|
||||
}
|
||||
#endif
|
||||
|
||||
if (dbfi.Size())
|
||||
if (domain_integs.Size())
|
||||
{
|
||||
for (int k = 0; k < dbfi.Size(); k++)
|
||||
for (int k = 0; k < domain_integs.Size(); k++)
|
||||
{
|
||||
if (dbfi_marker[k] != NULL)
|
||||
if (domain_integs_marker[k] != NULL)
|
||||
{
|
||||
MFEM_VERIFY(mesh->attributes.Size() == dbfi_marker[k]->Size(),
|
||||
MFEM_VERIFY(mesh->attributes.Size() ==
|
||||
domain_integs_marker[k]->Size(),
|
||||
"invalid element marker for domain integrator #"
|
||||
<< k << ", counting from zero");
|
||||
}
|
||||
@@ -430,14 +432,14 @@ void BilinearForm::Assemble(int skip_zeros)
|
||||
else
|
||||
{
|
||||
elmat.SetSize(0);
|
||||
for (int k = 0; k < dbfi.Size(); k++)
|
||||
for (int k = 0; k < domain_integs.Size(); k++)
|
||||
{
|
||||
if ( dbfi_marker[k] == NULL ||
|
||||
(*(dbfi_marker[k]))[elem_attr-1] == 1)
|
||||
if ( domain_integs_marker[k] == NULL ||
|
||||
(*(domain_integs_marker[k]))[elem_attr-1] == 1)
|
||||
{
|
||||
const FiniteElement &fe = *fes->GetFE(i);
|
||||
eltrans = fes->GetElementTransformation(i);
|
||||
dbfi[k]->AssembleElementMatrix(fe, *eltrans, elemmat);
|
||||
domain_integs[k]->AssembleElementMatrix(fe, *eltrans, elemmat);
|
||||
if (elmat.Size() == 0)
|
||||
{
|
||||
elmat = elemmat;
|
||||
@@ -472,20 +474,20 @@ void BilinearForm::Assemble(int skip_zeros)
|
||||
}
|
||||
}
|
||||
|
||||
if (bbfi.Size())
|
||||
if (boundary_integs.Size())
|
||||
{
|
||||
// Which boundary attributes need to be processed?
|
||||
Array<int> bdr_attr_marker(mesh->bdr_attributes.Size() ?
|
||||
mesh->bdr_attributes.Max() : 0);
|
||||
bdr_attr_marker = 0;
|
||||
for (int k = 0; k < bbfi.Size(); k++)
|
||||
for (int k = 0; k < boundary_integs.Size(); k++)
|
||||
{
|
||||
if (bbfi_marker[k] == NULL)
|
||||
if (boundary_integs_marker[k] == NULL)
|
||||
{
|
||||
bdr_attr_marker = 1;
|
||||
break;
|
||||
}
|
||||
Array<int> &bdr_marker = *bbfi_marker[k];
|
||||
Array<int> &bdr_marker = *boundary_integs_marker[k];
|
||||
MFEM_ASSERT(bdr_marker.Size() == bdr_attr_marker.Size(),
|
||||
"invalid boundary marker for boundary integrator #"
|
||||
<< k << ", counting from zero");
|
||||
@@ -504,21 +506,21 @@ void BilinearForm::Assemble(int skip_zeros)
|
||||
fes -> GetBdrElementVDofs (i, vdofs);
|
||||
eltrans = fes -> GetBdrElementTransformation (i);
|
||||
int k = 0;
|
||||
for (; k < bbfi.Size(); k++)
|
||||
for (; k < boundary_integs.Size(); k++)
|
||||
{
|
||||
if (bbfi_marker[k] &&
|
||||
(*bbfi_marker[k])[bdr_attr-1] == 0) { continue; }
|
||||
if (boundary_integs_marker[k] &&
|
||||
(*boundary_integs_marker[k])[bdr_attr-1] == 0) { continue; }
|
||||
|
||||
bbfi[k]->AssembleElementMatrix(be, *eltrans, elmat);
|
||||
boundary_integs[k]->AssembleElementMatrix(be, *eltrans, elmat);
|
||||
k++;
|
||||
break;
|
||||
}
|
||||
for (; k < bbfi.Size(); k++)
|
||||
for (; k < boundary_integs.Size(); k++)
|
||||
{
|
||||
if (bbfi_marker[k] &&
|
||||
(*bbfi_marker[k])[bdr_attr-1] == 0) { continue; }
|
||||
if (boundary_integs_marker[k] &&
|
||||
(*boundary_integs_marker[k])[bdr_attr-1] == 0) { continue; }
|
||||
|
||||
bbfi[k]->AssembleElementMatrix(be, *eltrans, elemmat);
|
||||
boundary_integs[k]->AssembleElementMatrix(be, *eltrans, elemmat);
|
||||
elmat += elemmat;
|
||||
}
|
||||
if (!static_cond)
|
||||
@@ -536,7 +538,7 @@ void BilinearForm::Assemble(int skip_zeros)
|
||||
}
|
||||
}
|
||||
|
||||
if (fbfi.Size())
|
||||
if (interior_face_integs.Size())
|
||||
{
|
||||
FaceElementTransformations *tr;
|
||||
Array<int> vdofs2;
|
||||
@@ -550,18 +552,19 @@ void BilinearForm::Assemble(int skip_zeros)
|
||||
fes -> GetElementVDofs (tr -> Elem1No, vdofs);
|
||||
fes -> GetElementVDofs (tr -> Elem2No, vdofs2);
|
||||
vdofs.Append (vdofs2);
|
||||
for (int k = 0; k < fbfi.Size(); k++)
|
||||
for (int k = 0; k < interior_face_integs.Size(); k++)
|
||||
{
|
||||
fbfi[k] -> AssembleFaceMatrix (*fes -> GetFE (tr -> Elem1No),
|
||||
*fes -> GetFE (tr -> Elem2No),
|
||||
*tr, elemmat);
|
||||
interior_face_integs[k]->
|
||||
AssembleFaceMatrix(*fes->GetFE(tr->Elem1No),
|
||||
*fes->GetFE(tr->Elem2No),
|
||||
*tr, elemmat);
|
||||
mat -> AddSubMatrix (vdofs, vdofs, elemmat, skip_zeros);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
if (bfbfi.Size())
|
||||
if (boundary_face_integs.Size())
|
||||
{
|
||||
FaceElementTransformations *tr;
|
||||
const FiniteElement *fe1, *fe2;
|
||||
@@ -570,14 +573,14 @@ void BilinearForm::Assemble(int skip_zeros)
|
||||
Array<int> bdr_attr_marker(mesh->bdr_attributes.Size() ?
|
||||
mesh->bdr_attributes.Max() : 0);
|
||||
bdr_attr_marker = 0;
|
||||
for (int k = 0; k < bfbfi.Size(); k++)
|
||||
for (int k = 0; k < boundary_face_integs.Size(); k++)
|
||||
{
|
||||
if (bfbfi_marker[k] == NULL)
|
||||
if (boundary_face_integs_marker[k] == NULL)
|
||||
{
|
||||
bdr_attr_marker = 1;
|
||||
break;
|
||||
}
|
||||
Array<int> &bdr_marker = *bfbfi_marker[k];
|
||||
Array<int> &bdr_marker = *boundary_face_integs_marker[k];
|
||||
MFEM_ASSERT(bdr_marker.Size() == bdr_attr_marker.Size(),
|
||||
"invalid boundary marker for boundary face integrator #"
|
||||
<< k << ", counting from zero");
|
||||
@@ -601,12 +604,14 @@ void BilinearForm::Assemble(int skip_zeros)
|
||||
// but we can't dereference a NULL pointer, and we don't want to
|
||||
// actually make a fake element.
|
||||
fe2 = fe1;
|
||||
for (int k = 0; k < bfbfi.Size(); k++)
|
||||
for (int k = 0; k < boundary_face_integs.Size(); k++)
|
||||
{
|
||||
if (bfbfi_marker[k] &&
|
||||
(*bfbfi_marker[k])[bdr_attr-1] == 0) { continue; }
|
||||
if (boundary_face_integs_marker[k] &&
|
||||
(*boundary_face_integs_marker[k])[bdr_attr-1] == 0)
|
||||
{ continue; }
|
||||
|
||||
bfbfi[k] -> AssembleFaceMatrix (*fe1, *fe2, *tr, elemmat);
|
||||
boundary_face_integs[k] -> AssembleFaceMatrix (*fe1, *fe2, *tr,
|
||||
elemmat);
|
||||
mat -> AddSubMatrix (vdofs, vdofs, elemmat, skip_zeros);
|
||||
}
|
||||
}
|
||||
@@ -857,7 +862,7 @@ void BilinearForm::RecoverFEMSolution(const Vector &X,
|
||||
|
||||
void BilinearForm::ComputeElementMatrices()
|
||||
{
|
||||
if (element_matrices || dbfi.Size() == 0 || fes->GetNE() == 0)
|
||||
if (element_matrices || domain_integs.Size() == 0 || fes->GetNE() == 0)
|
||||
{
|
||||
return;
|
||||
}
|
||||
@@ -886,11 +891,11 @@ void BilinearForm::ComputeElementMatrices()
|
||||
#endif
|
||||
fes->GetElementTransformation(i, &eltrans);
|
||||
|
||||
dbfi[0]->AssembleElementMatrix(fe, eltrans, elmat);
|
||||
for (int k = 1; k < dbfi.Size(); k++)
|
||||
domain_integs[0]->AssembleElementMatrix(fe, eltrans, elmat);
|
||||
for (int k = 1; k < domain_integs.Size(); k++)
|
||||
{
|
||||
// note: some integrators may not be thread-safe
|
||||
dbfi[k]->AssembleElementMatrix(fe, eltrans, tmp);
|
||||
domain_integs[k]->AssembleElementMatrix(fe, eltrans, tmp);
|
||||
elmat += tmp;
|
||||
}
|
||||
elmat.ClearExternalData();
|
||||
@@ -1105,10 +1110,12 @@ BilinearForm::~BilinearForm()
|
||||
if (!extern_bfs)
|
||||
{
|
||||
int k;
|
||||
for (k=0; k < dbfi.Size(); k++) { delete dbfi[k]; }
|
||||
for (k=0; k < bbfi.Size(); k++) { delete bbfi[k]; }
|
||||
for (k=0; k < fbfi.Size(); k++) { delete fbfi[k]; }
|
||||
for (k=0; k < bfbfi.Size(); k++) { delete bfbfi[k]; }
|
||||
for (k=0; k < domain_integs.Size(); k++) { delete domain_integs[k]; }
|
||||
for (k=0; k < boundary_integs.Size(); k++) { delete boundary_integs[k]; }
|
||||
for (k=0; k < interior_face_integs.Size(); k++)
|
||||
{ delete interior_face_integs[k]; }
|
||||
for (k=0; k < boundary_face_integs.Size(); k++)
|
||||
{ delete boundary_face_integs[k]; }
|
||||
}
|
||||
|
||||
delete ext;
|
||||
@@ -1141,13 +1148,13 @@ MixedBilinearForm::MixedBilinearForm (FiniteElementSpace *tr_fes,
|
||||
ext = NULL;
|
||||
|
||||
// Copy the pointers to the integrators
|
||||
dbfi = mbf->dbfi;
|
||||
bbfi = mbf->bbfi;
|
||||
tfbfi = mbf->tfbfi;
|
||||
btfbfi = mbf->btfbfi;
|
||||
domain_integs = mbf->domain_integs;
|
||||
boundary_integs = mbf->boundary_integs;
|
||||
trace_face_integs = mbf->trace_face_integs;
|
||||
boundary_trace_face_integs = mbf->boundary_trace_face_integs;
|
||||
|
||||
bbfi_marker = mbf->bbfi_marker;
|
||||
btfbfi_marker = mbf->btfbfi_marker;
|
||||
boundary_integs_marker = mbf->boundary_integs_marker;
|
||||
boundary_trace_face_integs_marker = mbf->boundary_trace_face_integs_marker;
|
||||
|
||||
assembly = AssemblyLevel::LEGACY;
|
||||
ext = NULL;
|
||||
@@ -1236,7 +1243,8 @@ MatrixInverse * MixedBilinearForm::Inverse() const
|
||||
{
|
||||
if (assembly != AssemblyLevel::LEGACY)
|
||||
{
|
||||
MFEM_WARNING("MixedBilinearForm::Inverse not possible with this assembly level!");
|
||||
MFEM_WARNING("MixedBilinearForm::Inverse not possible with this "
|
||||
"assembly level!");
|
||||
return NULL;
|
||||
}
|
||||
else
|
||||
@@ -1267,38 +1275,39 @@ void MixedBilinearForm::GetBlocks(Array2D<SparseMatrix *> &blocks) const
|
||||
|
||||
void MixedBilinearForm::AddDomainIntegrator (BilinearFormIntegrator * bfi)
|
||||
{
|
||||
dbfi.Append (bfi);
|
||||
domain_integs.Append (bfi);
|
||||
}
|
||||
|
||||
void MixedBilinearForm::AddBoundaryIntegrator (BilinearFormIntegrator * bfi)
|
||||
{
|
||||
bbfi.Append (bfi);
|
||||
bbfi_marker.Append(NULL); // NULL marker means apply everywhere
|
||||
boundary_integs.Append (bfi);
|
||||
boundary_integs_marker.Append(NULL); // NULL marker means apply everywhere
|
||||
}
|
||||
|
||||
void MixedBilinearForm::AddBoundaryIntegrator (BilinearFormIntegrator * bfi,
|
||||
Array<int> &bdr_marker)
|
||||
{
|
||||
bbfi.Append (bfi);
|
||||
bbfi_marker.Append(&bdr_marker);
|
||||
boundary_integs.Append (bfi);
|
||||
boundary_integs_marker.Append(&bdr_marker);
|
||||
}
|
||||
|
||||
void MixedBilinearForm::AddTraceFaceIntegrator (BilinearFormIntegrator * bfi)
|
||||
{
|
||||
tfbfi.Append (bfi);
|
||||
trace_face_integs.Append (bfi);
|
||||
}
|
||||
|
||||
void MixedBilinearForm::AddBdrTraceFaceIntegrator(BilinearFormIntegrator *bfi)
|
||||
{
|
||||
btfbfi.Append(bfi);
|
||||
btfbfi_marker.Append(NULL); // NULL marker means apply everywhere
|
||||
boundary_trace_face_integs.Append(bfi);
|
||||
// NULL marker means apply everywhere
|
||||
boundary_trace_face_integs_marker.Append(NULL);
|
||||
}
|
||||
|
||||
void MixedBilinearForm::AddBdrTraceFaceIntegrator(BilinearFormIntegrator *bfi,
|
||||
Array<int> &bdr_marker)
|
||||
{
|
||||
btfbfi.Append(bfi);
|
||||
btfbfi_marker.Append(&bdr_marker);
|
||||
boundary_trace_face_integs.Append(bfi);
|
||||
boundary_trace_face_integs_marker.Append(&bdr_marker);
|
||||
}
|
||||
|
||||
void MixedBilinearForm::Assemble (int skip_zeros)
|
||||
@@ -1320,37 +1329,37 @@ void MixedBilinearForm::Assemble (int skip_zeros)
|
||||
mat = new SparseMatrix(height, width);
|
||||
}
|
||||
|
||||
if (dbfi.Size())
|
||||
if (domain_integs.Size())
|
||||
{
|
||||
for (int i = 0; i < test_fes -> GetNE(); i++)
|
||||
{
|
||||
trial_fes -> GetElementVDofs (i, tr_vdofs);
|
||||
test_fes -> GetElementVDofs (i, te_vdofs);
|
||||
eltrans = test_fes -> GetElementTransformation (i);
|
||||
for (int k = 0; k < dbfi.Size(); k++)
|
||||
for (int k = 0; k < domain_integs.Size(); k++)
|
||||
{
|
||||
dbfi[k] -> AssembleElementMatrix2 (*trial_fes -> GetFE(i),
|
||||
*test_fes -> GetFE(i),
|
||||
*eltrans, elemmat);
|
||||
domain_integs[k] -> AssembleElementMatrix2 (*trial_fes -> GetFE(i),
|
||||
*test_fes -> GetFE(i),
|
||||
*eltrans, elemmat);
|
||||
mat -> AddSubMatrix (te_vdofs, tr_vdofs, elemmat, skip_zeros);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
if (bbfi.Size())
|
||||
if (boundary_integs.Size())
|
||||
{
|
||||
// Which boundary attributes need to be processed?
|
||||
Array<int> bdr_attr_marker(mesh->bdr_attributes.Size() ?
|
||||
mesh->bdr_attributes.Max() : 0);
|
||||
bdr_attr_marker = 0;
|
||||
for (int k = 0; k < bbfi.Size(); k++)
|
||||
for (int k = 0; k < boundary_integs.Size(); k++)
|
||||
{
|
||||
if (bbfi_marker[k] == NULL)
|
||||
if (boundary_integs_marker[k] == NULL)
|
||||
{
|
||||
bdr_attr_marker = 1;
|
||||
break;
|
||||
}
|
||||
Array<int> &bdr_marker = *bbfi_marker[k];
|
||||
Array<int> &bdr_marker = *boundary_integs_marker[k];
|
||||
MFEM_ASSERT(bdr_marker.Size() == bdr_attr_marker.Size(),
|
||||
"invalid boundary marker for boundary integrator #"
|
||||
<< k << ", counting from zero");
|
||||
@@ -1368,20 +1377,20 @@ void MixedBilinearForm::Assemble (int skip_zeros)
|
||||
trial_fes -> GetBdrElementVDofs (i, tr_vdofs);
|
||||
test_fes -> GetBdrElementVDofs (i, te_vdofs);
|
||||
eltrans = test_fes -> GetBdrElementTransformation (i);
|
||||
for (int k = 0; k < bbfi.Size(); k++)
|
||||
for (int k = 0; k < boundary_integs.Size(); k++)
|
||||
{
|
||||
if (bbfi_marker[k] &&
|
||||
(*bbfi_marker[k])[bdr_attr-1] == 0) { continue; }
|
||||
if (boundary_integs_marker[k] &&
|
||||
(*boundary_integs_marker[k])[bdr_attr-1] == 0) { continue; }
|
||||
|
||||
bbfi[k] -> AssembleElementMatrix2 (*trial_fes -> GetBE(i),
|
||||
*test_fes -> GetBE(i),
|
||||
*eltrans, elemmat);
|
||||
boundary_integs[k]->AssembleElementMatrix2 (*trial_fes -> GetBE(i),
|
||||
*test_fes -> GetBE(i),
|
||||
*eltrans, elemmat);
|
||||
mat -> AddSubMatrix (te_vdofs, tr_vdofs, elemmat, skip_zeros);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
if (tfbfi.Size())
|
||||
if (trace_face_integs.Size())
|
||||
{
|
||||
FaceElementTransformations *ftr;
|
||||
Array<int> te_vdofs2;
|
||||
@@ -1408,16 +1417,16 @@ void MixedBilinearForm::Assemble (int skip_zeros)
|
||||
// want to actually make a fake element.
|
||||
test_fe2 = test_fe1;
|
||||
}
|
||||
for (int k = 0; k < tfbfi.Size(); k++)
|
||||
for (int k = 0; k < trace_face_integs.Size(); k++)
|
||||
{
|
||||
tfbfi[k]->AssembleFaceMatrix(*trial_face_fe, *test_fe1, *test_fe2,
|
||||
*ftr, elemmat);
|
||||
trace_face_integs[k]->AssembleFaceMatrix(*trial_face_fe, *test_fe1,
|
||||
*test_fe2, *ftr, elemmat);
|
||||
mat->AddSubMatrix(te_vdofs, tr_vdofs, elemmat, skip_zeros);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
if (btfbfi.Size())
|
||||
if (boundary_trace_face_integs.Size())
|
||||
{
|
||||
FaceElementTransformations *ftr;
|
||||
Array<int> te_vdofs2;
|
||||
@@ -1427,17 +1436,17 @@ void MixedBilinearForm::Assemble (int skip_zeros)
|
||||
Array<int> bdr_attr_marker(mesh->bdr_attributes.Size() ?
|
||||
mesh->bdr_attributes.Max() : 0);
|
||||
bdr_attr_marker = 0;
|
||||
for (int k = 0; k < btfbfi.Size(); k++)
|
||||
for (int k = 0; k < boundary_trace_face_integs.Size(); k++)
|
||||
{
|
||||
if (btfbfi_marker[k] == NULL)
|
||||
if (boundary_trace_face_integs_marker[k] == NULL)
|
||||
{
|
||||
bdr_attr_marker = 1;
|
||||
break;
|
||||
}
|
||||
Array<int> &bdr_marker = *btfbfi_marker[k];
|
||||
Array<int> &bdr_marker = *boundary_trace_face_integs_marker[k];
|
||||
MFEM_ASSERT(bdr_marker.Size() == bdr_attr_marker.Size(),
|
||||
"invalid boundary marker for boundary trace face integrator #"
|
||||
<< k << ", counting from zero");
|
||||
"invalid boundary marker for boundary trace face"
|
||||
"integrator #" << k << ", counting from zero");
|
||||
for (int i = 0; i < bdr_attr_marker.Size(); i++)
|
||||
{
|
||||
bdr_attr_marker[i] |= bdr_marker[i];
|
||||
@@ -1460,13 +1469,16 @@ void MixedBilinearForm::Assemble (int skip_zeros)
|
||||
// boundaries, but we can't dereference a NULL pointer, and we don't
|
||||
// want to actually make a fake element.
|
||||
test_fe2 = test_fe1;
|
||||
for (int k = 0; k < btfbfi.Size(); k++)
|
||||
for (int k = 0; k < boundary_trace_face_integs.Size(); k++)
|
||||
{
|
||||
if (btfbfi_marker[k] &&
|
||||
(*btfbfi_marker[k])[bdr_attr-1] == 0) { continue; }
|
||||
if (boundary_trace_face_integs_marker[k] &&
|
||||
(*boundary_trace_face_integs_marker[k])[bdr_attr-1] == 0)
|
||||
{ continue; }
|
||||
|
||||
btfbfi[k]->AssembleFaceMatrix(*trial_face_fe, *test_fe1, *test_fe2,
|
||||
*ftr, elemmat);
|
||||
boundary_trace_face_integs[k]->AssembleFaceMatrix(*trial_face_fe,
|
||||
*test_fe1,
|
||||
*test_fe2,
|
||||
*ftr, elemmat);
|
||||
mat->AddSubMatrix(te_vdofs, tr_vdofs, elemmat, skip_zeros);
|
||||
}
|
||||
}
|
||||
@@ -1557,15 +1569,17 @@ void MixedBilinearForm::ConformingAssemble()
|
||||
|
||||
void MixedBilinearForm::ComputeElementMatrix(int i, DenseMatrix &elmat)
|
||||
{
|
||||
if (dbfi.Size())
|
||||
if (domain_integs.Size())
|
||||
{
|
||||
const FiniteElement &trial_fe = *trial_fes->GetFE(i);
|
||||
const FiniteElement &test_fe = *test_fes->GetFE(i);
|
||||
ElementTransformation *eltrans = test_fes->GetElementTransformation(i);
|
||||
dbfi[0]->AssembleElementMatrix2(trial_fe, test_fe, *eltrans, elmat);
|
||||
for (int k = 1; k < dbfi.Size(); k++)
|
||||
domain_integs[0]->AssembleElementMatrix2(trial_fe, test_fe, *eltrans,
|
||||
elmat);
|
||||
for (int k = 1; k < domain_integs.Size(); k++)
|
||||
{
|
||||
dbfi[k]->AssembleElementMatrix2(trial_fe, test_fe, *eltrans, elemmat);
|
||||
domain_integs[k]->AssembleElementMatrix2(trial_fe, test_fe, *eltrans,
|
||||
elemmat);
|
||||
elmat += elemmat;
|
||||
}
|
||||
}
|
||||
@@ -1580,15 +1594,17 @@ void MixedBilinearForm::ComputeElementMatrix(int i, DenseMatrix &elmat)
|
||||
|
||||
void MixedBilinearForm::ComputeBdrElementMatrix(int i, DenseMatrix &elmat)
|
||||
{
|
||||
if (bbfi.Size())
|
||||
if (boundary_integs.Size())
|
||||
{
|
||||
const FiniteElement &trial_be = *trial_fes->GetBE(i);
|
||||
const FiniteElement &test_be = *test_fes->GetBE(i);
|
||||
ElementTransformation *eltrans = test_fes->GetBdrElementTransformation(i);
|
||||
bbfi[0]->AssembleElementMatrix2(trial_be, test_be, *eltrans, elmat);
|
||||
for (int k = 1; k < bbfi.Size(); k++)
|
||||
boundary_integs[0]->AssembleElementMatrix2(trial_be, test_be, *eltrans,
|
||||
elmat);
|
||||
for (int k = 1; k < boundary_integs.Size(); k++)
|
||||
{
|
||||
bbfi[k]->AssembleElementMatrix2(trial_be, test_be, *eltrans, elemmat);
|
||||
boundary_integs[k]->AssembleElementMatrix2(trial_be, test_be, *eltrans,
|
||||
elemmat);
|
||||
elmat += elemmat;
|
||||
}
|
||||
}
|
||||
@@ -1688,10 +1704,10 @@ void MixedBilinearForm::EliminateTestDofs (const Array<int> &bdr_attr_is_ess)
|
||||
}
|
||||
}
|
||||
|
||||
void MixedBilinearForm::FormRectangularSystemMatrix(const Array<int>
|
||||
&trial_tdof_list,
|
||||
const Array<int> &test_tdof_list,
|
||||
OperatorHandle &A)
|
||||
void MixedBilinearForm::FormRectangularSystemMatrix(
|
||||
const Array<int> &trial_tdof_list,
|
||||
const Array<int> &test_tdof_list,
|
||||
OperatorHandle &A)
|
||||
|
||||
{
|
||||
if (ext)
|
||||
@@ -1729,17 +1745,17 @@ void MixedBilinearForm::FormRectangularSystemMatrix(const Array<int>
|
||||
A.Reset(mat, false);
|
||||
}
|
||||
|
||||
void MixedBilinearForm::FormRectangularLinearSystem(const Array<int>
|
||||
&trial_tdof_list,
|
||||
const Array<int> &test_tdof_list,
|
||||
Vector &x, Vector &b,
|
||||
OperatorHandle &A,
|
||||
Vector &X, Vector &B)
|
||||
void MixedBilinearForm::FormRectangularLinearSystem(
|
||||
const Array<int> &trial_tdof_list,
|
||||
const Array<int> &test_tdof_list,
|
||||
Vector &x, Vector &b,
|
||||
OperatorHandle &A,
|
||||
Vector &X, Vector &B)
|
||||
{
|
||||
if (ext)
|
||||
{
|
||||
ext->FormRectangularLinearSystem(trial_tdof_list, test_tdof_list, x, b, A, X,
|
||||
B);
|
||||
ext->FormRectangularLinearSystem(trial_tdof_list, test_tdof_list,
|
||||
x, b, A, X, B);
|
||||
return;
|
||||
}
|
||||
|
||||
@@ -1777,10 +1793,13 @@ MixedBilinearForm::~MixedBilinearForm()
|
||||
if (!extern_bfs)
|
||||
{
|
||||
int i;
|
||||
for (i = 0; i < dbfi.Size(); i++) { delete dbfi[i]; }
|
||||
for (i = 0; i < bbfi.Size(); i++) { delete bbfi[i]; }
|
||||
for (i = 0; i < tfbfi.Size(); i++) { delete tfbfi[i]; }
|
||||
for (i = 0; i < btfbfi.Size(); i++) { delete btfbfi[i]; }
|
||||
for (i = 0; i < domain_integs.Size(); i++) { delete domain_integs[i]; }
|
||||
for (i = 0; i < boundary_integs.Size(); i++)
|
||||
{ delete boundary_integs[i]; }
|
||||
for (i = 0; i < trace_face_integs.Size(); i++)
|
||||
{ delete trace_face_integs[i]; }
|
||||
for (i = 0; i < boundary_trace_face_integs.Size(); i++)
|
||||
{ delete boundary_trace_face_integs[i]; }
|
||||
}
|
||||
delete ext;
|
||||
}
|
||||
@@ -1830,7 +1849,7 @@ void DiscreteLinearOperator::Assemble(int skip_zeros)
|
||||
mat = new SparseMatrix(height, width);
|
||||
}
|
||||
|
||||
if (dbfi.Size() > 0)
|
||||
if (domain_integs.Size() > 0)
|
||||
{
|
||||
for (int i = 0; i < test_fes->GetNE(); i++)
|
||||
{
|
||||
@@ -1840,17 +1859,19 @@ void DiscreteLinearOperator::Assemble(int skip_zeros)
|
||||
dom_fe = trial_fes->GetFE(i);
|
||||
ran_fe = test_fes->GetFE(i);
|
||||
|
||||
dbfi[0]->AssembleElementMatrix2(*dom_fe, *ran_fe, *T, totelmat);
|
||||
for (int j = 1; j < dbfi.Size(); j++)
|
||||
domain_integs[0]->AssembleElementMatrix2(*dom_fe, *ran_fe, *T,
|
||||
totelmat);
|
||||
for (int j = 1; j < domain_integs.Size(); j++)
|
||||
{
|
||||
dbfi[j]->AssembleElementMatrix2(*dom_fe, *ran_fe, *T, elmat);
|
||||
domain_integs[j]->AssembleElementMatrix2(*dom_fe, *ran_fe, *T,
|
||||
elmat);
|
||||
totelmat += elmat;
|
||||
}
|
||||
mat->SetSubMatrix(ran_vdofs, dom_vdofs, totelmat, skip_zeros);
|
||||
}
|
||||
}
|
||||
|
||||
if (tfbfi.Size())
|
||||
if (trace_face_integs.Size())
|
||||
{
|
||||
const int nfaces = test_fes->GetMesh()->GetNumFaces();
|
||||
for (int i = 0; i < nfaces; i++)
|
||||
@@ -1861,10 +1882,12 @@ void DiscreteLinearOperator::Assemble(int skip_zeros)
|
||||
dom_fe = trial_fes->GetFaceElement(i);
|
||||
ran_fe = test_fes->GetFaceElement(i);
|
||||
|
||||
tfbfi[0]->AssembleElementMatrix2(*dom_fe, *ran_fe, *T, totelmat);
|
||||
for (int j = 1; j < tfbfi.Size(); j++)
|
||||
trace_face_integs[0]->AssembleElementMatrix2(*dom_fe, *ran_fe, *T,
|
||||
totelmat);
|
||||
for (int j = 1; j < trace_face_integs.Size(); j++)
|
||||
{
|
||||
tfbfi[j]->AssembleElementMatrix2(*dom_fe, *ran_fe, *T, elmat);
|
||||
trace_face_integs[j]->AssembleElementMatrix2(*dom_fe, *ran_fe, *T,
|
||||
elmat);
|
||||
totelmat += elmat;
|
||||
}
|
||||
mat->SetSubMatrix(ran_vdofs, dom_vdofs, totelmat, skip_zeros);
|
||||
|
||||
+36
-30
@@ -84,28 +84,29 @@ protected:
|
||||
the BilinearForm. */
|
||||
long sequence;
|
||||
|
||||
/** @brief Indicates the BilinearFormIntegrator%s stored in #dbfi, #bbfi,
|
||||
#fbfi, and #bfbfi are owned by another BilinearForm. */
|
||||
/** @brief Indicates the BilinearFormIntegrator%s stored in #domain_integs,
|
||||
#boundary_integs, #interior_face_integs, and #boundary_face_integs are
|
||||
owned by another BilinearForm. */
|
||||
int extern_bfs;
|
||||
|
||||
/// Set of Domain Integrators to be applied.
