more temporary memory in tmop

use temp mem in tmop+pa

.gitignore

@@ -175,7 +175,6 @@ miniapps/meshing/mesh-optimizer
 miniapps/meshing/pmesh-optimizer
 miniapps/meshing/minimal-surface
 miniapps/meshing/pminimal-surface
-miniapps/meshing/polar-nc
 
 miniapps/meshing/mobius-strip.mesh
 miniapps/meshing/klein-bottle.mesh
@@ -188,7 +187,6 @@ miniapps/meshing/extruder.mesh
 miniapps/meshing/trimmer.mesh
 miniapps/meshing/optimized*
 miniapps/meshing/perturbed*
-miniapps/meshing/polar-nc.mesh
 
 miniapps/performance/ex1
 miniapps/performance/ex1p
@@ -235,7 +233,6 @@ miniapps/nurbs/mode_*
 miniapps/nurbs/Example1*
 
 miniapps/gslib/field-diff
-miniapps/gslib/field-interp
 miniapps/gslib/findpts
 miniapps/gslib/pfindpts
 
@@ -256,16 +253,20 @@ tests/unit/unit_tests
 tests/unit/punit_tests
 tests/unit/sedov_tests_*
 tests/unit/psedov_tests_*
+tests/unit/tmop_tests_*
+tests/unit/ptmop_tests_*
+tests/unit/cube.mesh
+tests/unit/star.mesh
+tests/unit/blade.mesh
+tests/unit/square01.mesh
+tests/unit/toroid-hex.mesh
+tests/unit/beam-hex-nurbs.mesh
+tests/unit/square-disc-nurbs.mesh
 
 # Test script output
 tests/scripts/*.err
 tests/scripts/*.out
 tests/scripts/*.msg
 
-# Other tests
-tests/convergence/rates
-tests/convergence/prates
-tests/par-mesh-format/ex1p
-
 # VPATH builds
 build-*/*

.gitlab-ci.yml

@@ -71,8 +71,6 @@ stages:
   - build
   - test
   - deallocate
-  - lassen_build
-  - lassen_test
   - baseline_check
   - baseline_publish
 
@@ -81,11 +79,7 @@ stages:
 # 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:
-  tags:
-    - shell
-    - quartz
+.setup:
   stage: setup
   variables:
     GIT_STRATEGY: none
@@ -106,15 +100,6 @@ setup:
   before_script:
     - module load gcc/6.1.0
 
-# On lassen
-.with_gcc_8_3_1:
-  variables:
-    TOOLCHAIN: gcc_8_3_1
-    CXX: g++
-    CC: gcc
-  before_script:
-    - module load gcc/8.3.1
-
 .with_gcc_4_9_3:
   variables:
     TOOLCHAIN: gcc_4_9_3
@@ -305,4 +290,3 @@ setup:
 # The list on jobs is defined in machine-specific files.
 include:
   - local: .gitlab/quartz.yml
-  - local: .gitlab/lassen.yml

.gitlab/lassen.yml

@@ -1,57 +0,0 @@
-# Copyright (c) 2010-2020, 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.
-
-# GitLab pipelines configurations for the Lassen machine at LLNL
-
-.on_lassen:
-  tags:
-    - shell
-    - lassen
-  variables:
-    PLAT: lassen
-
-# Build MFEM
-build_mfem_ser_lassen:
-  extends: [.with_gcc_8_3_1, .on_lassen]
-  needs: [setup]
-  stage: lassen_build
-  script:
-    - mkdir -p ${BUILD_PATH}
-    - cp -r ${CI_PROJECT_DIR} ${BUILD_PATH}/${CI_PROJECT_NAME}_lassen_ser
-    - cd ${BUILD_PATH}/${CI_PROJECT_NAME}_lassen_ser
-    - lalloc 1 -W 5 -q pdebug make -j cuda CUDA_ARCH=sm_70
-
-build_mfem_debug_ser_lassen:
-  extends: [.with_gcc_8_3_1, .on_lassen]
-  needs: [setup]
-  stage: lassen_build
-  script:
-    - mkdir -p ${BUILD_PATH}
-    - cp -r ${CI_PROJECT_DIR} ${BUILD_PATH}/${CI_PROJECT_NAME}_lassen_ser_debug
-    - cd ${BUILD_PATH}/${CI_PROJECT_NAME}_lassen_ser_debug
-    - lalloc 1 -W 5 -q pdebug make -j cuda MFEM_DEBUG="YES" CUDA_ARCH=sm_70
-
-# Sanity check
-sanitycheck_mfem_ser_lassen:
-  extends: [.with_gcc_8_3_1, .on_lassen]
-  stage: lassen_test
-  needs: [build_mfem_ser_lassen]
-  script:
-    - cd ${BUILD_PATH}/${CI_PROJECT_NAME}_lassen_ser
-    - lalloc 1 -W 15 -q pdebug make -j test
-
-sanitycheck_mfem_debug_ser_lassen:
-  extends: [.with_gcc_8_3_1, .on_lassen]
-  stage: lassen_test
-  needs: [build_mfem_debug_ser_lassen]
-  script:
-    - cd ${BUILD_PATH}/${CI_PROJECT_NAME}_lassen_ser_debug
-    - lalloc 1 -W 30 -q pdebug make -j test

.gitlab/quartz.yml

@@ -22,6 +22,10 @@
     MAKE_PAR: 6
     BASELINE_PAR: 18
 
+# Setup
+setup_quartz:
+  extends: [.setup, .on_quartz]
+
 # Allocate
 allocate_quartz:
   variables:

CHANGELOG

@@ -16,12 +16,7 @@ Meshing improvements
 - The graph linear ordering library Gecko, previously an external dependency, is
   now included directly in MFEM. As a result, Mesh::GetGeckoElementOrdering is
   always available. The interface has also been improved, see for example the
-  Mesh Explorer miniapp.
-
-- Improved Gmsh reader (version 2.2), which now supports both high-order and
-  periodic meshes. Segments, triangles, quadrilaterals, and tetrahedra are
-  supported up to order 10. Wedges and hexahedra are supported up to order 9.
-  For sample periodic meshes, see the periodic*.msh files in the data directory.
+  mesh-explorer miniapp.
 
 - Added support for finite difference-based gradient and Hessian approximation
   in the TMOP mesh optimization algorithms. This improves the accuracy of the
@@ -32,17 +27,15 @@ Meshing improvements
   the user to specify different discrete functions for controlling the
   size, aspect-ratio, orientation, and skew of elements in the mesh.
 
-- Added TMOP capability for approximate tangential mesh relaxation. Added
-  support and examples for using TMOP on mixed meshes.
+- Added TMOP capability for approximate tangential mesh relaxation.
+
+- Added support for reading periodic meshes in Gmsh format (version 2.2). See
+  for example the periodic-annulus-sector and periodic-torus-sector files in
+  the data directory.
 
 - Added complete action of the TMOP Integrator to account for the spatial
   derivatives of discrete and analytic targets.
 
-- Added support for initialization of (serial) non-conforming meshes. Hanging
-  nodes can be marked with Mesh::AddVertexParents when building the mesh with
-  the "init" constructor. The usage is demonstrated in a new meshing miniapp
-  (polar-nc) which generates meshes that are non-conforming from the start.
-
 Performance improvements
 ------------------------
 - Added support for explicit vectorization in the high-performance templated
@@ -57,6 +50,11 @@ Performance improvements
 Improved GPU capabilities
 -------------------------
 - Added support for Chebyshev accelerated polynomial smoother on GPU.
+- The TMOP mesh optimization algorithms were extended to GPU:
+   - QualityMetric #1, #2 and #7 are available in 2D, #302, #303 and #321 in 3D
+   - Both AnalyticAdaptTC and DiscreteAdaptTC TargetConstructor are available
+   - Kernels for normalization and limiting have been added
+   - The AdvectorCG now also support AssemblyLevel::PARTIAL
 
 - Optimized AMD/HIP kernel support.
 
@@ -65,14 +63,8 @@ Improved GPU capabilities
   compute a global sparse matrix. All integrators supported by element assembly
   are also supported by full assembly. See the '-fa' option in Example 9.
 
-- Added CUDA support for sparse matrix-vector multiplication with cuSPARSE.
-
 - Added support for BlockOperator on GPU. See the updated Example 5.
 
-- Added partial assembly and GPU support for complex operators, including the
-  classes ComplexOperator, [Par]ComplexGridFunction, [Par]ComplexLinearForm, and
-  [Par]SesquilinearForm. See the updated Example 22.
-
 Discretization improvements
 ---------------------------
 - Added support for matrix-free interpolation and restriction operators between
@@ -101,14 +93,6 @@ Discretization improvements
 
 - Added support face integrals on the boundaries of NURBS meshes.
 
-- Added support for interpolation of functions in L2, H(div) and H(curl)
-  spaces using GSLIB-FindPoints.
-
-- Added support for computing asymptotic error estimates and convergence rates
-  for the whole de Rham sequence based on the new class ConvergenceStudy and new
-  member methods in GridFunction and ParGridFunction. See the rates.cpp file in
-  the tests/convergence directory for sample usage.
-
 Linear and nonlinear solvers
 ----------------------------
 - Added power method to iteratively estimate the largest eigenvalue and the
@@ -134,12 +118,6 @@ Linear and nonlinear solvers
 
 - Added support for the SLEPc eigensolver package.
 
-- Added partially assembled convergent diagonal preconditioner for adaptively
-  refined meshes (i.e. non-conforming finite element spaces), see Example 6/6p.
-
-- Added an interface to the Intel MKL Parallel Direct Sparse Solver for
-  Clusters. An example usage of the interface is shown in Example 11p.
-
 New and updated examples and miniapps
 -------------------------------------
 - Added a new example, Example 25/25p, to demonstrate the use of a Perfectly
@@ -176,9 +154,6 @@ New and updated examples and miniapps
 - Added a new meshing miniapp, Minimal Surface, which solves Plateau's problem:
   the Dirichlet problem for the minimal surface equation.
 
-- Added a new meshing miniapp, Polar NC, which demonstrates the construction of
-  polar non-conforming meshes.
-
 - Added partial assembly support to Example 4/4p and Example 5/5p, with diagonal
   preconditioning.
 
@@ -193,47 +168,28 @@ New and updated examples and miniapps
   mesh based on element attributes. Any newly exposed boundary elements are
   assigned attribute numbers related to the trimmed element attributes.
 
-- Added a new miniapp (field-interp) that demonstrates transfer of grid function
-  between different meshes using GSLIB-FindPoints.
-
-- Added diagonal preconditioner in Example 6/6p for partial assembly with AMR.
-
 - Added device support in Example 5/5p.
 
-- Added partial assembly and device support to Example 22/22p, with diagonal
-  preconditioning.
-
-- Added the option to plot a function in Mesh Explorer.
-
 Improved testing
 ----------------
-- Upgraded the Catch unit test framework from version 1.6.1 to version 2.13.0.
-
 - Added a GitLab pipeline that automates PR testing on supercomputing systems
   and Linux clusters at Lawrence Livermore National Lab (LLNL). This can be
   triggered only by LLNL developers, see .gitlab-ci.yml, the .gitlab directory
   and the updated CONTRIBUTING.md file.
 
-- Added testing of the parallel mesh format in tests/par-mesh-format.
-
 Miscellaneous
 -------------
 - Added support for ADIOS2 for parallel I/O with ParaView visualization. The
   classes adios2stream and ADIOS2DataCollection are introduced in mfem as the
   interfaces to generate ADIOS2 Binary Pack (BP4) directory datasets for the
-  entire spatial and temporal node data. Cell centered data is accessible by
-  ADIOS2 data readers (e.g. Python), but currently not yet implement as of
-  ParaView v5.8.1. In addition, ADIOS2 allows for setting a user-defined number
-  of data substreams/subfiles at scale. See examples 5, 9, 12, 16.
+  entire spatial and temporal data. In addition, ADIOS2 allows for setting a
+  user-defined number of data substreams/subfiles. See examples 5, 9, 12, 16.
 
 - The integration order used in the ComputeLpError and ComputeElementLpError
   methods of class GridFunction has been increased.
 
 - Various other simplifications, extensions, and bugfixes in the code.
 
-- Renamed "Backend::DEBUG" to "Backend::DEBUG_DEVICE" to avoid conflicts,
-  as DEBUG is sometimes used as a macro.
-
 
 Version 4.1, released on March 10, 2020
 =======================================

CMakeLists.txt

@@ -89,38 +89,8 @@ enable_language(CXX)
 if (MFEM_USE_CUDA)
    # MFEM_USE_CUDA requires CMake 3.8 or newer (for direct CUDA support)
    cmake_minimum_required(VERSION 3.8 FATAL_ERROR)
-   # Use ${CMAKE_CXX_COMPILER} as the cuda host compiler.
-   if (NOT CMAKE_CUDA_HOST_COMPILER)
-      set(CMAKE_CUDA_HOST_COMPILER ${CMAKE_CXX_COMPILER})
-   endif()
    enable_language(CUDA)
-   set(CMAKE_CUDA_STANDARD 11)
-   set(CMAKE_CUDA_STANDARD_REQUIRED ON)
-   set(CMAKE_CUDA_EXTENSIONS OFF)
-   set(CUDA_FLAGS "--expt-extended-lambda")
-   if (CMAKE_VERSION VERSION_LESS 3.18.0)
-      set(CUDA_FLAGS "-arch=${CUDA_ARCH} ${CUDA_FLAGS}")
-   elseif (NOT CMAKE_CUDA_ARCHITECTURES)
-      string(REGEX REPLACE "^sm_" "" ARCH_NUMBER "${CUDA_ARCH}")
-      if ("${CUDA_ARCH}" STREQUAL "sm_${ARCH_NUMBER}")
-         set(CMAKE_CUDA_ARCHITECTURES "${ARCH_NUMBER}")
-      else()
-         message(FATAL_ERROR "Unknown CUDA_ARCH: ${CUDA_ARCH}")
-      endif()
-   else()
-      set(CUDA_ARCH "CMAKE_CUDA_ARCHITECTURES: ${CMAKE_CUDA_ARCHITECTURES}")
-   endif()
    message(STATUS "Using CUDA architecture: ${CUDA_ARCH}")
-   if (CMAKE_VERSION VERSION_LESS 3.12.0)
-      # CMake versions 3.8 and 3.9 require this to work; 3.10 and 3.11 are not
-      # tested and may not actually need this (but should be ok to keep).
-      set(CUDA_FLAGS "-ccbin=${CMAKE_CXX_COMPILER} ${CUDA_FLAGS}")
-      set(CMAKE_CUDA_HOST_LINK_LAUNCHER ${CMAKE_CXX_COMPILER})
-   endif()
-   set(CMAKE_CUDA_FLAGS "${CUDA_FLAGS}" CACHE STRING
-       "CUDA flags set for MFEM" FORCE)
-   set(CUSPARSE_FOUND TRUE)
-   set(CUSPARSE_LIBRARIES "cusparse")
 endif()
 
 if (XSDK_ENABLE_C)
@@ -326,6 +296,22 @@ if (MFEM_USE_HIOP)
   # find_package updates HIOP_FOUND, HIOP_INCLUDE_DIRS, HIOP_LIBRARIES
 endif()
 
+# CUDA
+if (MFEM_USE_CUDA)
+   set(CMAKE_CUDA_STANDARD 11)
+   set(CMAKE_CUDA_STANDARD_REQUIRED ON)
+   set(CMAKE_CUDA_EXTENSIONS OFF)
+   set(CMAKE_CUDA_FLAGS "-arch=${CUDA_ARCH} --expt-extended-lambda"
+       CACHE STRING "CUDA flags set for MFEM" FORCE)
+   if (MFEM_USE_MPI)
+      set(CUDA_CCBIN_COMPILER ${MPI_CXX_COMPILER})
+   else()
+      set(CUDA_CCBIN_COMPILER ${CMAKE_CXX_COMPILER})
+   endif()
+   string(APPEND CMAKE_CUDA_FLAGS " -ccbin ${CUDA_CCBIN_COMPILER}")
+   set(CMAKE_CUDA_HOST_LINK_LAUNCHER ${CUDA_CCBIN_COMPILER})
+endif()
+
 # OCCA
 if (MFEM_USE_OCCA)
    find_package(OCCA REQUIRED)
@@ -346,12 +332,6 @@ if (MFEM_USE_ADIOS2)
    find_package(ADIOS2 REQUIRED)
 endif()
 
-if (MFEM_USE_MKL_CPARDISO)
-  if (MFEM_USE_MPI)
-   find_package(MKL_CPARDISO REQUIRED MKL_SEQUENTIAL MKL_LP64 MKL_MPI_WRAPPER)
-  endif()
-endif()
-
 # MFEM_TIMER_TYPE
 if (NOT DEFINED MFEM_TIMER_TYPE)
   if (APPLE)
@@ -377,8 +357,7 @@ endif()
 #    be before SuiteSparse.
 set(MFEM_TPLS MPI_CXX OPENMP BLAS LAPACK METIS HYPRE SuiteSparse SUNDIALS PETSC
     SLEPC MESQUITE SuperLUDist STRUMPACK AXOM CONDUIT Ginkgo GNUTLS GSLIB NETCDF
-    MPFR PUMI HIOP POSIXCLOCKS MFEMBacktrace ZLIB OCCA CEED RAJA UMPIRE ADIOS2
-    CUSPARSE MKL_CPARDISO)
+    MPFR PUMI HIOP POSIXCLOCKS MFEMBacktrace ZLIB OCCA CEED RAJA UMPIRE ADIOS2)
 # Add all *_FOUND libraries in the variable TPL_LIBRARIES.
 set(TPL_LIBRARIES "")
 set(TPL_INCLUDE_DIRS "")

INSTALL

@@ -486,13 +486,6 @@ MFEM_USE_CEED = YES/NO
    library for performant high-order operator evaluation developed by the Center
    for Efficient Exascale Discretizations in the Exascale Computing Project.
 
-MFEM_USE_MKL_CPARDISO = YES/NO
-   Enables the interface to the Intel MKL Parallel Direct Sparse Solver for
-   Clusters. Make sure to set the correct values for MKL_MPI_WRAPPER and
-   MKL_LIBRARY_SUBDIR as shown in defaults.mk. If you configure MFEM with
-   MFEM_USE_LAPACK=YES, verify that the MKL LAPACK libraries are used. The
-   OpenMP capabilities are disabled at link time.
-
 MFEM_BUILD_TAG = (any value)
    An optional tag to characterize the build. Exported to config/config.mk.
    Can be used to identify the MFEM build from other makefiles.
@@ -670,7 +663,7 @@ The specific libraries and their options are:
   URL: https://github.com/CEED/libCEED
        https://ceed.exascaleproject.org/libceed
   Options: CEED_DIR, CEED_OPT, CEED_LIB.
-  Versions: libCEED > 0.6, git-hash bdfed75.
+  Versions: libCEED >= 0.6, git-hash a970f63.
 
 - RAJA (optional), used when MFEM_USE_RAJA = YES.
   Beginning with MFEM v4.1, only RAJA v0.10.0+ is supported.

config/cmake/config.hpp.in

@@ -156,7 +156,4 @@
 // library.
 #cmakedefine MFEM_USE_SIMMETRIX
 
-// Enable interface to the MKL CPardiso library.
-#cmakedefine MFEM_USE_MKL_CPARDISO
-
 #endif // MFEM_CONFIG_HEADER

config/cmake/modules/FindADIOS2.cmake

@@ -38,19 +38,7 @@ if(NOT ADIOS2_FOUND)
   endif()
 
   find_path(ADIOS2_INCLUDE_DIR adios2.h ${ADIOS2_INCLUDE_OPTS})
-  
-  # adios2 version 2.5.0 
-  find_library(ADIOS2_LIBRARY NAMES adios2 ${ADIOS2_LIBRARY_OPTS})
-  
-  # adios2 version 2.6.0 and onwards
-  if(NOT ADIOS2_LIBRARY)
-  	find_library(ADIOS2_CXX11_MPI_LIBRARY NAMES adios2_cxx11_mpi ${ADIOS2_LIBRARY_OPTS})
-  	find_library(ADIOS2_CXX11_LIBRARY NAMES adios2_cxx11 ${ADIOS2_LIBRARY_OPTS})
-  	set(ADIOS2_LIBRARY ${ADIOS2_CXX11_MPI_LIBRARY} ${ADIOS2_CXX11_LIBRARY})
-  	if(MFEM_USE_MPI)
-  	  add_definitions(-DADIOS2_USE_MPI)
-  	endif()
-  endif() 
+  find_library(ADIOS2_LIBRARY NAMES adios2 ${ADIOS2_LIBRARY_OPTS}) 
 
   include(FindPackageHandleStandardArgs)
   find_package_handle_standard_args(ADIOS2

config/cmake/modules/FindMKL_CPARDISO.cmake

@@ -1,106 +0,0 @@
-# Copyright (c) 2010-2020, Lawrence Livermore National Security, LLC. Produced
-# at the Lawrence Livermore National Laboratory. All Rights reserved. See files
-# LICENSE and NOTICE for details. LLNL-CODE-806117.
-#
-# This file is part of the MFEM library. For more information and source code
-# availability visit https://mfem.org.
-#
-# MFEM is free software; you can redistribute it and/or modify it under the
-# terms of the BSD-3 license. We welcome feedback and contributions, see file
-# CONTRIBUTING.md for details.
-
-# Defines the following variables:
-#   - MKL_CPARDISO_FOUND
-#   - MKL_CPARDISO_LIBRARIES
-#   - MKL_CPARDISO_INCLUDE_DIRS
-
-if(NOT MKL_MPI_WRAPPER_LIB)
-  message(FATAL_ERROR "MKL CPardiso enabled but no MKL MPI Wrapper lib specified")
-endif()
-
-if(NOT MKL_LIBRARY_DIR)
-  message(WARNING "Using default MKL library path. Double check the variable MKL_LIBRARY_DIR")
-  set(MKL_LIBRARY_DIR "lib")
-endif()
-
-include(MfemCmakeUtilities)
-mfem_find_package(MKL_CPARDISO MKL_CPARDISO
-    MKL_CPARDISO_DIR "include" mkl_cluster_sparse_solver.h ${MKL_LIBRARY_DIR} mkl_core
-  "Paths to headers required by MKL CPardiso." "Libraries required by MKL CPARDISO."
-  ADD_COMPONENT MKL_LP64 "include" "" ${MKL_LIBRARY_DIR} mkl_intel_lp64
-  ADD_COMPONENT MKL_SEQUENTIAL "include" "" ${MKL_LIBRARY_DIR} mkl_sequential
-  ADD_COMPONENT MKL_MPI_WRAPPER "include" "" ${MKL_LIBRARY_DIR} ${MKL_MPI_WRAPPER_LIB}
-  CHECK_BUILD MKL_CPARDISO_VERSION_OK TRUE
-  "
-#include <mpi.h>
-#include <mkl.h>
-#include <mkl_cluster_sparse_solver.h>
-int main (void)
-{
-    MKL_INT n = 5;
-    MKL_INT ia[6]  = { 1, 4, 6, 9, 12, 14};
-    MKL_INT ja[13] = {  1, 2, 4,        /* index of non-zeros in 1 row*/
-                        1, 2,           /* index of non-zeros in 2 row*/
-                        3, 4, 5,        /* index of non-zeros in 3 row*/
-                        1, 3, 4,        /* index of non-zeros in 4 row*/
-                        2, 5            /* index of non-zeros in 5 row*/
-    };
-    double a[13] = {
-                         1.0, -1.0, /*0*/ -3.0, /*0*/
-                        -2.0,  5.0, /*0*/ /*0*/ /*0*/
-                        /*0*/  4.0,  6.0,  4.0, /*0*/
-                        -4.0, /*0*/  2.0,  7.0, /*0*/
-                        /*0*/  8.0, /*0*/ /*0*/ -5.0
-                    };
-
-    MKL_INT mtype = 11; /* set matrix type to \"real unsymmetric matrix\" */
-    MKL_INT nrhs  = 1;  /* Number of right hand sides. */
-    double b[5], x[5], bs[5], res, res0; /* RHS and solution vectors. */
-
-    /* Internal solver memory pointer pt
-     *       32-bit:      int pt[64] or void *pt[64];
-     *       64-bit: long int pt[64] or void *pt[64]; */
-    void *pt[64] = { 0 };
-
-    /* Cluster Sparse Solver control parameters. */
-    MKL_INT iparm[64] = { 0 };
-    MKL_INT maxfct, mnum, phase, msglvl, error;
-
-    /* Auxiliary variables. */
-    double  ddum; /* Double dummy   */
-    MKL_INT idum; /* Integer dummy. */
-    MKL_INT i, j;
-    int     mpi_stat = 0;
-    int     argc = 0;
-    int     comm, rank;
-    char*   uplo;
-    char**  argv;
-
-    mpi_stat = MPI_Init( &argc, &argv );
-    mpi_stat = MPI_Comm_rank( MPI_COMM_WORLD, &rank );
-    comm =  MPI_Comm_c2f( MPI_COMM_WORLD );
-
-    iparm[ 0] =  1; /* Solver default parameters overriden with provided by iparm */
-    iparm[ 1] =  2; /* Use METIS for fill-in reordering */
-    iparm[ 5] =  0; /* Write solution into x */
-    iparm[ 7] =  2; /* Max number of iterative refinement steps */
-    iparm[ 9] = 13; /* Perturb the pivot elements with 1E-13 */
-    iparm[10] =  1; /* Use nonsymmetric permutation and scaling MPS */
-    iparm[12] =  1; /* Switch on Maximum Weighted Matching algorithm (default for non-symmetric) */
-    iparm[17] = -1; /* Output: Number of nonzeros in the factor LU */
-    iparm[18] = -1; /* Output: Mflops for LU factorization */
-    iparm[26] =  1; /* Check input data for correctness */
-    iparm[39] =  0; /* Input: matrix/rhs/solution stored on master */
-    maxfct = 1; /* Maximum number of numerical factorizations. */
-    mnum   = 1; /* Which factorization to use. */
-    msglvl = 1; /* Print statistical information in file */
-    error  = 0; /* Initialize error flag */
-
-    phase = 11;
-    cluster_sparse_solver ( pt, &maxfct, &mnum, &mtype, &phase,
-                &n, a, ia, ja, &idum, &nrhs, iparm, &msglvl, &ddum, &ddum, &comm, &error );
-
-    mpi_stat = MPI_Finalize();
-    return error;
-}
-")

config/cmake/modules/MfemCmakeUtilities.cmake

@@ -128,15 +128,7 @@ function(add_mfem_miniapp MFEM_EXE_NAME)
   if (MFEM_USE_CUDA)
     set_property(SOURCE ${MAIN_LIST} ${EXTRA_SOURCES_LIST}
       PROPERTY LANGUAGE CUDA)
-    if (CMAKE_VERSION VERSION_GREATER_EQUAL 3.12.0)
-      list(TRANSFORM EXTRA_OPTIONS_LIST PREPEND "-Xcompiler=")
-    else()
-      set(LIST_)
-      foreach(item IN LISTS EXTRA_OPTIONS_LIST)
-        list(APPEND LIST_ "-Xcompiler=${item}")
-      endforeach()
-      set(EXTRA_OPTIONS_LIST ${LIST_})
-    endif()
+    list(TRANSFORM EXTRA_OPTIONS_LIST PREPEND "-Xcompiler=")
   endif()
 