|
||||
Array<BilinearFormIntegrator*> dbfi;
|
||||
Array<BilinearFormIntegrator*> domain_integs;
|
||||
/// Element attribute marker (should be of length mesh->attributes)
|
||||
/// Includes all by default.
|
||||
/// 0 - ignore attribute
|
||||
/// 1 - include attribute
|
||||
Array<Array<int>*> dbfi_marker;
|
||||
Array<Array<int>*> domain_integs_marker;
|
||||
|
||||
/// Set of Boundary Integrators to be applied.
|
||||
Array<BilinearFormIntegrator*> bbfi;
|
||||
Array<Array<int>*> bbfi_marker; ///< Entries are not owned.
|
||||
Array<BilinearFormIntegrator*> boundary_integs;
|
||||
Array<Array<int>*> boundary_integs_marker; ///< Entries are not owned.
|
||||
|
||||
/// Set of interior face Integrators to be applied.
|
||||
Array<BilinearFormIntegrator*> fbfi;
|
||||
Array<BilinearFormIntegrator*> interior_face_integs;
|
||||
|
||||
/// Set of boundary face Integrators to be applied.
|
||||
Array<BilinearFormIntegrator*> bfbfi;
|
||||
Array<Array<int>*> bfbfi_marker; ///< Entries are not owned.
|
||||
Array<BilinearFormIntegrator*> boundary_face_integs;
|
||||
Array<Array<int>*> boundary_face_integs_marker; ///< Entries are not owned.
|
||||
|
||||
DenseMatrix elemmat;
|
||||
Array<int> vdofs;
|
||||
@@ -231,24 +232,25 @@ public:
|
||||
void AllocateMatrix() { if (mat == NULL) { AllocMat(); } }
|
||||
|
||||
/// Access all the integrators added with AddDomainIntegrator().
|
||||
Array<BilinearFormIntegrator*> *GetDBFI() { return &dbfi; }
|
||||
Array<BilinearFormIntegrator*> *GetDBFI() { return &domain_integs; }
|
||||
|
||||
/// Access all the integrators added with AddBoundaryIntegrator().
|
||||
Array<BilinearFormIntegrator*> *GetBBFI() { return &bbfi; }
|
||||
Array<BilinearFormIntegrator*> *GetBBFI() { return &boundary_integs; }
|
||||
/** @brief Access all boundary markers added with AddBoundaryIntegrator().
|
||||
If no marker was specified when the integrator was added, the
|
||||
corresponding pointer (to Array<int>) will be NULL. */
|
||||
Array<Array<int>*> *GetBBFI_Marker() { return &bbfi_marker; }
|
||||
Array<Array<int>*> *GetBBFI_Marker() { return &boundary_integs_marker; }
|
||||
|
||||
/// Access all integrators added with AddInteriorFaceIntegrator().
|
||||
Array<BilinearFormIntegrator*> *GetFBFI() { return &fbfi; }
|
||||
Array<BilinearFormIntegrator*> *GetFBFI() { return &interior_face_integs; }
|
||||
|
||||
/// Access all integrators added with AddBdrFaceIntegrator().
|
||||
Array<BilinearFormIntegrator*> *GetBFBFI() { return &bfbfi; }
|
||||
Array<BilinearFormIntegrator*> *GetBFBFI() { return &boundary_face_integs; }
|
||||
/** @brief Access all boundary markers added with AddBdrFaceIntegrator().
|
||||
If no marker was specified when the integrator was added, the
|
||||
corresponding pointer (to Array<int>) will be NULL. */
|
||||
Array<Array<int>*> *GetBFBFI_Marker() { return &bfbfi_marker; }
|
||||
Array<Array<int>*> *GetBFBFI_Marker()
|
||||
{ return &boundary_face_integs_marker; }
|
||||
|
||||
/// Returns a reference to: \f$ M_{ij} \f$
|
||||
const double &operator()(int i, int j) { return (*mat)(i,j); }
|
||||
@@ -652,23 +654,25 @@ protected:
|
||||
Partial Assembly (PA), or Matrix Free assembly (MF). */
|
||||
MixedBilinearFormExtension *ext;
|
||||
|
||||
/** @brief Indicates the BilinearFormIntegrator%s stored in #dbfi, #bbfi,
|
||||
#tfbfi and #btfbfi are owned by another MixedBilinearForm. */
|
||||
/** @brief Indicates the BilinearFormIntegrator%s stored in #domain_integs,
|
||||
#boundary_integs, #trace_face_integs and #boundary_trace_face_integs
|
||||
are owned by another MixedBilinearForm. */
|
||||
int extern_bfs;
|
||||
|
||||
/// Domain integrators.
|
||||
Array<BilinearFormIntegrator*> dbfi;
|
||||
Array<BilinearFormIntegrator*> domain_integs;
|
||||
|
||||
/// Boundary integrators.
|
||||
Array<BilinearFormIntegrator*> bbfi;
|
||||
Array<Array<int>*> bbfi_marker;///< Entries are not owned.
|
||||
Array<BilinearFormIntegrator*> boundary_integs;
|
||||
Array<Array<int>*> boundary_integs_marker; ///< Entries are not owned.
|
||||
|
||||
/// Trace face (skeleton) integrators.
|
||||
Array<BilinearFormIntegrator*> tfbfi;
|
||||
Array<BilinearFormIntegrator*> trace_face_integs;
|
||||
|
||||
/// Boundary trace face (skeleton) integrators.
|
||||
Array<BilinearFormIntegrator*> btfbfi;
|
||||
Array<Array<int>*> btfbfi_marker;///< Entries are not owned.
|
||||
Array<BilinearFormIntegrator*> boundary_trace_face_integs;
|
||||
/// Entries are not owned.
|
||||
Array<Array<int>*> boundary_trace_face_integs_marker;
|
||||
|
||||
DenseMatrix elemmat;
|
||||
Array<int> trial_vdofs, test_vdofs;
|
||||
@@ -762,24 +766,26 @@ public:
|
||||
Array<int> &bdr_marker);
|
||||
|
||||
/// Access all integrators added with AddDomainIntegrator().
|
||||
Array<BilinearFormIntegrator*> *GetDBFI() { return &dbfi; }
|
||||
Array<BilinearFormIntegrator*> *GetDBFI() { return &domain_integs; }
|
||||
|
||||
/// Access all integrators added with AddBoundaryIntegrator().
|
||||
Array<BilinearFormIntegrator*> *GetBBFI() { return &bbfi; }
|
||||
Array<BilinearFormIntegrator*> *GetBBFI() { return &boundary_integs; }
|
||||
/** @brief Access all boundary markers added with AddBoundaryIntegrator().
|
||||
If no marker was specified when the integrator was added, the
|
||||
corresponding pointer (to Array<int>) will be NULL. */
|
||||
Array<Array<int>*> *GetBBFI_Marker() { return &bbfi_marker; }
|
||||
Array<Array<int>*> *GetBBFI_Marker() { return &boundary_integs_marker; }
|
||||
|
||||
/// Access all integrators added with AddTraceFaceIntegrator().
|
||||
Array<BilinearFormIntegrator*> *GetTFBFI() { return &tfbfi; }
|
||||
Array<BilinearFormIntegrator*> *GetTFBFI() { return &trace_face_integs; }
|
||||
|
||||
/// Access all integrators added with AddBdrTraceFaceIntegrator().
|
||||
Array<BilinearFormIntegrator*> *GetBTFBFI() { return &btfbfi; }
|
||||
Array<BilinearFormIntegrator*> *GetBTFBFI()
|
||||
{ return &boundary_trace_face_integs; }
|
||||
/** @brief Access all boundary markers added with AddBdrTraceFaceIntegrator().
|
||||
If no marker was specified when the integrator was added, the
|
||||
corresponding pointer (to Array<int>) will be NULL. */
|
||||
Array<Array<int>*> *GetBTFBFI_Marker() { return &btfbfi_marker; }
|
||||
Array<Array<int>*> *GetBTFBFI_Marker()
|
||||
{ return &boundary_trace_face_integs_marker; }
|
||||
|
||||
/// Sets all sparse values of \f$ M \f$ to @a a.
|
||||
void operator=(const double a) { *mat = a; }
|
||||
@@ -1004,7 +1010,7 @@ public:
|
||||
{ AddTraceFaceIntegrator(di); }
|
||||
|
||||
/// Access all interpolators added with AddDomainInterpolator().
|
||||
Array<BilinearFormIntegrator*> *GetDI() { return &dbfi; }
|
||||
Array<BilinearFormIntegrator*> *GetDI() { return &domain_integs; }
|
||||
|
||||
/// Set the desired assembly level. The default is AssemblyLevel::FULL.
|
||||
/** This method must be called before assembly. */
|
||||
|
||||
+12
-12
@@ -160,7 +160,7 @@ void MFBilinearFormExtension::Mult(const Vector &x, Vector &y) const
|
||||
{
|
||||
intFaceIntegrators[i]->AddMultMF(faceIntX, faceIntY);
|
||||
}
|
||||
int_face_restrict_lex->MultTranspose(faceIntY, y);
|
||||
int_face_restrict_lex->AddMultTranspose(faceIntY, y);
|
||||
}
|
||||
}
|
||||
|
||||
@@ -176,7 +176,7 @@ void MFBilinearFormExtension::Mult(const Vector &x, Vector &y) const
|
||||
{
|
||||
bdrFaceIntegrators[i]->AddMultMF(faceBdrX, faceBdrY);
|
||||
}
|
||||
bdr_face_restrict_lex->MultTranspose(faceBdrY, y);
|
||||
bdr_face_restrict_lex->AddMultTranspose(faceBdrY, y);
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -217,7 +217,7 @@ void MFBilinearFormExtension::MultTranspose(const Vector &x, Vector &y) const
|
||||
{
|
||||
intFaceIntegrators[i]->AddMultTransposeMF(faceIntX, faceIntY);
|
||||
}
|
||||
int_face_restrict_lex->MultTranspose(faceIntY, y);
|
||||
int_face_restrict_lex->AddMultTranspose(faceIntY, y);
|
||||
}
|
||||
}
|
||||
|
||||
@@ -233,7 +233,7 @@ void MFBilinearFormExtension::MultTranspose(const Vector &x, Vector &y) const
|
||||
{
|
||||
bdrFaceIntegrators[i]->AddMultTransposeMF(faceBdrX, faceBdrY);
|
||||
}
|
||||
bdr_face_restrict_lex->MultTranspose(faceBdrY, y);
|
||||
bdr_face_restrict_lex->AddMultTranspose(faceBdrY, y);
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -417,7 +417,7 @@ void PABilinearFormExtension::Mult(const Vector &x, Vector &y) const
|
||||
{
|
||||
intFaceIntegrators[i]->AddMultPA(faceIntX, faceIntY);
|
||||
}
|
||||
int_face_restrict_lex->MultTranspose(faceIntY, y);
|
||||
int_face_restrict_lex->AddMultTranspose(faceIntY, y);
|
||||
}
|
||||
}
|
||||
|
||||
@@ -433,7 +433,7 @@ void PABilinearFormExtension::Mult(const Vector &x, Vector &y) const
|
||||
{
|
||||
bdrFaceIntegrators[i]->AddMultPA(faceBdrX, faceBdrY);
|
||||
}
|
||||
bdr_face_restrict_lex->MultTranspose(faceBdrY, y);
|
||||
bdr_face_restrict_lex->AddMultTranspose(faceBdrY, y);
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -474,7 +474,7 @@ void PABilinearFormExtension::MultTranspose(const Vector &x, Vector &y) const
|
||||
{
|
||||
intFaceIntegrators[i]->AddMultTransposePA(faceIntX, faceIntY);
|
||||
}
|
||||
int_face_restrict_lex->MultTranspose(faceIntY, y);
|
||||
int_face_restrict_lex->AddMultTranspose(faceIntY, y);
|
||||
}
|
||||
}
|
||||
|
||||
@@ -490,7 +490,7 @@ void PABilinearFormExtension::MultTranspose(const Vector &x, Vector &y) const
|
||||
{
|
||||
bdrFaceIntegrators[i]->AddMultTransposePA(faceBdrX, faceBdrY);
|
||||
}
|
||||
bdr_face_restrict_lex->MultTranspose(faceBdrY, y);
|
||||
bdr_face_restrict_lex->AddMultTranspose(faceBdrY, y);
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -657,7 +657,7 @@ void EABilinearFormExtension::Mult(const Vector &x, Vector &y) const
|
||||
Y(j, 0, f) += res;
|
||||
});
|
||||
// Apply the Interior Face Restriction transposed
|
||||
int_face_restrict_lex->MultTranspose(faceIntY, y);
|
||||
int_face_restrict_lex->AddMultTranspose(faceIntY, y);
|
||||
}
|
||||
}
|
||||
|
||||
@@ -688,7 +688,7 @@ void EABilinearFormExtension::Mult(const Vector &x, Vector &y) const
|
||||
Y(j, f) += res;
|
||||
});
|
||||
// Apply the Boundary Face Restriction transposed
|
||||
bdr_face_restrict_lex->MultTranspose(faceBdrY, y);
|
||||
bdr_face_restrict_lex->AddMultTranspose(faceBdrY, y);
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -783,7 +783,7 @@ void EABilinearFormExtension::MultTranspose(const Vector &x, Vector &y) const
|
||||
Y(j, 0, f) += res;
|
||||
});
|
||||
// Apply the Interior Face Restriction transposed
|
||||
int_face_restrict_lex->MultTranspose(faceIntY, y);
|
||||
int_face_restrict_lex->AddMultTranspose(faceIntY, y);
|
||||
}
|
||||
}
|
||||
|
||||
@@ -814,7 +814,7 @@ void EABilinearFormExtension::MultTranspose(const Vector &x, Vector &y) const
|
||||
Y(j, f) += res;
|
||||
});
|
||||
// Apply the Boundary Face Restriction transposed
|
||||
bdr_face_restrict_lex->MultTranspose(faceBdrY, y);
|
||||
bdr_face_restrict_lex->AddMultTranspose(faceBdrY, y);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
@@ -72,8 +72,8 @@ protected:
|
||||
mutable Vector faceIntX, faceIntY;
|
||||
mutable Vector faceBdrX, faceBdrY;
|
||||
const Operator *elem_restrict; // Not owned
|
||||
const Operator *int_face_restrict_lex; // Not owned
|
||||
const Operator *bdr_face_restrict_lex; // Not owned
|
||||
const FaceRestriction *int_face_restrict_lex; // Not owned
|
||||
const FaceRestriction *bdr_face_restrict_lex; // Not owned
|
||||
|
||||
public:
|
||||
PABilinearFormExtension(BilinearForm*);
|
||||
@@ -143,8 +143,8 @@ protected:
|
||||
mutable Vector faceIntX, faceIntY;
|
||||
mutable Vector faceBdrX, faceBdrY;
|
||||
const Operator *elem_restrict; // Not owned
|
||||
const Operator *int_face_restrict_lex; // Not owned
|
||||
const Operator *bdr_face_restrict_lex; // Not owned
|
||||
const FaceRestriction *int_face_restrict_lex; // Not owned
|
||||
const FaceRestriction *bdr_face_restrict_lex; // Not owned
|
||||
|
||||
public:
|
||||
MFBilinearFormExtension(BilinearForm *form);
|
||||
|
||||
+29
-2
@@ -175,6 +175,11 @@ void BilinearFormIntegrator::AssembleFaceVector(
|
||||
elmat.Mult(elfun, elvect);
|
||||
}
|
||||
|
||||
void TransposeIntegrator::SetIntRule(const IntegrationRule *ir)
|
||||
{
|
||||
IntRule = ir;
|
||||
bfi->SetIntRule(ir);
|
||||
}
|
||||
|
||||
void TransposeIntegrator::AssembleElementMatrix (
|
||||
const FiniteElement &el, ElementTransformation &Trans, DenseMatrix &elmat)
|
||||
@@ -202,6 +207,12 @@ void TransposeIntegrator::AssembleFaceMatrix (
|
||||
elmat.Transpose (bfi_elmat);
|
||||
}
|
||||
|
||||
void LumpedIntegrator::SetIntRule(const IntegrationRule *ir)
|
||||
{
|
||||
IntRule = ir;
|
||||
bfi->SetIntRule(ir);
|
||||
}
|
||||
|
||||
void LumpedIntegrator::AssembleElementMatrix (
|
||||
const FiniteElement &el, ElementTransformation &Trans, DenseMatrix &elmat)
|
||||
{
|
||||
@@ -209,6 +220,12 @@ void LumpedIntegrator::AssembleElementMatrix (
|
||||
elmat.Lump();
|
||||
}
|
||||
|
||||
void InverseIntegrator::SetIntRule(const IntegrationRule *ir)
|
||||
{
|
||||
IntRule = ir;
|
||||
integrator->SetIntRule(ir);
|
||||
}
|
||||
|
||||
void InverseIntegrator::AssembleElementMatrix(
|
||||
const FiniteElement &el, ElementTransformation &Trans, DenseMatrix &elmat)
|
||||
{
|
||||
@@ -216,6 +233,15 @@ void InverseIntegrator::AssembleElementMatrix(
|
||||
elmat.Invert();
|
||||
}
|
||||
|
||||
void SumIntegrator::SetIntRule(const IntegrationRule *ir)
|
||||
{
|
||||
IntRule = ir;
|
||||
for (int i = 0; i < integrators.Size(); i++)
|
||||
{
|
||||
integrators[i]->SetIntRule(ir);
|
||||
}
|
||||
}
|
||||
|
||||
void SumIntegrator::AssembleElementMatrix(
|
||||
const FiniteElement &el, ElementTransformation &Trans, DenseMatrix &elmat)
|
||||
{
|
||||
@@ -1751,15 +1777,16 @@ void DerivativeIntegrator::AssembleElementMatrix2 (
|
||||
int dim = trial_fe.GetDim();
|
||||
int trial_nd = trial_fe.GetDof();
|
||||
int test_nd = test_fe.GetDof();
|
||||
int spaceDim = Trans.GetSpaceDim();
|
||||
|
||||
int i, l;
|
||||
double det;
|
||||
|
||||
elmat.SetSize (test_nd,trial_nd);
|
||||
dshape.SetSize (trial_nd,dim);
|
||||
dshapedxt.SetSize(trial_nd,dim);
|
||||
dshapedxt.SetSize(trial_nd, spaceDim);
|
||||
dshapedxi.SetSize(trial_nd);
|
||||
invdfdx.SetSize(dim);
|
||||
invdfdx.SetSize(dim, spaceDim);
|
||||
shape.SetSize (test_nd);
|
||||
|
||||
const IntegrationRule *ir = IntRule;
|
||||
|
||||
@@ -261,6 +261,8 @@ public:
|
||||
TransposeIntegrator (BilinearFormIntegrator *bfi_, int own_bfi_ = 1)
|
||||
{ bfi = bfi_; own_bfi = own_bfi_; }
|
||||
|
||||
virtual void SetIntRule(const IntegrationRule *ir);
|
||||
|
||||
virtual void AssembleElementMatrix(const FiniteElement &el,
|
||||
ElementTransformation &Trans,
|
||||
DenseMatrix &elmat);
|
||||
@@ -328,6 +330,8 @@ public:
|
||||
LumpedIntegrator (BilinearFormIntegrator *bfi_, int own_bfi_ = 1)
|
||||
{ bfi = bfi_; own_bfi = own_bfi_; }
|
||||
|
||||
virtual void SetIntRule(const IntegrationRule *ir);
|
||||
|
||||
virtual void AssembleElementMatrix(const FiniteElement &el,
|
||||
ElementTransformation &Trans,
|
||||
DenseMatrix &elmat);
|
||||
@@ -346,6 +350,8 @@ public:
|
||||
InverseIntegrator(BilinearFormIntegrator *integ, int own_integ = 1)
|
||||
{ integrator = integ; own_integrator = own_integ; }
|
||||
|
||||
virtual void SetIntRule(const IntegrationRule *ir);
|
||||
|
||||
virtual void AssembleElementMatrix(const FiniteElement &el,
|
||||
ElementTransformation &Trans,
|
||||
DenseMatrix &elmat);
|
||||
@@ -364,6 +370,8 @@ private:
|
||||
public:
|
||||
SumIntegrator(int own_integs = 1) { own_integrators = own_integs; }
|
||||
|
||||
virtual void SetIntRule(const IntegrationRule *ir);
|
||||
|
||||
void AddIntegrator(BilinearFormIntegrator *integ)
|
||||
{ integrators.Append(integ); }
|
||||
|
||||
|
||||
@@ -143,7 +143,7 @@ Solver *BuildSmootherFromCeed(ConstrainedOperator &op, bool chebyshev)
|
||||
if (chebyshev)
|
||||
{
|
||||
const int cheb_order = 3;
|
||||
out = new OperatorChebyshevSmoother(&op, t_diag, ess_tdofs, cheb_order);
|
||||
out = new OperatorChebyshevSmoother(op, t_diag, ess_tdofs, cheb_order);
|
||||
}
|
||||
else
|
||||
{
|
||||
|
||||
@@ -629,27 +629,6 @@ void MatrixVectorProductCoefficient::Eval(Vector &V, ElementTransformation &T,
|
||||
ma.Mult(vb, V);
|
||||
}
|
||||
|
||||
|
||||
MatrixMatrixProductCoefficient::MatrixMatrixProductCoefficient(MatrixCoefficient &A,
|
||||
MatrixCoefficient &B)
|
||||
: MatrixCoefficient(A.GetHeight(), A.GetWidth()),
|
||||
a(&A), b(&B),
|
||||
ma(A.GetHeight(), A.GetWidth()),
|
||||
mb(B.GetHeight(), B.GetWidth())
|
||||
{
|
||||
MFEM_ASSERT(A.GetWidth() == B.GetHeight(),
|
||||
"MatrixMatrixProductCoefficient: "
|
||||
"Arguments must have the same dimensions.");
|
||||
}
|
||||
|
||||
void MatrixMatrixProductCoefficient::Eval(DenseMatrix &M, ElementTransformation &T,
|
||||
const IntegrationPoint &ip)
|
||||
{
|
||||
a->Eval(ma, T, ip);
|
||||
b->Eval(mb, T, ip);
|
||||
Mult(ma, mb, M);
|
||||
}
|
||||
|
||||
void IdentityMatrixCoefficient::Eval(DenseMatrix &M, ElementTransformation &T,
|
||||
const IntegrationPoint &ip)
|
||||
{
|
||||
|
||||
@@ -1470,42 +1470,6 @@ public:
|
||||
const IntegrationPoint &ip);
|
||||
};
|
||||
|
||||
/** @brief Matrix coefficient defined as a product of two
|
||||
matrix coefficients */
|
||||
class MatrixMatrixProductCoefficient : public MatrixCoefficient
|
||||
{
|
||||
private:
|
||||
MatrixCoefficient * a;
|
||||
MatrixCoefficient * b;
|
||||
|
||||
mutable DenseMatrix ma;
|
||||
mutable DenseMatrix mb;
|
||||
|
||||
public:
|
||||
/// Constructor with two coefficients. Result is A*B.
|
||||
MatrixMatrixProductCoefficient(MatrixCoefficient &A, MatrixCoefficient &B);
|
||||
|
||||
/// Reset the matrix coefficient
|
||||
void SetACoef(MatrixCoefficient &A) { a = &A; }
|
||||
/// Return the matrix coefficient
|
||||
MatrixCoefficient * GetACoef() const { return a; }
|
||||
|
||||
/// Reset the vector coefficient
|
||||
void SetBCoef(MatrixCoefficient &B) { b = &B; }
|
||||
/// Return the vector coefficient
|
||||
MatrixCoefficient * GetBCoef() const { return b; }
|
||||
|
||||
/// Evaluate the vector coefficient at @a ip.
|
||||
virtual void Eval(DenseMatrix &M, ElementTransformation &T,
|
||||
const IntegrationPoint &ip);
|
||||
};
|
||||
|
||||
/// Convenient alias for the MatrixVectorProductCoefficient
|
||||
typedef MatrixMatrixProductCoefficient MatMatCoefficient;
|
||||
|
||||
|
||||
|
||||
|
||||
/// Matrix coefficient defined as the linear combination of two matrices
|
||||
class MatrixSumCoefficient : public MatrixCoefficient
|
||||
{
|
||||
|
||||
+58
-8
@@ -17,6 +17,7 @@
|
||||
|
||||
#include <cerrno> // errno
|
||||
#include <sstream>
|
||||
#include <regex>
|
||||
|
||||
#ifndef _WIN32
|
||||
#include <sys/stat.h> // mkdir
|
||||
@@ -764,7 +765,8 @@ ParaViewDataCollection::ParaViewDataCollection(const std::string&
|
||||
: DataCollection(collection_name, mesh_),
|
||||
levels_of_detail(1),
|
||||
pv_data_format(VTKFormat::BINARY),
|
||||
high_order_output(false)
|
||||
high_order_output(false),
|
||||
restart_mode(false)
|
||||
{
|
||||
#ifdef MFEM_USE_ZLIB
|
||||
compression = -1; // default zlib compression level, equivalent to 6
|
||||
@@ -842,17 +844,60 @@ void ParaViewDataCollection::Save()
|
||||
}
|
||||
// the directory is created
|
||||
|
||||
// create pvd file if needed
|
||||
// create pvd file if needed. If we are not in restart mode, a new pvd file
|
||||
// is always created. In restart mode, we keep any previously defined
|
||||
// timestep values as long as they are less than the currently defined time.
|
||||
|
||||
if (myid == 0 && !pvd_stream.is_open())
|
||||
{
|
||||
std::string dpath=GenerateCollectionPath();
|
||||
std::string pvdname=dpath+"/"+GeneratePVDFileName();
|
||||
pvd_stream.open(pvdname.c_str(),std::ios::out);
|
||||
// initialize the file
|
||||
pvd_stream << "<?xml version=\"1.0\"?>\n";
|
||||
pvd_stream << "<VTKFile type=\"Collection\" version=\"0.1\"";
|
||||
pvd_stream << " byte_order=\"" << VTKByteOrder() << "\">\n";
|
||||
pvd_stream << "<Collection>" << std::endl;
|
||||
|
||||
std::ifstream pvd_in;
|
||||
if (restart_mode && (pvd_in.open(pvdname,std::ios::binary),pvd_in.good()))
|
||||
{
|
||||
// PVD file exists and restart mode enabled: preserve existing time
|
||||
// steps less than the current time.
|
||||
std::fstream::pos_type pos_begin = pvd_in.tellg();
|
||||
std::fstream::pos_type pos_end = pos_begin;
|
||||
|
||||
std::regex regexp("timestep=\"([^[:space:]]+)\".*file=\"Cycle(\\d+)");
|
||||
std::smatch match;
|
||||
|
||||
std::string line;
|
||||
while (getline(pvd_in,line))
|
||||
{
|
||||
if (regex_search(line,match,regexp))
|
||||
{
|
||||
MFEM_ASSERT(match.size() == 3, "Unable to parse DataSet");
|
||||
double tvalue = std::stod(match[1]);
|
||||
if (tvalue >= GetTime()) { break; }
|
||||
int cvalue = std::stoi(match[2]);
|
||||
MFEM_VERIFY(cvalue < GetCycle(), "Cycle " << GetCycle() <<
|
||||
" is too small for restart mode: trying to overwrite"
|
||||
" existing data.");
|
||||
pos_end = pvd_in.tellg();
|
||||
}
|
||||
}
|
||||
size_t count = pos_end - pos_begin;
|
||||
std::vector<char> buf(count);
|
||||
pvd_in.clear();
|
||||
pvd_in.seekg(pos_begin);
|
||||
pvd_in.read(buf.data(), count);
|
||||
pvd_in.close();
|
||||
pvd_stream.open(pvdname.c_str(),std::ios::out);
|
||||
pvd_stream.write(buf.data(), count);
|
||||
}
|
||||
else
|
||||
{
|
||||
// initialize new pvd file
|
||||
pvd_stream.open(pvdname.c_str(),std::ios::out);
|
||||
// initialize the file
|
||||
pvd_stream << "<?xml version=\"1.0\"?>\n";
|
||||
pvd_stream << "<VTKFile type=\"Collection\" version=\"0.1\"";
|
||||
pvd_stream << " byte_order=\"" << VTKByteOrder() << "\">\n";
|
||||
pvd_stream << "<Collection>" << std::endl;
|
||||
}
|
||||
}
|
||||
|
||||
// define the vtu file
|
||||
@@ -1091,6 +1136,11 @@ void ParaViewDataCollection::SetCompression(bool compression_)
|
||||
}
|
||||
}
|
||||
|
||||
void ParaViewDataCollection::UseRestartMode(bool restart_mode_)
|
||||
{
|
||||
restart_mode = restart_mode_;
|
||||
}
|
||||
|
||||
const char *ParaViewDataCollection::GetDataFormatString() const
|
||||
{
|
||||
if (pv_data_format == VTKFormat::ASCII)
|
||||
|
||||
@@ -488,6 +488,7 @@ private:
|
||||
std::fstream pvd_stream;
|
||||
VTKFormat pv_data_format;
|
||||
bool high_order_output;
|
||||
bool restart_mode;
|
||||
|
||||
protected:
|
||||
void SaveDataVTU(std::ostream &out, int ref);
|
||||
@@ -545,6 +546,11 @@ public:
|
||||
/// by default). Reading high-order data requires ParaView 5.5 or later.
|
||||
void SetHighOrderOutput(bool high_order_output_);
|
||||
|
||||
/// Enable or disable restart mode. If restart is enabled, new writes will
|
||||
/// preserve timestep metadata for any solutions prior to the currently
|
||||
/// defined time.
|
||||
void UseRestartMode(bool restart_mode_);
|
||||
|
||||
/// Load the collection - not implemented in the ParaView writer
|
||||
virtual void Load(int cycle_ = 0) override;
|
||||
};
|
||||
|
||||
+23
-4
@@ -7948,7 +7948,27 @@ VectorTensorFiniteElement::VectorTensorFiniteElement(const int dims,
|
||||
p, M, FunctionSpace::Qk),
|
||||
TensorBasisElement(dims, p, VerifyNodal(cbtype), dmtype),
|
||||
cbasis1d(poly1d.GetBasis(p, VerifyClosed(cbtype))),
|
||||
obasis1d(poly1d.GetBasis(p - 1, VerifyOpen(obtype))) { }
|
||||
obasis1d(poly1d.GetBasis(p - 1, VerifyOpen(obtype)))
|
||||
{
|
||||
MFEM_VERIFY(dims > 1, "Constructor for VectorTensorFiniteElement with both "
|
||||
"open and closed bases is not valid for 1D elements.");
|
||||
}
|
||||
|
||||
VectorTensorFiniteElement::VectorTensorFiniteElement(const int dims,
|
||||
const int d,
|
||||
const int p,
|
||||
const int obtype,
|
||||
const int M,
|
||||
const DofMapType dmtype)
|
||||
: VectorFiniteElement(dims, GetTensorProductGeometry(dims), d,
|
||||
p, M, FunctionSpace::Pk),
|
||||
TensorBasisElement(dims, p, obtype, dmtype),
|
||||
cbasis1d(poly1d.GetBasis(p, VerifyOpen(obtype))),
|
||||
obasis1d(poly1d.GetBasis(p, VerifyOpen(obtype)))
|
||||
{
|
||||
MFEM_VERIFY(dims == 1, "Constructor for VectorTensorFiniteElement without "
|
||||
"closed basis is only valid for 1D elements.");
|
||||
}
|
||||
|
||||
H1_SegmentElement::H1_SegmentElement(const int p, const int btype)
|
||||
: NodalTensorFiniteElement(1, p, VerifyClosed(btype), H1_DOF_MAP)
|
||||
@@ -13055,9 +13075,8 @@ void ND_TriangleElement::CalcCurlShape(const IntegrationPoint &ip,
|
||||
const double ND_SegmentElement::tk[1] = { 1. };
|
||||
|
||||
ND_SegmentElement::ND_SegmentElement(const int p, const int ob_type)
|
||||
: VectorFiniteElement(1, Geometry::SEGMENT, p, p - 1,
|
||||
H_CURL, FunctionSpace::Pk),
|
||||
obasis1d(poly1d.GetBasis(p - 1, VerifyOpen(ob_type))),
|
||||
: VectorTensorFiniteElement(1, p, p - 1, ob_type, H_CURL,
|
||||
DofMapType::L2_DOF_MAP),
|
||||
dof2tk(dof)
|
||||
{
|
||||
if (obasis1d.IsIntegratedType()) { is_nodal = false; }
|
||||
|
||||
+6
-3
@@ -2239,6 +2239,11 @@ public:
|
||||
const int cbtype, const int obtype,
|
||||
const int M, const DofMapType dmtype);
|
||||
|
||||
// For 1D elements: there is only an "open basis", no "closed basis"
|
||||
VectorTensorFiniteElement(const int dims, const int d, const int p,
|
||||
const int obtype, const int M,
|
||||
const DofMapType dmtype);
|
||||
|
||||
const DofToQuad &GetDofToQuad(const IntegrationRule &ir,
|
||||
DofToQuad::Mode mode) const;
|
||||
|
||||
@@ -3311,11 +3316,9 @@ public:
|
||||
|
||||
|
||||
/// Arbitrary order Nedelec elements in 1D on a segment
|
||||
class ND_SegmentElement : public VectorFiniteElement
|
||||
class ND_SegmentElement : public VectorTensorFiniteElement
|
||||
{
|
||||
static const double tk[1];
|
||||
|
||||
Poly_1D::Basis &obasis1d;
|
||||
Array<int> dof2tk;
|
||||
|
||||
public:
|
||||
|
||||
+2
-2
@@ -1225,7 +1225,7 @@ const Operator *FiniteElementSpace::GetElementRestriction(
|
||||
return L2E_nat.Ptr();
|
||||
}
|
||||
|
||||
const Operator *FiniteElementSpace::GetFaceRestriction(
|
||||
const FaceRestriction *FiniteElementSpace::GetFaceRestriction(
|
||||
ElementDofOrdering e_ordering, FaceType type, L2FaceValues mul) const
|
||||
{
|
||||
const bool is_dg_space = IsDGSpace();
|
||||
@@ -1239,7 +1239,7 @@ const Operator *FiniteElementSpace::GetFaceRestriction(
|
||||
}
|
||||
else
|
||||
{
|
||||
Operator* res;
|
||||
FaceRestriction *res;
|
||||
if (is_dg_space)
|
||||
{
|
||||
res = new L2FaceRestriction(*this, e_ordering, type, m);
|
||||
|
||||
+2
-2
@@ -164,7 +164,7 @@ protected:
|
||||
+ 8 * (int)std::get<3>(k);
|
||||
}
|
||||
};
|
||||
using map_L2F = std::unordered_map<const key_face,Operator*,key_hash>;
|
||||
using map_L2F = std::unordered_map<const key_face,FaceRestriction*,key_hash>;
|
||||
mutable map_L2F L2F;
|
||||
|
||||
mutable Array<QuadratureInterpolator*> E2Q_array;
|
||||
@@ -488,7 +488,7 @@ public:
|
||||
const Operator *GetElementRestriction(ElementDofOrdering e_ordering) const;
|
||||
|
||||
/// Return an Operator that converts L-vectors to E-vectors on each face.
|
||||
virtual const Operator *GetFaceRestriction(
|
||||
virtual const FaceRestriction *GetFaceRestriction(
|
||||
ElementDofOrdering e_ordering, FaceType,
|
||||
L2FaceValues mul = L2FaceValues::DoubleValued) const;
|
||||
|
||||
|
||||
+9
-4
@@ -34,6 +34,7 @@ IntegrationRule::IntegrationRule(IntegrationRule &irx, IntegrationRule &iry)
|
||||
nx = irx.GetNPoints();
|
||||
ny = iry.GetNPoints();
|
||||
SetSize(nx * ny);
|
||||
SetPointIndices();
|
||||
|
||||
for (j = 0; j < ny; j++)
|
||||
{
|
||||
@@ -48,8 +49,6 @@ IntegrationRule::IntegrationRule(IntegrationRule &irx, IntegrationRule &iry)
|
||||
ip.weight = ipx.weight * ipy.weight;
|
||||
}
|
||||
}
|
||||
|
||||
SetPointIndices();
|
||||
}
|
||||
|
||||
IntegrationRule::IntegrationRule(IntegrationRule &irx, IntegrationRule &iry,
|
||||
@@ -59,6 +58,7 @@ IntegrationRule::IntegrationRule(IntegrationRule &irx, IntegrationRule &iry,
|
||||
const int ny = iry.GetNPoints();
|
||||
const int nz = irz.GetNPoints();
|
||||
SetSize(nx*ny*nz);
|
||||
SetPointIndices();
|
||||
|
||||
for (int iz = 0; iz < nz; ++iz)
|
||||
{
|
||||
@@ -78,8 +78,6 @@ IntegrationRule::IntegrationRule(IntegrationRule &irx, IntegrationRule &iry,
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
SetPointIndices();
|
||||
}
|
||||
|
||||
const Array<double> &IntegrationRule::GetWeights() const
|
||||
@@ -125,6 +123,7 @@ void IntegrationRule::GrundmannMollerSimplexRule(int s, int n)
|
||||
}
|
||||
np /= f;
|
||||
SetSize(np);
|
||||
SetPointIndices();
|
||||
|
||||
int pt = 0;
|
||||
for (int i = 0; i <= s; i++)
|
||||
@@ -375,6 +374,7 @@ public:
|
||||
void QuadratureFunctions1D::GaussLegendre(const int np, IntegrationRule* ir)
|
||||
{
|
||||
ir->SetSize(np);
|
||||
ir->SetPointIndices();
|
||||
|
||||
switch (np)
|
||||
{
|
||||
@@ -477,6 +477,7 @@ void QuadratureFunctions1D::GaussLobatto(const int np, IntegrationRule* ir)
|
||||
*/
|
||||
|
||||
ir->SetSize(np);
|
||||
ir->SetPointIndices();
|
||||
if ( np == 1 )
|
||||
{
|
||||
ir->IntPoint(0).Set1w(0.5, 1.0);
|
||||
@@ -576,6 +577,7 @@ void QuadratureFunctions1D::GaussLobatto(const int np, IntegrationRule* ir)
|
||||
void QuadratureFunctions1D::OpenUniform(const int np, IntegrationRule* ir)
|
||||
{
|
||||
ir->SetSize(np);
|
||||
ir->SetPointIndices();
|
||||
|
||||
// The Newton-Cotes quadrature is based on weights that integrate exactly the
|
||||
// interpolatory polynomial through the equally spaced quadrature points.
|
||||
@@ -591,6 +593,7 @@ void QuadratureFunctions1D::ClosedUniform(const int np,
|
||||
IntegrationRule* ir)
|
||||
{
|
||||
ir->SetSize(np);
|
||||
ir->SetPointIndices();
|
||||
if ( np == 1 ) // allow this case as "closed"
|
||||
{
|
||||
ir->IntPoint(0).Set1w(0.5, 1.0);
|
||||
@@ -608,6 +611,7 @@ void QuadratureFunctions1D::ClosedUniform(const int np,
|
||||
void QuadratureFunctions1D::OpenHalfUniform(const int np, IntegrationRule* ir)
|
||||
{
|
||||
ir->SetSize(np);
|
||||
ir->SetPointIndices();
|
||||
|
||||
// Open half points: the centers of np uniform intervals
|
||||
for (int i = 0; i < np ; ++i)
|
||||
@@ -621,6 +625,7 @@ void QuadratureFunctions1D::OpenHalfUniform(const int np, IntegrationRule* ir)
|
||||
void QuadratureFunctions1D::ClosedGL(const int np, IntegrationRule* ir)
|
||||
{
|
||||
ir->SetSize(np);
|
||||
ir->SetPointIndices();
|
||||
ir->IntPoint(0).x = 0.0;
|
||||
ir->IntPoint(np-1).x = 1.0;
|
||||
|
||||
|
||||
+5
-3
@@ -96,9 +96,6 @@ private:
|
||||
by request with the method GetWeights(). */
|
||||
mutable Array<double> weights;
|
||||
|
||||
/// Sets the indices of each quadrature point on initialization.
|
||||
void SetPointIndices();
|
||||
|
||||
/// Define n-simplex rule (triangle/tetrahedron for n=2/3) of order (2s+1)
|
||||
void GrundmannMollerSimplexRule(int s, int n = 3);
|
||||
|
||||
@@ -227,6 +224,11 @@ public:
|
||||
}
|
||||
}
|
||||
|
||||
/// Sets the indices of each quadrature point on initialization.
|
||||
/** Note that most calls to IntegrationRule::SetSize should be paired with a
|
||||
call to SetPointIndices in order for the indices to be set correctly. */
|
||||
void SetPointIndices();
|
||||
|
||||
/// Tensor product of two 1D integration rules
|
||||
IntegrationRule(IntegrationRule &irx, IntegrationRule &iry);
|
||||
|
||||
|
||||
+74
-64
@@ -26,14 +26,14 @@ LinearForm::LinearForm(FiniteElementSpace *f, LinearForm *lf)
|
||||
extern_lfs = 1;
|
||||
|
||||
// Copy the pointers to the integrators
|
||||
dlfi = lf->dlfi;
|
||||
domain_integs = lf->domain_integs;
|
||||
|
||||
dlfi_delta = lf->dlfi_delta;
|
||||
domain_delta_integs = lf->domain_delta_integs;
|
||||
|
||||
blfi = lf->blfi;
|
||||
boundary_integs = lf->boundary_integs;
|
||||
|
||||
flfi = lf->flfi;
|
||||
flfi_marker = lf->flfi_marker;
|
||||
boundary_face_integs = lf->boundary_face_integs;
|
||||
boundary_face_integs_marker = lf->boundary_face_integs_marker;
|
||||
}
|
||||
|
||||
void LinearForm::AddDomainIntegrator(LinearFormIntegrator *lfi)
|
||||
@@ -42,13 +42,13 @@ void LinearForm::AddDomainIntegrator(LinearFormIntegrator *lfi)
|
||||
dynamic_cast<DeltaLFIntegrator *>(lfi);
|
||||
if (!maybe_delta || !maybe_delta->IsDelta())
|
||||
{
|
||||
dlfi.Append(lfi);
|
||||
domain_integs.Append(lfi);
|
||||
}
|
||||
else
|
||||
{
|
||||
dlfi_delta.Append(maybe_delta);
|
||||
domain_delta_integs.Append(maybe_delta);
|
||||
}
|
||||
dlfi_marker.Append(NULL);
|
||||
domain_integs_marker.Append(NULL);
|
||||
}
|
||||
|
||||
void LinearForm::AddDomainIntegrator(LinearFormIntegrator *lfi,
|
||||
@@ -58,44 +58,45 @@ void LinearForm::AddDomainIntegrator(LinearFormIntegrator *lfi,
|
||||
dynamic_cast<DeltaLFIntegrator *>(lfi);
|
||||
if (!maybe_delta || !maybe_delta->IsDelta())
|
||||
{
|
||||
dlfi.Append(lfi);
|
||||
domain_integs.Append(lfi);
|
||||
}
|
||||
else
|
||||
{
|
||||
dlfi_delta.Append(maybe_delta);
|
||||
domain_delta_integs.Append(maybe_delta);
|
||||
}
|
||||
dlfi_marker.Append(&elem_marker);
|
||||
domain_integs_marker.Append(&elem_marker);
|
||||
}
|
||||
|
||||
void LinearForm::AddBoundaryIntegrator (LinearFormIntegrator * lfi)
|
||||
{
|
||||
blfi.Append (lfi);
|
||||
blfi_marker.Append(NULL); // NULL -> all attributes are active
|
||||
boundary_integs.Append (lfi);
|
||||
boundary_integs_marker.Append(NULL); // NULL -> all attributes are active
|
||||
}
|
||||
|
||||
void LinearForm::AddBoundaryIntegrator (LinearFormIntegrator * lfi,
|
||||
Array<int> &bdr_attr_marker)
|
||||
{
|
||||
blfi.Append (lfi);
|
||||
blfi_marker.Append(&bdr_attr_marker);
|
||||
boundary_integs.Append (lfi);
|
||||
boundary_integs_marker.Append(&bdr_attr_marker);
|
||||
}
|
||||
|
||||
void LinearForm::AddBdrFaceIntegrator (LinearFormIntegrator * lfi)
|
||||
{
|
||||
flfi.Append(lfi);
|
||||
flfi_marker.Append(NULL); // NULL -> all attributes are active
|
||||
boundary_face_integs.Append(lfi);
|
||||
// NULL -> all attributes are active
|
||||
boundary_face_integs_marker.Append(NULL);
|
||||
}
|
||||
|
||||
void LinearForm::AddBdrFaceIntegrator(LinearFormIntegrator *lfi,
|
||||
Array<int> &bdr_attr_marker)
|
||||
{
|
||||
flfi.Append(lfi);
|
||||
flfi_marker.Append(&bdr_attr_marker);
|
||||
boundary_face_integs.Append(lfi);
|
||||
boundary_face_integs_marker.Append(&bdr_attr_marker);
|
||||
}
|
||||
|
||||
void LinearForm::AddInteriorFaceIntegrator(LinearFormIntegrator *lfi)
|
||||
{
|
||||
iflfi.Append(lfi);
|
||||
interior_face_integs.Append(lfi);
|
||||
}
|
||||
|
||||
void LinearForm::Assemble()
|
||||
@@ -112,14 +113,14 @@ void LinearForm::Assemble()
|
||||
// The first use of AddElementVector() below will move it back to host
|
||||
// because both 'vdofs' and 'elemvect' are on host.
|
||||
|
||||
if (dlfi.Size())
|
||||
if (domain_integs.Size())
|
||||
{
|
||||
for (int k = 0; k < dlfi.Size(); k++)
|
||||
for (int k = 0; k < domain_integs.Size(); k++)
|
||||
{
|
||||
if (dlfi_marker[k] != NULL)
|
||||
if (domain_integs_marker[k] != NULL)
|
||||
{
|
||||
MFEM_VERIFY(fes->GetMesh()->attributes.Size() ==
|
||||
dlfi_marker[k]->Size(),
|
||||
domain_integs_marker[k]->Size(),
|
||||
"invalid element marker for domain linear form "
|
||||
"integrator #" << k << ", counting from zero");
|
||||
}
|
||||
@@ -128,14 +129,15 @@ void LinearForm::Assemble()
|
||||
for (i = 0; i < fes -> GetNE(); i++)
|
||||
{
|
||||
int elem_attr = fes->GetMesh()->GetAttribute(i);
|
||||
for (int k = 0; k < dlfi.Size(); k++)
|
||||
for (int k = 0; k < domain_integs.Size(); k++)
|
||||
{
|
||||
if ( dlfi_marker[k] == NULL ||
|
||||
(*(dlfi_marker[k]))[elem_attr-1] == 1 )
|
||||
if ( domain_integs_marker[k] == NULL ||
|
||||
(*(domain_integs_marker[k]))[elem_attr-1] == 1 )
|
||||
{
|
||||
fes -> GetElementVDofs (i, vdofs);
|
||||
eltrans = fes -> GetElementTransformation (i);
|
||||
dlfi[k]->AssembleRHSElementVect(*fes->GetFE(i), *eltrans, elemvect);
|
||||
domain_integs[k]->AssembleRHSElementVect(*fes->GetFE(i),
|
||||
*eltrans, elemvect);
|
||||
AddElementVector (vdofs, elemvect);
|
||||
}
|
||||
}
|
||||
@@ -143,7 +145,7 @@ void LinearForm::Assemble()
|
||||
}
|
||||
AssembleDelta();
|
||||
|
||||
if (blfi.Size())
|
||||
if (boundary_integs.Size())
|
||||
{
|
||||
Mesh *mesh = fes->GetMesh();
|
||||
|
||||
@@ -151,14 +153,14 @@ void LinearForm::Assemble()
|
||||
Array<int> bdr_attr_marker(mesh->bdr_attributes.Size() ?
|
||||
mesh->bdr_attributes.Max() : 0);
|
||||
bdr_attr_marker = 0;
|
||||
for (int k = 0; k < blfi.Size(); k++)
|
||||
for (int k = 0; k < boundary_integs.Size(); k++)
|
||||
{
|
||||
if (blfi_marker[k] == NULL)
|
||||
if (boundary_integs_marker[k] == NULL)
|
||||
{
|
||||
bdr_attr_marker = 1;
|
||||
break;
|
||||
}
|
||||
Array<int> &bdr_marker = *blfi_marker[k];
|
||||
Array<int> &bdr_marker = *boundary_integs_marker[k];
|
||||
MFEM_ASSERT(bdr_marker.Size() == bdr_attr_marker.Size(),
|
||||
"invalid boundary marker for boundary integrator #"
|
||||
<< k << ", counting from zero");
|
||||
@@ -174,18 +176,19 @@ void LinearForm::Assemble()
|
||||
if (bdr_attr_marker[bdr_attr-1] == 0) { continue; }
|
||||
fes -> GetBdrElementVDofs (i, vdofs);
|
||||
eltrans = fes -> GetBdrElementTransformation (i);
|
||||
for (int k=0; k < blfi.Size(); k++)
|
||||
for (int k=0; k < boundary_integs.Size(); k++)
|
||||
{
|
||||
if (blfi_marker[k] &&
|
||||
(*blfi_marker[k])[bdr_attr-1] == 0) { continue; }
|
||||
if (boundary_integs_marker[k] &&
|
||||
(*boundary_integs_marker[k])[bdr_attr-1] == 0) { continue; }
|
||||
|
||||
blfi[k]->AssembleRHSElementVect(*fes->GetBE(i), *eltrans, elemvect);
|
||||
boundary_integs[k]->AssembleRHSElementVect(*fes->GetBE(i),
|
||||
*eltrans, elemvect);
|
||||
|
||||
AddElementVector (vdofs, elemvect);
|
||||
}
|
||||
}
|
||||
}
|
||||
if (flfi.Size())
|
||||
if (boundary_face_integs.Size())
|
||||
{
|
||||
FaceElementTransformations *tr;
|
||||
Mesh *mesh = fes->GetMesh();
|
||||
@@ -194,14 +197,14 @@ void LinearForm::Assemble()
|
||||
Array<int> bdr_attr_marker(mesh->bdr_attributes.Size() ?
|
||||
mesh->bdr_attributes.Max() : 0);
|
||||
bdr_attr_marker = 0;
|
||||
for (int k = 0; k < flfi.Size(); k++)
|
||||
for (int k = 0; k < boundary_face_integs.Size(); k++)
|
||||
{
|
||||
if (flfi_marker[k] == NULL)
|
||||
if (boundary_face_integs_marker[k] == NULL)
|
||||
{
|
||||
bdr_attr_marker = 1;
|
||||
break;
|
||||
}
|
||||
Array<int> &bdr_marker = *flfi_marker[k];
|
||||
Array<int> &bdr_marker = *boundary_face_integs_marker[k];
|
||||
MFEM_ASSERT(bdr_marker.Size() == bdr_attr_marker.Size(),
|
||||
"invalid boundary marker for boundary face integrator #"
|
||||
<< k << ", counting from zero");
|
||||
@@ -220,24 +223,26 @@ void LinearForm::Assemble()
|
||||
if (tr != NULL)
|
||||
{
|
||||
fes -> GetElementVDofs (tr -> Elem1No, vdofs);
|
||||
for (int k = 0; k < flfi.Size(); k++)
|
||||
for (int k = 0; k < boundary_face_integs.Size(); k++)
|
||||
{
|
||||
if (flfi_marker[k] &&
|
||||
(*flfi_marker[k])[bdr_attr-1] == 0) { continue; }
|
||||
if (boundary_face_integs_marker[k] &&
|
||||
(*boundary_face_integs_marker[k])[bdr_attr-1] == 0)
|
||||
{ continue; }
|
||||
|
||||
flfi[k] -> AssembleRHSElementVect (*fes->GetFE(tr -> Elem1No),
|
||||
*tr, elemvect);
|
||||
boundary_face_integs[k]->
|
||||
AssembleRHSElementVect(*fes->GetFE(tr->Elem1No),
|
||||
*tr, elemvect);
|
||||
AddElementVector (vdofs, elemvect);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
if (iflfi.Size())
|
||||
if (interior_face_integs.Size())
|
||||
{
|
||||
Mesh *mesh = fes->GetMesh();
|
||||
|
||||
for (int k = 0; k < iflfi.Size(); k++)
|
||||
for (int k = 0; k < interior_face_integs.Size(); k++)
|
||||
{
|
||||
for (i = 0; i < mesh->GetNumFaces(); i++)
|
||||
{
|
||||
@@ -249,9 +254,10 @@ void LinearForm::Assemble()
|
||||
Array<int> vdofs2;
|
||||
fes -> GetElementVDofs (tr -> Elem2No, vdofs2);
|
||||
vdofs.Append(vdofs2);
|
||||
iflfi[k] -> AssembleRHSElementVect (*fes->GetFE(tr -> Elem1No),
|
||||
*fes->GetFE(tr -> Elem2No),
|
||||
*tr, elemvect);
|
||||
interior_face_integs[k]->
|
||||
AssembleRHSElementVect(*fes->GetFE(tr->Elem1No),
|
||||
*fes->GetFE(tr->Elem2No),
|
||||
*tr, elemvect);
|
||||
AddElementVector (vdofs, elemvect);
|
||||
}
|
||||
}
|
||||
@@ -277,40 +283,41 @@ void LinearForm::MakeRef(FiniteElementSpace *f, Vector &v, int v_offset)
|
||||
|
||||
void LinearForm::AssembleDelta()
|
||||
{
|
||||
if (dlfi_delta.Size() == 0) { return; }
|
||||
if (domain_delta_integs.Size() == 0) { return; }
|
||||
|
||||
if (!HaveDeltaLocations())
|
||||
{
|
||||
int sdim = fes->GetMesh()->SpaceDimension();
|
||||
Vector center;
|
||||
DenseMatrix centers(sdim, dlfi_delta.Size());
|
||||
DenseMatrix centers(sdim, domain_delta_integs.Size());
|
||||
for (int i = 0; i < centers.Width(); i++)
|
||||
{
|
||||
centers.GetColumnReference(i, center);
|
||||
dlfi_delta[i]->GetDeltaCenter(center);
|
||||
domain_delta_integs[i]->GetDeltaCenter(center);
|
||||
MFEM_VERIFY(center.Size() == sdim,
|
||||
"Point dim " << center.Size() <<
|
||||
" does not match space dim " << sdim);
|
||||
}
|
||||
fes->GetMesh()->FindPoints(centers, dlfi_delta_elem_id, dlfi_delta_ip);
|
||||
fes->GetMesh()->FindPoints(centers, domain_delta_integs_elem_id,
|
||||
domain_delta_integs_ip);
|
||||
}
|
||||
|
||||
Array<int> vdofs;
|
||||
Vector elemvect;
|
||||
for (int i = 0; i < dlfi_delta.Size(); i++)
|
||||
for (int i = 0; i < domain_delta_integs.Size(); i++)
|
||||
{
|
||||
int elem_id = dlfi_delta_elem_id[i];
|
||||
int elem_id = domain_delta_integs_elem_id[i];
|
||||
// The delta center may be outside of this sub-domain, or
|
||||
// (Par)Mesh::FindPoints() failed to find this point:
|
||||
if (elem_id < 0) { continue; }
|
||||
|
||||
const IntegrationPoint &ip = dlfi_delta_ip[i];
|
||||
const IntegrationPoint &ip = domain_delta_integs_ip[i];
|
||||
ElementTransformation &Trans = *fes->GetElementTransformation(elem_id);
|
||||
Trans.SetIntPoint(&ip);
|
||||
|
||||
fes->GetElementVDofs(elem_id, vdofs);
|
||||
dlfi_delta[i]->AssembleDeltaElementVect(*fes->GetFE(elem_id), Trans,
|
||||
elemvect);
|
||||
domain_delta_integs[i]->AssembleDeltaElementVect(*fes->GetFE(elem_id),
|
||||
Trans, elemvect);
|
||||
AddElementVector(vdofs, elemvect);
|
||||
}
|
||||
}
|
||||
@@ -333,11 +340,14 @@ LinearForm::~LinearForm()
|
||||
if (!extern_lfs)
|
||||
{
|
||||
int k;
|
||||
for (k=0; k < dlfi_delta.Size(); k++) { delete dlfi_delta[k]; }
|
||||
for (k=0; k < dlfi.Size(); k++) { delete dlfi[k]; }
|
||||
for (k=0; k < blfi.Size(); k++) { delete blfi[k]; }
|
||||
for (k=0; k < flfi.Size(); k++) { delete flfi[k]; }
|
||||
for (k=0; k < iflfi.Size(); k++) { delete iflfi[k]; }
|
||||
for (k=0; k < domain_delta_integs.Size(); k++)
|
||||
{ delete domain_delta_integs[k]; }
|
||||
for (k=0; k < domain_integs.Size(); k++) { delete domain_integs[k]; }
|
||||
for (k=0; k < boundary_integs.Size(); k++) { delete boundary_integs[k]; }
|
||||
for (k=0; k < boundary_face_integs.Size(); k++)
|
||||
{ delete boundary_face_integs[k]; }
|
||||
for (k=0; k < interior_face_integs.Size(); k++)
|
||||
{ delete interior_face_integs[k]; }
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
+25
-19
@@ -26,43 +26,46 @@ protected:
|
||||
/// FE space on which the LinearForm lives. Not owned.
|
||||
FiniteElementSpace *fes;
|
||||
|
||||
/** @brief Indicates the LinearFormIntegrator%s stored in #dlfi, #dlfi_delta,
|
||||
#blfi, and #flfi are owned by another LinearForm. */
|
||||
/** @brief Indicates the LinearFormIntegrator%s stored in #domain_integs,
|
||||
#domain_delta_integs, #boundary_integs, and #boundary_face_integs are
|
||||
owned by another LinearForm. */
|
||||
int extern_lfs;
|
||||
|
||||
/// Set of Domain Integrators to be applied.
|
||||
Array<LinearFormIntegrator*> dlfi;
|
||||
Array<LinearFormIntegrator*> domain_integs;
|
||||
/// Element attribute marker (should be of length mesh->attributes)
|
||||
/// Includes all by default.
|
||||
/// 0 - ignore attribute
|
||||
/// 1 - include attribute
|
||||
Array<Array<int>*> dlfi_marker;
|
||||
Array<Array<int>*> domain_integs_marker;
|
||||
|
||||
/// Separate array for integrators with delta function coefficients.
|
||||
Array<DeltaLFIntegrator*> dlfi_delta;
|
||||
Array<DeltaLFIntegrator*> domain_delta_integs;
|
||||
|
||||
/// Set of Boundary Integrators to be applied.
|
||||
Array<LinearFormIntegrator*> blfi;
|
||||
Array<Array<int>*> blfi_marker; ///< Entries are not owned.
|
||||
Array<LinearFormIntegrator*> boundary_integs;
|
||||
/// Entries are not owned.
|
||||
Array<Array<int>*> boundary_integs_marker;
|
||||
|
||||
/// Set of Boundary Face Integrators to be applied.