   # Actually add the executable

config/config.hpp

@@ -42,15 +42,9 @@
 #ifdef MFEM_USE_SUPERLU
 #error Building with SuperLU_DIST (MFEM_USE_SUPERLU=YES) requires MPI (MFEM_USE_MPI=YES)
 #endif
-#ifdef MFEM_USE_MUMPS
-#error Building with MUMPS (MFEM_USE_MUMPS=YES) requires MPI (MFEM_USE_MPI=YES)
-#endif
 #ifdef MFEM_USE_STRUMPACK
 #error Building with STRUMPACK (MFEM_USE_STRUMPACK=YES) requires MPI (MFEM_USE_MPI=YES)
 #endif
-#ifdef MFEM_USE_MKL_CPARDISO
-#error Building with MKL CPARDISO (MFEM_USE_MKL_CPARDISO=YES) requires MPI (MFEM_USE_MPI=YES)
-#endif
 #ifdef MFEM_USE_PETSC
 #error Building with PETSc (MFEM_USE_PETSC=YES) requires MPI (MFEM_USE_MPI=YES)
 #endif

config/config.hpp.in

@@ -94,10 +94,6 @@
 // Enable MFEM functionality based on the SuperLU library.
 // #define MFEM_USE_SUPERLU
 
-// Enable MFEM functionality based on the MUMPS library.
-// #define MFEM_USE_MUMPS
-// #define MFEM_MUMPS_VERSION @MFEM_MUMPS_VERSION@
-
 // Enable MFEM functionality based on the STRUMPACK library.
 // #define MFEM_USE_STRUMPACK
 
@@ -167,7 +163,4 @@
 // library.
 // #define MFEM_USE_SIMMETRIX
 
-// Enable interface to the MKL CPardiso library.
-// #define MFEM_USE_MKL_CPARDISO
-
 #endif // MFEM_CONFIG_HEADER

config/config.mk.in

@@ -32,7 +32,6 @@ MFEM_USE_SUNDIALS      = @MFEM_USE_SUNDIALS@
 MFEM_USE_MESQUITE      = @MFEM_USE_MESQUITE@
 MFEM_USE_SUITESPARSE   = @MFEM_USE_SUITESPARSE@
 MFEM_USE_SUPERLU       = @MFEM_USE_SUPERLU@
-MFEM_USE_MUMPS         = @MFEM_USE_MUMPS@
 MFEM_USE_STRUMPACK     = @MFEM_USE_STRUMPACK@
 MFEM_USE_GINKGO        = @MFEM_USE_GINKGO@
 MFEM_USE_GNUTLS        = @MFEM_USE_GNUTLS@
@@ -53,7 +52,6 @@ MFEM_USE_CEED          = @MFEM_USE_CEED@
 MFEM_USE_UMPIRE        = @MFEM_USE_UMPIRE@
 MFEM_USE_SIMD          = @MFEM_USE_SIMD@
 MFEM_USE_ADIOS2        = @MFEM_USE_ADIOS2@
-MFEM_USE_MKL_CPARDISO  = @MFEM_USE_MKL_CPARDISO@
 
 # Compiler, compile options, and link options
 MFEM_CXX       = @MFEM_CXX@

config/defaults.cmake

@@ -52,7 +52,6 @@ option(MFEM_USE_CEED "Enable CEED" OFF)
 option(MFEM_USE_UMPIRE "Enable Umpire" OFF)
 option(MFEM_USE_SIMD "Enable use of SIMD intrinsics" OFF)
 option(MFEM_USE_ADIOS2 "Enable ADIOS2" OFF)
-option(MFEM_USE_MKL_CPARDISO "Enable MKL CPardiso" OFF)
 
 set(MFEM_MPI_NP 4 CACHE STRING "Number of processes used for MPI tests")
 
@@ -181,10 +180,6 @@ set(HIOP_DIR "${MFEM_DIR}/../hiop/install" CACHE STRING
 set(HIOP_REQUIRED_PACKAGES "BLAS" "LAPACK" CACHE STRING
     "Packages that HiOp depends on.")
 
-set(MKL_CPARDISO_DIR "" CACHE STRING "MKL installation path.")
-set(MKL_MPI_WRAPPER_LIB "mkl_blacs_mpich_lp64" CACHE STRING "MKL MPI wrapper library")
-set(MKL_LIBRARY_DIR "" CACHE STRING "Custom library subdirectory")
-
 set(OCCA_DIR "${MFEM_DIR}/../occa" CACHE PATH "Path to OCCA")
 set(RAJA_DIR "${MFEM_DIR}/../raja" CACHE PATH "Path to RAJA")
 set(CEED_DIR "${MFEM_DIR}/../libCEED" CACHE PATH "Path to libCEED")

config/defaults.mk

@@ -120,7 +120,6 @@ MFEM_USE_SUNDIALS      = NO
 MFEM_USE_MESQUITE      = NO
 MFEM_USE_SUITESPARSE   = NO
 MFEM_USE_SUPERLU       = NO
-MFEM_USE_MUMPS         = NO
 MFEM_USE_STRUMPACK     = NO
 MFEM_USE_GINKGO        = NO
 MFEM_USE_GNUTLS        = NO
@@ -139,9 +138,9 @@ MFEM_USE_RAJA          = NO
 MFEM_USE_OCCA          = NO
 MFEM_USE_CEED          = NO
 MFEM_USE_UMPIRE        = NO
+MFEM_USE_CAMP          = NO
 MFEM_USE_SIMD          = NO
 MFEM_USE_ADIOS2        = NO
-MFEM_USE_MKL_CPARDISO  = NO
 
 # Compile and link options for zlib.
 ZLIB_DIR =
@@ -158,7 +157,7 @@ HYPRE_OPT = -I$(HYPRE_DIR)/include
 HYPRE_LIB = -L$(HYPRE_DIR)/lib -lHYPRE
 
 # METIS library configuration
-ifeq ($(MFEM_USE_SUPERLU)$(MFEM_USE_STRUMPACK)$(MFEM_USE_MUMPS),NONONO)
+ifeq ($(MFEM_USE_SUPERLU)$(MFEM_USE_STRUMPACK),NONO)
    ifeq ($(MFEM_USE_METIS_5),NO)
      METIS_DIR = @MFEM_DIR@/../metis-4.0
      METIS_OPT =
@@ -234,7 +233,7 @@ SCALAPACK_DIR = @MFEM_DIR@/../scalapack-2.0.2
 SCALAPACK_OPT = -I$(SCALAPACK_DIR)/SRC
 SCALAPACK_LIB = -L$(SCALAPACK_DIR)/lib -lscalapack $(LAPACK_LIB)
 
-# MPI Fortran library, needed e.g. by STRUMPACK or MUMPS
+# MPI Fortran library, needed e.g. by STRUMPACK
 # MPICH:
 MPI_FORTRAN_LIB = -lmpifort
 # OpenMPI:
@@ -242,11 +241,6 @@ MPI_FORTRAN_LIB = -lmpifort
 # Additional Fortan library:
 # MPI_FORTRAN_LIB += -lgfortran
 
-# MUMPS library configuration
-MUMPS_DIR =
-MUMPS_OPT = -I$(MUMPS_DIR)/include
-MUMPS_LIB = -Wl,-rpath,$(MUMPS_DIR)/lib -L$(MUMPS_DIR)/lib -ldmumps -lmumps_common -lpord $(SCALAPACK_LIB) $(LAPACK_LIB) $(MPI_FORTRAN_LIB)
-
 # STRUMPACK library configuration
 STRUMPACK_DIR = @MFEM_DIR@/../STRUMPACK-build
 STRUMPACK_OPT = -I$(STRUMPACK_DIR)/include $(SCOTCH_OPT)
@@ -379,14 +373,10 @@ UMPIRE_DIR = @MFEM_DIR@/../umpire
 UMPIRE_OPT = -I$(UMPIRE_DIR)/include
 UMPIRE_LIB = -L$(UMPIRE_DIR)/lib -lumpire
 
-# MKL CPardiso library configuration
-MKL_CPARDISO_DIR ?=
-MKL_MPI_WRAPPER ?= mkl_blacs_mpich_lp64
-MKL_LIBRARY_SUBDIR ?= lib
-MKL_CPARDISO_OPT = -I$(MKL_CPARDISO_DIR)/include
-MKL_CPARDISO_LIB = -Wl,-rpath,$(MKL_CPARDISO_DIR)/$(MKL_LIBRARY_SUBDIR)\
-                   -L$(MKL_CPARDISO_DIR)/$(MKL_LIBRARY_SUBDIR) -l$(MKL_MPI_WRAPPER)\
-                   -lmkl_intel_lp64 -lmkl_sequential -lmkl_core
+# CAMP library configuration
+CAMP_DIR = @MFEM_DIR@/../camp
+CAMP_OPT = -I$(CAMP_DIR)/include
+CAMP_LIB = -L$(CAMP_DIR)/lib
 
 # If YES, enable some informational messages
 VERBOSE = NO

config/get_mumps_version.cpp

@@ -1,33 +0,0 @@
-// Copyright (c) 2010-2020, 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 "dmumps_c.h"
-#include <string>
-#include <iostream>
-#include <algorithm>
-
-// Macros to expand a macro as a string
-#define STR_EXPAND(s) #s
-#define STR(s) STR_EXPAND(s)
-
-int main()
-{
-#ifdef MUMPS_VERSION
-   const char *ptr = STR(MUMPS_VERSION);
-   std::string s(ptr);
-   s.erase(std::remove(s.begin(), s.end(), '"'), s.end());
-   s.erase(std::remove(s.begin(), s.end(), '.'), s.end());
-   std::cout << s << "\n";
-   return 0;
-#else
-   return -1;
-#endif
-}

config/makefile

@@ -42,10 +42,6 @@ GHV_FLAGS = $(subst @MFEM_DIR@,$(if $(MFEM_DIR),$(MFEM_DIR),..),$(HYPRE_OPT))
 SMX = $(if $(MFEM_USE_PUMI:NO=),MFEM_USE_SIMMETRIX)
 SMX_PATH = $(PUMI_DIR)/include/gmi_sim.h
 SMX_FILE = $(subst @MFEM_DIR@,$(if $(MFEM_DIR),$(MFEM_DIR),..),$(SMX_PATH))
-MUMPS = $(MFEM_USE_MUMPS:NO=)
-GMV_CXX ?= $(MFEM_CXX)
-GMV = get_mumps_version
-GMV_FLAGS = $(subst @MFEM_DIR@,$(if $(MFEM_DIR),$(MFEM_DIR),..),$(MUMPS_OPT))
 
 $(GHV): $(SRC)$(GHV).cpp
 	$(call mfem-info, Determining HYPRE version ...)
@@ -54,13 +50,6 @@ $(GHV).out: $(GHV)
 	./$(GHV) > $(GHV).out
 .INTERMEDIATE: $(GHV) $(GHV).out
 
-$(GMV): $(SRC)$(GMV).cpp
-	$(call mfem-info, Determining MUMPS version ...)
-	$(GMV_CXX) ${GMV_FLAGS} $(SRC)$(GMV).cpp -o $(GMV)
-$(GMV).out: $(GMV)
-	./$(GMV) > $(GMV).out
-.INTERMEDIATE: $(GMV) $(GMV).out
-
 get-hypre-version: $(GHV).out
 	$(eval MFEM_HYPRE_VERSION:=$(shell cat $(GHV).out))
 	$(if $(MFEM_HYPRE_VERSION),$(eval export MFEM_HYPRE_VERSION)\
@@ -73,16 +62,10 @@ check-smx:
 	$(call mfem-info, MFEM_USE_SIMMETRIX = $(MFEM_USE_SIMMETRIX))
 	$(eval export MFEM_USE_SIMMETRIX)
 