|
||||
Array<LinearFormIntegrator*> flfi;
|
||||
Array<Array<int>*> flfi_marker; ///< Entries are not owned.
|
||||
Array<LinearFormIntegrator*> boundary_face_integs;
|
||||
Array<Array<int>*> boundary_face_integs_marker; ///< Entries not owned.
|
||||
|
||||
/// Set of Internal Face Integrators to be applied.
|
||||
Array<LinearFormIntegrator*> iflfi;
|
||||
Array<LinearFormIntegrator*> interior_face_integs;
|
||||
|
||||
/// The element ids where the centers of the delta functions lie
|
||||
Array<int> dlfi_delta_elem_id;
|
||||
Array<int> domain_delta_integs_elem_id;
|
||||
|
||||
/// The reference coordinates where the centers of the delta functions lie
|
||||
Array<IntegrationPoint> dlfi_delta_ip;
|
||||
Array<IntegrationPoint> domain_delta_integs_ip;
|
||||
|
||||
/// If true, the delta locations are not (re)computed during assembly.
|
||||
bool HaveDeltaLocations() { return (dlfi_delta_elem_id.Size() != 0); }
|
||||
bool HaveDeltaLocations()
|
||||
{ return (domain_delta_integs_elem_id.Size() != 0); }
|
||||
|
||||
/// Force (re)computation of delta locations.
|
||||
void ResetDeltaLocations() { dlfi_delta_elem_id.SetSize(0); }
|
||||
void ResetDeltaLocations() { domain_delta_integs_elem_id.SetSize(0); }
|
||||
|
||||
private:
|
||||
/// Copy construction is not supported; body is undefined.
|
||||
@@ -150,22 +153,25 @@ public:
|
||||
/** @brief Access all integrators added with AddDomainIntegrator() which are
|
||||
not DeltaLFIntegrator%s or they are DeltaLFIntegrator%s with non-delta
|
||||
coefficients. */
|
||||
Array<LinearFormIntegrator*> *GetDLFI() { return &dlfi; }
|
||||
Array<LinearFormIntegrator*> *GetDLFI() { return &domain_integs; }
|
||||
|
||||
/** @brief Access all integrators added with AddDomainIntegrator() which are
|
||||
DeltaLFIntegrator%s with delta coefficients. */
|
||||
Array<DeltaLFIntegrator*> *GetDLFI_Delta() { return &dlfi_delta; }
|
||||
Array<DeltaLFIntegrator*> *GetDLFI_Delta() { return &domain_delta_integs; }
|
||||
|
||||
/// Access all integrators added with AddBoundaryIntegrator().
|
||||
Array<LinearFormIntegrator*> *GetBLFI() { return &blfi; }
|
||||
Array<LinearFormIntegrator*> *GetBLFI() { return &boundary_integs; }
|
||||
|
||||
/// Access all integrators added with AddBdrFaceIntegrator().
|
||||
Array<LinearFormIntegrator*> *GetFLFI() { return &flfi; }
|
||||
Array<LinearFormIntegrator*> *GetFLFI() { return &boundary_face_integs; }
|
||||
|
||||
/// Access all integrators added with AddInteriorFaceIntegrator().
|
||||
Array<LinearFormIntegrator*> *GetIFLFI() { return &interior_face_integs; }
|
||||
|
||||
/** @brief Access all boundary markers added with AddBdrFaceIntegrator().
|
||||
If no marker was specified when the integrator was added, the
|
||||
corresponding pointer (to Array<int>) will be NULL. */
|
||||
Array<Array<int>*> *GetFLFI_Marker() { return &flfi_marker; }
|
||||
Array<Array<int>*> *GetFLFI_Marker() { return &boundary_face_integs_marker; }
|
||||
|
||||
/// Assembles the linear form i.e. sums over all domain/bdr integrators.
|
||||
void Assemble();
|
||||
|
||||
@@ -979,6 +979,8 @@ void BlockNonlinearForm::ComputeGradientBlocked(const BlockVector &bx) const
|
||||
|
||||
for (int k = 0; k < bfnfi.Size(); ++k)
|
||||
{
|
||||
if (bfnfi_marker[k] &&
|
||||
(*bfnfi_marker[k])[bdr_attr-1] == 0) { continue; }
|
||||
bfnfi[k]->AssembleFaceGrad(fe, fe2, *tr, el_x_const, elmats);
|
||||
for (int l=0; l<fes.Size(); ++l)
|
||||
{
|
||||
|
||||
+2
-2
@@ -40,10 +40,10 @@ protected:
|
||||
public:
|
||||
/** @brief Prescribe a fixed IntegrationRule to use (when @a ir != NULL) or
|
||||
let the integrator choose (when @a ir == NULL). */
|
||||
void SetIntRule(const IntegrationRule *ir) { IntRule = ir; }
|
||||
virtual void SetIntRule(const IntegrationRule *ir) { IntRule = ir; }
|
||||
|
||||
/// Prescribe a fixed IntegrationRule to use.
|
||||
void SetIntegrationRule(const IntegrationRule &irule) { IntRule = &irule; }
|
||||
void SetIntegrationRule(const IntegrationRule &ir) { SetIntRule(&ir); }
|
||||
|
||||
/// Set the memory type used for GeometricFactors and other large allocations
|
||||
/// in PA extensions.
|
||||
|
||||
+10
-8
@@ -130,7 +130,7 @@ void ParBilinearForm::ParallelAssemble(OperatorHandle &A, SparseMatrix *A_local)
|
||||
|
||||
OperatorHandle dA(A.Type()), Ph(A.Type()), hdA;
|
||||
|
||||
if (fbfi.Size() == 0)
|
||||
if (interior_face_integs.Size() == 0)
|
||||
{
|
||||
// construct a parallel block-diagonal matrix 'A' based on 'a'
|
||||
dA.MakeSquareBlockDiag(pfes->GetComm(), pfes->GlobalVSize(),
|
||||
@@ -214,11 +214,12 @@ void ParBilinearForm::AssembleSharedFaces(int skip_zeros)
|
||||
}
|
||||
}
|
||||
vdofs_all.Append(vdofs2);
|
||||
for (int k = 0; k < fbfi.Size(); k++)
|
||||
for (int k = 0; k < interior_face_integs.Size(); k++)
|
||||
{
|
||||
fbfi[k]->AssembleFaceMatrix(*pfes->GetFE(T->Elem1No),
|
||||
*pfes->GetFaceNbrFE(Elem2NbrNo),
|
||||
*T, elemmat);
|
||||
interior_face_integs[k]->
|
||||
AssembleFaceMatrix(*pfes->GetFE(T->Elem1No),
|
||||
*pfes->GetFaceNbrFE(Elem2NbrNo),
|
||||
*T, elemmat);
|
||||
if (keep_nbr_block)
|
||||
{
|
||||
mat->AddSubMatrix(vdofs_all, vdofs_all, elemmat, skip_zeros);
|
||||
@@ -233,7 +234,7 @@ void ParBilinearForm::AssembleSharedFaces(int skip_zeros)
|
||||
|
||||
void ParBilinearForm::Assemble(int skip_zeros)
|
||||
{
|
||||
if (fbfi.Size())
|
||||
if (interior_face_integs.Size())
|
||||
{
|
||||
pfes->ExchangeFaceNbrData();
|
||||
if (!ext && mat == NULL)
|
||||
@@ -244,7 +245,7 @@ void ParBilinearForm::Assemble(int skip_zeros)
|
||||
|
||||
BilinearForm::Assemble(skip_zeros);
|
||||
|
||||
if (!ext && fbfi.Size() > 0)
|
||||
if (!ext && interior_face_integs.Size() > 0)
|
||||
{
|
||||
AssembleSharedFaces(skip_zeros);
|
||||
}
|
||||
@@ -316,7 +317,8 @@ ParallelEliminateEssentialBC(const Array<int> &bdr_attr_is_ess,
|
||||
void ParBilinearForm::TrueAddMult(const Vector &x, Vector &y, const double a)
|
||||
const
|
||||
{
|
||||
MFEM_VERIFY(fbfi.Size() == 0, "the case of interior face integrators is not"
|
||||
MFEM_VERIFY(interior_face_integs.Size() == 0,
|
||||
"the case of interior face integrators is not"
|
||||
" implemented");
|
||||
|
||||
if (X.ParFESpace() != pfes)
|
||||
|
||||
+2
-2
@@ -515,7 +515,7 @@ const FiniteElement *ParFiniteElementSpace::GetFE(int i) const
|
||||
else { return FiniteElementSpace::GetFE(i); }
|
||||
}
|
||||
|
||||
const Operator *ParFiniteElementSpace::GetFaceRestriction(
|
||||
const FaceRestriction *ParFiniteElementSpace::GetFaceRestriction(
|
||||
ElementDofOrdering e_ordering, FaceType type, L2FaceValues mul) const
|
||||
{
|
||||
const bool is_dg_space = IsDGSpace();
|
||||
@@ -529,7 +529,7 @@ const Operator *ParFiniteElementSpace::GetFaceRestriction(
|
||||
}
|
||||
else
|
||||
{
|
||||
Operator* res;
|
||||
FaceRestriction *res;
|
||||
if (is_dg_space)
|
||||
{
|
||||
res = new ParL2FaceRestriction(*this, e_ordering, type, m);
|
||||
|
||||
+1
-1
@@ -299,7 +299,7 @@ public:
|
||||
presence of shared faces. Shared faces are treated as interior faces,
|
||||
the returned operator handles the communication needed to get the
|
||||
shared face values from other MPI ranks */
|
||||
virtual const Operator *GetFaceRestriction(
|
||||
virtual const FaceRestriction *GetFaceRestriction(
|
||||
ElementDofOrdering e_ordering, FaceType type,
|
||||
L2FaceValues mul = L2FaceValues::DoubleValued) const;
|
||||
|
||||
|
||||
+7
-6
@@ -47,7 +47,7 @@ void ParLinearForm::Assemble()
|
||||
{
|
||||
LinearForm::Assemble();
|
||||
|
||||
if (iflfi.Size())
|
||||
if (interior_face_integs.Size())
|
||||
{
|
||||
pfes->ExchangeFaceNbrData();
|
||||
AssembleSharedFaces();
|
||||
@@ -59,10 +59,10 @@ void ParLinearForm::AssembleSharedFaces()
|
||||
Array<int> vdofs;
|
||||
Vector elemvect;
|
||||
|
||||
if (iflfi.Size())
|
||||
if (interior_face_integs.Size())
|
||||
{
|
||||
ParMesh *pmesh = pfes->GetParMesh();
|
||||
for (int k = 0; k < iflfi.Size(); k++)
|
||||
for (int k = 0; k < interior_face_integs.Size(); k++)
|
||||
{
|
||||
for (int i = 0; i < pmesh->GetNSharedFaces(); i++)
|
||||
{
|
||||
@@ -73,9 +73,10 @@ void ParLinearForm::AssembleSharedFaces()
|
||||
{
|
||||
int Elem2Nbr = tr->Elem2No - pmesh->GetNE();
|
||||
fes -> GetElementVDofs (tr -> Elem1No, vdofs);
|
||||
iflfi[0] -> AssembleRHSElementVect (*fes->GetFE(tr -> Elem1No),
|
||||
*pfes->GetFaceNbrFE(Elem2Nbr),
|
||||
*tr, elemvect);
|
||||
interior_face_integs[k]->
|
||||
AssembleRHSElementVect(*fes->GetFE(tr->Elem1No),
|
||||
*pfes->GetFaceNbrFE(Elem2Nbr),
|
||||
*tr, elemvect);
|
||||
AddElementVector (vdofs, elemvect);
|
||||
}
|
||||
}
|
||||
|
||||
+5
-5
@@ -847,7 +847,7 @@ void H1FaceRestriction::Mult(const Vector& x, Vector& y) const
|
||||
});
|
||||
}
|
||||
|
||||
void H1FaceRestriction::MultTranspose(const Vector& x, Vector& y) const
|
||||
void H1FaceRestriction::AddMultTranspose(const Vector& x, Vector& y) const
|
||||
{
|
||||
// Assumes all elements have the same number of dofs
|
||||
const int nd = dof;
|
||||
@@ -856,7 +856,7 @@ void H1FaceRestriction::MultTranspose(const Vector& x, Vector& y) const
|
||||
auto d_offsets = offsets.Read();
|
||||
auto d_indices = gather_indices.Read();
|
||||
auto d_x = Reshape(x.Read(), nd, vd, nf);
|
||||
auto d_y = Reshape(y.Write(), t?vd:ndofs, t?ndofs:vd);
|
||||
auto d_y = Reshape(y.ReadWrite(), t?vd:ndofs, t?ndofs:vd);
|
||||
MFEM_FORALL(i, ndofs,
|
||||
{
|
||||
const int offset = d_offsets[i];
|
||||
@@ -1267,7 +1267,7 @@ void L2FaceRestriction::Mult(const Vector& x, Vector& y) const
|
||||
}
|
||||
}
|
||||
|
||||
void L2FaceRestriction::MultTranspose(const Vector& x, Vector& y) const
|
||||
void L2FaceRestriction::AddMultTranspose(const Vector& x, Vector& y) const
|
||||
{
|
||||
// Assumes all elements have the same number of dofs
|
||||
const int nd = dof;
|
||||
@@ -1280,7 +1280,7 @@ void L2FaceRestriction::MultTranspose(const Vector& x, Vector& y) const
|
||||
if (m == L2FaceValues::DoubleValued)
|
||||
{
|
||||
auto d_x = Reshape(x.Read(), nd, vd, 2, nf);
|
||||
auto d_y = Reshape(y.Write(), t?vd:ndofs, t?ndofs:vd);
|
||||
auto d_y = Reshape(y.ReadWrite(), t?vd:ndofs, t?ndofs:vd);
|
||||
MFEM_FORALL(i, ndofs,
|
||||
{
|
||||
const int offset = d_offsets[i];
|
||||
@@ -1304,7 +1304,7 @@ void L2FaceRestriction::MultTranspose(const Vector& x, Vector& y) const
|
||||
else
|
||||
{
|
||||
auto d_x = Reshape(x.Read(), nd, vd, nf);
|
||||
auto d_y = Reshape(y.Write(), t?vd:ndofs, t?ndofs:vd);
|
||||
auto d_y = Reshape(y.ReadWrite(), t?vd:ndofs, t?ndofs:vd);
|
||||
MFEM_FORALL(i, ndofs,
|
||||
{
|
||||
const int offset = d_offsets[i];
|
||||
|
||||
+121
-15
@@ -21,10 +21,6 @@ namespace mfem
|
||||
class FiniteElementSpace;
|
||||
enum class ElementDofOrdering;
|
||||
|
||||
/** An enum type to specify if only e1 value is requested (SingleValued) or both
|
||||
e1 and e2 (DoubleValued). */
|
||||
enum class L2FaceValues : bool {SingleValued, DoubleValued};
|
||||
|
||||
/// Operator that converts FiniteElementSpace L-vectors to E-vectors.
|
||||
/** Objects of this type are typically created and owned by FiniteElementSpace
|
||||
objects, see FiniteElementSpace::GetElementRestriction(). */
|
||||
@@ -104,10 +100,75 @@ public:
|
||||
void FillJAndData(const Vector &ea_data, SparseMatrix &mat) const;
|
||||
};
|
||||
|
||||
/// Operator that extracts Face degrees of freedom.
|
||||
/** An enum type to specify if only e1 value is requested (SingleValued) or both
|
||||
e1 and e2 (DoubleValued). */
|
||||
enum class L2FaceValues : bool {SingleValued, DoubleValued};
|
||||
|
||||
/** @brief Base class for operators that extracts Face degrees of freedom.
|
||||
|
||||
In order to compute quantities on the faces of a mesh, it is often useful to
|
||||
extract the degrees of freedom on the faces of the elements. This class
|
||||
provides an interface for such operations.
|
||||
|
||||
If the FiniteElementSpace is ordered by Ordering::byVDIM, then the expected
|
||||
format for the L-vector is (vdim x ndofs), otherwise if Ordering::byNODES
|
||||
the expected format is (ndofs x vdim), where ndofs is the total number of
|
||||
degrees of freedom.
|
||||
Since FiniteElementSpace can either be continuous or discontinuous, the
|
||||
degrees of freedom on a face can either be single valued or double valued,
|
||||
this is what we refer to as the multiplicity and is represented by the
|
||||
L2FaceValues enum type.
|
||||
The format of the output face E-vector of degrees of freedom is
|
||||
(face_dofs x vdim x multiplicity x nfaces), where face_dofs is the number of
|
||||
degrees of freedom on each face, and nfaces the number of faces of the
|
||||
requested FaceType (see FiniteElementSpace::GetNFbyType).
|
||||
|
||||
@note Objects of this type are typically created and owned by
|
||||
FiniteElementSpace objects, see FiniteElementSpace::GetFaceRestriction(). */
|
||||
class FaceRestriction : public Operator
|
||||
{
|
||||
public:
|
||||
FaceRestriction(): Operator() { }
|
||||
|
||||
FaceRestriction(int h, int w): Operator(h, w) { }
|
||||
|
||||
virtual ~FaceRestriction() { }
|
||||
|
||||
/** @brief Extract the face degrees of freedom from @a x into @a y.
|
||||
|
||||
@param[in] x The L-vector of degrees of freedom.
|
||||
@param[out] y The degrees of freedom on the face, corresponding to a face
|
||||
E-vector.
|
||||
*/
|
||||
void Mult(const Vector &x, Vector &y) const override = 0;
|
||||
|
||||
/** @brief Add the face degrees of freedom @a x to the element degrees of
|
||||
freedom @a y.
|
||||
|
||||
@param[in] x The face degrees of freedom on the face.
|
||||
@param[in,out] y The L-vector of degrees of freedom to which we add the
|
||||
face degrees of freedom.
|
||||
*/
|
||||
virtual void AddMultTranspose(const Vector &x, Vector &y) const = 0;
|
||||
|
||||
/** @brief Set the face degrees of freedom in the element degrees of freedom
|
||||
@a y to the values given in @a x.
|
||||
|
||||
@param[in] x The face degrees of freedom on the face.
|
||||
@param[in,out] y The L-vector of degrees of freedom to which we add the
|
||||
face degrees of freedom.
|
||||
*/
|
||||
void MultTranspose(const Vector &x, Vector &y) const override
|
||||
{
|
||||
y = 0.0;
|
||||
AddMultTranspose(x, y);
|
||||
}
|
||||
};
|
||||
|
||||
/// Operator that extracts Face degrees of freedom for H1 FiniteElementSpaces.
|
||||
/** Objects of this type are typically created and owned by FiniteElementSpace
|
||||
objects, see FiniteElementSpace::GetFaceRestriction(). */
|
||||
class H1FaceRestriction : public Operator
|
||||
class H1FaceRestriction : public FaceRestriction
|
||||
{
|
||||
protected:
|
||||
const FiniteElementSpace &fes;
|
||||
@@ -122,16 +183,42 @@ protected:
|
||||
Array<int> gather_indices;
|
||||
|
||||
public:
|
||||
H1FaceRestriction(const FiniteElementSpace&, const ElementDofOrdering,
|
||||
const FaceType);
|
||||
void Mult(const Vector &x, Vector &y) const;
|
||||
void MultTranspose(const Vector &x, Vector &y) const;
|
||||
/** @brief Constructor for a H1FaceRestriction.
|
||||
|
||||
@param[in] fes The FiniteElementSpace on which this H1FaceRestriction
|
||||
operates.
|
||||
@param[in] ordering The requested output ordering of the
|
||||
H1FaceRestriction, either Native or Lexicographic.
|
||||
@param[in] type The requested type of faces on which this operator
|
||||
extracts the degrees of freedom, either Interior or
|
||||
Boundary.
|
||||
*/
|
||||
H1FaceRestriction(const FiniteElementSpace& fes,
|
||||
const ElementDofOrdering ordering,
|
||||
const FaceType type);
|
||||
|
||||
/** @brief Extract the face degrees of freedom from @a x into @a y.
|
||||
|
||||
@param[in] x The L-vector of degrees of freedom.
|
||||
@param[out] y The degrees of freedom on the face, corresponding to a face
|
||||
E-vector.
|
||||
*/
|
||||
void Mult(const Vector &x, Vector &y) const override;
|
||||
|
||||
/** @brief Add the face degrees of freedom @a x to the element degrees of
|
||||
freedom @a y.
|
||||
|
||||
@param[in] x The face degrees of freedom on the face.
|
||||
@param[in,out] y The L-vector of degrees of freedom to which we add the
|
||||
face degrees of freedom.
|
||||
*/
|
||||
void AddMultTranspose(const Vector &x, Vector &y) const override;
|
||||
};
|
||||
|
||||
/// Operator that extracts Face degrees of freedom.
|
||||
/// Operator that extracts Face degrees of freedom on L2 FiniteElementSpaces.
|
||||
/** Objects of this type are typically created and owned by FiniteElementSpace
|
||||
objects, see FiniteElementSpace::GetFaceRestriction(). */
|
||||
class L2FaceRestriction : public Operator
|
||||
class L2FaceRestriction : public FaceRestriction
|
||||
{
|
||||
protected:
|
||||
const FiniteElementSpace &fes;
|
||||
@@ -154,19 +241,38 @@ protected:
|
||||
const L2FaceValues m = L2FaceValues::DoubleValued);
|
||||
|
||||
public:
|
||||
L2FaceRestriction(const FiniteElementSpace&, const ElementDofOrdering,
|
||||
L2FaceRestriction(const FiniteElementSpace&,
|
||||
const ElementDofOrdering,
|
||||
const FaceType,
|
||||
const L2FaceValues m = L2FaceValues::DoubleValued);
|
||||
virtual void Mult(const Vector &x, Vector &y) const;
|
||||
void MultTranspose(const Vector &x, Vector &y) const;
|
||||
|
||||
/** @brief Extract the face degrees of freedom from @a x into @a y.
|
||||
|
||||
@param[in] x The L-vector of degrees of freedom.
|
||||
@param[out] y The degrees of freedom on the face, corresponding to a face
|
||||
E-vector.
|
||||
*/
|
||||
void Mult(const Vector &x, Vector &y) const override;
|
||||
|
||||
/** @brief Add the face degrees of freedom @a x to the element degrees of
|
||||
freedom @a y.
|
||||
|
||||
@param[in] x The face degrees of freedom on the face.
|
||||
@param[in,out] y The L-vector of degrees of freedom to which we add the
|
||||
face degrees of freedom.
|
||||
*/
|
||||
void AddMultTranspose(const Vector &x, Vector &y) const override;
|
||||
|
||||
/** Fill the I array of SparseMatrix corresponding to the sparsity pattern
|
||||
given by this L2FaceRestriction. */
|
||||
virtual void FillI(SparseMatrix &mat, const bool keep_nbr_block = false) const;
|
||||
|
||||
/** Fill the J and Data arrays of SparseMatrix corresponding to the sparsity
|
||||
pattern given by this L2FaceRestriction, and the values of ea_data. */
|
||||
virtual void FillJAndData(const Vector &ea_data,
|
||||
SparseMatrix &mat,
|
||||
const bool keep_nbr_block = false) const;
|
||||
|
||||
/// This methods adds the DG face matrices to the element matrices.
|
||||
void AddFaceMatricesToElementMatrices(Vector &fea_data,
|
||||
Vector &ea_data) const;
|
||||
|
||||
@@ -34,6 +34,7 @@ void TMOP_Combo_QualityMetric::EvalP(const DenseMatrix &Jpt,
|
||||
DenseMatrix &P) const
|
||||
{
|
||||
DenseMatrix Pt(P.Size());
|
||||
P = 0.0;
|
||||
for (int i = 0; i < tmop_q_arr.Size(); i++)
|
||||
{
|
||||
tmop_q_arr[i]->EvalP(Jpt, Pt);
|
||||
@@ -50,6 +51,7 @@ void TMOP_Combo_QualityMetric::AssembleH(const DenseMatrix &Jpt,
|
||||
DenseMatrix At(A.Size());
|
||||
for (int i = 0; i < tmop_q_arr.Size(); i++)
|
||||
{
|
||||
At = 0.0;
|
||||
tmop_q_arr[i]->AssembleH(Jpt, DS, weight, At);
|
||||
At *= wt_arr[i];
|
||||
A += At;
|
||||
|
||||
@@ -371,6 +371,8 @@ public:
|
||||
AddQualityMetric(sh_metric, 1.-gamma_);
|
||||
AddQualityMetric(sz_metric, gamma_);
|
||||
}
|
||||
virtual int Id() const { return 80; }
|
||||
double GetGamma() const { return gamma; }
|
||||
|
||||
virtual ~TMOP_Metric_080() { delete sh_metric; delete sz_metric; }
|
||||
};
|
||||
@@ -590,6 +592,52 @@ public:
|
||||
virtual int Id() const { return 321; }
|
||||
};
|
||||
|
||||
/// 3D barrier Shape+Size (VS) metric (polyconvex).
|
||||
class TMOP_Metric_332 : public TMOP_Combo_QualityMetric
|
||||
{
|
||||
protected:
|
||||
double gamma;
|
||||
TMOP_QualityMetric *sh_metric, *sz_metric;
|
||||
|
||||
public:
|
||||
TMOP_Metric_332(double gamma_) : gamma(gamma_),
|
||||
sh_metric(new TMOP_Metric_302),
|
||||
sz_metric(new TMOP_Metric_315)
|
||||
{
|
||||
// (1-gamma) mu_302 + gamma mu_315
|
||||
AddQualityMetric(sh_metric, 1.-gamma_);
|
||||
AddQualityMetric(sz_metric, gamma_);
|
||||
}
|
||||
|
||||
virtual int Id() const { return 332; }
|
||||
double GetGamma() const { return gamma; }
|
||||
|
||||
virtual ~TMOP_Metric_332() { delete sh_metric; delete sz_metric; }
|
||||
};
|
||||
|
||||
/// 3D barrier Shape+Size (VS) metric (polyconvex).
|
||||
class TMOP_Metric_333 : public TMOP_Combo_QualityMetric
|
||||
{
|
||||
protected:
|
||||
double gamma;
|
||||
TMOP_QualityMetric *sh_metric, *sz_metric;
|
||||
|
||||
public:
|
||||
TMOP_Metric_333(double gamma_) : gamma(gamma_),
|
||||
sh_metric(new TMOP_Metric_302),
|
||||
sz_metric(new TMOP_Metric_316)
|
||||
{
|
||||
// (1-gamma) mu_302 + gamma mu_316
|
||||
AddQualityMetric(sh_metric, 1.-gamma_);
|
||||
AddQualityMetric(sz_metric, gamma_);
|
||||
}
|
||||
|
||||
virtual int Id() const { return 333; }
|
||||
double GetGamma() const { return gamma; }
|
||||
|
||||
virtual ~TMOP_Metric_333() { delete sh_metric; delete sz_metric; }
|
||||
};
|
||||
|
||||
/// Shifted barrier form of 3D metric 16 (volume, ideal barrier metric), 3D
|
||||
class TMOP_Metric_352 : public TMOP_QualityMetric
|
||||
{
|
||||
|
||||
@@ -150,9 +150,49 @@ void EvalH_077(const int e, const int qx, const int qy,
|
||||
}
|
||||
}
|
||||
|
||||
static MFEM_HOST_DEVICE inline
|
||||
void EvalH_080(const int e, const int qx, const int qy,
|
||||
const double weight, const double gamma, const double *Jpt,
|
||||
DeviceTensor<7,double> H)
|
||||
{
|
||||
// h_80 = (1-gamma) h_2 + gamma h_77.