-get-mumps-version: $(GMV).out
-	$(eval MFEM_MUMPS_VERSION:=$(shell cat $(GMV).out))
-	$(if $(MFEM_MUMPS_VERSION),$(eval export MFEM_MUMPS_VERSION)\
-	$(info MUMPS version: $(MFEM_MUMPS_VERSION)),\
-	$(error Unable to determine MUMPS version))
-
-header: $(if $(MPI),get-hypre-version,) $(if $(SMX),check-smx,) $(if $(MUMPS),get-mumps-version,)
+header: $(if $(MPI),get-hypre-version,) $(if $(SMX),check-smx)
 	$(call mfem-info, Writing $(CONFIG_HPP) ...)
 	@set -- && \
-	for def in $${MFEM_DEFINES} $(if $(MPI),MFEM_HYPRE_VERSION) $(SMX) $(if $(MUMPS),MFEM_MUMPS_VERSION); do \
+	for def in $${MFEM_DEFINES} $(if $(MPI),MFEM_HYPRE_VERSION) $(SMX); do \
 	   eval var=\$$$$def && \
 	   if [ "NO" != "$${var}" ]; then \
 	      set -- "$$@" -e "s|// \(#define $${def} \)|\1|" && \

config/sample-runs.sh

@@ -78,14 +78,6 @@ groups_parallel=(
    "miniapps/electromagnetics"
    "joule.cpp"'
 #   "{volta,tesla,joule}.cpp"' # todo: multiline sample runs
-'"convergence"
-   "Convergence tests:"
-   "tests/convergence"
-   "diffusion.cpp"'
-'"par-mesh-format"
-   "Parallel mesh tests:"
-   "tests/par-mesh-format"
-   "ex1p.cpp"'
 )
 # All groups serial + parallel runs mixed in the same group:
 groups_all=(
@@ -115,14 +107,6 @@ groups_all=(
    "miniapps/electromagnetics"
    "joule.cpp"'
 #   "{volta,tesla,joule}.cpp"' # todo: multiline sample runs
-'"convergence"
-   "Convergence tests:"
-   "tests/convergence"
-   "diffusion.cpp"'
-'"par-mesh-format"
-   "Parallel mesh tests:"
-   "tests/par-mesh-format"
-   "ex1p.cpp"'
 )
 make_all="all"
 base_timeformat=$'real: %3Rs  user: %3Us  sys: %3Ss  %%cpu: %P'
@@ -396,15 +380,10 @@ function timed_run()
 # This function is used to execute the sample runs
 function go()
 {
-   # Strip leading and trailing spaces from $1 and store the result in cmd_line
-   shopt -s extglob
-   local cmd_line="${1##+( )}"
-   cmd_line="${cmd_line%%+( )}"
-   shopt -u extglob
-   eval local cmd=(${cmd_line})
+   local cmd=("$@")
    local res=""
    echo $sep
-   echo "<${group}>" "${cmd_line}"
+   echo "<${group}>" "${cmd[@]}"
    echo $sep
    if [ "${timing}" == "yes" ]; then
       timed_run "${cmd[@]}"
@@ -416,15 +395,15 @@ function go()
    else
       res="${red}FAILED${none}"
    fi
-   printf "[${res}] <${group}> ${cmd_line}\n"
+   printf "[${res}] <${group}> ${cmd[*]}\n"
    if [ "${timing}" == "yes" ]; then
       printf "Run time: %s\n" "${timer}"
       timer=(${timer})
       timer="${timer[1]}"
-      printf -v line "[$res](%8s) ${cmd_line}" "$timer"
+      printf -v line "[$res](%8s) ${cmd[*]}" "$timer"
       summary=("${summary[@]}" "$line")
    else
-      summary=("${summary[@]}" "[${res}] ${cmd_line}")
+      summary=("${summary[@]}" "[${res}] ${cmd[*]}")
    fi
    echo $sep
 }
@@ -459,7 +438,7 @@ function go_group()
       fi
       for run in "${runs[@]}"; do
          if [ "${run}" == "" ]; then continue; fi
-         eval go \"\${run_prefix} \${run} \${run_suffix}\" $output
+         eval go \${run_prefix} \${run} \${run_suffix} $output
       done
    done
    ${make} clean-exec
@@ -525,7 +504,7 @@ function echo_run()
 {
    echo "   $@"
    { echo "   $@"; echo "$sep";
-     eval "$@"
+     "$@"
      echo "$sep"; } >> "$echo_log" 2>&1
 }
 
@@ -545,28 +524,6 @@ function build_all()
    echo_run ${make} config ${mfem_config} || exit 1
    echo_run ${make} ${make_j} || exit 1
    echo_run ${make} ${make_all} ${make_j} || exit 1
-   # Build groups in directories other than the directories built by 'make all':
-   for group_params in "${groups[@]}"; do
-      eval params=(${group_params})
-      group_dir="${params[2]}"
-      case "$group_dir" in
-         (examples*|miniapps*)
-            # Built by 'make all'
-            ;;
-         (*)
-            if [ "${mfem_dir}" != "${mfem_build_dir}" ]; then
-               echo_run mkdir -p "${group_dir}" || exit 1
-               echo_run cd "${group_dir}" || exit 1
-               echo_run cp -af "${mfem_dir}/${group_dir}/makefile" . || exit 1
-            else
-               echo_run cd "${group_dir}" || exit 1
-            fi
-            echo_run ${make} clean || exit 1
-            echo_run ${make} MFEM_DIR="${mfem_dir}" ${make_j} || exit 1
-            echo_run cd "${mfem_build_dir}" || exit 1
-            ;;
-      esac
-   done
 }
 
 # Function that runs all sample runs, given by the array variable "groups".

data/periodic-annulus-sector.geo

@@ -1,38 +1,13 @@
 SetFactory("OpenCASCADE");
 
-// Select periodic mesh by setting this to either 0 - standard, 1 - periodic
-periodic = 1;
-
-// Set the geometry order (1, 2, ..., 9)
-order = 3;
-
-// Set the element type (3 - triangles, 4 - quadrilaterals)
-type = 3;
-
-// Number of radial elements
-nrad = 2;
-
-// Number of azimuthal elements on inner arc
-nazm1 = 3;
-
-// Number of azimuthal elements on outer arc
-nazm2 = 5;
-
-// Note: Using type = 4 with nazm1 != nazm2 can lead to mixed meshes
-//       containing both triangles and quadrilaterals.
-
-// Inner and outer radii
 R1 = 1.0;
 R2 = 2.0;
 
-// Angular size of the sector
-Phi = Pi/3.0;
-
 Point(1) = {0.0, 0, 0, 1.0};
 Point(2) = {R1, 0, 0, 1.0};
 Point(3) = {R2, 0, 0, 1.0};
-Point(4) = {R1*Cos(Phi), R1*Sin(Phi), 0, 1.0};
-Point(5) = {R2*Cos(Phi), R2*Sin(Phi), 0, 1.0};
+Point(4) = {R1*Cos(Pi/3), R1*Sin(Pi/3), 0, 1.0};
+Point(5) = {R2*Cos(Pi/3), R2*Sin(Pi/3), 0, 1.0};
 Line(1) = {2, 3};
 Line(2) = {4, 5};
 Circle(3) = {2, 1, 4};
@@ -40,23 +15,13 @@ Circle(4) = {3, 1, 5};
 Curve Loop(5) = {1, 4, -2, -3};
 Plane Surface(1) = {5};
 
-Transfinite Curve{1} = nrad+1;
-Transfinite Curve{2} = nrad+1;
-Transfinite Curve{3} = nazm1+1;
-Transfinite Curve{4} = nazm2+1;
-
-If (nazm1 == nazm2)
-   Transfinite Surface{1};
-EndIf
-
-If (type == 4)
-   Recombine Surface {1};
-EndIf
+Transfinite Curve{1} = 7;
+Transfinite Curve{2} = 7;
+Transfinite Curve{3} = 4;
+Transfinite Curve{4} = 10;
 
 // Set a rotation periodicity constraint:
-If (periodic)
-   Periodic Line{1} = {2} Rotate{{0,0,1}, {0,0,0}, -Phi};
-EndIf
+Periodic Line{1} = {2} Rotate{{0,0,1}, {0,0,0}, -Pi/3};
 
 // Tag surfaces and volumes with positive integers
 Physical Curve(1) = {3};
@@ -65,22 +30,8 @@ Physical Curve(3) = {1};
 Physical Curve(4) = {2};
 Physical Surface(1) = {1};
 
-// Optimize the high-order mesh
-// See https://gmsh.info/doc/texinfo/gmsh.html#index-Mesh_002eHighOrderOptimize
-// Mesh.ElementOrder = order;
-// Mesh.HighOrderOptimize = 1;
-
 // Generate 2D mesh
 Mesh 2;
-SetOrder order;
 Mesh.MshFileVersion = 2.2;
 
-// Check the element quality (the Plugin may be called AnalyseCurvedMesh)
-// Plugin(AnalyseMeshQuality).JacobianDeterminant = 1;
-// Plugin(AnalyseMeshQuality).Run;
-
-If (periodic)
-   Save Sprintf("periodic-annulus-sector-t%01g-o%01g.msh", type, order);
-Else
-   Save Sprintf("annulus-sector-t%01g-o%01g.msh", type, order);
-EndIf
+Save "periodic-annulus-sector.msh";

data/periodic-annulus-sector.msh

@@ -2,191 +2,184 @@ $MeshFormat
 2.2 0 8
 $EndMeshFormat
 $Nodes
-136
+55
 1 1 0 0
 2 2 0 0
 3 0.5000000000000001 0.8660254037844386 0
 4 1 1.732050807568877 0
-5 1.5 0 0
-6 1.166666666666667 0 0
-7 1.333333333333333 0 0
+5 1.166666666666667 0 0
+6 1.333333333333333 0 0
+7 1.5 0 0
 8 1.666666666666667 0 0
 9 1.833333333333333 0 0
-10 0.7500000000000002 1.299038105676658 0
-11 0.5833333333333335 1.010362971081845 0
-12 0.6666666666666667 1.154700538379251 0
+10 0.5833333333333335 1.010362971081845 0
+11 0.6666666666666667 1.154700538379251 0
+12 0.7500000000000002 1.299038105676658 0
 13 0.8333333333333335 1.443375672974064 0
 14 0.9166666666666669 1.587713240271471 0
 15 0.9396926207859085 0.3420201433256683 0
 16 0.7660444431189786 0.6427876096865386 0
-17 0.993238357741943 0.1160929141252301 0
-18 0.9730448705798238 0.2306158707424401 0
-19 0.8936326403234125 0.4487991802004617 0
-20 0.8354878114129367 0.5495089780708056 0
-21 0.6862416378687343 0.7273736415730481 0
-22 0.597158591702787 0.8021231927550432 0
-23 1.956295201467611 0.4158233816355181 0
-24 1.827090915285202 0.8134732861515996 0
-25 1.618033988749896 1.175570504584944 0
-26 1.338261212717719 1.486289650954786 0
-27 1.995128100519648 0.1395129474882505 0
-28 1.980536137483141 0.278346201920131 0
-29 1.922523391876638 0.551274711633998 0
-30 1.879385241571817 0.6840402866513373 0
-31 1.765895185717855 0.9389431255717802 0
-32 1.696096192312853 1.059838528466408 0
-33 1.532088886237958 1.285575219373077 0
-34 1.438679600677305 1.389316740917992 0
-35 1.231322950651319 1.576021507213442 0
-36 1.118385806941496 1.658075145110082 0
-37 1.162276263405681 0.6710405135499813 0
-38 1.248615852873337 1.079531485311822 0
-39 1.559209616901855 0.5415673055003691 0
-40 1.478306597054007 0.8535007117539289 0
-41 0.9210953433941653 0.9653302893212266 0
-42 1.296548225291847 0.3150268220262836 0
-43 1.055002035226811 1.358510675893086 0
-44 1.704005774249187 0.2344032256041583 0
-45 0.6403651144647218 0.8991270322967013 0
-46 0.7807302289294435 0.9322286608089638 0
-47 0.864063562262777 1.07656622810637 0
-48 0.8070317811313885 1.187802166891514 0
-49 0.7236984477980553 1.043464599594108 0
-50 1.432182741763949 0.1050089406754279 0
-51 1.364365483527898 0.2100178813508558 0
-52 1.197698816861231 0.2100178813508558 0
-53 1.098849408430616 0.1050089406754279 0
-54 1.265516075097282 0.1050089406754278 0
-55 0.8177280765440409 0.7503018362314346 0
-56 0.869411709969103 0.8578160627763305 0
-57 0.7348318576552288 0.8316090412164392 0
-58 1.177596357123201 0.3240245957927452 0
-59 1.058644488954555 0.3330223695592067 0
-60 1.087610484537871 0.2205785356313179 0
-61 1.267619707955123 0.7318605796179638 0
-62 1.372963152504565 0.7926806456859463 0
-63 1.40174301566045 0.9288443029398934 0
-64 1.325179434266893 1.004187894125858 0
-65 1.219835989717452 0.9433678280578752 0
-66 1.191056126561566 0.8072041708039283 0
-67 1.296399571111008 0.8680242368719107 0
-68 1.532241943619239 0.645545107584889 0
-69 1.505274270336623 0.7495229096694089 0
-70 1.294587381237739 0.6278827775334439 0
-71 1.426898499069797 0.5847250415169065 0
-72 1.399930825787181 0.6887028436014264 0
-73 1.139442349713613 1.041464419981624 0
-74 1.030268846553889 1.003397354651425 0
-75 1.001488983398004 0.8672336973974781 0
-76 1.081882623401843 0.7691371054737297 0
-77 1.110662486557728 0.9053007627276766 0
-78 1.207033584034403 0.5523692830420821 0
-79 1.251790904663125 0.4336980525341829 0
-80 1.384102022495183 0.3905403165176455 0
-81 1.471655819698519 0.4660538110090073 0
-82 1.339344701866461 0.5092115470255447 0
-83 1.73779714915742 0.7228379592678562 0
-84 1.648503383029637 0.6322026323841127 0
-85 1.691571478423774 0.4996526642120855 0
-86 1.823933339945692 0.4577380229238018 0
-87 1.787039416783436 0.5922941012619014 0
-88 1.308379426102925 1.350703595740465 0
-89 1.278497639488131 1.215117540526144 0
-90 1.371755231498856 1.111544491736196 0
-91 1.494894610124376 1.14355749816057 0
-92 1.4064614465962 1.25147448186763 0
-93 1.013887168325833 0.451693600067106 0
-94 1.088081715865757 0.5613670568085436 0
-95 1.03019898997678 0.6616228789288338 0
-96 0.8981217165478794 0.6522052443076862 0
-97 0.9637989050473432 0.5564495572737495 0
-98 1.432367408277627 0.2881522898855752 0
-99 1.568186591263407 0.2612777577448668 0
-100 1.655740388466743 0.3367912522362286 0
-101 1.607475002684299 0.4391792788682989 0
-102 1.519921205480963 0.3636657843769371 0
-103 1.184077913657828 1.17252454883891 0
-104 1.119539974442319 1.265517612365998 0
-105 1.010366471282595 1.2274505470358 0
-106 0.9657309073383804 1.096390418178513 0
-107 1.074904410498104 1.134457483508712 0
-108 1.901335258083062 0.07813440853471942 0
-109 1.802670516166124 0.1562688170694388 0
-110 1.636003849499458 0.156268817069439 0
-111 1.568001924749729 0.07813440853471942 0
-112 1.734668591416396 0.07813440853471944 0
-113 0.8516673450756037 1.318862295748801 0
-114 0.9533346901512071 1.338686485820944 0
-115 1.03666802348454 1.48302405311835 0
-116 1.01833401174227 1.607537430343614 0
-117 0.9350006784089369 1.463199863046207 0
-118 1.710829475874804 0.8268157613523761 0
-119 1.594568036464405 0.8401582365531526 0
-120 1.621535709747021 0.7361804344686326 0
-121 1.52488239428597 0.9608573093642675 0
-122 1.571458191517933 1.068213906974606 0
-123 1.448318812892413 1.036200900550232 0
-124 0.908699126206992 1.207626356963657 0
-125 1.500184666513678 0.1831433492101474 0
-126 1.646765991694905 0.9507607885973723 0
-127 0.9498053499729417 0.7597194708525821 0
-128 1.132839036494479 0.4426958263006443 0
-129 1.149421761057113 1.40110366758032 0
-130 1.243841486887416 1.443696659267553 0
-131 1.134903597606542 1.530869109405537 0
-132 1.872198725728137 0.3553499962917315 0
-133 1.788102249988661 0.2948766109479449 0
-134 1.893223337417325 0.2174207916708467 0
-135 1.213959700272622 1.308110604053232 0
-136 1.739836864206217 0.3972646375800152 0
+17 1.986476715483886 0.2321858282504602 0
+18 1.946089741159648 0.4612317414848793 0
+19 1.879385241571817 0.6840402866513365 0
+20 1.787265280646825 0.8975983604009234 0
+21 1.670975622825874 1.09901795614161 0
+22 1.532088886237958 1.285575219373077 0
+23 1.372483275737469 1.454747283146095 0
+24 1.194317183405575 1.604246385510085 0
+25 1.425989114816062 0.1915326920916892 0
+26 0.8788667344146573 1.13917645290495 0
+27 1.630372059110754 0.7154531062316609 0
+28 1.436395769298814 1.053728612482506 0
+29 1.081023776188756 0.6241293681829633 0
+30 1.168737372335971 1.428012728596308 0
+31 1.821063986059922 0.298149890497067 0
+32 1.234707097211386 0.3469796339295647 0
+33 1.377747393186519 0.6200150626754309 0
+34 1.457047681210906 0.3890895843559762 0
+35 0.917846726184522 0.8957978954532204 0
+36 1.218335619030348 0.9017812086952638 0
+37 1.066623110765233 1.061857005744772 0
+38 1.587029716281926 0.1355955181472859 0
+39 1.744445799211916 0.1441515753740107 0
+40 1.25 0.1443375672974065 0
+41 1.453660070628011 0.8435769396609902 0
+42 1.741367044061892 0.499612708014486 0
+43 1.30550638526547 1.257610469847477 0
+44 1.118213276932792 0.1666674689105279 0
+45 0.9109440214958271 1.306610291787315 0
+46 0.9970618258753989 1.438658589955562 0
+47 0.7499999999999998 1.010362971081845 0
+48 0.7034449005273667 0.8850673702175776 0
+49 1.605449512513618 0.9269067082200894 0
+50 1.561654019115059 0.5298592532912715 0
+51 1.229782222487711 1.096820457143683 0
+52 1.617066998712459 0.3090202662210922 0
+53 1.079645953234324 1.246963713711438 0
+54 1.877063966817811 0.1348974588243076 0
+55 1.055356609656722 1.558136350380461 0
 $EndNodes
 $Elements
-38
-1 26 2 3 1 1 5 6 7
-2 26 2 3 1 5 2 8 9
-3 26 2 4 2 3 10 11 12
-4 26 2 4 2 10 4 13 14
-5 26 2 1 3 1 15 17 18
-6 26 2 1 3 15 16 19 20
-7 26 2 1 3 16 3 21 22
-8 26 2 2 4 2 23 27 28
-9 26 2 2 4 23 24 29 30
-10 26 2 2 4 24 25 31 32
-11 26 2 2 4 25 26 33 34
-12 26 2 2 4 26 4 35 36
-13 21 2 1 1 3 41 10 45 46 47 48 12 11 49
-14 21 2 1 1 5 42 1 50 51 52 53 6 7 54
-15 21 2 1 1 16 41 3 55 56 46 45 22 21 57
-16 21 2 1 1 1 42 15 53 52 58 59 18 17 60
-17 21 2 1 1 37 40 38 61 62 63 64 65 66 67
-18 21 2 1 1 39 40 37 68 69 62 61 70 71 72
-19 21 2 1 1 38 41 37 73 74 75 76 66 65 77
-20 21 2 1 1 37 42 39 78 79 80 81 71 70 82
-21 21 2 1 1 24 39 23 83 84 85 86 29 30 87
-22 21 2 1 1 26 38 25 88 89 90 91 33 34 92
-23 21 2 1 1 15 37 16 93 94 95 96 20 19 97
-24 21 2 1 1 42 44 39 98 99 100 101 81 80 102
-25 21 2 1 1 38 43 41 103 104 105 106 74 73 107
-26 21 2 1 1 2 44 5 108 109 110 111 8 9 112
-27 21 2 1 1 10 43 4 113 114 115 116 14 13 117
-28 21 2 1 1 24 40 39 118 119 69 68 84 83 120
-29 21 2 1 1 38 40 25 64 63 121 122 91 90 123
-30 21 2 1 1 41 43 10 106 105 114 113 48 47 124
-31 21 2 1 1 5 44 42 111 110 99 98 51 50 125
-32 21 2 1 1 25 40 24 122 121 119 118 31 32 126
-33 21 2 1 1 37 41 16 76 75 56 55 96 95 127
-34 21 2 1 1 15 42 37 59 58 79 78 94 93 128
-35 21 2 1 1 4 43 26 116 115 129 130 35 36 131
-36 21 2 1 1 23 44 2 132 133 109 108 27 28 134
-37 21 2 1 1 26 43 38 130 129 104 103 89 88 135
-38 21 2 1 1 39 44 23 101 100 133 132 86 85 136
+108
+1 1 2 3 1 1 5
+2 1 2 3 1 5 6
+3 1 2 3 1 6 7
+4 1 2 3 1 7 8
+5 1 2 3 1 8 9
+6 1 2 3 1 9 2
+7 1 2 4 2 3 10
+8 1 2 4 2 10 11
+9 1 2 4 2 11 12
+10 1 2 4 2 12 13
+11 1 2 4 2 13 14
+12 1 2 4 2 14 4
+13 1 2 1 3 1 15
+14 1 2 1 3 15 16
+15 1 2 1 3 16 3
+16 1 2 2 4 2 17
+17 1 2 2 4 17 18
+18 1 2 2 4 18 19
+19 1 2 2 4 19 20
+20 1 2 2 4 20 21
+21 1 2 2 4 21 22
+22 1 2 2 4 22 23
+23 1 2 2 4 23 24
+24 1 2 2 4 24 4
+25 2 2 1 1 32 40 25
+26 2 2 1 1 25 34 32
+27 2 2 1 1 33 41 36
+28 2 2 1 1 38 52 25
+29 2 2 1 1 33 36 29
+30 2 2 1 1 26 47 35
+31 2 2 1 1 35 37 26
+32 2 2 1 1 25 52 34
+33 2 2 1 1 32 44 40
+34 2 2 1 1 15 32 29
+35 2 2 1 1 15 29 16
+36 2 2 1 1 36 41 28
+37 2 2 1 1 32 33 29
+38 2 2 1 1 50 52 42
+39 2 2 1 1 32 34 33
+40 2 2 1 1 42 52 31
+41 2 2 1 1 43 53 51
+42 2 2 1 1 27 41 33
+43 2 2 1 1 26 53 45
+44 2 2 1 1 18 31 17
+45 2 2 1 1 29 35 16
+46 2 2 1 1 29 36 35
+47 2 2 1 1 24 30 23
+48 2 2 1 1 30 53 43
+49 2 2 1 1 17 54 2
+50 2 2 1 1 4 55 24
+51 2 2 1 1 28 51 36
+52 2 2 1 1 47 48 35
+53 2 2 1 1 36 37 35
+54 2 2 1 1 37 53 26
+55 2 2 1 1 22 28 21
+56 2 2 1 1 20 27 19
+57 2 2 1 1 33 50 27
+58 2 2 1 1 15 44 32
+59 2 2 1 1 18 42 31
+60 2 2 1 1 30 43 23
+61 2 2 1 1 35 48 16
+62 2 2 1 1 31 54 17
+63 2 2 1 1 9 39 8
+64 2 2 1 1 8 38 7
+65 2 2 1 1 7 25 6
+66 2 2 1 1 22 43 28
+67 2 2 1 1 23 43 22
+68 2 2 1 1 39 54 31
+69 2 2 1 1 19 42 18
+70 2 2 1 1 24 55 30
+71 2 2 1 1 27 42 19
+72 2 2 1 1 13 46 14
+73 2 2 1 1 51 53 37
+74 2 2 1 1 39 52 38
+75 2 2 1 1 6 40 5
+76 2 2 1 1 34 52 50
+77 2 2 1 1 12 45 13
+78 2 2 1 1 30 55 46
+79 2 2 1 1 10 47 11
+80 2 2 1 1 8 39 38
+81 2 2 1 1 28 49 21
+82 2 2 1 1 7 38 25
+83 2 2 1 1 41 49 28
+84 2 2 1 1 20 49 27
+85 2 2 1 1 11 26 12
+86 2 2 1 1 27 49 41
+87 2 2 1 1 31 52 39
+88 2 2 1 1 25 40 6
+89 2 2 1 1 2 54 9
+90 2 2 1 1 14 55 4
+91 2 2 1 1 45 53 46
+92 2 2 1 1 45 46 13
+93 2 2 1 1 5 44 1
+94 2 2 1 1 21 49 20
+95 2 2 1 1 46 53 30
+96 2 2 1 1 3 48 10
+97 2 2 1 1 34 50 33
+98 2 2 1 1 36 51 37
+99 2 2 1 1 26 45 12
+100 2 2 1 1 11 47 26
+101 2 2 1 1 27 50 42
+102 2 2 1 1 40 44 5
+103 2 2 1 1 43 51 28
+104 2 2 1 1 10 48 47
+105 2 2 1 1 9 54 39
+106 2 2 1 1 46 55 14
+107 2 2 1 1 1 44 15
+108 2 2 1 1 16 48 3
 $EndElements
 $Periodic
 1
 1 1 2
 Affine 0.5000000000000001 0.8660254037844386 0 0 -0.8660254037844386 0.5000000000000001 0 0 0 0 1 0 0 0 0 1
-3
+7
+9 14
+6 11
+8 13
 5 10
-1 3
+7 12
 2 4
+1 3
 $EndPeriodic

data/periodic-torus-sector.geo

@@ -1,141 +1,25 @@
-// Select periodic mesh by setting this to either 0 - standard, 1 - periodic
-periodic = 1;
+SetFactory("OpenCASCADE");
 