|
||||
|
||||
constexpr int DIM = 2;
|
||||
double ddI1[4], ddI1b[4], dI2[4], dI2b[4], ddI2[4];
|
||||
kernels::InvariantsEvaluator2D ie(Args()
|
||||
.J(Jpt)
|
||||
.dI2(dI2)
|
||||
.ddI1(ddI1)
|
||||
.ddI1b(ddI1b)
|
||||
.dI2b(dI2b)
|
||||
.ddI2(ddI2));
|
||||
|
||||
const double I2 = ie.Get_I2(), I2inv_sq = 1.0 / (I2 * I2);
|
||||
ConstDeviceMatrix di2(ie.Get_dI2(),DIM,DIM);
|
||||
for (int i = 0; i < DIM; i++)
|
||||
{
|
||||
for (int j = 0; j < DIM; j++)
|
||||
{
|
||||
ConstDeviceMatrix ddi1b(ie.Get_ddI1b(i,j),DIM,DIM);
|
||||
ConstDeviceMatrix ddi2(ie.Get_ddI2(i,j),DIM,DIM);
|
||||
for (int r = 0; r < DIM; r++)
|
||||
{
|
||||
for (int c = 0; c < DIM; c++)
|
||||
{
|
||||
H(r,c,i,j,qx,qy,e) =
|
||||
(1.0 - gamma) * 0.5 * weight * ddi1b(r,c) +
|
||||
gamma * ( weight * 0.5 * (1.0 - I2inv_sq) * ddi2(r,c) +
|
||||
weight * (I2inv_sq / I2) * di2(r,c) * di2(i,j) );
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
MFEM_REGISTER_TMOP_KERNELS(void, SetupGradPA_2D,
|
||||
const Vector &x_,
|
||||
const double metric_normal,
|
||||
const double metric_param,
|
||||
const int mid,
|
||||
const int NE,
|
||||
const Array<double> &w_,
|
||||
@@ -163,7 +203,7 @@ MFEM_REGISTER_TMOP_KERNELS(void, SetupGradPA_2D,
|
||||
const int d1d,
|
||||
const int q1d)
|
||||
{
|
||||
MFEM_VERIFY(mid == 1 || mid == 2 || mid == 7 || mid == 77,
|
||||
MFEM_VERIFY(mid == 1 || mid == 2 || mid == 7 || mid == 77 || mid == 80,
|
||||
"Metric not yet implemented!");
|
||||
|
||||
constexpr int DIM = 2;
|
||||
@@ -222,6 +262,7 @@ MFEM_REGISTER_TMOP_KERNELS(void, SetupGradPA_2D,
|
||||
if (mid == 2) { EvalH_002(e,qx,qy,weight,Jpt,H); }
|
||||
if (mid == 7) { EvalH_007(e,qx,qy,weight,Jpt,H); }
|
||||
if (mid == 77) { EvalH_077(e,qx,qy,weight,Jpt,H); }
|
||||
if (mid == 80) { EvalH_080(e,qx,qy,weight,metric_param,Jpt,H); }
|
||||
} // qx
|
||||
} // qy
|
||||
});
|
||||
@@ -241,7 +282,10 @@ void TMOP_Integrator::AssembleGradPA_2D(const Vector &X) const
|
||||
const Array<double> &G = PA.maps->G;
|
||||
Vector &H = PA.H;
|
||||
|
||||
MFEM_LAUNCH_TMOP_KERNEL(SetupGradPA_2D,id,X,mn,M,N,W,B,G,J,H);
|
||||
double mp = 0.0;
|
||||
if (auto m = dynamic_cast<TMOP_Metric_080 *>(metric)) { mp = m->GetGamma(); }
|
||||
|
||||
MFEM_LAUNCH_TMOP_KERNEL(SetupGradPA_2D,id,X,mn,mp,M,N,W,B,G,J,H);
|
||||
}
|
||||
|
||||
} // namespace mfem
|
||||
|
||||
@@ -181,8 +181,58 @@ void EvalH_321(const int e, const int qx, const int qy, const int qz,
|
||||
}
|
||||
}
|
||||
|
||||
// H_332 = (1-gamma) H_302 + gamma H_315
|
||||
static MFEM_HOST_DEVICE inline
|
||||
void EvalH_332(const int e, const int qx, const int qy, const int qz,
|
||||
const double weight, const double gamma,
|
||||
const double *J, DeviceTensor<8,double> dP)
|
||||
{
|
||||
double B[9];
|
||||
double dI1b[9], ddI1b[9];
|
||||
double dI2[9], dI2b[9], ddI2[9], ddI2b[9];
|
||||
double dI3b[9], ddI3b[9];
|
||||
constexpr int DIM = 3;
|
||||
kernels::InvariantsEvaluator3D ie(Args()
|
||||
.J(J).B(B)
|
||||
.dI1b(dI1b).ddI1b(ddI1b)
|
||||
.dI2(dI2).dI2b(dI2b).ddI2(ddI2).ddI2b(ddI2b)
|
||||
.dI3b(dI3b).ddI3b(ddI3b));
|
||||
double sign_detJ;
|
||||
const double c1 = weight/9.;
|
||||
const double I1b = ie.Get_I1b();
|
||||
const double I2b = ie.Get_I2b();
|
||||
const double I3b = ie.Get_I3b(sign_detJ);
|
||||
ConstDeviceMatrix di1b(ie.Get_dI1b(),DIM,DIM);
|
||||
ConstDeviceMatrix di2b(ie.Get_dI2b(),DIM,DIM);
|
||||
ConstDeviceMatrix di3b(ie.Get_dI3b(sign_detJ),DIM,DIM);
|
||||
for (int i = 0; i < DIM; i++)
|
||||
{
|
||||
for (int j = 0; j < DIM; j++)
|
||||
{
|
||||
ConstDeviceMatrix ddi1b(ie.Get_ddI1b(i,j),DIM,DIM);
|
||||
ConstDeviceMatrix ddi2b(ie.Get_ddI2b(i,j),DIM,DIM);
|
||||
ConstDeviceMatrix ddi3b(ie.Get_ddI3b(i,j),DIM,DIM);
|
||||
for (int r = 0; r < DIM; r++)
|
||||
{
|
||||
for (int c = 0; c < DIM; c++)
|
||||
{
|
||||
const double dp_302 =
|
||||
(di2b(r,c)*di1b(i,j) + di1b(r,c)*di2b(i,j))
|
||||
+ ddi2b(r,c)*I1b
|
||||
+ ddi1b(r,c)*I2b;
|
||||
const double dp_315 = 2.0 * weight * (I3b - 1.0) * ddi3b(r,c) +
|
||||
2.0 * weight * di3b(r,c) * di3b(i,j);
|
||||
dP(r,c,i,j,qx,qy,qz,e) = (1.0 - gamma) * c1 * dp_302 +
|
||||
gamma * dp_315;
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
MFEM_REGISTER_TMOP_KERNELS(void, SetupGradPA_3D,
|
||||
const double metric_normal,
|
||||
const double metric_param,
|
||||
const int mid,
|
||||
const Vector &x_,
|
||||
const int NE,
|
||||
@@ -194,8 +244,8 @@ MFEM_REGISTER_TMOP_KERNELS(void, SetupGradPA_3D,
|
||||
const int d1d,
|
||||
const int q1d)
|
||||
{
|
||||
MFEM_VERIFY(mid == 302 || mid == 303 || mid == 315 || mid == 321 ,
|
||||
"3D metric not yet implemented!");
|
||||
MFEM_VERIFY(mid == 302 || mid == 303 || mid == 315 ||
|
||||
mid == 321 || mid == 332, "3D metric not yet implemented!");
|
||||
|
||||
constexpr int DIM = 3;
|
||||
const int D1D = T_D1D ? T_D1D : d1d;
|
||||
@@ -255,6 +305,7 @@ MFEM_REGISTER_TMOP_KERNELS(void, SetupGradPA_3D,
|
||||
if (mid == 303) { EvalH_303(e,qx,qy,qz,weight,Jpt,H); }
|
||||
if (mid == 315) { EvalH_315(e,qx,qy,qz,weight,Jpt,H); }
|
||||
if (mid == 321) { EvalH_321(e,qx,qy,qz,weight,Jpt,H); }
|
||||
if (mid == 332) { EvalH_332(e,qx,qy,qz,weight,metric_param,Jpt,H); }
|
||||
} // qx
|
||||
} // qy
|
||||
} // qz
|
||||
@@ -275,7 +326,10 @@ void TMOP_Integrator::AssembleGradPA_3D(const Vector &X) const
|
||||
const Array<double> &G = PA.maps->G;
|
||||
Vector &H = PA.H;
|
||||
|
||||
MFEM_LAUNCH_TMOP_KERNEL(SetupGradPA_3D,id,mn,M,X,N,W,B,G,J,H);
|
||||
double mp = 0.0;
|
||||
if (auto m = dynamic_cast<TMOP_Metric_332 *>(metric)) { mp = m->GetGamma(); }
|
||||
|
||||
MFEM_LAUNCH_TMOP_KERNEL(SetupGradPA_3D,id,mn,mp,M,X,N,W,B,G,J,H);
|
||||
}
|
||||
|
||||
} // namespace mfem
|
||||
|
||||
+22
-2
@@ -58,8 +58,24 @@ void EvalP_077(const double *Jpt, double *P)
|
||||
kernels::Set(2,2, 0.5 * (1.0 - 1.0 / (I2 * I2)), ie.Get_dI2(), P);
|
||||
}
|
||||
|
||||
static MFEM_HOST_DEVICE inline
|
||||
void EvalP_080(const double *Jpt, double gamma, double *P)
|
||||
{
|
||||
// p_80 = (1-gamma) p_2 + gamma p_77.
|
||||
|
||||
double dI1b[4], dI2[4], dI2b[4];
|
||||
kernels::InvariantsEvaluator2D ie(Args().J(Jpt).
|
||||
dI1b(dI1b).dI2(dI2).dI2b(dI2b));
|
||||
|
||||
kernels::Set(2,2, (1.0 - gamma) * 1./2., ie.Get_dI1b(), P);
|
||||
|
||||
const double I2 = ie.Get_I2();
|
||||
kernels::Add(2,2, gamma * 0.5 * (1.0 - 1.0 / (I2 * I2)), ie.Get_dI2(), P);
|
||||
}
|
||||
|
||||
MFEM_REGISTER_TMOP_KERNELS(void, AddMultPA_Kernel_2D,
|
||||
const double metric_normal,
|
||||
const double metric_param,
|
||||
const int mid,
|
||||
const int NE,
|
||||
const DenseTensor &j_,
|
||||
@@ -71,7 +87,7 @@ MFEM_REGISTER_TMOP_KERNELS(void, AddMultPA_Kernel_2D,
|
||||
const int d1d,
|
||||
const int q1d)
|
||||
{
|
||||
MFEM_VERIFY(mid == 1 || mid == 2 || mid == 7 || mid == 77,
|
||||
MFEM_VERIFY(mid == 1 || mid == 2 || mid == 7 || mid == 77 || mid == 80,
|
||||
"Metric not yet implemented!");
|
||||
|
||||
constexpr int DIM = 2;
|
||||
@@ -132,6 +148,7 @@ MFEM_REGISTER_TMOP_KERNELS(void, AddMultPA_Kernel_2D,
|
||||
if (mid == 2) { EvalP_002(Jpt, P); }
|
||||
if (mid == 7) { EvalP_007(Jpt, P); }
|
||||
if (mid == 77) { EvalP_077(Jpt, P); }
|
||||
if (mid == 80) { EvalP_080(Jpt, metric_param, P); }
|
||||
for (int i = 0; i < 4; i++) { P[i] *= weight; }
|
||||
|
||||
// PMatO += DS . P^t += DSh . (Jrt . P^t)
|
||||
@@ -160,7 +177,10 @@ void TMOP_Integrator::AddMultPA_2D(const Vector &X, Vector &Y) const
|
||||
const Array<double> &G = PA.maps->G;
|
||||
const double mn = metric_normal;
|
||||
|
||||
MFEM_LAUNCH_TMOP_KERNEL(AddMultPA_Kernel_2D,id,mn,M,N,J,W,B,G,X,Y);
|
||||
double mp = 0.0;
|
||||
if (auto m = dynamic_cast<TMOP_Metric_080 *>(metric)) { mp = m->GetGamma(); }
|
||||
|
||||
MFEM_LAUNCH_TMOP_KERNEL(AddMultPA_Kernel_2D,id,mn,mp,M,N,J,W,B,G,X,Y);
|
||||
}
|
||||
|
||||
} // namespace mfem
|
||||
|
||||
+32
-7
@@ -75,8 +75,29 @@ void EvalP_321(const double *J, double *P)
|
||||
kernels::Add(3,3, ie.Get_dI1(), P);
|
||||
}
|
||||
|
||||
// P_332 = (1-gamma) P_302 + gamma P_315.
|
||||
static MFEM_HOST_DEVICE inline
|
||||
void EvalP_332(const double *J, double gamma, double *P)
|
||||
{
|
||||
double B[9];
|
||||
double dI1b[9], dI2[9], dI2b[9], dI3b[9];
|
||||
kernels::InvariantsEvaluator3D ie(Args()
|
||||
.J(J).B(B)
|
||||
.dI1b(dI1b)
|
||||
.dI2(dI2).dI2b(dI2b)
|
||||
.dI3b(dI3b));
|
||||
const double alpha = (1.0 - gamma) * ie.Get_I1b()/9.;
|
||||
const double beta = (1.0 - gamma) * ie.Get_I2b()/9.;
|
||||
kernels::Add(3,3, alpha, ie.Get_dI2b(), beta, ie.Get_dI1b(), P);
|
||||
|
||||
double sign_detJ;
|
||||
const double I3b = ie.Get_I3b(sign_detJ);
|
||||
kernels::Add(3,3, gamma * 2.0 * (I3b - 1.0), ie.Get_dI3b(sign_detJ), P);
|
||||
}
|
||||
|
||||
MFEM_REGISTER_TMOP_KERNELS(void, AddMultPA_Kernel_3D,
|
||||
const double metric_normal,
|
||||
double metric_param,
|
||||
const int mid,
|
||||
const int NE,
|
||||
const DenseTensor &j_,
|
||||
@@ -88,8 +109,8 @@ MFEM_REGISTER_TMOP_KERNELS(void, AddMultPA_Kernel_3D,
|
||||
const int d1d,
|
||||
const int q1d)
|
||||
{
|
||||
MFEM_VERIFY(mid == 302 || mid == 303 || mid == 315 || mid == 321 ,
|
||||
"3D metric not yet implemented!");
|
||||
MFEM_VERIFY(mid == 302 || mid == 303 || mid == 315 ||
|
||||
mid == 321 || mid == 332, "3D metric not yet implemented!");
|
||||
|
||||
constexpr int DIM = 3;
|
||||
const int D1D = T_D1D ? T_D1D : d1d;
|
||||
@@ -146,10 +167,11 @@ MFEM_REGISTER_TMOP_KERNELS(void, AddMultPA_Kernel_3D,
|
||||
|
||||
// metric->EvalP(Jpt, P);
|
||||
double P[9];
|
||||
if (mid == 302) { EvalP_302(Jpt,P); }
|
||||
if (mid == 303) { EvalP_303(Jpt,P); }
|
||||
if (mid == 315) { EvalP_315(Jpt,P); }
|
||||
if (mid == 321) { EvalP_321(Jpt,P); }
|
||||
if (mid == 302) { EvalP_302(Jpt, P); }
|
||||
if (mid == 303) { EvalP_303(Jpt, P); }
|
||||
if (mid == 315) { EvalP_315(Jpt, P); }
|
||||
if (mid == 321) { EvalP_321(Jpt, P); }
|
||||
if (mid == 332) { EvalP_332(Jpt, metric_param, P); }
|
||||
for (int i = 0; i < 9; i++) { P[i] *= weight; }
|
||||
|
||||
// Y += DS . P^t += DSh . (Jrt . P^t)
|
||||
@@ -180,7 +202,10 @@ void TMOP_Integrator::AddMultPA_3D(const Vector &X, Vector &Y) const
|
||||
const Array<double> &G = PA.maps->G;
|
||||
const double mn = metric_normal;
|
||||
|
||||
MFEM_LAUNCH_TMOP_KERNEL(AddMultPA_Kernel_3D,id,mn,M,N,J,W,B,G,X,Y);
|
||||
double mp = 0.0;
|
||||
if (auto m = dynamic_cast<TMOP_Metric_332 *>(metric)) { mp = m->GetGamma(); }
|
||||
|
||||
MFEM_LAUNCH_TMOP_KERNEL(AddMultPA_Kernel_3D,id,mn,mp,M,N,J,W,B,G,X,Y);
|
||||
}
|
||||
|
||||
} // namespace mfem
|
||||
|
||||
+15
-4
@@ -50,8 +50,15 @@ double EvalW_077(const double *Jpt)
|
||||
return 0.5*(I2b*I2b + 1./(I2b*I2b) - 2.);
|
||||
}
|
||||
|
||||
static MFEM_HOST_DEVICE inline
|
||||
double EvalW_080(const double *Jpt, double gamma)
|
||||
{
|
||||
return (1.0 - gamma) * EvalW_002(Jpt) + gamma * EvalW_077(Jpt);
|
||||
}
|
||||
|
||||
MFEM_REGISTER_TMOP_KERNELS(double, EnergyPA_2D,
|
||||
const double metric_normal,
|
||||
const double metric_param,
|
||||
const int mid,
|
||||
const int NE,
|
||||
const DenseTensor &j_,
|
||||
@@ -64,7 +71,7 @@ MFEM_REGISTER_TMOP_KERNELS(double, EnergyPA_2D,
|
||||
const int d1d,
|
||||
const int q1d)
|
||||
{
|
||||
MFEM_VERIFY(mid == 1 || mid == 2 || mid == 7 || mid == 77,
|
||||
MFEM_VERIFY(mid == 1 || mid == 2 || mid == 7 || mid == 77 || mid == 80,
|
||||
"2D metric not yet implemented!");
|
||||
|
||||
constexpr int DIM = 2;
|
||||
@@ -125,7 +132,8 @@ MFEM_REGISTER_TMOP_KERNELS(double, EnergyPA_2D,
|
||||
mid == 1 ? EvalW_001(Jpt) :
|
||||
mid == 2 ? EvalW_002(Jpt) :
|
||||
mid == 7 ? EvalW_007(Jpt) :
|
||||
mid == 77 ? EvalW_077(Jpt) : 0.0;
|
||||
mid == 77 ? EvalW_077(Jpt) :
|
||||
mid == 80 ? EvalW_080(Jpt, metric_param) : 0.0;
|
||||
|
||||
E(qx,qy,e) = weight * EvalW;
|
||||
}
|
||||
@@ -141,7 +149,7 @@ double TMOP_Integrator::GetLocalStateEnergyPA_2D(const Vector &X) const
|
||||
const int D1D = PA.maps->ndof;
|
||||
const int Q1D = PA.maps->nqpt;
|
||||
const int id = (D1D << 4 ) | Q1D;
|
||||
const double m = metric_normal;
|
||||
const double mn = metric_normal;
|
||||
const DenseTensor &J = PA.Jtr;
|
||||
const Array<double> &W = PA.ir->GetWeights();
|
||||
const Array<double> &B = PA.maps->B;
|
||||
@@ -149,7 +157,10 @@ double TMOP_Integrator::GetLocalStateEnergyPA_2D(const Vector &X) const
|
||||
const Vector &O = PA.O;
|
||||
Vector &E = PA.E;
|
||||
|
||||
MFEM_LAUNCH_TMOP_KERNEL(EnergyPA_2D,id,m,M,N,J,W,B,G,X,O,E);
|
||||
double mp = 0.0;
|
||||
if (auto m = dynamic_cast<TMOP_Metric_080 *>(metric)) { mp = m->GetGamma(); }
|
||||
|
||||
MFEM_LAUNCH_TMOP_KERNEL(EnergyPA_2D,id,mn,mp,M,N,J,W,B,G,X,O,E);
|
||||
}
|
||||
|
||||
} // namespace mfem
|
||||
|
||||
+15
-4
@@ -58,8 +58,15 @@ double EvalW_321(const double *J)
|
||||
return ie.Get_I1() + ie.Get_I2()/ie.Get_I3() - 6.0;
|
||||
}
|
||||
|
||||
static MFEM_HOST_DEVICE inline
|
||||
double EvalW_332(const double *J, double gamma)
|
||||
{
|
||||
return (1.0 - gamma) * EvalW_302(J) + gamma * EvalW_315(J);
|
||||
}
|
||||
|
||||
MFEM_REGISTER_TMOP_KERNELS(double, EnergyPA_3D,
|
||||
const double metric_normal,
|
||||
const double metric_param,
|
||||
const int mid,
|
||||
const int NE,
|
||||
const DenseTensor &j_,
|
||||
@@ -72,8 +79,8 @@ MFEM_REGISTER_TMOP_KERNELS(double, EnergyPA_3D,
|
||||
const int d1d,
|
||||
const int q1d)
|
||||
{
|
||||
MFEM_VERIFY(mid == 302 || mid == 303 || mid == 315 || mid == 321 ,
|
||||
"3D metric not yet implemented!");
|
||||
MFEM_VERIFY(mid == 302 || mid == 303 || mid == 315 ||
|
||||
mid == 321 || mid == 332, "3D metric not yet implemented!");
|
||||
|
||||
constexpr int DIM = 3;
|
||||
const int D1D = T_D1D ? T_D1D : d1d;
|
||||
@@ -134,7 +141,8 @@ MFEM_REGISTER_TMOP_KERNELS(double, EnergyPA_3D,
|
||||
mid == 302 ? EvalW_302(Jpt) :
|
||||
mid == 303 ? EvalW_303(Jpt) :
|
||||
mid == 315 ? EvalW_315(Jpt) :
|
||||
mid == 321 ? EvalW_321(Jpt) : 0.0;
|
||||
mid == 321 ? EvalW_321(Jpt) :
|
||||
mid == 332 ? EvalW_332(Jpt, metric_param) : 0.0;
|
||||
|
||||
E(qx,qy,qz,e) = weight * EvalW;
|
||||
}
|
||||
@@ -159,7 +167,10 @@ double TMOP_Integrator::GetLocalStateEnergyPA_3D(const Vector &X) const
|
||||
const Vector &O = PA.O;
|
||||
Vector &E = PA.E;
|
||||
|
||||
MFEM_LAUNCH_TMOP_KERNEL(EnergyPA_3D,id,mn,M,N,J,W,B,G,O,X,E);
|
||||
double mp = 0.0;
|
||||
if (auto m = dynamic_cast<TMOP_Metric_332 *>(metric)) { mp = m->GetGamma(); }
|
||||
|
||||
MFEM_LAUNCH_TMOP_KERNEL(EnergyPA_3D,id,mn,mp,M,N,J,W,B,G,O,X,E);
|
||||
}
|
||||
|
||||
} // namespace mfem
|
||||
|
||||
@@ -71,7 +71,7 @@ static const unsigned char b64table[] =
|
||||
255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255
|
||||
};
|
||||
|
||||
void DecodeBase64(const char *src, size_t len, std::vector<unsigned char> &buf)
|
||||
void DecodeBase64(const char *src, size_t len, std::vector<char> &buf)
|
||||
{
|
||||
const unsigned char *in = (const unsigned char *)src;
|
||||
buf.clear();
|
||||
@@ -79,7 +79,7 @@ void DecodeBase64(const char *src, size_t len, std::vector<unsigned char> &buf)
|
||||
for (size_t i=0; i<len; ++i) { if (b64table[in[i]] != 255) { ++count; } }
|
||||
if (count % 4 != 0) { return; }
|
||||
buf.resize(3*len/4);
|
||||
unsigned char *out = buf.data();
|
||||
unsigned char *out = (unsigned char *)buf.data();
|
||||
count = 0;
|
||||
int pad = 0;
|
||||
unsigned char c[4];
|
||||
@@ -97,8 +97,10 @@ void DecodeBase64(const char *src, size_t len, std::vector<unsigned char> &buf)
|
||||
count = pad = 0;
|
||||
}
|
||||
}
|
||||
buf.resize(out - buf.data());
|
||||
buf.resize(out - (unsigned char *)buf.data());
|
||||
}
|
||||
|
||||
size_t NumBase64Chars(size_t nbytes) { return ((4*nbytes/3) + 3) & ~3; }
|
||||
|
||||
} // namespace mfem::bin_io
|
||||
} // namespace mfem
|
||||
|
||||
+11
-2
@@ -50,6 +50,7 @@ inline T read(const char *buf)
|
||||
return value;
|
||||
}
|
||||
|
||||
/// Append the binary representation of @a val to the byte buffer @a vec.
|
||||
template <typename T>
|
||||
void AppendBytes(std::vector<char> &vec, const T &val)
|
||||
{
|
||||
@@ -57,9 +58,17 @@ void AppendBytes(std::vector<char> &vec, const T &val)
|
||||
vec.insert(vec.end(), ptr, ptr + sizeof(T));
|
||||
}
|
||||
|
||||
void WriteBase64(std::ostream &out, const void *bytes, size_t length);
|
||||
/// Given a buffer @a buf of length @a nbytes, encode the data in base-64
|
||||
/// format, and write the encoded data to the output stream @a out.
|
||||
void WriteBase64(std::ostream &out, const void *bytes, size_t nbytes);
|
||||
|
||||
void DecodeBase64(const char *src, size_t len, std::vector<unsigned char> &buf);
|
||||
/// Decode @a len base-64 encoded characters in the buffer @a src, and store the
|
||||
/// resulting decoded data in @a buf. @a buf will be resized as needed.
|
||||
void DecodeBase64(const char *src, size_t len, std::vector<char> &buf);
|
||||
|
||||
/// Return the number of characters needed to encode @a nbytes in base-64. This
|
||||
/// is equal to 4*nbytes/3, rounded up to the nearest multiple of 4.