-// Set the geometry order (1, 2, ..., 10 for tetrahedra or 9 for other types)
-order = 3;
+R = 1.5;
+r = 0.5;
 
-// Set the element type (4 - tetrahedra, 6 - wedges, 8 - hexahedra)
-type = 8;
+Torus(1) = {0,0,0, R, r, Pi/3};
 
-// Minor and major radii
-R1 = 1.0;
-R2 = 2.0;
+pts() = PointsOf{ Volume{1}; };
 
-// Side length of interior square
-A1 = 0.8;
-
-// Angular size of the sector
-Phi = Pi/3.0;
-
-// Number of azimuthal elements
-nazm = 3;
-
-// Number of elements around a quarter of the circle
-narc = 2;
-
-// Number of elements between surface and interior square
-nshl = 1;
-
-lc = 0.5;
-a1 = A1 / Sqrt(2.0);
-
-Point(1) = {R2+R1, 0, 0, lc};
-Point(2) = {R2, 0, R1, lc};
-Point(3) = {R2-R1, 0, 0, lc};
-Point(4) = {R2, 0, -R1, lc};
-Point(5) = {R2, 0, 0, lc};
-Point(6) = {R2+a1, 0, 0, lc};
-Point(7) = {R2, 0, a1, lc};
-Point(8) = {R2-a1, 0, 0, lc};
-Point(9) = {R2, 0, -a1, lc};
-
-Circle(1) = {1,5,2};
-Circle(2) = {2,5,3};
-Circle(3) = {3,5,4};
-Circle(4) = {4,5,1};
-
-Line(5) = {6,1};
-Line(6) = {7,2};
-Line(7) = {8,3};
-Line(8) = {9,4};
-
-Line(9) = {6, 7};
-Line(10) = {7, 8};
-Line(11) = {8, 9};
-Line(12) = {9, 6};
-
-Line Loop(101) = {1, -6, -9, 5};
-Line Loop(102) = {2, -7, -10, 6};
-Line Loop(103) = {3, -8, -11, 7};
-Line Loop(104) = {4, -5, -12, 8};
-Line Loop(105) = {9, 10, 11, 12};
-
-Plane Surface(201) = {101};
-Plane Surface(202) = {102};
-Plane Surface(203) = {103};
-Plane Surface(204) = {104};
-Plane Surface(205) = {105};
-
-Transfinite Curve{1} = narc+1;
-Transfinite Curve{2} = narc+1;
-Transfinite Curve{3} = narc+1;
-Transfinite Curve{4} = narc+1;
-
-Transfinite Curve{5} = nshl+1;
-Transfinite Curve{6} = nshl+1;
-Transfinite Curve{7} = nshl+1;
-Transfinite Curve{8} = nshl+1;
-
-Transfinite Curve{9} = narc+1;
-Transfinite Curve{10} = narc+1;
-Transfinite Curve{11} = narc+1;
-Transfinite Curve{12} = narc+1;
-
-If (type == 8)
-   Recombine Surface {201};
-   Recombine Surface {202};
-   Recombine Surface {203};
-   Recombine Surface {204};
-   Recombine Surface {205};
-
-   Transfinite Surface {201} = {1,2,7,6};
-   Transfinite Surface {202} = {2,3,8,7};
-   Transfinite Surface {203} = {3,4,9,8};
-   Transfinite Surface {204} = {4,1,6,9};
-   Transfinite Surface {205} = {6,7,8,9};
-EndIf
-
-If (type == 4)
-   Extrude { {0,0,1} , {0,0,0} , Phi} {
-      Surface{201,202,203,204,205}; Layers{nazm};
-}
-Else
-   Extrude { {0,0,1} , {0,0,0} , Phi} {
-      Surface{201,202,203,204,205}; Layers{nazm}; Recombine;
-}
-EndIf
+Characteristic Length{ pts() } = 0.25;
 
 // Set a rotation periodicity constraint:
-If (periodic)
-   Periodic Surface{227} = {201} Rotate{{0,0,1}, {0,0,0}, Phi};
-   Periodic Surface{249} = {202} Rotate{{0,0,1}, {0,0,0}, Phi};
-   Periodic Surface{271} = {203} Rotate{{0,0,1}, {0,0,0}, Phi};
-   Periodic Surface{293} = {204} Rotate{{0,0,1}, {0,0,0}, Phi};
-   Periodic Surface{315} = {205} Rotate{{0,0,1}, {0,0,0}, Phi};
-EndIf
+Periodic Surface{3} = {2} Rotate{{0,0,1}, {0,0,0}, Pi/3};
 
 // Tag surfaces and volumes with positive integers
-Physical Surface(1) = {201,202,203,204,205};
-Physical Surface(2) = {227,249,271,293,315};
-Physical Surface(3) = {214,236,258,280};
-Physical Volume(1) = {1,2,3,4,5};
-
-// Optimize the high-order mesh
-// See https://gmsh.info/doc/texinfo/gmsh.html#index-Mesh_002eHighOrderOptimize
-// Mesh.ElementOrder = order;
-// Mesh.HighOrderOptimize = 1;
+Physical Surface(1) = {1};
+Physical Surface(2) = {2};
+Physical Surface(3) = {3};
+Physical Volume(1) = {1};
 
 // Generate 3D mesh
 Mesh 3;
-SetOrder order;
+
 Mesh.MshFileVersion = 2.2;
-
-// Check the element quality (the Plugin may be called AnalyseCurvedMesh)
-// Plugin(AnalyseMeshQuality).JacobianDeterminant = 1;
-// Plugin(AnalyseMeshQuality).Run;
-
-If (periodic)
-   Save Sprintf("periodic-torus-sector-t%01g-o%01g.msh", type, order);
-Else
-   Save Sprintf("torus-sector-t%01g-o%01g.msh", type, order);
-EndIf
+Save "periodic-torus-sector.msh";

data/square-mixed.mesh

@@ -1,118 +0,0 @@
-MFEM mesh v1.0
-
-#
-# MFEM Geometry Types (see mesh/geom.hpp):
-#
-# POINT       = 0
-# SEGMENT     = 1
-# TRIANGLE    = 2
-# SQUARE      = 3
-# TETRAHEDRON = 4
-# CUBE        = 5
-# PRISM       = 6
-#
-
-dimension
-2
-
-elements
-20
-1 3 0 1 6 5
-1 3 1 2 7 6
-1 3 2 3 8 7
-1 3 3 4 9 8
-1 3 5 6 11 10
-1 2 6 7 11
-1 2 7 12 11
-1 2 7 8 13
-1 2 7 13 12
-1 3 8 9 14 13
-1 3 10 11 16 15
-1 2 11 12 17
-1 2 11 17 16
-1 2 12 13 17
-1 2 13 18 17
-1 3 13 14 19 18
-1 3 15 16 21 20
-1 3 16 17 22 21
-1 3 17 18 23 22
-1 3 18 19 24 23
-
-boundary
-16
-2 1 0 1
-2 1 1 2
-2 1 2 3
-2 1 3 4
-2 1 21 20
-2 1 22 21
-2 1 23 22
-2 1 24 23
-1 1 5 0
-1 1 10 5
-1 1 15 10
-1 1 20 15
-1 1 4 9
-1 1 9 14
-1 1 14 19
-1 1 19 24
-
-vertices
-25
-
-nodes
-FiniteElementSpace
-FiniteElementCollection: H1_2D_P1
-VDim: 2
-Ordering: 0
-
-0
-0.25
-0.5
-0.75
-1
-0
-0.25
-0.5
-0.75
-1
-0
-0.25
-0.5
-0.75
-1
-0
-0.25
-0.5
-0.75
-1
-0
-0.25
-0.5
-0.75
-1
-0
-0
-0
-0
-0
-0.25
-0.25
-0.25
-0.25
-0.25
-0.5
-0.5
-0.5
-0.5
-0.5
-0.75
-0.75
-0.75
-0.75
-0.75
-1
-1
-1
-1
-1

doc/CodeDocumentation.dox

@@ -144,12 +144,12 @@ namespace mfem {
  * - <a class="el" href="maxwell_8cpp_source.html">Maxwell</a>: simple transient full-wave electromagnetics simulation code
  * - <a class="el" href="joule_8cpp_source.html">Joule</a>: transient magnetics and Joule heating miniapp
  * - <a class="el" href="classmfem_1_1navier_1_1NavierSolver.html">Navier</a>: solve the transient incompressible Navier-Stokes equations
- * - <a class="el" href="mobius-strip_8cpp_source.html">Mobius Strip</a>: generate various Mobius strip-like meshes
+
+* - <a class="el" href="mobius-strip_8cpp_source.html">Mobius Strip</a>: generate various Mobius strip-like meshes
  * - <a class="el" href="klein-bottle_8cpp_source.html">Klein Bottle</a>: generate three types of Klein bottle surfaces
  * - <a class="el" href="toroid_8cpp_source.html">Toroid</a>: generate simple toroidal meshes
  * - <a class="el" href="twist_8cpp_source.html">Twist</a>: generate simple periodic meshes
  * - <a class="el" href="minimal-surface_8cpp_source.html">Minimal Surface</a>: compute minimal surfaces, <a class="el" href="minimal-surface_8cpp_source.html">serial</a> and <a class="el" href="pminimal-surface_8cpp_source.html">parallel</a> versions
- * - <a class="el" href="polar-nc_8cpp_source.html">Polar NC</a>: generate polar non-conforming meshes
  * - <a class="el" href="shaper_8cpp_source.html">Shaper</a>: resolve material interfaces by mesh refinement
  * - <a class="el" href="extruder_8cpp_source.html">Extruder</a>: extrude a low-dimensional mesh into a higher dimension
  * - <a class="el" href="mesh-explorer_8cpp_source.html">Mesh Explorer</a>: visualize and manipulate meshes
@@ -162,7 +162,6 @@ namespace mfem {
  * - <a class="el" href="lor-transfer_8cpp_source.html">LOR Transfer</a>: map functions between high-order and low-order refined spaces
  * - <a class="el" href="findpts_8cpp_source.html">Find Points</a>: evaluate grid function in physical space, <a class="el" href="findpts_8cpp_source.html">serial</a> and <a class="el" href="pfindpts_8cpp_source.html">parallel</a> versions
  * - <a class="el" href="field-diff_8cpp_source.html">Field Diff</a>: compare grid functions on different meshes
- * - <a class="el" href="field-interp_8cpp_source.html">Field Interp</a>: transfer a grid functions betwen meshes
  * - <a class="el" href="miniapps_2performance_2ex1_8cpp_source.html">HPC Example 1</a>: high-performance nodal H1 FEM for the Laplace problem
  * - <a class="el" href="miniapps_2performance_2ex1p_8cpp_source.html">HPC Example 1p</a>: high-performance parallel nodal H1 FEM for the Laplace problem
  *

doc/makefile

@@ -19,7 +19,7 @@ html: $(DOXYGEN_CONF)
 	@# Generate the html documentation
 	@doxygen $(DOXYGEN_CONF)
 	@echo "<meta http-equiv=\"REFRESH\" content=\"0;URL=CodeDocumentation/html/index.html\">" > CodeDocumentation.html
-	@cat warnings.log 1>&2
+	@cat warnings.log
 	@# Generate the log of undocumented methods
 	@( cat $(DOXYGEN_CONF) ; echo "GENERATE_HTML=NO" ; echo "EXTRACT_ALL=NO" ; echo "WARN_LOGFILE=undoc.log" ; echo "QUIET=YES" ) | doxygen - &> /dev/null
 

examples/ex11p.cpp

@@ -72,7 +72,6 @@ int main(int argc, char *argv[])
    int seed = 75;
    bool slu_solver  = false;
    bool sp_solver = false;
-   bool pardiso_solver = false;
    bool visualization = 1;
 