|
||||
size_t NumBase64Chars(size_t nbytes);
|
||||
|
||||
} // namespace mfem::bin_io
|
||||
|
||||
|
||||
@@ -17,6 +17,7 @@ list(APPEND SRCS
|
||||
complex_operator.cpp
|
||||
constraints.cpp
|
||||
densemat.cpp
|
||||
fdsolver.cpp
|
||||
symmat.cpp
|
||||
handle.cpp
|
||||
matrix.cpp
|
||||
@@ -39,6 +40,7 @@ list(APPEND HDRS
|
||||
dinvariants.hpp
|
||||
symmat.hpp
|
||||
dtensor.hpp
|
||||
fdsolver.hpp
|
||||
handle.hpp
|
||||
invariants.hpp
|
||||
kernels.hpp
|
||||
|
||||
+10
-8
@@ -208,11 +208,14 @@ void AmgXSolver::DefaultParameters(const AMGX_MODE amgxMode_,
|
||||
" \"config_version\": 2, \n"
|
||||
" \"solver\": { \n"
|
||||
" \"solver\": \"AMG\", \n"
|
||||
" \"scope\": \"main\", \n"
|
||||
" \"smoother\": \"JACOBI_L1\", \n"
|
||||
" \"presweeps\": 1, \n"
|
||||
" \"postsweeps\": 1, \n"
|
||||
" \"interpolator\": \"D2\", \n"
|
||||
" \"max_iters\": 2, \n"
|
||||
" \"convergence\": \"ABSOLUTE\", \n"
|
||||
" \"max_row_sum\" : 0.9, \n"
|
||||
" \"strength_threshold\" : 0.25, \n"
|
||||
" \"postsweeps\": 1, \n"
|
||||
" \"max_iters\": 1, \n"
|
||||
" \"cycle\": \"V\"";
|
||||
if (verbose)
|
||||
{
|
||||
@@ -239,22 +242,21 @@ void AmgXSolver::DefaultParameters(const AMGX_MODE amgxMode_,
|
||||
" \"solver\": \"AMG\", \n"
|
||||
" \"smoother\": { \n"
|
||||
" \"scope\": \"jacobi\", \n"
|
||||
" \"solver\": \"BLOCK_JACOBI\", \n"
|
||||
" \"relaxation_factor\": 0.7 \n"
|
||||
" \"solver\": \"JACOBI_L1\" \n"
|
||||
" }, \n"
|
||||
" \"presweeps\": 1, \n"
|
||||
" \"interpolator\": \"D2\", \n"
|
||||
" \"max_row_sum\" : 0.9, \n"
|
||||
" \"strength_threshold\" : 0.25, \n"
|
||||
" \"max_iters\": 2, \n"
|
||||
" \"max_iters\": 1, \n"
|
||||
" \"scope\": \"amg\", \n"
|
||||
" \"max_levels\": 100, \n"
|
||||
" \"cycle\": \"V\", \n"
|
||||
" \"postsweeps\": 1 \n"
|
||||
" }, \n"
|
||||
" \"solver\": \"PCG\", \n"
|
||||
" \"max_iters\": 100, \n"
|
||||
" \"convergence\": \"RELATIVE_MAX\", \n"
|
||||
" \"max_iters\": 150, \n"
|
||||
" \"convergence\": \"RELATIVE_INI_CORE\", \n"
|
||||
" \"scope\": \"main\", \n"
|
||||
" \"tolerance\": 1e-12, \n"
|
||||
" \"monitor_residual\": 1, \n"
|
||||
|
||||
+276
-17
@@ -50,6 +50,10 @@ dsyevr_(char *JOBZ, char *RANGE, char *UPLO, int *N, double *A, int *LDA,
|
||||
double *W, double *Z, int *LDZ, int *ISUPPZ, double *WORK, int *LWORK,
|
||||
int *IWORK, int *LIWORK, int *INFO);
|
||||
extern "C" void
|
||||
dgeev_(const char * jobvl, const char * jobvr, int *n, double * A, int * lda,
|
||||
double * wr, double * wl, double * vl, int * ldvl, double * vr, int * ldvr,
|
||||
double * work, int * lwork, int * info);
|
||||
extern "C" void
|
||||
dsyev_(char *JOBZ, char *UPLO, int *N, double *A, int *LDA, double *W,
|
||||
double *WORK, int *LWORK, int *INFO);
|
||||
extern "C" void
|
||||
@@ -174,7 +178,14 @@ const double &DenseMatrix::Elem(int i, int j) const
|
||||
|
||||
void DenseMatrix::Mult(const double *x, double *y) const
|
||||
{
|
||||
kernels::Mult(height, width, Data(), x, y);
|
||||
const double *data = Read();
|
||||
const int h = height;
|
||||
const int w = width;
|
||||
|
||||
MFEM_FORALL(i, 1,
|
||||
{
|
||||
kernels::Mult(h, w, data, x, y);
|
||||
});
|
||||
}
|
||||
|
||||
void DenseMatrix::Mult(const Vector &x, Vector &y) const
|
||||
@@ -182,7 +193,9 @@ void DenseMatrix::Mult(const Vector &x, Vector &y) const
|
||||
MFEM_ASSERT(height == y.Size() && width == x.Size(),
|
||||
"incompatible dimensions");
|
||||
|
||||
Mult((const double *)x, (double *)y);
|
||||
const double *dx = x.Read();
|
||||
double *dy = y.ReadWrite();
|
||||
Mult(dx, dy);
|
||||
}
|
||||
|
||||
double DenseMatrix::operator *(const DenseMatrix &m) const
|
||||
@@ -2003,7 +2016,7 @@ void Mult(const DenseMatrix &b, const DenseMatrix &c, DenseMatrix &a)
|
||||
MFEM_ASSERT(a.Height() == b.Height() && a.Width() == c.Width() &&
|
||||
b.Width() == c.Height(), "incompatible dimensions");
|
||||
|
||||
#ifdef MFEM_USE_LAPACK
|
||||
#if defined(MFEM_USE_LAPACK) && !defined(MFEM_USE_CUDA) && !defined(MFEM_USE_HIP)
|
||||
static char transa = 'N', transb = 'N';
|
||||
static double alpha = 1.0, beta = 0.0;
|
||||
int m = b.Height(), n = c.Width(), k = b.Width();
|
||||
@@ -2014,10 +2027,13 @@ void Mult(const DenseMatrix &b, const DenseMatrix &c, DenseMatrix &a)
|
||||
const int ah = a.Height();
|
||||
const int aw = a.Width();
|
||||
const int bw = b.Width();
|
||||
double *ad = a.Data();
|
||||
const double *bd = b.Data();
|
||||
const double *cd = c.Data();
|
||||
kernels::Mult(ah,aw,bw,bd,cd,ad);
|
||||
double *ad = a.ReadWrite();
|
||||
const double *bd = b.Read();
|
||||
const double *cd = c.Read();
|
||||
MFEM_FORALL(i, 1,
|
||||
{
|
||||
kernels::Mult(ah, aw, bw, bd, cd, ad);
|
||||
});
|
||||
#endif
|
||||
}
|
||||
|
||||
@@ -2859,6 +2875,155 @@ void AddMult_a_VVt(const double a, const Vector &v, DenseMatrix &VVt)
|
||||
}
|
||||
}
|
||||
|
||||
void KronProd(const DenseMatrix & A, const DenseMatrix & B, DenseMatrix & C)
|
||||
{
|
||||
const int ah = A.Height();
|
||||
const int aw = A.Width();
|
||||
const int bh = B.Height();
|
||||
const int bw = B.Width();
|
||||
|
||||
C.SetSize(ah*bh,aw*bw);
|
||||
const double * ad = A.Read();
|
||||
const double * bd = B.Read();
|
||||
double * cd = C.ReadWrite();
|
||||
|
||||
MFEM_FORALL(i, 1,
|
||||
{
|
||||
for (int ja = 0; ja<aw; ++ja)
|
||||
for (int jb = 0; jb<bw; ++jb)
|
||||
for (int ia = 0; ia<ah; ++ia)
|
||||
for (int ib = 0; ib<bh; ++ib)
|
||||
cd[bh*ia + ib + ah*bh*(bw*ja + jb)]
|
||||
= ad[ia + ja * ah] * bd[ib + jb*bh];
|
||||
});
|
||||
}
|
||||
|
||||
#if 0 // this is a finer level parallel KronProd
|
||||
void KronProd2(const DenseMatrix & A, const DenseMatrix & B, DenseMatrix & C)
|
||||
{
|
||||
const int ah = A.Height();
|
||||
const int aw = A.Width();
|
||||
const int bh = B.Height();
|
||||
const int bw = B.Width();
|
||||
const int ch = ah * bh;
|
||||
const int cw = aw * bw;
|
||||
|
||||
C.SetSize(ch, cw);
|
||||
const double *ad = A.Read();
|
||||
const double *bd = B.Read();
|
||||
double *cd = C.ReadWrite();
|
||||
|
||||
MFEM_FORALL(i, ch * cw,
|
||||
{
|
||||
const int jc = i / ch;
|
||||
const int ic = i - jc * ch;
|
||||
const int ja = jc / bw;
|
||||
const int jb = jc - ja * bw;
|
||||
const int ia = ic / bh;
|
||||
const int ib = ic - ia * bh;
|
||||
cd[jc * ch + ic] = ad[ja * ah + ia] * bd[jb * bh + ib];
|
||||
});
|
||||
}
|
||||
#endif
|
||||
|
||||
void KronMult(const DenseMatrix &A, const DenseMatrix &B, const Vector &r,
|
||||
Vector &z)
|
||||
{
|
||||
const int nA = A.Height();
|
||||
const int mA = A.Width();
|
||||
const int nB = B.Height();
|
||||
const int mB = B.Width();
|
||||
const int nr = r.Size();
|
||||
MFEM_VERIFY(nr == mA*mB, "Wrong size of Vector r");
|
||||
z.SetSize(nA*nB);
|
||||
#if !defined(MFEM_USE_CUDA)
|
||||
DenseMatrix R(r.GetData(),mB,mA);
|
||||
DenseMatrix X(nB,mA);
|
||||
DenseMatrix Y(z.GetData(),nB,nA);
|
||||
Mult(B,R,X);
|
||||
MultABt(X,A,Y);
|
||||
#else
|
||||
const double *ad = A.Read();
|
||||
const double *bd = B.Read();
|
||||
const double *rd = r.Read();
|
||||
double *zd = z.Write();
|
||||
MFEM_FORALL(i, 1,
|
||||
{
|
||||
kernels::KronMult(nA, mA, ad, nB, mB, bd, rd, zd);
|
||||
});
|
||||
#endif
|
||||
}
|
||||
|
||||
void KronMult(const DenseMatrix &A, const DenseMatrix &B, const DenseMatrix &R,
|
||||
DenseMatrix & Z)
|
||||
{
|
||||
const int nA = A.Height();
|
||||
const int nB = B.Height();
|
||||
const int nR = R.Height();
|
||||
const int mR = R.Width();
|
||||
Z.SetSize(nA*nB,mR);
|
||||
|
||||
Vector r,z;
|
||||
double * dataR = R.Data();
|
||||
for (int i = 0; i<mR; i++)
|
||||
{
|
||||
r.SetDataAndSize(&dataR[i*nR],nR);
|
||||
KronMult(A,B,r,z);
|
||||
Z.SetCol(i,z);
|
||||
}
|
||||
}
|
||||
|
||||
void KronMult(const DenseMatrix &A, const DenseMatrix &B, const DenseMatrix &C,
|
||||
const Vector &r, Vector &z)
|
||||
{
|
||||
const int nA = A.Height();
|
||||
const int mA = A.Width();
|
||||
const int nB = B.Height();
|
||||
const int mB = B.Width();
|
||||
const int nC = C.Height();
|
||||
const int mC = C.Width();
|
||||
const int nr = r.Size();
|
||||
MFEM_VERIFY(nr == mA*mB*mC, "Wrong size of Vector r");
|
||||
z.SetSize(nA*nB*nC);
|
||||
|
||||
#if !defined(MFEM_USE_CUDA)
|
||||
double * dataR = r.GetData();
|
||||
DenseMatrix R(dataR,mC,mA*mB);
|
||||
DenseMatrix X(nC,mA*mB);
|
||||
Mult(C,R,X);
|
||||
X.Transpose();
|
||||
DenseMatrix Z(z.GetData(),mA*mB,nC);
|
||||
KronMult(A,B,X,Z);
|
||||
Z.Transpose();
|
||||
#else
|
||||
const double *ad = A.Read();
|
||||
const double *bd = B.Read();
|
||||
const double *cd = C.Read();
|
||||
const double *rd = r.Read();
|
||||
double *zd = z.Write();
|
||||
MFEM_FORALL(i, 1,
|
||||
{
|
||||
kernels::KronMult(nA, mA, ad, nB, mB, bd, nC, mC, cd, rd, zd);
|
||||
});
|
||||
#endif
|
||||
}
|
||||
|
||||
void KronMult(const Array<DenseMatrix *> & A, const Vector & r, Vector & z)
|
||||
{
|
||||
int dim = A.Size();
|
||||
if (dim == 2)
|
||||
{
|
||||
KronMult(*A[0],*A[1],r,z);
|
||||
}
|
||||
else if (dim == 3)
|
||||
{
|
||||
KronMult(*A[0],*A[1],*A[2], r,z);
|
||||
}
|
||||
else
|
||||
{
|
||||
MFEM_ABORT("KronMult::Wrong dimension");
|
||||
}
|
||||
}
|
||||
|
||||
bool LUFactors::Factor(int m, double TOL)
|
||||
{
|
||||
@@ -3310,23 +3475,105 @@ DenseMatrixInverse::~DenseMatrixInverse()
|
||||
delete [] lu.ipiv;
|
||||
}
|
||||
|
||||
void KronMult(const DenseMatrixInverse &A, const DenseMatrixInverse &B,
|
||||
const Vector &r, Vector & z)
|
||||
{
|
||||
// A and B are square matrices
|
||||
z.SetSize(r.Size());
|
||||
int nA = A.Height();
|
||||
int nB = B.Height();
|
||||
DenseMatrix R(r.GetData(),nB,nA);
|
||||
DenseMatrix X(nB,nA);
|
||||
B.Mult(R,X);
|
||||
X.Transpose();
|
||||
DenseMatrix Y(z.GetData(),nA,nB);
|
||||
A.Mult(X,Y);
|
||||
Y.Transpose();
|
||||
}
|
||||
|
||||
DenseMatrixEigensystem::DenseMatrixEigensystem(DenseMatrix &m)
|
||||
void KronMult(const DenseMatrixInverse &A, const DenseMatrixInverse &B,
|
||||
const DenseMatrix &R, DenseMatrix & Z)
|
||||
{
|
||||
// A and B are square matrices
|
||||
int nR = R.Height();
|
||||
int mR = R.Width();
|
||||
Z.SetSize(nR,mR);
|
||||
Vector r(nR);
|
||||
Vector z(nR);
|
||||
double * dataR = R.GetData();
|
||||
double * dataZ = Z.GetData();
|
||||
for (int i = 0; i<mR; i++)
|
||||
{
|
||||
r.SetData(&dataR[i*nR]);
|
||||
z.SetData(&dataZ[i*nR]);
|
||||
KronMult(A,B,r,z);
|
||||
}
|
||||
}
|
||||
|
||||
void KronMult(const DenseMatrixInverse &A, const DenseMatrixInverse &B,
|
||||
const DenseMatrixInverse &C, const Vector &r, Vector & z)
|
||||
{
|
||||
// A, B and C are square matrices
|
||||
int n = r.Size();
|
||||
z.SetSize(n);
|
||||
int nA = A.Height();
|
||||
int nB = B.Height();
|
||||
int nC = C.Height();
|
||||
double * dataR = r.GetData();
|
||||
DenseMatrix R(dataR,nC,nA*nB);
|
||||
DenseMatrix X(nC,nA*nB);
|
||||
C.Mult(R,X);
|
||||
X.Transpose();
|
||||
DenseMatrix Z(z.GetData(),nC,nA*nB);
|
||||
KronMult(A,B,X,Z);
|
||||
Z.Transpose();
|
||||
}
|
||||
|
||||
void KronMult(const Array<DenseMatrixInverse *> & A, const Vector & r,
|
||||
Vector & z)
|
||||
{
|
||||
int dim = A.Size();
|
||||
if (dim == 2)
|
||||
{
|
||||
KronMult(*A[0],*A[1],r,z);
|
||||
}
|
||||
else if (dim == 3)
|
||||
{
|
||||
KronMult(*A[0],*A[1],*A[2], r,z);
|
||||
}
|
||||
else
|
||||
{
|
||||
MFEM_ABORT("KronMult::Wrong dimension");
|
||||
}
|
||||
}
|
||||
|
||||
DenseMatrixEigensystem::DenseMatrixEigensystem(DenseMatrix &m, bool sym_)
|
||||
: mat(m)
|
||||
{
|
||||
n = mat.Width();
|
||||
EVal.SetSize(n);
|
||||
EVali.SetSize(n);
|
||||
EVect.SetSize(n);
|
||||
ev.SetDataAndSize(NULL, n);
|
||||
|
||||
#ifdef MFEM_USE_LAPACK
|
||||
sym = sym_;
|
||||
jobz = 'V';
|
||||
uplo = 'U';
|
||||
lwork = -1;
|
||||
double qwork;
|
||||
dsyev_(&jobz, &uplo, &n, EVect.Data(), &n, EVal.GetData(),
|
||||
&qwork, &lwork, &info);
|
||||
|
||||
if (sym)
|
||||
{
|
||||
uplo = 'U';
|
||||
dsyev_(&jobz, &uplo, &n, EVect.Data(), &n, EVal.GetData(),
|
||||
&qwork, &lwork, &info);
|
||||
}
|
||||
else
|
||||
{
|
||||
char jobvl = 'N';
|
||||
int ldvl = 1;
|
||||
dgeev_(&jobvl,&jobz,&n, mat.GetData(), &n, EVal.GetData(), EVali.GetData(),
|
||||
nullptr, &ldvl, EVect.GetData(), &n, &qwork, &lwork, &info);
|
||||
}
|
||||
lwork = (int) qwork;
|
||||
work = new double[lwork];
|
||||
#endif
|
||||
@@ -3338,6 +3585,7 @@ DenseMatrixEigensystem::DenseMatrixEigensystem(
|
||||
n(other.n)
|
||||
{
|
||||
#ifdef MFEM_USE_LAPACK
|
||||
sym = other.sym;
|
||||
jobz = other.jobz;
|
||||
uplo = other.uplo;
|
||||
lwork = other.lwork;
|
||||
@@ -3356,13 +3604,24 @@ void DenseMatrixEigensystem::Eval()
|
||||
#endif
|
||||
|
||||
#ifdef MFEM_USE_LAPACK
|
||||
EVect = mat;
|
||||
dsyev_(&jobz, &uplo, &n, EVect.Data(), &n, EVal.GetData(),
|
||||
work, &lwork, &info);
|
||||
|
||||
if (sym)
|
||||
{
|
||||
EVect = mat;
|
||||
dsyev_(&jobz, &uplo, &n, EVect.Data(), &n, EVal.GetData(),
|
||||
work, &lwork, &info);
|
||||
}
|
||||
else
|
||||
{
|
||||
char jobvl = 'N';
|
||||
int ldvl = 1;
|
||||
DenseMatrix T = mat; // mat is overwritten by dgeev
|
||||
dgeev_(&jobvl,&jobz,&n, T.GetData(), &n, EVal.GetData(), EVali.GetData(),
|
||||
nullptr, &ldvl, EVect.GetData(), &n, work, &lwork, &info);
|
||||
}
|
||||
if (info != 0)
|
||||
{
|
||||
mfem::err << "DenseMatrixEigensystem::Eval(): DSYEV error code: "
|
||||
string lpck = (sym) ? "DSYEV" : "DGEEV";
|
||||
mfem::err << "DenseMatrixEigensystem::Eval(): " << lpck << "error code: "
|
||||
<< info << endl;
|
||||
mfem_error();
|
||||
}
|
||||
|
||||
+36
-2
@@ -523,6 +523,23 @@ void AddMult_a_VWt(const double a, const Vector &v, const Vector &w,
|
||||
/// VVt += a * v v^t
|
||||
void AddMult_a_VVt(const double a, const Vector &v, DenseMatrix &VVt);
|
||||
|
||||
/// C = A ⊗ B
|
||||
void KronProd(const DenseMatrix & A, const DenseMatrix & B, DenseMatrix & C);
|
||||
void KronProd2(const DenseMatrix & A, const DenseMatrix & B, DenseMatrix & C);
|
||||
|
||||
/// z = (A ⊗ B) r = vec(B R A^T), where R := vec^-1 (r)
|
||||
void KronMult(const DenseMatrix &A, const DenseMatrix &B, const Vector &r,
|
||||
Vector & z);
|
||||
|
||||
/// z = (A ⊗ B) R
|
||||
void KronMult(const DenseMatrix &A, const DenseMatrix &B, const DenseMatrix &R,
|
||||
DenseMatrix & Z);
|
||||
|
||||
/// z = ( A ⊗ B ⊗ C ) r
|
||||
void KronMult(const DenseMatrix &A, const DenseMatrix &B, const DenseMatrix &C,
|
||||
const Vector &r, Vector & z);
|
||||
|
||||
void KronMult(const Array<DenseMatrix *> & A, const Vector & r, Vector & z);
|
||||
|
||||
/** Class that can compute LU factorization of external data and perform various
|
||||
operations with the factored data. */
|
||||
@@ -685,16 +702,33 @@ public:
|
||||
virtual ~DenseMatrixInverse();
|
||||
};
|
||||
|
||||
/// z = (A^-1 ⊗ B^-1) r = vec(B^-1 R A^-T), where R := vec^-1 (r)
|
||||
void KronMult(const DenseMatrixInverse &Ainv, const DenseMatrixInverse &Binv,
|
||||
const Vector &r, Vector & z);
|
||||
|
||||
/// z = (A^-1 ⊗ B^-1) R
|
||||
void KronMult(const DenseMatrixInverse &Ainv, const DenseMatrixInverse &Binv,
|
||||
const DenseMatrix &R, DenseMatrix & Z);
|
||||
|
||||
/// z = ( A^-1 ⊗ B^-1 ⊗ C^-1 ) r
|
||||
void KronMult(const DenseMatrixInverse &Ainv, const DenseMatrixInverse &Binv,
|
||||
const DenseMatrixInverse &Cinv, const Vector &r, Vector & z);
|
||||
|
||||
void KronMult(const Array<DenseMatrixInverse *> & A, const Vector & r,
|
||||
Vector & z);
|
||||
|
||||
class DenseMatrixEigensystem
|
||||
{
|
||||
DenseMatrix &mat;
|
||||
Vector EVal;
|
||||
// Possible non zero imaginary part of Eigenvalues
|
||||
Vector EVali;
|
||||
DenseMatrix EVect;
|
||||
Vector ev;
|
||||
int n;
|
||||
|
||||
#ifdef MFEM_USE_LAPACK
|
||||
bool sym;
|
||||
double *work;
|
||||
char jobz, uplo;
|
||||
int lwork, info;
|
||||
@@ -702,10 +736,10 @@ class DenseMatrixEigensystem
|
||||
|
||||
public:
|
||||
|
||||
DenseMatrixEigensystem(DenseMatrix &m);
|
||||
DenseMatrixEigensystem(DenseMatrix &m, bool sym_ = false);
|
||||
DenseMatrixEigensystem(const DenseMatrixEigensystem &other);
|
||||
void Eval();
|
||||
Vector &Eigenvalues() { return EVal; }
|
||||
Vector &Eigenvalues(bool imag = false) { return imag ? EVali : EVal; }
|
||||
DenseMatrix &Eigenvectors() { return EVect; }
|
||||
double Eigenvalue(int i) { return EVal(i); }
|
||||
const Vector &Eigenvector(int i)
|
||||
|
||||
@@ -0,0 +1,143 @@
|
||||
// Copyright (c) 2010-2021, 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 "linalg.hpp"
|
||||
|
||||
namespace mfem
|
||||
{
|
||||
|
||||
void KronProdInvDiag(const Vector & a, const Vector & b, Vector & dinv)
|
||||
{
|
||||
int n = a.Size(), m = b.Size();
|
||||
dinv.SetSize(n*m);
|
||||
|
||||
for (int j = 0; j<m; j++)
|
||||
for (int i = 0; i<n; i++)
|
||||
{
|
||||
dinv(i*m+j) = 1./(a(i) + b(j));
|
||||
}
|
||||
}
|
||||
|
||||
void KronProdInvDiag(const Vector & a, const Vector & b,
|
||||
const Vector & c, Vector & dinv)
|
||||
{
|
||||
int n = a.Size(), m = b.Size(), l = c.Size();
|
||||
dinv.SetSize(n*m*l);
|
||||
|
||||
for (int k = 0; k<l; k++)
|
||||
for (int j = 0; j<m; j++)
|
||||
for (int i = 0; i<n; i++)
|
||||
{
|
||||
dinv(i*m*l+j*l+k) = 1./(a(i) + b(j) + c(k));
|
||||
}
|
||||
}
|
||||
|
||||
void KronProdInvDiag(const Array<Vector *> & X, Vector & dinv)
|
||||
{
|
||||
int dim = X.Size();
|
||||
if (dim == 1)
|
||||
{
|
||||
int n = X[0]->Size();
|
||||
dinv.SetSize(n);
|
||||
for (int i = 0; i<n; i++) { dinv(i) = 1./(*X[0])(i); }
|
||||
}
|
||||
else if (dim == 2)
|
||||
{
|
||||
KronProdInvDiag(*X[0], *X[1], dinv);
|
||||
}
|
||||
else if (dim == 3)
|
||||
{
|
||||
KronProdInvDiag(*X[0], *X[1], *X[2], dinv);
|
||||
}
|
||||
else
|
||||
{
|
||||
MFEM_ABORT("KronProdInvDiag::Wrong dimension");
|
||||
}
|
||||
}
|
||||
|
||||
#ifdef MFEM_USE_LAPACK
|
||||
|
||||
FDSolver::FDSolver(const Array<DenseMatrix *> & A,
|
||||
const Array<DenseMatrix *> & B)
|
||||
{
|
||||
MFEM_ASSERT(A.Size() == B.Size(), "DenseFDSolver: Incompatible Dimensions");
|
||||
dim = A.Size();
|
||||
|
||||
int solver_size = 1;
|
||||
for (int i = 0; i<dim; i++)
|
||||
{
|
||||
MFEM_ASSERT(A[i]->Height() == A[i]->Width(),
|
||||
"DenseFDSolver: Matrix is not square");
|
||||
MFEM_ASSERT(B[i]->Height() == B[i]->Width(),
|
||||
"DenseFDSolver: Matrix is not square");
|
||||
MFEM_ASSERT(A[i]->Height() == B[i]->Height(),
|
||||
"DenseFDSolver: Matrices A and B have incompatible size");
|
||||
solver_size *= A[i]->Height();
|
||||
}
|
||||
this->height = solver_size;
|
||||
this->width = solver_size;
|
||||
if (solver_size) { Setup(A,B); }
|
||||
}
|
||||
|
||||
void FDSolver::Setup(const Array<DenseMatrix *> & A,
|
||||
const Array<DenseMatrix *> & B)
|
||||
{
|
||||
EigSystem.SetSize(dim);
|
||||
eigv.SetSize(dim);
|
||||
Array<Vector *> evalues(dim);
|
||||
SQ.SetSize(dim);
|
||||
DenseMatrix D;
|
||||
for (int i = 0; i<dim; i++)
|
||||
{
|
||||
DenseMatrixInverse Minv(*B[i]);
|
||||
Minv.Mult(*A[i],D);
|
||||
EigSystem[i] = new DenseMatrixEigensystem(D);
|
||||
EigSystem[i]->Eval();
|
||||
evalues[i] = &EigSystem[i]->Eigenvalues();
|
||||
eigv[i] = &EigSystem[i]->Eigenvectors();
|
||||
DenseMatrixInverse Qinv(*eigv[i]);
|
||||
DenseMatrix Sdinv;
|
||||
Minv.GetInverseMatrix(Sdinv);
|
||||
SQ[i] = new DenseMatrix;
|
||||
Qinv.Mult(Sdinv,*SQ[i]);
|
||||
}
|
||||
KronProdInvDiag(evalues,dinv);
|
||||
}
|
||||
|
||||
|
||||
void FDSolver::Mult(const Vector & r,Vector & z) const
|
||||
{
|
||||
MFEM_ASSERT(height == r.Size(),
|
||||
"DenseFDSolver::Mult: Inconsistent vector size");
|
||||
if (r.Size() == 0) { return; }
|
||||
Vector rtemp;
|
||||
KronMult(SQ,r,rtemp);
|
||||
// 2. Diagonal solve;
|
||||
rtemp *= dinv;
|
||||
// 3. Modify RHS; z <-- (Q1 x Q2) rtemp
|
||||
KronMult(eigv,rtemp,z);
|
||||
}
|
||||
|
||||
FDSolver::~FDSolver()
|
||||
{
|
||||
if (height)
|
||||
{
|
||||
for (int i=0; i<dim; i++)
|
||||
{
|
||||
delete SQ[i];
|
||||
delete EigSystem[i];
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
#endif // MFEM_USE_LAPACK
|
||||
|
||||
} // namespace mfem
|
||||
@@ -0,0 +1,60 @@
|
||||
// Copyright (c) 2010-2021, 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_FDSOLVER
|
||||
#define MFEM_FDSOLVER
|
||||
|
||||
#include "../config/config.hpp"
|
||||
#include "densemat.hpp"
|
||||
|
||||
namespace mfem
|
||||
{
|
||||
|
||||
/// Computes the inverse diagonal dinv = (a⊗I + I⊗b)^-1
|
||||
/// where a, b are diagonal matrices and I is the identity of the
|
||||
/// appropriate size
|
||||
void KronProdInvDiag(const Vector & a, const Vector & b, Vector & dinv);
|
||||
|
||||
/// Computes the inverse diagonal dinv = (a⊗I⊗I + I⊗b⊗I + I⊗I⊗c)^-1
|
||||
/// where a, b, c are diagonal matrices and I is the identity of the
|
||||
/// appropriate size
|
||||
void KronProdInvDiag(const Vector & a, const Vector & b,
|
||||
const Vector & c, Vector & dinv);
|
||||
|
||||
void KronProdInvDiag(const Array<Vector *> & X, Vector & dinv);
|
||||
|
||||
#ifdef MFEM_USE_LAPACK
|
||||
|
||||
/// In 2D it solves the system (A_0 ⊗ B_1 + B_0 ⊗ A_1) z = r
|
||||
/// In 3D it solves the system
|
||||
/// (A_0 ⊗ B_1 ⊗ B_2 + B_0 ⊗ A_1 ⊗ B_2 + B_0 ⊗ B_1 ⊗ A_2) z = r
|
||||
class FDSolver: public Solver
|
||||
{
|
||||
private:
|
||||
int dim = 2;
|
||||
Array<DenseMatrixEigensystem *> EigSystem;
|
||||
Array<DenseMatrix *> eigv; // eigenvectors
|
||||
Array<DenseMatrix *> SQ;
|
||||
mutable Vector dinv;
|
||||
void Setup(const Array<DenseMatrix *> & A, const Array<DenseMatrix *> & B);
|
||||
public:
|
||||
FDSolver(const Array<DenseMatrix *> & A, const Array<DenseMatrix *> & B);
|
||||
virtual void SetOperator(const Operator &op) {}
|
||||
virtual void Mult(const Vector &r, Vector &z) const;
|
||||
virtual ~FDSolver();
|
||||
};
|
||||
|
||||
#endif // MFEM_USE_LAPACK
|
||||
|
||||
} // mfem name space
|
||||
|
||||
|
||||
#endif // MFEM_FDSOLVER
|
||||
+8
-1
@@ -516,7 +516,14 @@ public:
|
||||
|
||||
/// Initialize all entries with value.
|
||||
HypreParMatrix &operator=(double value)
|
||||
{ internal::hypre_ParCSRMatrixSetConstantValues(A, value); return *this; }
|
||||
{
|
||||
#if MFEM_HYPRE_VERSION < 22200
|
||||
internal::hypre_ParCSRMatrixSetConstantValues(A, value);
|
||||
#else
|
||||
hypre_ParCSRMatrixSetConstantValues(A, value);
|
||||
#endif
|
||||
return *this;
|
||||
}
|
||||
|
||||
/** Perform the operation `*this += B`, assuming that both matrices use the
|
||||
same row and column partitions and the same col_map_offd arrays, or B has
|
||||
|
||||
@@ -1942,8 +1942,8 @@ HYPRE_Int
|
||||
hypre_ParCSRMatrixSetConstantValues(hypre_ParCSRMatrix *A,
|
||||
HYPRE_Complex value)
|
||||
{
|
||||
hypre_CSRMatrixSetConstantValues(hypre_ParCSRMatrixDiag(A), value);
|
||||
hypre_CSRMatrixSetConstantValues(hypre_ParCSRMatrixOffd(A), value);
|
||||
internal::hypre_CSRMatrixSetConstantValues(hypre_ParCSRMatrixDiag(A), value);
|
||||
internal::hypre_CSRMatrixSetConstantValues(hypre_ParCSRMatrixOffd(A), value);
|
||||
|
||||
return 0;
|
||||
}
|
||||
|
||||
@@ -198,6 +198,16 @@ hypre_CSRMatrixSum(hypre_CSRMatrix *A,
|
||||
HYPRE_Complex beta,
|
||||
hypre_CSRMatrix *B);
|
||||
|
||||
#if MFEM_HYPRE_VERSION >= 22200
|
||||
/** Provide an overloaded function for code consistency between HYPRE API
|
||||
versions. */
|
||||
inline hypre_CSRMatrix *hypre_CSRMatrixAdd(hypre_CSRMatrix *A,
|
||||
hypre_CSRMatrix *B)
|
||||
{
|
||||
return ::hypre_CSRMatrixAdd(1.0, A, 1.0, B);
|
||||
}
|
||||
#endif
|
||||
|
||||
/** Return a new matrix containing the sum of A and B, assuming that both
|
||||
matrices use the same row and column partitions. The col_map_offd do not
|
||||
need to be the same, but a more efficient algorithm is used if that's the
|
||||
|
||||
+63
-3
@@ -160,7 +160,7 @@ double Norml2(const int size, const T *data)
|
||||
data of the input and output vectors. */
|
||||
template<typename TA, typename TX, typename TY>
|
||||
MFEM_HOST_DEVICE inline
|
||||
void Mult(const int height, const int width, TA *data, const TX *x, TY *y)
|
||||
void Mult(const int height, const int width, const TA *data, const TX *x, TY *y)
|
||||
{
|
||||
if (width == 0)
|
||||
{
|
||||
@@ -170,7 +170,8 @@ void Mult(const int height, const int width, TA *data, const TX *x, TY *y)
|
||||
}
|
||||
return;
|
||||
}
|
||||
TA *d_col = data;
|
||||
|
||||
TA *d_col = (TA *) data;
|
||||
TX x_col = x[0];
|
||||
for (int row = 0; row < height; row++)
|
||||
{
|
||||
@@ -188,6 +189,52 @@ void Mult(const int height, const int width, TA *data, const TX *x, TY *y)
|
||||
}
|
||||
}
|
||||
|
||||
template<typename TA, typename TB, typename TR, typename TZ>
|
||||
MFEM_HOST_DEVICE inline
|
||||
void KronMult(const int ah, const int aw, const TA *ad, const int bh, const int bw, const TB *bd, TR *r, TZ *z)
|
||||
{
|
||||
for (int i = 0; i < bh; i++)
|
||||
{
|
||||
for (int l = 0; l < ah; l++)
|
||||
{
|
||||
TZ t1 = 0.0;
|
||||
for (int j = 0; j < bw; j++)
|
||||
{
|
||||
const TB t2 = bd[i + j * bh];
|
||||
for (int k = 0; k < aw; k++)
|
||||
{
|
||||
t1 += t2 * r[j + k * bw] * ad[l + k * ah];
|
||||
}
|
||||
}
|
||||
z[i + l * bh] = t1;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
template<typename TA, typename TB, typename TC, typename TR, typename TZ>
|
||||
MFEM_HOST_DEVICE inline
|
||||
void KronMult(const int ah, const int aw, const TA *ad, const int bh, const int bw, const TB *bd, const int ch, const int cw, const TC *cd, TR *r, TZ *z)
|
||||
{
|
||||
for (int i = 0; i < ch; i++)
|
||||
{
|
||||
for (int l = 0; l < ah * bh; l++)
|
||||
{
|
||||
TZ t1 = 0.0;
|
||||
for (int j = 0; j < cw; j++)
|
||||
{
|
||||
const TB t2 = cd[i + j * ch];
|
||||
for (int k = 0; k < aw * bw; k++)
|
||||
{
|
||||
const TA ta = ad[(l / bh) + (k / bw) * ah];
|
||||
const TB tb = bd[(l % bh) + (k % bw) * bh];
|
||||
t1 += t2 * r[j + k * cw] * ta * tb;
|
||||
}
|
||||
}
|
||||
z[i + l * ch] = t1;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
/// Symmetrize a square matrix with given @a size and @a data: A -> (A+A^T)/2.