    OptionsParser args(argc, argv);
@@ -96,14 +95,6 @@ int main(int argc, char *argv[])
 #ifdef MFEM_USE_STRUMPACK
    args.AddOption(&sp_solver, "-sp", "--strumpack", "-no-sp",
                   "--no-strumpack", "Use the STRUMPACK Solver.");
-#endif
-#ifdef MFEM_USE_MKL_CPARDISO
-   args.AddOption(&pardiso_solver,
-                  "-pardiso",
-                  "--pardiso",
-                  "-no-pardiso",
-                  "--no-pardiso",
-                  "Use the MKL Cluster Pardiso Solver.");
 #endif
    args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
                   "--no-visualization",
@@ -245,7 +236,7 @@ int main(int argc, char *argv[])
    //    preconditioner for A to be used within the solver. Set the matrices
    //    which define the generalized eigenproblem A x = lambda M x.
    Solver * precond = NULL;
-   if (!slu_solver && !sp_solver && !pardiso_solver)
+   if (!slu_solver && !sp_solver)
    {
       HypreBoomerAMG * amg = new HypreBoomerAMG(*A);
       amg->SetPrintLevel(0);
@@ -277,19 +268,10 @@ int main(int argc, char *argv[])
          strumpack->SetFromCommandLine();
          precond = strumpack;
       }
-#endif
-#ifdef MFEM_USE_MKL_CPARDISO
-      if (pardiso_solver)
-      {
-         auto pardiso = new CPardisoSolver(A->GetComm());
-         pardiso->SetMatrixType(CPardisoSolver::MatType::REAL_STRUCTURE_SYMMETRIC);
-         pardiso->SetPrintLevel(1);
-         pardiso->SetOperator(*A);
-         precond = pardiso;
-      }
 #endif
    }
 
+
    HypreLOBPCG * lobpcg = new HypreLOBPCG(MPI_COMM_WORLD);
    lobpcg->SetNumModes(nev);
    lobpcg->SetRandomSeed(seed);

examples/ex1p_mumps.cpp

@@ -1,234 +0,0 @@
-//                       MFEM Example 1 - Parallel Version
-//
-// Compile with: make ex1p
-//
-// Sample runs:  mpirun -np 4 ex1p -m ../data/square-disc.mesh
-//
-
-#include "mfem.hpp"
-#include <fstream>
-#include <iostream>
-
-using namespace std;
-using namespace mfem;
-
-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.
-   const char *mesh_file = "../data/inline-quad.mesh";
-   int order = 1;
-   bool static_cond = false;
-   bool visualization = true;
-   int sr = 1;
-   int pr = 1;
-
-   OptionsParser args(argc, argv);
-   args.AddOption(&mesh_file, "-m", "--mesh",
-                  "Mesh file to use.");
-   args.AddOption(&order, "-o", "--order",
-                  "Finite element order (polynomial degree) or -1 for"
-                  " isoparametric space.");
-   args.AddOption(&sr, "-sr", "--serial_ref",
-                  "Number of serial refinements");        
-   args.AddOption(&pr, "-pr", "--parallel_ref",
-                  "Number of parallel refinements");                               
-   args.AddOption(&static_cond, "-sc", "--static-condensation", "-no-sc",
-                  "--no-static-condensation", "Enable static condensation.");
-   args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
-                  "--no-visualization",
-                  "Enable or disable GLVis visualization.");
-   args.Parse();
-   if (!args.Good())
-   {
-      if (myid == 0)
-      {
-         args.PrintUsage(cout);
-      }
-      MPI_Finalize();
-      return 1;
-   }
-   if (myid == 0)
-   {
-      args.PrintOptions(cout);
-   }
-
-   // 4. Read the (serial) mesh from the given mesh file on all processors.  We
-   //    can handle triangular, quadrilateral, tetrahedral, hexahedral, surface
-   //    and volume meshes with the same code.
-   Mesh mesh(mesh_file, 1, 1);
-   int dim = mesh.Dimension();
-
-   // 5. Refine the serial mesh on all processors to increase the resolution. In
-   //    this example we do 'ref_levels' of uniform refinement. We choose
-   //    'ref_levels' to be the largest number that gives a final mesh with no
-   //    more than 10,000 elements.
-   {
-      for (int l = 0; l < sr; l++)
-      {
-         mesh.UniformRefinement();
-      }
-   }
-
-   // 6. Define a parallel mesh by a partitioning of the serial mesh. Refine
-   //    this mesh further in parallel to increase the resolution. Once the
-   //    parallel mesh is defined, the serial mesh can be deleted.
-   ParMesh pmesh(MPI_COMM_WORLD, mesh);
-   mesh.Clear();
-   {
-      for (int l = 0; l < pr; l++)
-      {
-         pmesh.UniformRefinement();
-      }
-   }
-
-   // 7. Define a parallel finite element space on the parallel mesh. Here we
-   //    use continuous Lagrange finite elements of the specified order. If
-   //    order < 1, we instead use an isoparametric/isogeometric space.
-   FiniteElementCollection *fec = new H1_FECollection(order, dim);
-   ParFiniteElementSpace fespace(&pmesh, fec);
-   HYPRE_Int size = fespace.GlobalTrueVSize();
-   if (myid == 0)
-   {
-      cout << "Number of finite element unknowns: " << size << endl;
-   }
-
-   // 8. Determine the list of true (i.e. parallel conforming) essential
-   //    boundary dofs. In this example, the boundary conditions are defined
-   //    by marking all the boundary attributes from the mesh as essential
-   //    (Dirichlet) and converting them to a list of true dofs.
-   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);
-   }
-
-   // 9. Set up the parallel linear form b(.) which corresponds to the
-   //    right-hand side of the FEM linear system, which in this case is
-   //    (1,phi_i) where phi_i are the basis functions in fespace.
-   ParLinearForm b(&fespace);
-   ConstantCoefficient one(1.0);
-   b.AddDomainIntegrator(new DomainLFIntegrator(one));
-   b.Assemble();
-
-   // 10. Define the solution vector x as a parallel finite element grid function
-   //     corresponding to fespace. Initialize x with initial guess of zero,
-   //     which satisfies the boundary conditions.
-   ParGridFunction x(&fespace);
-   x = 0.0;
-
-   // 11. Set up the parallel bilinear form a(.,.) on the finite element space
-   //     corresponding to the Laplacian operator -Delta, by adding the Diffusion
-   //     domain integrator.
-   ParBilinearForm a(&fespace);
-   a.AddDomainIntegrator(new DiffusionIntegrator(one));
-
-   if (static_cond) { a.EnableStaticCondensation(); }
-   a.Assemble();
-
-   HypreParMatrix A;
-   Vector B, X;
-   a.FormLinearSystem(ess_tdof_list, x, b, A, X, B);
-
-   // // 13. Solve the linear system A X = B.
-   // //     * With full assembly, use the BoomerAMG preconditioner from hypre.
-   // //     * With partial assembly, use Jacobi smoothing, for now.
-   StopWatch chrono;
-   chrono.Clear();
-   chrono.Start();
-   HypreBoomerAMG *prec = new HypreBoomerAMG;
-   prec->SetPrintLevel(0);
-   CGSolver cg(MPI_COMM_WORLD);
-   cg.SetRelTol(1e-13);
-   cg.SetMaxIter(2000);
-   cg.SetPrintLevel(0);
-   if (prec) { cg.SetPreconditioner(*prec); }
-   cg.SetOperator(A);
-   cg.Mult(B, X);
-   delete prec;
-
-   if (myid == 0)
-   {
-      cout << "PCG-AMG time: " << chrono.RealTime() << endl;
-   }
-
-   chrono.Clear();
-   chrono.Start();
-   {
-      MUMPSSolver MA;
-      MA.SetMatrixSymType(0);
-      MA.SetOperator(A);
-      Vector Y(X.Size());
-      MA.Mult(B,Y);
-      Y-=X;
-      cout << "Mumps Diff norm = " << Y.Norml2() << endl;
-   }
-   if (myid == 0)
-   {
-      cout << "mumps time: " << chrono.RealTime() << endl;
-   }
-
-   chrono.Clear();
-   chrono.Start();
-
-   {
-      CPardisoSolver pardiso(A.GetComm());
-      // pardiso.SetMatrixType(CPardisoSolver::MatType::REAL_STRUCTURE_SYMMETRIC);
-      pardiso.SetMatrixType(CPardisoSolver::MatType::REAL_UNSYMMETRIC);
-      pardiso.SetPrintLevel(0);
-      pardiso.SetOperator(A);
-      Vector Y(X.Size());
-      pardiso.Mult(B, Y);
-      Y-=X;
-      cout << "Pardiso Diff norm = " << Y.Norml2() << endl;
-   }
-
-   if (myid == 0)
-   {
-      cout << "pardiso time: " << chrono.RealTime() << endl;
-   }
-
-   {
-      SuperLURowLocMatrix SA(A);
-      SuperLUSolver superlu(MPI_COMM_WORLD);
-      superlu.SetPrintStatistics(false);
-      superlu.SetSymmetricPattern(false);
-      superlu.SetColumnPermutation(superlu::PARMETIS);
-      superlu.SetOperator(SA);
-      Vector Y(X.Size());
-      superlu.Mult(B, Y);
-      Y-=X;
-      cout << "Superlu Diff norm = " << Y.Norml2() << endl;
-   }
-
-   if (myid == 0)
-   {
-      cout << "superlu time: " << chrono.RealTime() << endl;
-   }
-
-   a.RecoverFEMSolution(X, b, x);
-
-   // 16. Send the solution by socket to a GLVis server.
-   if (visualization)
-   {
-      char vishost[] = "localhost";
-      int  visport   = 19916;
-      socketstream sol_sock(vishost, visport);
-      sol_sock << "parallel " << num_procs << " " << myid << "\n";
-      sol_sock.precision(8);
-      sol_sock << "solution\n" << pmesh << x << flush;
-   }
-
-   // 17. Free the used memory.
-   delete fec;
-   MPI_Finalize();
-
-   return 0;
-}

examples/ex22.cpp

@@ -6,19 +6,17 @@
 //               ex22 -m ../data/inline-tri.mesh -o 3
 //               ex22 -m ../data/inline-quad.mesh -o 3
 //               ex22 -m ../data/inline-quad.mesh -o 3 -p 1
-//               ex22 -m ../data/inline-quad.mesh -o 3 -p 1 -pa
 //               ex22 -m ../data/inline-quad.mesh -o 3 -p 2
 //               ex22 -m ../data/inline-tet.mesh -o 2
 //               ex22 -m ../data/inline-hex.mesh -o 2
 //               ex22 -m ../data/inline-hex.mesh -o 2 -p 1
 //               ex22 -m ../data/inline-hex.mesh -o 2 -p 2
-//               ex22 -m ../data/inline-hex.mesh -o 2 -p 2 -pa
 //               ex22 -m ../data/star.mesh -r 1 -o 2 -sigma 10.0
 //
-// Device sample runs:
-//               ex22 -m ../data/inline-quad.mesh -o 3 -p 1 -pa -d cuda
-//               ex22 -m ../data/inline-hex.mesh -o 2 -p 2 -pa -d cuda
-//               ex22 -m ../data/star.mesh -r 1 -o 2 -sigma 10.0 -pa -d cuda
+// With partial assembly:
+//               ex22 -m ../data/inline-quad.mesh -o 3 -p 1 -pa
+//               ex22 -m ../data/inline-hex.mesh -o 2 -p 2 -pa
+//               ex22 -m ../data/star.mesh -r 1 -o 2 -sigma 10.0 -pa
 //
 // Description:  This example code demonstrates the use of MFEM to define and
 //               solve simple complex-valued linear systems. It implements three
@@ -84,7 +82,6 @@ int main(int argc, char *argv[])
    bool herm_conv = true;
    bool exact_sol = true;
    bool pa = false;
-   const char *device_config = "cpu";
 
    OptionsParser args(argc, argv);
    args.AddOption(&mesh_file, "-m", "--mesh",
@@ -117,8 +114,6 @@ int main(int argc, char *argv[])
                   "Enable or disable GLVis visualization.");
    args.AddOption(&pa, "-pa", "--partial-assembly", "-no-pa",
                   "--no-partial-assembly", "Enable Partial Assembly.");
-   args.AddOption(&device_config, "-d", "--device",
-                  "Device configuration string, see Device::Configure().");
    args.Parse();
    if (!args.Good())
    {
@@ -148,18 +143,13 @@ int main(int argc, char *argv[])
    ComplexOperator::Convention conv =
       herm_conv ? ComplexOperator::HERMITIAN : ComplexOperator::BLOCK_SYMMETRIC;
 
-   // 2. Enable hardware devices such as GPUs, and programming models such as
-   //    CUDA, OCCA, RAJA and OpenMP based on command line options.
-   Device device(device_config);
-   device.Print();
-
-   // 3. Read the mesh from the given mesh file. We can handle triangular,
+   // 2. Read the mesh from the given mesh file. We can handle triangular,
    //    quadrilateral, tetrahedral, hexahedral, surface and volume meshes
    //    with the same code.
    Mesh *mesh = new Mesh(mesh_file, 1, 1);
    int dim = mesh->Dimension();
 
-   // 4. Refine the mesh to increase resolution. In this example we do
+   // 3. Refine the mesh to increase resolution. In this example we do
    //    'ref_levels' of uniform refinement where the user specifies
    //    the number of levels with the '-r' option.
    for (int l = 0; l < ref_levels; l++)
@@ -167,7 +157,7 @@ int main(int argc, char *argv[])
       mesh->UniformRefinement();
    }
 