|
||||
template<typename T>
|
||||
MFEM_HOST_DEVICE inline
|
||||
@@ -276,6 +323,20 @@ void Add(const int height, const int width, const TA *Adata, TB *Bdata)
|
||||
}
|
||||
}
|
||||
|
||||
/** @brief Compute B +=alpha*A, where the matrices A and B are of size
|
||||
@a height x @a width with data @a Adata and @a Bdata. */
|
||||
template<typename TA, typename TB>
|
||||
MFEM_HOST_DEVICE inline
|
||||
void Add(const int height, const int width,
|
||||
const double alpha, const TA *Adata, TB *Bdata)
|
||||
{
|
||||
const int m = height * width;
|
||||
for (int i = 0; i < m; i++)
|
||||
{
|
||||
Bdata[i] += alpha * Adata[i];
|
||||
}
|
||||
}
|
||||
|
||||
/** @brief Compute B = alpha*A, where the matrices A and B are of size
|
||||
@a height x @a width with data @a Adata and @a Bdata. */
|
||||
template<typename TA, typename TB>
|
||||
@@ -290,7 +351,6 @@ void Set(const int height, const int width,
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
/** @brief Matrix-matrix multiplication: A = B * C, where the matrices A, B and
|
||||
C are of sizes @a Aheight x @a Awidth, @a Aheight x @a Bwidth and @a Bwidth
|
||||
x @a Awidth, respectively. */
|
||||
|
||||
@@ -31,6 +31,7 @@
|
||||
#include "invariants.hpp"
|
||||
#include "constraints.hpp"
|
||||
#include "auxiliary.hpp"
|
||||
#include "fdsolver.hpp"
|
||||
|
||||
#ifdef MFEM_USE_AMGX
|
||||
#include "amgxsolver.hpp"
|
||||
|
||||
+28
-6
@@ -232,7 +232,7 @@ void OperatorJacobiSmoother::Mult(const Vector &x, Vector &y) const
|
||||
MFEM_FORALL(i, height, Y[i] += DI[i] * R[i]; );
|
||||
}
|
||||
|
||||
OperatorChebyshevSmoother::OperatorChebyshevSmoother(Operator* oper_,
|
||||
OperatorChebyshevSmoother::OperatorChebyshevSmoother(const Operator &oper_,
|
||||
const Vector &d,
|
||||
const Array<int>& ess_tdofs,
|
||||
int order_, double max_eig_estimate_)
|
||||
@@ -246,15 +246,15 @@ OperatorChebyshevSmoother::OperatorChebyshevSmoother(Operator* oper_,
|
||||
coeffs(order),
|
||||
ess_tdof_list(ess_tdofs),
|
||||
residual(N),
|
||||
oper(oper_) { Setup(); }
|
||||
oper(&oper_) { Setup(); }
|
||||
|
||||
#ifdef MFEM_USE_MPI
|
||||
OperatorChebyshevSmoother::OperatorChebyshevSmoother(Operator* oper_,
|
||||
OperatorChebyshevSmoother::OperatorChebyshevSmoother(const Operator &oper_,
|
||||
const Vector &d,
|
||||
const Array<int>& ess_tdofs,
|
||||
int order_, MPI_Comm comm, int power_iterations, double power_tolerance)
|
||||
#else
|
||||
OperatorChebyshevSmoother::OperatorChebyshevSmoother(Operator* oper_,
|
||||
OperatorChebyshevSmoother::OperatorChebyshevSmoother(const Operator &oper_,
|
||||
const Vector &d,
|
||||
const Array<int>& ess_tdofs,
|
||||
int order_, int power_iterations, double power_tolerance)
|
||||
@@ -267,7 +267,7 @@ OperatorChebyshevSmoother::OperatorChebyshevSmoother(Operator* oper_,
|
||||
coeffs(order),
|
||||
ess_tdof_list(ess_tdofs),
|
||||
residual(N),
|
||||
oper(oper_)
|
||||
oper(&oper_)
|
||||
{
|
||||
OperatorJacobiSmoother invDiagOperator(diag, ess_tdofs, 1.0);
|
||||
ProductOperator diagPrecond(&invDiagOperator, oper, false, false);
|
||||
@@ -284,6 +284,28 @@ OperatorChebyshevSmoother::OperatorChebyshevSmoother(Operator* oper_,
|
||||
Setup();
|
||||
}
|
||||
|
||||
OperatorChebyshevSmoother::OperatorChebyshevSmoother(const Operator* oper_,
|
||||
const Vector &d,
|
||||
const Array<int>& ess_tdofs,
|
||||
int order_, double max_eig_estimate_)
|
||||
: OperatorChebyshevSmoother(*oper_, d, ess_tdofs, order_, max_eig_estimate_) { }
|
||||
|
||||
#ifdef MFEM_USE_MPI
|
||||
OperatorChebyshevSmoother::OperatorChebyshevSmoother(const Operator* oper_,
|
||||
const Vector &d,
|
||||
const Array<int>& ess_tdofs,
|
||||
int order_, MPI_Comm comm, int power_iterations, double power_tolerance)
|
||||
: OperatorChebyshevSmoother(*oper_, d, ess_tdofs, order_, comm,
|
||||
power_iterations, power_tolerance) { }
|
||||
#else
|
||||
OperatorChebyshevSmoother::OperatorChebyshevSmoother(const Operator* oper_,
|
||||
const Vector &d,
|
||||
const Array<int>& ess_tdofs,
|
||||
int order_, int power_iterations, double power_tolerance)
|
||||
: OperatorChebyshevSmoother(*oper_, d, ess_tdofs, order_, power_iterations,
|
||||
power_tolerance) { }
|
||||
#endif
|
||||
|
||||
void OperatorChebyshevSmoother::Setup()
|
||||
{
|
||||
// Invert diagonal
|
||||
@@ -2533,7 +2555,7 @@ BlockILU::BlockILU(int block_size_,
|
||||
reordering(reordering_)
|
||||
{ }
|
||||
|
||||
BlockILU::BlockILU(Operator &op,
|
||||
BlockILU::BlockILU(const Operator &op,
|
||||
int block_size_,
|
||||
Reordering reordering_,
|
||||
int k_fill_)
|
||||
|
||||
+25
-4
@@ -208,7 +208,13 @@ public:
|
||||
the matrix-free setting. The estimated largest eigenvalue of the
|
||||
diagonally preconditoned operator must be provided via
|
||||
max_eig_estimate. */
|
||||
OperatorChebyshevSmoother(Operator* oper_, const Vector &d,
|
||||
OperatorChebyshevSmoother(const Operator &oper_, const Vector &d,
|
||||
const Array<int>& ess_tdof_list,
|
||||
int order, double max_eig_estimate);
|
||||
|
||||
/// Deprecated: see pass-by-reference version above
|
||||
MFEM_DEPRECATED
|
||||
OperatorChebyshevSmoother(const Operator* oper_, const Vector &d,
|
||||
const Array<int>& ess_tdof_list,
|
||||
int order, double max_eig_estimate);
|
||||
|
||||
@@ -220,13 +226,28 @@ public:
|
||||
accuracy of the estimated eigenvalue may be controlled via
|
||||
power_iterations and power_tolerance. */
|
||||
#ifdef MFEM_USE_MPI
|
||||
OperatorChebyshevSmoother(Operator* oper_, const Vector &d,
|
||||
OperatorChebyshevSmoother(const Operator &oper_, const Vector &d,
|
||||
const Array<int>& ess_tdof_list,
|
||||
int order, MPI_Comm comm = MPI_COMM_NULL,
|
||||
int power_iterations = 10,
|
||||
double power_tolerance = 1e-8);
|
||||
|
||||
/// Deprecated: see pass-by-reference version above
|
||||
MFEM_DEPRECATED
|
||||
OperatorChebyshevSmoother(const Operator* oper_, const Vector &d,
|
||||
const Array<int>& ess_tdof_list,
|
||||
int order, MPI_Comm comm = MPI_COMM_NULL,
|
||||
int power_iterations = 10,
|
||||
double power_tolerance = 1e-8);
|
||||
#else
|
||||
OperatorChebyshevSmoother(Operator* oper_, const Vector &d,
|
||||
OperatorChebyshevSmoother(const Operator &oper_, const Vector &d,
|
||||
const Array<int>& ess_tdof_list,
|
||||
int order, int power_iterations = 10,
|
||||
double power_tolerance = 1e-8);
|
||||
|
||||
/// Deprecated: see pass-by-reference version above
|
||||
MFEM_DEPRECATED
|
||||
OperatorChebyshevSmoother(const Operator* oper_, const Vector &d,
|
||||
const Array<int>& ess_tdof_list,
|
||||
int order, int power_iterations = 10,
|
||||
double power_tolerance = 1e-8);
|
||||
@@ -731,7 +752,7 @@ public:
|
||||
* case that @a op is a HypreParMatrix, the ILU factorization is performed
|
||||
* on the diagonal blocks of the parallel decomposition.
|
||||
*/
|
||||
BlockILU(Operator &op, int block_size_ = 1,
|
||||
BlockILU(const Operator &op, int block_size_ = 1,
|
||||
Reordering reordering_ = Reordering::MINIMUM_DISCARDED_FILL,
|
||||
int k_fill_ = 0);
|
||||
|
||||
|
||||
@@ -37,6 +37,7 @@ MFEM makefile targets:
|
||||
make distclean
|
||||
make style
|
||||
make tags
|
||||
make hooks
|
||||
|
||||
Examples:
|
||||
|
||||
@@ -96,6 +97,8 @@ make style
|
||||
make tags
|
||||
Generate a vi or Emacs compatible TAGS file in ${MFEM_DIR}/TAGS. Requires
|
||||
functional "etags" and "egrep" in the user ${PATH}.
|
||||
make hooks
|
||||
Creates symlinks to the hooks in the `.git/hooks` directory.
|
||||
endef
|
||||
|
||||
# Save the MAKEOVERRIDES for cases where we explicitly want to pass the command
|
||||
@@ -759,6 +762,15 @@ endif
|
||||
@cd $(MFEM_REAL_DIR) && $(ETAGS_BIN) --class-qualify \
|
||||
--declarations -o $(MFEM_REAL_DIR)/TAGS $(MFEM_TRACKED_SOURCE)
|
||||
|
||||
# Creates symlinks to the hooks in the `.git/hooks` directory. Individual
|
||||
# hooks can be enabled by manually creating symlinks. Hooks can be customized
|
||||
# using hard copies (trading off with automated updates).
|
||||
.PHONY: hooks
|
||||
hooks:
|
||||
@cd $(MFEM_DIR)/.git/hooks && \
|
||||
ln -s ../../config/githooks/pre-commit pre-commit; \
|
||||
ln -s ../../config/githooks/pre-push pre-push;
|
||||
|
||||
# Print the contents of a makefile variable, e.g.: 'make print-MFEM_LIBS'.
|
||||
print-%:
|
||||
$(info [ variable name]: $*)
|
||||
|
||||
+4
-4
@@ -11478,10 +11478,10 @@ FaceGeometricFactors::FaceGeometricFactors(const Mesh *mesh,
|
||||
const int NF = fespace->GetNFbyType(type);
|
||||
const int NQ = ir.GetNPoints();
|
||||
|
||||
const Operator *face_restr = fespace->GetFaceRestriction(
|
||||
ElementDofOrdering::LEXICOGRAPHIC,
|
||||
type,
|
||||
L2FaceValues::SingleValued );
|
||||
const FaceRestriction *face_restr = fespace->GetFaceRestriction(
|
||||
ElementDofOrdering::LEXICOGRAPHIC,
|
||||
type,
|
||||
L2FaceValues::SingleValued );
|
||||
Vector Fnodes(face_restr->Height());
|
||||
face_restr->Mult(*nodes, Fnodes);
|
||||
|
||||
|
||||
+64
-35
@@ -691,16 +691,31 @@ struct BufferReader : BufferReaderBase
|
||||
BufferReader(bool compressed_, HeaderType header_type_)
|
||||
: compressed(compressed_), header_type(header_type_) { }
|
||||
|
||||
/// Return the number of bytes in the header. The header consists of one
|
||||
/// integer if the data is uncompressed, and four integers if the data is
|
||||
/// compressed. The integers are either 32 or 64 bytes depending on the value
|
||||
/// of @a header_type.
|
||||
int NumHeaderBytes() const
|
||||
/// Return the number of bytes of each header entry.
|
||||
size_t HeaderEntrySize() const
|
||||
{
|
||||
int num_entries = compressed ? 4 : 1;
|
||||
int entry_size = (header_type == UINT64_HEADER)
|
||||
? sizeof(uint64_t) : sizeof(uint32_t);
|
||||
return num_entries*entry_size;
|
||||
return header_type == UINT64_HEADER ? sizeof(uint64_t) : sizeof(uint32_t);
|
||||
}
|
||||
|
||||
/// Return the value of the header entry pointer to by @a header_buf. The
|
||||
/// value is stored as either uint32_t or uint64_t, according to the @a
|
||||
/// header_type, and is returned as uint64_t.
|
||||
uint64_t ReadHeaderEntry(const char *header_buf) const
|
||||
{
|
||||
return (header_type == UINT64_HEADER) ? bin_io::read<uint64_t>(header_buf)
|
||||
: bin_io::read<uint32_t>(header_buf);
|
||||
}
|
||||
|
||||
/// Return the number of bytes in the header. The header consists of one
|
||||
/// integer if the data is uncompressed, and @a N + 3 integers if the data is
|
||||
/// compressed, where @a N is the number of blocks. The integers are either
|
||||
/// 32 or 64 bytes depending on the value of @a header_type. The number of
|
||||
/// blocks is determined by reading the first integer (of type @a
|
||||
/// header_type) pointed to by @a header_buf.
|
||||
int NumHeaderBytes(const char *header_buf) const
|
||||
{
|
||||
if (!compressed) { return HeaderEntrySize(); }
|
||||
return (3 + ReadHeaderEntry(header_buf))*HeaderEntrySize();
|
||||
}
|
||||
|
||||
/// Read @a n elements of type @a F from the source buffer @a buf into the
|
||||
@@ -711,32 +726,42 @@ struct BufferReader : BufferReaderBase
|
||||
void ReadBinaryWithHeader(const char *header_buf, const char *buf,
|
||||
void *dest_void, int n) const
|
||||
{
|
||||
std::vector<unsigned char> uncompressed_data;
|
||||
std::vector<char> uncompressed_data;
|
||||
T *dest = static_cast<T*>(dest_void);
|
||||
|
||||
if (compressed)
|
||||
{
|
||||
#ifdef MFEM_USE_ZLIB
|
||||
uint64_t header[4];
|
||||
if (header_type == UINT32_HEADER)
|
||||
// The header has format (where header_t is uint32_t or uint64_t):
|
||||
// header_t number_of_blocks;
|
||||
// header_t uncompressed_block_size;
|
||||
// header_t uncompressed_last_block_size;
|
||||
// header_t compressed_size[number_of_blocks];
|
||||
int header_entry_size = HeaderEntrySize();
|
||||
int nblocks = ReadHeaderEntry(header_buf);
|
||||
header_buf += header_entry_size;
|
||||
std::vector<int> header(nblocks + 2);
|
||||
for (int i=0; i<nblocks+2; ++i)
|
||||
{
|
||||
uint32_t *header_32 = (uint32_t *)header_buf;
|
||||
for (int i=0; i<4; ++i) { header[i] = header_32[i]; }
|
||||
header[i] = ReadHeaderEntry(header_buf);
|
||||
header_buf += header_entry_size;
|
||||
}
|
||||
else
|
||||
uncompressed_data.resize((nblocks-1)*header[0] + header[1]);
|
||||
Bytef *dest_ptr = (Bytef *)uncompressed_data.data();
|
||||
Bytef *dest_start = dest_ptr;
|
||||
const Bytef *source_ptr = (const Bytef *)buf;
|
||||
for (int i=0; i<nblocks; ++i)
|
||||
{
|
||||
uint64_t *header_64 = (uint64_t *)header_buf;
|
||||
for (int i=0; i<4; ++i) { header[i] = header_64[i]; }
|
||||
uLongf source_len = header[i+2];
|
||||
uLong dest_len = (i == nblocks-1) ? header[1] : header[0];
|
||||
int res = uncompress(dest_ptr, &dest_len, source_ptr, source_len);
|
||||
MFEM_VERIFY(res == Z_OK, "Error uncompressing");
|
||||
dest_ptr += dest_len;
|
||||
source_ptr += source_len;
|
||||
}
|
||||
|
||||
MFEM_VERIFY(header[0] == 1, "Multiple compressed blocks not supported");
|
||||
uLongf dest_len = header[1];
|
||||
uncompressed_data.resize(dest_len);
|
||||
int res = uncompress(uncompressed_data.data(), &dest_len,
|
||||
(const Bytef *)buf, header[3]);
|
||||
MFEM_VERIFY(res == Z_OK, "Error uncompressing");
|
||||
MFEM_VERIFY(sizeof(F)*n == dest_len, "AppendedData: wrong data size");
|
||||
buf = (const char *)uncompressed_data.data();
|
||||
MFEM_VERIFY(int(sizeof(F)*n) == (dest_ptr - dest_start),
|
||||
"AppendedData: wrong data size");
|
||||
buf = uncompressed_data.data();
|
||||
#else
|
||||
MFEM_ABORT("MFEM must be compiled with zlib enabled to uncompress.")
|
||||
#endif
|
||||
@@ -779,7 +804,7 @@ struct BufferReader : BufferReaderBase
|
||||
/// buffer @a dest. The input buffer contains both the header and the data.
|
||||
void ReadBinary(const char *buf, void *dest, int n) const override
|
||||
{
|
||||
ReadBinaryWithHeader(buf, buf + NumHeaderBytes(), dest, n);
|
||||
ReadBinaryWithHeader(buf, buf + NumHeaderBytes(buf), dest, n);
|
||||
}
|
||||
|
||||
/// Read @a n elements of type @a F from base-64 encoded source buffer into
|
||||
@@ -796,21 +821,25 @@ struct BufferReader : BufferReaderBase
|
||||
}
|
||||
if (compressed)
|
||||
{
|
||||
std::vector<unsigned char> data, header;
|
||||
// Decode the first entry of the header, which we need to determine
|
||||
// how long the rest of the header is.
|
||||
std::vector<char> nblocks_buf;
|
||||
int nblocks_b64 = bin_io::NumBase64Chars(HeaderEntrySize());
|
||||
bin_io::DecodeBase64(txt, nblocks_b64, nblocks_buf);
|
||||
std::vector<char> data, header;
|
||||
// Compute number of characters needed to encode header in base 64,
|
||||
// then round to nearest multiple of 4 to take padding into account.
|
||||
int b64_header = ((4*NumHeaderBytes()/3) + 3) & ~3;
|
||||
int header_b64 = bin_io::NumBase64Chars(NumHeaderBytes(nblocks_buf.data()));
|
||||
// If data is compressed, header is encoded separately
|
||||
bin_io::DecodeBase64(txt, b64_header, header);
|
||||
bin_io::DecodeBase64(txt + b64_header, strlen(txt)-b64_header, data);
|
||||
ReadBinaryWithHeader((const char *)header.data(),
|
||||
(const char *)data.data(), dest, n);
|
||||
bin_io::DecodeBase64(txt, header_b64, header);
|
||||
bin_io::DecodeBase64(txt + header_b64, strlen(txt)-header_b64, data);
|
||||
ReadBinaryWithHeader(header.data(), data.data(), dest, n);
|
||||
}
|
||||
else
|
||||
{
|
||||
std::vector<unsigned char> data;
|
||||
std::vector<char> data;
|
||||
bin_io::DecodeBase64(txt, strlen(txt), data);
|
||||
ReadBinary((const char *)data.data(), dest, n);
|
||||
ReadBinary(data.data(), dest, n);
|
||||
}
|
||||
}
|
||||
};
|
||||
|
||||
+20
-2
@@ -4321,7 +4321,7 @@ const CoarseFineTransformations& NCMesh::GetRefinementTransforms()
|
||||
if (!transforms.embeddings.Size())
|
||||
{
|
||||
transforms.Clear();
|
||||
transforms.embeddings.SetSize(leaf_elements.Size());
|
||||
transforms.embeddings.SetSize(NElements);
|
||||
|
||||
std::string ref_path;
|
||||
ref_path.reserve(100);
|
||||
@@ -4455,7 +4455,8 @@ struct RefType
|
||||
void CoarseFineTransformations::GetCoarseToFineMap(
|
||||
const mfem::Mesh &fine_mesh, Table &coarse_to_fine,
|
||||
Array<int> &coarse_to_ref_type, Table &ref_type_to_matrix,
|
||||
Array<mfem::Geometry::Type> &ref_type_to_geom) const
|
||||
Array<mfem::Geometry::Type> &ref_type_to_geom,
|
||||
bool get_coarse_to_fine_only) const
|
||||
{
|
||||
const int fine_ne = embeddings.Size();
|
||||
int coarse_ne = -1;
|
||||
@@ -4495,6 +4496,11 @@ void CoarseFineTransformations::GetCoarseToFineMap(
|
||||
coarse_to_fine.GetJ()[i] = cf_j[i].two;
|
||||
}
|
||||
|
||||
if (get_coarse_to_fine_only) { return; }
|
||||
MFEM_VERIFY(fine_mesh.GetLastOperation() != Mesh::Operation::DEREFINE,
|
||||
"GetCoarseToFineMap is not fully supported for derefined meshes."
|
||||
" Set 'get_coarse_to_fine_only=true'.")
|
||||
|
||||
using internal::RefType;
|
||||
using std::map;
|
||||
using std::pair;
|
||||
@@ -4536,6 +4542,18 @@ void CoarseFineTransformations::GetCoarseToFineMap(
|
||||
ref_type_to_matrix.ShiftUpI();
|
||||
}
|
||||
|
||||
void CoarseFineTransformations::GetCoarseToFineMap(const Mesh &fine_mesh,
|
||||
Table &coarse_to_fine) const
|
||||
{
|
||||
Array<int> coarse_to_ref_type;
|
||||
Table ref_type_to_matrix;
|
||||
Array<mfem::Geometry::Type> ref_type_to_geom;
|
||||
bool get_coarse_to_fine_only = true;
|
||||
GetCoarseToFineMap(fine_mesh, coarse_to_fine, coarse_to_ref_type,
|
||||
ref_type_to_matrix, ref_type_to_geom,
|
||||
get_coarse_to_fine_only);
|
||||
}
|
||||
|
||||
void NCMesh::ClearTransforms()
|
||||
{
|
||||
coarse_elements.DeleteAll();
|
||||
|
||||
+5
-1
@@ -68,7 +68,11 @@ struct CoarseFineTransformations
|
||||
Table &coarse_to_fine,
|
||||
Array<int> &coarse_to_ref_type,
|
||||
Table &ref_type_to_matrix,
|
||||
Array<Geometry::Type> &ref_type_to_geom) const;
|
||||
Array<Geometry::Type> &ref_type_to_geom,
|
||||
bool get_coarse_to_fine_only = false) const;
|
||||
|
||||
void GetCoarseToFineMap(const Mesh &fine_mesh,
|
||||
Table &coarse_to_fine) const;
|
||||
|
||||
void Clear();
|
||||
bool IsInitialized() const;
|
||||
|
||||
@@ -297,9 +297,6 @@ protected: // implementation
|
||||
|
||||
virtual void Update();
|
||||
|
||||
virtual int GetNumGhostElements() const { return NGhostElements; }
|
||||
virtual int GetNumGhostVertices() const { return NGhostVertices; }
|
||||
|
||||
/// Return the processor number for a global element number.
|
||||
int Partition(long index, long total_elements) const
|
||||
{ return index * NRanks / total_elements; }
|
||||
|
||||
@@ -109,15 +109,19 @@ int main(int argc, char *argv[])
|
||||
args.Parse();
|
||||
if (!args.Good())
|
||||
{
|
||||
args.PrintUsage(cout);
|
||||
if (myid == 0)
|
||||
{
|
||||
args.PrintUsage(cout);
|
||||
}
|
||||
MPI_Finalize();
|
||||
return 1;
|
||||
}
|
||||
args.PrintOptions(cout);
|
||||
if (myid == 0) { args.PrintOptions(cout); }
|
||||
|
||||
// Enable hardware devices such as GPUs, and programming models such as CUDA,
|
||||
// OCCA, RAJA and OpenMP based on command line options.
|
||||
Device device("cpu");
|
||||
device.Print();
|
||||
if (myid == 0) { device.Print(); }
|
||||
|
||||
// Refine the mesh.
|
||||
Mesh mesh(mesh_file, 1, 1);
|
||||
|
||||
@@ -96,7 +96,7 @@ SparseMatrix* AggToInteriorDof(const Array<int>& bdr_truedofs,
|
||||
const SparseMatrix& agg_elem,
|
||||
const SparseMatrix& elem_dof,
|
||||
const HypreParMatrix& dof_truedof,
|
||||
Array<int>& agg_starts)
|
||||
Array<HYPRE_Int>& agg_starts)
|
||||
{
|
||||
OperatorPtr agg_dof(Mult(agg_elem, elem_dof));
|
||||
SparseMatrix& agg_dof_ref = *agg_dof.As<SparseMatrix>();
|
||||
@@ -131,7 +131,7 @@ SparseMatrix* AggToInteriorDof(const Array<int>& bdr_truedofs,
|
||||
|
||||
void DFSSpaces::MakeDofRelationTables(int level)
|
||||
{
|
||||
Array<int> agg_starts(Array<int>(l2_0_fes_->GetDofOffsets(), 2));
|
||||
Array<HYPRE_Int> agg_starts(Array<HYPRE_Int>(l2_0_fes_->GetDofOffsets(), 2));
|
||||
auto& elem_agg = (const SparseMatrix&)*l2_0_fes_->GetUpdateOperator();
|
||||
OperatorPtr agg_elem(Transpose(elem_agg));
|
||||
SparseMatrix& agg_el = *agg_elem.As<SparseMatrix>();
|
||||
|
||||
@@ -110,6 +110,7 @@ then
|
||||
echo "~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~"
|
||||
|
||||
cp ${hostconfig_path} ${project_dir}/config/
|
||||
ln -sf ${build_root}/data ${project_dir}/../
|
||||
|
||||
make all -j ${threads}
|
||||
fi
|
||||
|
||||
@@ -28,6 +28,7 @@ set(UNIT_TESTS_SRCS
|
||||
linalg/test_complex_operator.cpp
|
||||
linalg/test_constrainedsolver.cpp
|
||||
linalg/test_direct_solvers.cpp
|
||||
linalg/test_fdsolver.cpp
|
||||
linalg/test_hypre_ilu.cpp
|
||||
linalg/test_ilu.cpp
|
||||
linalg/test_matrix_block.cpp
|
||||
|
||||
@@ -11,48 +11,14 @@
|
||||
|
||||
#define CATCH_CONFIG_RUNNER
|
||||
#include "mfem.hpp"
|
||||
#include "unit_tests.hpp"
|
||||
#include "run_unit_tests.hpp"
|
||||
|
||||
bool launch_all_non_regression_tests = false;
|
||||
std::string mfem_data_dir;
|
||||
|
||||
int main(int argc, char *argv[])
|
||||
{
|
||||
mfem::Device device("cuda");
|
||||
|
||||
// There must be exactly one instance.
|
||||
Catch::Session session;
|
||||
|
||||
// Build a new command line parser on top of Catch's
|
||||
using namespace Catch::clara;
|
||||
auto cli = session.cli() |
|
||||
Opt(launch_all_non_regression_tests) ["--all"] ("all tests");
|
||||
session.cli(cli);
|
||||
|
||||
// For floating point comparisons, print 8 digits for single precision
|
||||
// values, and 16 digits for double precision values.
|
||||
Catch::StringMaker<float>::precision = 8;
|
||||
Catch::StringMaker<double>::precision = 16;
|
||||
|
||||
// Apply provided command line arguments.
|
||||
int r = session.applyCommandLine(argc, argv);
|
||||
if (r != 0) { return r; }
|
||||
|
||||
auto cfg = session.configData();
|
||||
|
||||
cfg.testsOrTags.push_back("[CUDA]");
|
||||
|
||||
#ifdef MFEM_USE_MPI
|
||||
// Exclude tests marked as Parallel in a serial run, even when compiled with
|
||||
// MPI. This is done because there is no MPI session initialized.
|
||||
cfg.testsOrTags.push_back("~[Parallel]");
|
||||
#endif
|
||||
|
||||
std::cout << "INFO: Test filter: [CUDA] ~[Parallel]" << std::endl;
|
||||
device.Print();
|
||||
|
||||
session.useConfigData(cfg);
|
||||
|
||||
int result = session.run();
|
||||
|
||||
return result;
|
||||
// Include only tests labeled with CUDA. Exclude parallel tests.
|
||||
return RunCatchSession(argc, argv, {"[CUDA]", "~[Parallel]"});
|
||||
}
|
||||
|
||||
@@ -11,6 +11,7 @@
|
||||
|
||||
#include "mfem.hpp"
|
||||
#include "unit_tests.hpp"
|
||||
#include "general/tinyxml2.h"
|
||||
#include <stdio.h>
|
||||
|
||||
#ifndef _WIN32
|
||||
@@ -235,3 +236,86 @@ TEST_CASE("Save and load from collections", "[DataCollection]")
|
||||
}
|
||||
|
||||
}
|
||||
|
||||
void SaveDataCollection(DataCollection &dc, int cycle, double t)
|
||||
{
|
||||
dc.SetCycle(cycle);
|
||||
dc.SetTime(t);
|
||||
dc.Save();
|
||||
}
|
||||
|
||||
TEST_CASE("ParaView restart mode", "[ParaView]")
|
||||
{
|
||||
Mesh mesh = Mesh::MakeCartesian2D(2, 3, Element::QUADRILATERAL);
|
||||
H1_FECollection fec(1, mesh.Dimension());
|
||||
FiniteElementSpace fes(&mesh, &fec);
|
||||
GridFunction u(&fes);
|
||||
u = 0.0;
|
||||
|
||||
// Write initial dataset with three timesteps: 0, 1, 2.
|
||||
{
|
||||
ParaViewDataCollection dc("ParaView", &mesh);
|
||||
dc.RegisterField("u", &u);
|
||||
SaveDataCollection(dc, 0, 0);
|
||||
SaveDataCollection(dc, 1, 1);
|
||||
SaveDataCollection(dc, 2, 2);
|
||||
}
|
||||
|
||||
// Using restart mode, append to the existing dataset, overwriting timesteps
|
||||
// 1 and 2 with 1 and 1.5.
|
||||
{
|
||||
ParaViewDataCollection dc("ParaView", &mesh);
|
||||
dc.UseRestartMode(true);
|
||||
dc.RegisterField("u", &u);
|
||||
SaveDataCollection(dc, 1, 1.0);
|
||||
SaveDataCollection(dc, 2, 1.5);
|
||||
}
|
||||
|
||||
// Parse the resulting PVD file, and verify that the structure is correct,
|
||||
// and that it contains three timesteps: 0, 1, and 1.5.
|
||||
using namespace tinyxml2;
|
||||
auto StringCompare = [](const char *s1, const char *s2)
|
||||
{
|
||||
if (s1 == NULL || s2 == NULL) { return false; }
|
||||
return strcmp(s1, s2) == 0;
|
||||
};
|
||||
auto VerifyDataset = [StringCompare](const XMLElement *ds, double t_ref)
|
||||
{
|
||||
REQUIRE(ds);
|
||||
REQUIRE(StringCompare(ds->Name(), "DataSet"));
|
||||
const char *timestep = ds->Attribute("timestep");
|
||||
REQUIRE(timestep);
|
||||
double t = std::stod(timestep);
|
||||
REQUIRE(t == MFEM_Approx(t_ref));
|
||||
};
|
||||
|
||||
XMLDocument xml;
|
||||
xml.LoadFile("ParaView/ParaView.pvd");
|
||||
REQUIRE(xml.ErrorID() == XML_SUCCESS);
|
||||
|
||||
const XMLElement *vtkfile = xml.FirstChildElement();
|
||||
REQUIRE(vtkfile);
|
||||
REQUIRE(StringCompare(vtkfile->Name(), "VTKFile"));
|
||||
const XMLElement *collection = vtkfile->FirstChildElement();
|
||||
REQUIRE(collection);
|
||||
REQUIRE(StringCompare(collection->Name(), "Collection"));
|
||||
|
||||
const XMLElement *dataset = collection->FirstChildElement();
|
||||
VerifyDataset(dataset, 0.0);
|
||||
dataset = dataset->NextSiblingElement();
|
||||
VerifyDataset(dataset, 1.0);
|
||||
dataset = dataset->NextSiblingElement();
|
||||
VerifyDataset(dataset, 1.5);
|
||||
REQUIRE(dataset->NextSiblingElement() == NULL);
|
||||
|
||||
// Clean up
|
||||
for (int c=0; c<=2; ++c)
|
||||
{
|
||||
std::string prefix = "ParaView/Cycle00000" + std::to_string(c);
|
||||
REQUIRE(remove((prefix + "/data.pvtu").c_str()) == 0);
|
||||
REQUIRE(remove((prefix + "/proc000000.vtu").c_str()) == 0);
|
||||
REQUIRE(rmdir(prefix.c_str()) == 0);
|
||||
}
|
||||
REQUIRE(remove("ParaView/ParaView.pvd") == 0);
|
||||
REQUIRE(rmdir("ParaView") == 0);
|
||||
}
|
||||
|
||||
@@ -167,9 +167,9 @@ TEST_CASE("DG SumIntegrator", "[SumIntegrator][PartialAssembly]")
|
||||
integ2.AssemblePAInteriorFaces(fes);
|
||||
integ_sum.AssemblePAInteriorFaces(fes);
|
||||
|
||||
const Operator *R_int = fes.GetFaceRestriction(
|
||||
ElementDofOrdering::LEXICOGRAPHIC,
|
||||
FaceType::Interior);
|
||||
const FaceRestriction *R_int = fes.GetFaceRestriction(
|
||||
ElementDofOrdering::LEXICOGRAPHIC,
|
||||
FaceType::Interior);
|
||||
|
||||
int n_int = R_int->Height();
|
||||
Vector x(n_int), y1(n_int), y2(n_int);
|
||||
@@ -198,10 +198,10 @@ TEST_CASE("DG SumIntegrator", "[SumIntegrator][PartialAssembly]")
|
||||
integ2.AssemblePABoundaryFaces(fes);
|
||||
integ_sum.AssemblePABoundaryFaces(fes);
|
||||
|
||||
const Operator *R_bdr = fes.GetFaceRestriction(
|
||||
ElementDofOrdering::LEXICOGRAPHIC,
|
||||
FaceType::Boundary,
|
||||
L2FaceValues::DoubleValued);
|
||||
const FaceRestriction *R_bdr = fes.GetFaceRestriction(
|
||||
ElementDofOrdering::LEXICOGRAPHIC,
|
||||
FaceType::Boundary,
|
||||
L2FaceValues::DoubleValued);
|
||||
|
||||
int n_bdr = R_bdr->Height();
|
||||
x.SetSize(n_bdr);
|
||||
|
||||
@@ -38,7 +38,7 @@ TEST_CASE("OperatorChebyshevSmoother", "[Chebyshev symmetry]")
|
||||
Vector diag(fespace.GetTrueVSize());
|
||||
aform.AssembleDiagonal(diag);
|
||||
|
||||
Solver* smoother = new OperatorChebyshevSmoother(opr.Ptr(), diag, ess_tdof_list,
|
||||
Solver* smoother = new OperatorChebyshevSmoother(*opr, diag, ess_tdof_list,
|
||||
cheb_order);
|
||||
|
||||
int n = smoother->Width();
|
||||
|
||||
@@ -0,0 +1,126 @@
|
||||
// Copyright (c) 2010-2021, 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 "mfem.hpp"
|
||||
#include "unit_tests.hpp"
|
||||
|
||||
using namespace mfem;
|
||||
|
||||
#ifdef MFEM_USE_LAPACK
|
||||
|
||||
TEST_CASE("FDSolver",
|
||||
"[FDSolver]")
|
||||
{
|
||||
double tol = 1e-10;
|
||||
|
||||
// SPD matrices
|
||||
DenseMatrix A0(
|
||||
{
|
||||
{
|
||||
1.29919, 0.61256, 0.82545
|
||||
},
|
||||
{0.61256, 0.57891, 0.39662},
|
||||
{0.82545, 0.39662, 0.57541}
|
||||
});
|
||||
|
||||
DenseMatrix A1(
|
||||
{
|
||||
{0.748236, 0.701663, 0.607517, 0.236740},
|
||||
{0.701663, 0.809316, 0.713186, 0.256070},
|
||||
{0.607517, 0.713186, 0.794221, 0.233943},
|
||||
{0.236740, 0.256070, 0.233943, 0.083129}
|
||||
});
|
||||
|
||||
DenseMatrix B0(
|
||||
{
|
||||
{0.13483, 0.51389, 0.43052},
|
||||
{0.51389, 2.26750, 1.86331},
|
||||
{0.43052, 1.86331, 1.59869}
|
||||
});
|
||||
|
||||
DenseMatrix B1(
|
||||
{
|
||||
{0.94177, 1.02400, 1.14743, 0.35723},
|
||||
{1.02400, 1.79087, 1.78708, 0.78304},
|
||||
{1.14743, 1.78708, 2.06259, 0.80837},
|
||||
{0.35723, 0.78304, 0.80837, 1.01798}
|
||||
});
|
||||
|
||||
|
||||
SECTION("2D")
|
||||
{
|
||||
Array<DenseMatrix *> A(2), B(2);
|
||||
A[0] = &A0; A[1] = &A1;
|
||||
B[0] = &B0; B[1] = &B1;
|
||||
|
||||
Vector y(12); y.Randomize(1);
|
||||
Vector x(12), diff(12);
|
||||
|
||||
FDSolver S(A,B);
|
||||
S.Mult(y,x);
|
||||
|
||||
DenseMatrix C1, C;
|
||||
KronProd(A0, B1, C1);
|
||||
KronProd(B0, A1, C);
|
||||
|
||||
C.Add(1., C1);
|
||||
|
||||
DenseMatrixInverse Cinv(C);
|
||||
Cinv.Mult(y,diff);
|
||||
|
||||
diff-=x;
|
||||
REQUIRE(diff.Norml2() < tol);
|
||||
}
|
||||
|
||||
|
||||
SECTION("3D")
|
||||
{
|
||||
DenseMatrix A2(
|
||||
{
|
||||
{1.14593, 0.76119},
|
||||
{0.76119, 0.78993}
|
||||
});
|
||||
DenseMatrix B2(
|
||||
{
|
||||
{0.88088, 0.37899},
|
||||
{0.37899, 0.45096}
|
||||
});
|
||||
Array<DenseMatrix *> A(3), B(3);
|
||||
A[0] = &A0; A[1] = &A1; A[2] = &A2;
|
||||
B[0] = &B0; B[1] = &B1; B[2] = &B2;
|
||||
|
||||
Vector y(24); y.Randomize(1);
|
||||
Vector x(24), diff(24);
|
||||
|
||||
FDSolver S(A,B);
|
||||
S.Mult(y,x);
|
||||
|
||||
DenseMatrix Temp, C0, C1, C;
|
||||
|
||||
KronProd(A0, B1, Temp);
|
||||
KronProd(Temp, B2, C0);
|
||||
KronProd(B0, A1, Temp);
|
||||
KronProd(Temp, B2, C1);
|
||||
KronProd(B0, B1, Temp);
|
||||
KronProd(Temp, A2, C);
|
||||
|
||||
C.Add(1.,C0);
|
||||
C.Add(1.,C1);
|
||||
|
||||
DenseMatrixInverse Cinv(C);
|
||||
Cinv.Mult(y,diff);
|
||||
|
||||
diff-=x;
|
||||
REQUIRE(diff.Norml2() < tol);
|
||||
}
|
||||
}
|
||||
|
||||
#endif // if MFEM_USE_LAPACK
|
||||
@@ -12,6 +12,8 @@
|
||||
#include "mfem.hpp"
|
||||
#include "unit_tests.hpp"
|
||||
#include "linalg/dtensor.hpp"
|
||||
#include "nvToolsExt.h"
|
||||
#include "nvToolsExtCudaRt.h"
|
||||
|
||||
using namespace mfem;
|
||||
|
||||
@@ -239,6 +241,212 @@ TEST_CASE("DenseMatrix A*B^T methods",
|
||||
}
|
||||
}
|
||||
|
||||
TEST_CASE("KronMult methods",
|
||||
"[DenseMatrix], [CUDA]")
|
||||
{
|
||||
double tol = 1e-12;
|
||||
int nA = 3, mA = 4;
|
||||
int nB = 5, mB = 6;
|
||||
DenseMatrix A(nA,mA);
|
||||
DenseMatrix B(nB,mB);
|
||||
|
||||
for (int i = 0; i<nA; i++)
|
||||
for (int j = 0; j<mA; j++)
|
||||
{
|
||||
A(i,j) = ((double)rand()/(double)RAND_MAX);
|
||||
}
|
||||
|
||||
for (int i = 0; i<nB; i++)
|
||||
for (int j = 0; j<mB; j++)
|
||||
{
|
||||
B(i,j) = ((double)rand()/(double)RAND_MAX);
|
||||
}
|
||||
|
||||
DenseMatrix AB;
|
||||
KronProd(A,B,AB);
|
||||
AB.HostRead();
|
||||
|
||||
// (A ⊗ B) r
|
||||
SECTION("KronMultABr")
|
||||
{
|
||||
nvtxRangePush("KronMultABr");
|
||||
Vector r(mA*mB);
|
||||
MFEM_VERIFY(r.Size() == AB.Width(), "Check r size");
|
||||
|
||||
r.HostReadWrite();
|
||||
r.Randomize();
|
||||
|
||||
Vector z0(AB.Height());
|
||||
AB.Mult(r,z0);
|
||||
//z0.HostRead();
|
||||
Vector z1;
|
||||
KronMult(A,B,r,z1);
|
||||
MFEM_VERIFY(z0.Size() == z1.Size(), "Check z1 size");
|
||||
z0-=z1;
|
||||
REQUIRE(z0.Norml2() < tol);
|
||||
nvtxRangePop();
|
||||
}
|
||||
|
||||
|
||||
// (A ⊗ B) R
|
||||
SECTION("KronMultABR")
|
||||
{
|
||||
nvtxRangePush("KronMultABR");
|
||||
int nR = mA*mB;
|
||||
int mR = 7;
|
||||
DenseMatrix R(nR, mR);
|
||||
R.HostReadWrite();
|
||||
|
||||
for (int i = 0; i<nR; i++)
|
||||
for (int j = 0; j<mR; j++)
|
||||
{
|
||||
R(i,j) = ((double)rand()/(double)RAND_MAX);
|
||||
}
|
||||
|
||||
DenseMatrix Z0(nA*nB,mR);
|
||||
Mult(AB,R,Z0);
|
||||
|
||||
DenseMatrix Z1;
|
||||
KronMult(A,B,R,Z1);
|
||||
MFEM_VERIFY(Z0.Height() == Z1.Height() &&
|
||||
Z0.Width() == Z1.Width(), "Check z1 size");
|
||||
Z0-=Z1;
|
||||
|
||||
REQUIRE(Z0.MaxMaxNorm() < tol);
|
||||
|
||||
nvtxRangePop();
|
||||
}
|
||||
|
||||
// (A ⊗ B ⊗ C) r
|
||||
SECTION("KronMultABCr")
|
||||
{
|
||||
nvtxRangePush("KronMultABCr");
|
||||
|
||||
int nC = 7, mC = 2;
|
||||
DenseMatrix C(nC, mC);
|
||||
|
||||
C.HostReadWrite();
|
||||
|
||||
for (int i = 0; i<nC; i++)
|
||||
for (int j = 0; j<mC; j++)
|
||||
{
|
||||
C(i,j) = ((double)rand()/(double)RAND_MAX);
|
||||
}
|
||||
|
||||
DenseMatrix ABC;
|
||||
KronProd(AB,C,ABC);
|
||||
|
||||
Vector r(mA*mB*mC);
|
||||
r.HostReadWrite();
|
||||
r.Randomize();
|
||||
|
||||
MFEM_VERIFY(r.Size() == ABC.Width(), "Check r size");
|
||||
Vector z0(nA*nB*nC);
|
||||
ABC.Mult(r,z0);
|
||||
|
||||
Vector z1;
|
||||
KronMult(A,B,C,r,z1);
|
||||
MFEM_VERIFY(z0.Size() == z1.Size(), "Check z1 size");
|
||||
z0-=z1;
|
||||
REQUIRE(z0.Norml2() < tol);
|
||||
|
||||
nvtxRangePop();
|
||||
}
|
||||
}
|
||||
|
||||
TEST_CASE("KronMultInv methods",
|
||||
"[DenseMatrixInverse]")
|
||||
{
|
||||
double tol = 1e-12;
|
||||
int nA = 3;
|
||||
int nB = 2;
|
||||
DenseMatrix A(
|
||||
{
|
||||
{ 1.0, 0.2, 3.4},
|
||||
{-2.0, -1.0, 3.1},
|
||||
{ 0.7, 1.4,-0.9}
|
||||
});
|
||||
DenseMatrix B(
|
||||
{
|
||||
{-10.1, 5.7},
|
||||
{-3.0, 4.2}
|
||||
});
|
||||
|
||||
DenseMatrixInverse Ainv(A);
|
||||
DenseMatrixInverse Binv(B);
|
||||
|
||||
DenseMatrix AB;
|
||||
KronProd(A,B,AB);
|
||||
|
||||
// (A^-1 ⊗ B^-1) r
|
||||
SECTION("KronMultInvABr")
|
||||
{
|
||||
|
||||
Vector r(nA*nB); r.Randomize();
|
||||
MFEM_VERIFY(r.Size() == AB.Width(), "Check r size");
|
||||
Vector z0(AB.Height());
|
||||
DenseMatrixInverse ABinv(AB);
|
||||
ABinv.Mult(r,z0);
|
||||
|
||||
Vector z1;
|
||||
KronMult(Ainv,Binv,r,z1);
|
||||
MFEM_VERIFY(z0.Size() == z1.Size(), "Check z1 size");
|
||||
z0-=z1;
|
||||
REQUIRE(z0.Norml2() < tol);
|
||||
}
|
||||
|
||||
// (A^-1 ⊗ B^-1) R
|
||||
SECTION("KronMultInvABR")
|
||||
{
|
||||
int nR = nA*nB;
|
||||
int mR = 7;
|
||||
DenseMatrix R(nR, mR);
|
||||
for (int i = 0; i<nR; i++)
|
||||
for (int j = 0; j<mR; j++)
|
||||
{
|
||||
R(i,j) = ((double)rand()/(double)RAND_MAX);
|
||||
}
|
||||
|
||||
DenseMatrixInverse ABinv(AB);
|
||||
DenseMatrix Z0(nA*nB,mR);
|
||||
ABinv.Mult(R,Z0);
|
||||
|
||||
DenseMatrix Z1;
|
||||
KronMult(Ainv,Binv,R,Z1);
|
||||
MFEM_VERIFY(Z0.Height() == Z1.Height() &&
|
||||
Z0.Width() == Z1.Width(), "Check z1 size");
|
||||
Z0-=Z1;
|
||||
|
||||
REQUIRE(Z0.MaxMaxNorm() < tol);
|
||||
}
|
||||
|
||||
// (A^-1 ⊗ B^-1 ⊗ C^-1) r
|
||||
SECTION("KronMultInvABCr")
|
||||
{
|
||||
int nC = 4;
|
||||
DenseMatrix C(
|
||||
{
|
||||
{-2.1, 1.6, -3.4, 17.5},
|
||||
{-7.1, 1.3, -7.5, -12.5},
|
||||
{ 0.5, 5.7, -6.0, -0.5},
|
||||
{ 9.2, 0.3, -1.4, -14.9}
|
||||
});
|
||||
DenseMatrix ABC;
|
||||
KronProd(AB,C,ABC);
|
||||
DenseMatrixInverse ABCInv(ABC);
|
||||
Vector r(nA*nB*nC); r.Randomize();
|
||||
MFEM_VERIFY(r.Size() == ABC.Width(), "Check r size");
|
||||
Vector z0(nA*nB*nC);
|
||||
ABCInv.Mult(r,z0);
|
||||
|
||||
DenseMatrixInverse Cinv(C);
|
||||
Vector z1;
|
||||
KronMult(Ainv,Binv,Cinv,r,z1);
|
||||
MFEM_VERIFY(z0.Size() == z1.Size(), "Check z1 size");
|
||||
z0-=z1;
|
||||
REQUIRE(z0.Norml2() < tol);
|
||||
}
|
||||
}
|
||||
|
||||
TEST_CASE("LUFactors RightSolve", "[DenseMatrix]")
|
||||
{
|
||||
@@ -314,6 +522,88 @@ TEST_CASE("DenseTensor LinearSolve methods",
|
||||
}
|
||||
}
|
||||
|
||||
#ifdef MFEM_USE_LAPACK
|
||||
|
||||
TEST_CASE("EigenSystem methods",
|
||||
"[DenseMatrix]")
|
||||
{
|
||||
double tol = 1e-12;
|
||||
SECTION("SPD Matrix")
|
||||
{
|
||||
DenseMatrix A({{0.56806, 0.29211, 0.48315, 0.70024},
|
||||
{0.29211, 0.85147, 0.68123, 0.70689},
|
||||
{0.48315, 0.68123, 1.07229, 1.02681},
|
||||
{0.70024, 0.70689, 1.02681, 1.15468}
|
||||
});
|
||||
DenseMatrix V, AV(4);
|
||||
Vector Lambda;
|
||||
for (bool sym: { false, true })
|
||||
{
|
||||
DenseMatrixEigensystem EigA(A,sym);
|
||||
EigA.Eval();
|
||||
V = EigA.Eigenvectors();
|
||||
Lambda = EigA.Eigenvalues();
|
||||
Mult(A,V,AV);
|
||||
V.RightScaling(Lambda);
|
||||
AV -= V;
|
||||
REQUIRE(AV.MaxMaxNorm() < tol);
|
||||
}
|
||||
}
|
||||
|
||||
SECTION("Indefinite Matrix")
|
||||
{
|
||||
DenseMatrix A({{0.486278, 0.041135, 0.480727, 0.616026},
|
||||
{0.523599, 0.119827, 0.087808, 0.415241},
|
||||
{0.214454, 0.661631, 0.909626, 0.744259},
|
||||
{0.107007, 0.630604, 0.077862, 0.221006}
|
||||
});
|
||||
DenseMatrixEigensystem EigA(A);
|
||||
EigA.Eval();
|
||||
|
||||
Vector Lambda_r, Lambda_i;
|
||||
// Real part of eigenvalues
|
||||
Lambda_r = EigA.Eigenvalues();
|
||||
// Imag part of eigenvalues
|
||||
Lambda_i = EigA.Eigenvalues(true);
|
||||
|
||||
DenseMatrix V;
|
||||
V = EigA.Eigenvectors();
|
||||
// Real part of eigenvectors
|
||||
DenseMatrix Vr(4), Vi(4);
|
||||
Vr.SetCol(0,V.GetColumn(0));
|
||||
Vr.SetCol(1,V.GetColumn(1));
|
||||
Vr.SetCol(2,V.GetColumn(1));
|
||||
Vr.SetCol(3,V.GetColumn(3));
|
||||
|
||||
// Imag part of eigenvectors
|
||||
Vector vi(4); V.GetColumn(2,vi);
|
||||
Vi.SetCol(0,0.);
|
||||
Vi.SetCol(1,vi); vi *= -1.;
|
||||
Vi.SetCol(2,vi);
|
||||
Vi.SetCol(3,0.);
|
||||
|
||||
// Check that A*V = V * Lambda
|
||||
// or A * (V_r + i V_i ) = (V_r + i V_i)*(Lamda_r + i Lambda_i)
|
||||
// or A * V_r = V_r * Lambda_r - V_i * Lambda_i
|
||||
// and A * V_i = V_r ( Lambda_i + V_i * Lambda_r
|
||||
DenseMatrix AVr(4), AVi(4);
|
||||
Mult(A,Vr, AVr);
|
||||
Mult(A,Vi, AVi);
|
||||
|
||||
DenseMatrix Vrlr = Vr; Vrlr.RightScaling(Lambda_r);
|
||||
DenseMatrix Vrli = Vr; Vrli.RightScaling(Lambda_i);
|
||||
DenseMatrix Vilr = Vi; Vilr.RightScaling(Lambda_r);
|
||||
DenseMatrix Vili = Vi; Vili.RightScaling(Lambda_i);
|
||||
|
||||
AVr -= Vrlr; AVr+= Vili;
|
||||
AVi -= Vrli; AVi-= Vilr;
|
||||
|
||||
REQUIRE(AVr.MaxMaxNorm() < tol);
|
||||
REQUIRE(AVi.MaxMaxNorm() < tol);
|
||||
}
|
||||
}
|
||||
|
||||
#endif // if MFEM_USE_LAPACK
|
||||
TEST_CASE("DenseTensor copy", "[DenseMatrix][DenseTensor]")
|
||||
{
|
||||
DenseTensor t1(2,3,4);
|
||||
|
||||
+13
-8
@@ -214,21 +214,27 @@ $(DATA_DIR):
|
||||
MFEM_TESTS = UNIT_TESTS
|
||||
include $(MFEM_TEST_MK)
|
||||
|
||||
ifeq (,$(wildcard $(MFEM_DIR)/../data))
|
||||
MFEM_DATA_FLAG =
|
||||
else
|
||||
MFEM_DATA_FLAG = --data $(MFEM_DIR)/../data
|
||||
endif
|
||||
|
||||
%-test-seq: %
|
||||
@$(call mfem-test,$<,, Unit tests,,SKIP-NO-VIS)
|
||||
@$(call mfem-test,$<,, Unit tests,$(MFEM_DATA_FLAG),SKIP-NO-VIS)
|
||||
|
||||
ceed_tests-test-seq: ceed_tests
|
||||
@$(call mfem-test,$<,, CEED Unit tests (cpu),--device ceed-cpu,SKIP-NO-VIS)
|
||||
@$(call mfem-test,$<,, CEED Unit tests (cpu),--device ceed-cpu $(MFEM_DATA_FLAG),SKIP-NO-VIS)
|
||||
ifeq ($(MFEM_USE_CUDA),YES)
|
||||
@$(call mfem-test,$<,, CEED Unit tests (cuda-ref),--device ceed-cuda:/gpu/cuda/ref,SKIP-NO-VIS)
|
||||
@$(call mfem-test,$<,, CEED Unit tests (cuda-shared),--device ceed-cuda:/gpu/cuda/shared,SKIP-NO-VIS)
|
||||
@$(call mfem-test,$<,, CEED Unit tests (cuda-gen),--device ceed-cuda:/gpu/cuda/gen,SKIP-NO-VIS)
|
||||
@$(call mfem-test,$<,, CEED Unit tests (cuda-ref),--device ceed-cuda:/gpu/cuda/ref $(MFEM_DATA_FLAG),SKIP-NO-VIS)
|
||||
@$(call mfem-test,$<,, CEED Unit tests (cuda-shared),--device ceed-cuda:/gpu/cuda/shared $(MFEM_DATA_FLAG),SKIP-NO-VIS)
|
||||
@$(call mfem-test,$<,, CEED Unit tests (cuda-gen),--device ceed-cuda:/gpu/cuda/gen $(MFEM_DATA_FLAG),SKIP-NO-VIS)
|
||||
endif
|
||||
|
||||
RUN_MPI = $(MFEM_MPIEXEC) $(MFEM_MPIEXEC_NP)
|
||||
%-test-par: %
|
||||
@$(call mfem-test,$<, $(RUN_MPI) 1, Parallel unit tests,,SKIP-NO-VIS)
|
||||
@$(call mfem-test,$<, $(RUN_MPI) $(MFEM_MPI_NP), Parallel unit tests,,SKIP-NO-VIS)
|
||||
@$(call mfem-test,$<, $(RUN_MPI) 1, Parallel unit tests,$(MFEM_DATA_FLAG),SKIP-NO-VIS)
|
||||
@$(call mfem-test,$<, $(RUN_MPI) $(MFEM_MPI_NP), Parallel unit tests,$(MFEM_DATA_FLAG),SKIP-NO-VIS)
|
||||
|
||||
# Generate an error message if the MFEM library is not built and exit
|
||||
$(MFEM_LIB_FILE):
|
||||
@@ -237,4 +243,3 @@ $(MFEM_LIB_FILE):
|
||||
clean:
|
||||
rm -f $(SEQ_UNIT_TESTS) $(PAR_UNIT_TESTS) *.o */*.o */*~ *~
|
||||
rm -rf *.dSYM output_meshes
|
||||
|
||||
|
||||
@@ -41,3 +41,21 @@ TEST_CASE("VTU XML Reader", "[Mesh][VTU][XML]")
|
||||
REQUIRE(mesh.GetNumGeometries(2) == 1);
|
||||
}
|
||||
}
|
||||
|
||||
TEST_CASE("VTU XML Compressed Blocks", "[VTU][XML][MFEMData]")
|
||||
{
|
||||
auto filename = GENERATE(
|
||||
"bracket_appended_compressed.vtu",
|
||||
"bracket_appended_encoded_compressed.vtu",
|
||||
"bracket_inline_compressed.vtu"
|
||||
);
|
||||
|
||||
std::string mesh_path = mfem_data_dir + "/vtk/" + filename;
|
||||
Mesh mesh = Mesh::LoadFromFile(mesh_path.c_str());
|
||||
|
||||
REQUIRE(mesh.Dimension() == 3);
|
||||
REQUIRE(mesh.GetNE() == 206208);
|
||||
REQUIRE(mesh.GetNV() == 50000);
|
||||
REQUIRE(mesh.HasGeometry(Geometry::TETRAHEDRON));
|
||||
REQUIRE(mesh.GetNumGeometries(3) == 1);
|
||||
}
|
||||
|
||||
@@ -186,10 +186,12 @@ int tmop(int id, Req &res, int argc, char *argv[])
|
||||
case 2: metric = new TMOP_Metric_002; break;
|
||||
case 7: metric = new TMOP_Metric_007; break;
|
||||
case 77: metric = new TMOP_Metric_077; break;
|
||||
case 80: metric = new TMOP_Metric_080(0.5); break;
|
||||
case 302: metric = new TMOP_Metric_302; break;
|
||||
case 303: metric = new TMOP_Metric_303; break;
|
||||
case 315: metric = new TMOP_Metric_315; break;
|
||||
case 321: metric = new TMOP_Metric_321; break;
|
||||
case 332: metric = new TMOP_Metric_332(0.5); break;
|
||||
default:
|
||||
{
|
||||
if (id == 0) { cout << "Unknown metric_id: " << metric_id << endl; }
|
||||
@@ -740,6 +742,13 @@ static void tmop_tests(int id = 0, bool all = false)
|
||||
POR({1,2}).QOR({2,4}).
|
||||
TID({4}).MID({1,2})).Run(id,all);
|
||||
|
||||
Launch(Launch::Args("Square01 + Adapted discrete size").
|
||||
MESH("../../miniapps/meshing/square01.mesh").REFINE(1).
|
||||
NORMALIZATION(true).
|
||||
POR({1,2}).QOR({4,6}).
|
||||
LINEAR_ITERATIONS(150).
|
||||
TID({5}).MID({80}).LS({3})).Run(id,all);
|
||||
|
||||
Launch(Launch::Args("Blade").
|
||||
MESH("../../miniapps/meshing/blade.mesh").
|
||||
POR({1,2}).QOR({2,4}).
|
||||
@@ -768,6 +777,12 @@ static void tmop_tests(int id = 0, bool all = false)
|
||||
POR({1,2}).QOR({4,2}).
|
||||
TID({7}).MID({302,321})).Run(id,all);
|
||||
|
||||
Launch(Launch::Args("Cube + Discrete size + normalization").
|
||||
MESH("../../miniapps/meshing/cube.mesh").
|
||||
NORMALIZATION(true).
|
||||
POR({1,2}).QOR({4,2}).
|
||||
TID({5}).MID({332})).Run(id,all);
|
||||
|
||||
// Note: order 1 has no interior nodes, so all residuals are zero and the
|
||||
// Newton iteration exits immediately.
|
||||
Launch(Launch::Args("Toroid-Hex").
|
||||
|
||||
Some files were not shown because too many files have changed in this diff Show More
Reference in New Issue
Block a user