-   // 5. Define a finite element space on the mesh. Here we use continuous
+   // 4. Define a finite element space on the mesh. Here we use continuous
    //    Lagrange, Nedelec, or Raviart-Thomas finite elements of the specified
    //    order.
    if (dim == 1 && prob != 0 )
@@ -189,7 +179,7 @@ int main(int argc, char *argv[])
    cout << "Number of finite element unknowns: " << fespace->GetTrueVSize()
         << endl;
 
-   // 6. Determine the list of true (i.e. conforming) essential boundary dofs.
+   // 5. Determine the list of true (i.e. conforming) essential boundary dofs.
    //    In this example, the boundary conditions are defined based on the type
    //    of mesh and the problem type.
    Array<int> ess_tdof_list;
@@ -201,12 +191,12 @@ int main(int argc, char *argv[])
       fespace->GetEssentialTrueDofs(ess_bdr, ess_tdof_list);
    }
 
-   // 7. Set up the linear form b(.) which corresponds to the right-hand side of
+   // 6. Set up the linear form b(.) which corresponds to the right-hand side of
    //    the FEM linear system.
    ComplexLinearForm b(fespace, conv);
    b.Vector::operator=(0.0);
 
-   // 8. Define the solution vector u as a complex finite element grid function
+   // 7. Define the solution vector u as a complex finite element grid function
    //    corresponding to fespace. Initialize u with initial guess of 1+0i or
    //    the exact solution if it is known.
    ComplexGridFunction u(fespace);
@@ -228,6 +218,7 @@ int main(int argc, char *argv[])
    VectorConstantCoefficient zeroVecCoef(zeroVec);
    VectorConstantCoefficient oneVecCoef(oneVec);
 
+   u = 0.0;
    switch (prob)
    {
       case 0:
@@ -280,7 +271,7 @@ int main(int argc, char *argv[])
                  << "window_title 'Exact: Imaginary Part'" << flush;
    }
 
-   // 9. Set up the sesquilinear form a(.,.) on the finite element space
+   // 8. Set up the sesquilinear form a(.,.) on the finite element space
    //    corresponding to the damped harmonic oscillator operator of the
    //    appropriate type:
    //
@@ -323,7 +314,7 @@ int main(int argc, char *argv[])
       default: break; // This should be unreachable
    }
 
-   // 9a. Set up the bilinear form for the preconditioner corresponding to the
+   // 8a. Set up the bilinear form for the preconditioner corresponding to the
    //     appropriate operator
    //
    //      0) A scalar H1 field
@@ -358,9 +349,9 @@ int main(int argc, char *argv[])
       default: break; // This should be unreachable
    }
 
-   // 10. Assemble the form and the corresponding linear system, applying any
-   //     necessary transformations such as: assembly, eliminating boundary
-   //     conditions, conforming constraints for non-conforming AMR, etc.
+   // 9. Assemble the form and the corresponding linear system, applying any
+   //    necessary transformations such as: assembly, eliminating boundary
+   //    conditions, conforming constraints for non-conforming AMR, etc.
    a->Assemble();
    pcOp->Assemble();
 
@@ -371,7 +362,7 @@ int main(int argc, char *argv[])
 
    cout << "Size of linear system: " << A->Width() << endl << endl;
 
-   // 11. Define and apply a GMRES solver for AU=B with a block diagonal
+   // 10. Define and apply a GMRES solver for AU=B with a block diagonal
    //     preconditioner based on the appropriate sparse smoother.
    {
       Array<int> blockOffsets;
@@ -428,7 +419,7 @@ int main(int argc, char *argv[])
       gmres.Mult(B, U);
    }
 
-   // 12. Recover the solution as a finite element grid function and compute the
+   // 11. Recover the solution as a finite element grid function and compute the
    //     errors if the exact solution is known.
    a->RecoverFEMSolution(U, b, u);
 
@@ -460,7 +451,7 @@ int main(int argc, char *argv[])
       cout << endl;
    }
 
-   // 13. Save the refined mesh and the solution. This output can be viewed
+   // 12. Save the refined mesh and the solution. This output can be viewed
    //     later using GLVis: "glvis -m mesh -g sol".
    {
       ofstream mesh_ofs("refined.mesh");
@@ -475,7 +466,7 @@ int main(int argc, char *argv[])
       u.imag().Save(sol_i_ofs);
    }
 
-   // 14. Send the solution by socket to a GLVis server.
+   // 13. Send the solution by socket to a GLVis server.
    if (visualization)
    {
       char vishost[] = "localhost";
@@ -534,7 +525,7 @@ int main(int argc, char *argv[])
       }
    }
 
-   // 15. Free the used memory.
+   // 14. Free the used memory.
    delete a;
    delete u_exact;
    delete pcOp;

examples/ex22p.cpp

@@ -7,18 +7,16 @@
 //               mpirun -np 4 ex22p -m ../data/inline-quad.mesh -o 3
 //               mpirun -np 4 ex22p -m ../data/inline-quad.mesh -o 3 -p 1
 //               mpirun -np 4 ex22p -m ../data/inline-quad.mesh -o 3 -p 2
-//               mpirun -np 4 ex22p -m ../data/inline-quad.mesh -o 1 -p 1 -pa
 //               mpirun -np 4 ex22p -m ../data/inline-tet.mesh -o 2
 //               mpirun -np 4 ex22p -m ../data/inline-hex.mesh -o 2
 //               mpirun -np 4 ex22p -m ../data/inline-hex.mesh -o 2 -p 1
 //               mpirun -np 4 ex22p -m ../data/inline-hex.mesh -o 2 -p 2
-//               mpirun -np 4 ex22p -m ../data/inline-hex.mesh -o 1 -p 2 -pa
 //               mpirun -np 4 ex22p -m ../data/star.mesh -o 2 -sigma 10.0
 //
-// Device sample runs:
-//               mpirun -np 4 ex22p -m ../data/inline-quad.mesh -o 1 -p 1 -pa -d cuda
-//               mpirun -np 4 ex22p -m ../data/inline-hex.mesh -o 1 -p 2 -pa -d cuda
-//               mpirun -np 4 ex22p -m ../data/star.mesh -o 2 -sigma 10.0 -pa -d cuda
+// With partial assembly:
+//               mpirun -np 4 ex22p -m ../data/inline-quad.mesh -o 1 -p 1 -pa
+//               mpirun -np 4 ex22p -m ../data/inline-hex.mesh -o 1 -p 2 -pa
+//               mpirun -np 4 ex22p -m ../data/star.mesh -o 2 -sigma 10.0 -pa
 //
 // Description:  This example code demonstrates the use of MFEM to define and
 //               solve simple complex-valued linear systems. It implements three
@@ -48,6 +46,7 @@
 //               We recommend viewing examples 1, 3 and 4 before viewing this
 //               example.
 
+
 #include "mfem.hpp"
 #include <fstream>
 #include <iostream>
@@ -91,7 +90,6 @@ int main(int argc, char *argv[])
    bool herm_conv = true;
    bool exact_sol = true;
    bool pa = false;
-   const char *device_config = "cpu";
 
    OptionsParser args(argc, argv);
    args.AddOption(&mesh_file, "-m", "--mesh",
@@ -126,8 +124,6 @@ int main(int argc, char *argv[])
                   "Enable or disable GLVis visualization.");
    args.AddOption(&pa, "-pa", "--partial-assembly", "-no-pa",
                   "--no-partial-assembly", "Enable Partial Assembly.");
-   args.AddOption(&device_config, "-d", "--device",
-                  "Device configuration string, see Device::Configure().");
    args.Parse();
    if (!args.Good())
    {
@@ -164,24 +160,19 @@ int main(int argc, char *argv[])
    ComplexOperator::Convention conv =
       herm_conv ? ComplexOperator::HERMITIAN : ComplexOperator::BLOCK_SYMMETRIC;
 
-   // 3. Enable hardware devices such as GPUs, and programming models such as
-   //    CUDA, OCCA, RAJA and OpenMP based on command line options.
-   Device device(device_config);
-   if (myid == 0) { device.Print(); }
-
-   // 4. Read the (serial) mesh from the given mesh file on all processors. We
+   // 3. Read the (serial) mesh from the given mesh file on all processors. We
    //    can handle triangular, quadrilateral, tetrahedral, hexahedral, surface
    //    and volume meshes with the same code.
    Mesh *mesh = new Mesh(mesh_file, 1, 1);
    int dim = mesh->Dimension();
 
-   // 5. Refine the serial mesh on all processors to increase the resolution.
+   // 4. Refine the serial mesh on all processors to increase the resolution.
    for (int l = 0; l < ser_ref_levels; l++)
    {
       mesh->UniformRefinement();
    }
 
-   // 6. Define a parallel mesh by a partitioning of the serial mesh. Refine
+   // 5. Define a parallel mesh by a partitioning of the serial mesh. Refine
    //    this mesh further in parallel to increase the resolution. Once the
    //    parallel mesh is defined, the serial mesh can be deleted.
    ParMesh *pmesh = new ParMesh(MPI_COMM_WORLD, *mesh);
@@ -191,7 +182,7 @@ int main(int argc, char *argv[])
       pmesh->UniformRefinement();
    }
 
-   // 7. Define a parallel finite element space on the parallel mesh. Here we
+   // 6. Define a parallel finite element space on the parallel mesh. Here we
    //    use continuous Lagrange, Nedelec, or Raviart-Thomas finite elements of
    //    the specified order.
    if (dim == 1 && prob != 0 )
@@ -219,7 +210,7 @@ int main(int argc, char *argv[])
       cout << "Number of finite element unknowns: " << size << endl;
    }
 
-   // 8. Determine the list of true (i.e. parallel conforming) essential
+   // 7. Determine the list of true (i.e. parallel conforming) essential
    //    boundary dofs. In this example, the boundary conditions are defined
    //    based on the type of mesh and the problem type.
    Array<int> ess_tdof_list;
@@ -231,14 +222,14 @@ int main(int argc, char *argv[])
       fespace->GetEssentialTrueDofs(ess_bdr, ess_tdof_list);
    }
 
-   // 9. Set up the parallel linear form b(.) which corresponds to the
+   // 8. Set up the parallel linear form b(.) which corresponds to the
    //    right-hand side of the FEM linear system.
    ParComplexLinearForm b(fespace, conv);
    b.Vector::operator=(0.0);
 
-   // 10. Define the solution vector u as a parallel complex finite element grid
-   //     function corresponding to fespace. Initialize u with initial guess of
-   //     1+0i or the exact solution if it is known.
+   // 9. Define the solution vector u as a parallel complex finite element grid
+   //    function corresponding to fespace. Initialize u with initial guess of
+   //    1+0i or the exact solution if it is known.
    ParComplexGridFunction u(fespace);
    ParComplexGridFunction * u_exact = NULL;
    if (exact_sol) { u_exact = new ParComplexGridFunction(fespace); }
@@ -258,6 +249,7 @@ int main(int argc, char *argv[])
    VectorConstantCoefficient zeroVecCoef(zeroVec);
    VectorConstantCoefficient oneVecCoef(oneVec);
 
+   u = 0.0;
    switch (prob)
    {
       case 0:
@@ -312,7 +304,7 @@ int main(int argc, char *argv[])
                  << "window_title 'Exact: Imaginary Part'" << flush;
    }
 
-   // 11. Set up the parallel sesquilinear form a(.,.) on the finite element
+   // 10. Set up the parallel sesquilinear form a(.,.) on the finite element
    //     space corresponding to the damped harmonic oscillator operator of the
    //     appropriate type:
    //
@@ -355,7 +347,7 @@ int main(int argc, char *argv[])
       default: break; // This should be unreachable
    }
 
-   // 11a. Set up the parallel bilinear form for the preconditioner
+   // 10a. Set up the parallel bilinear form for the preconditioner
    //      corresponding to the appropriate operator
    //
    //      0) A scalar H1 field
@@ -389,7 +381,7 @@ int main(int argc, char *argv[])
       default: break; // This should be unreachable
    }
 
-   // 12. Assemble the parallel bilinear form and the corresponding linear
+   // 11. Assemble the parallel bilinear form and the corresponding linear
    //     system, applying any necessary transformations such as: parallel
    //     assembly, eliminating boundary conditions, applying conforming
    //     constraints for non-conforming AMR, etc.
@@ -407,7 +399,7 @@ int main(int argc, char *argv[])
            << 2 * fespace->GlobalTrueVSize() << endl << endl;
    }
 
-   // 13. Define and apply a parallel FGMRES solver for AU=B with a block
+   // 12. Define and apply a parallel FGMRES solver for AU=B with a block
    //     diagonal preconditioner based on the appropriate multigrid
    //     preconditioner from hypre.
    {
@@ -468,7 +460,7 @@ int main(int argc, char *argv[])
       fgmres.SetPrintLevel(1);
       fgmres.Mult(B, U);
    }
-   // 14. Recover the parallel grid function corresponding to U. This is the
+   // 13. Recover the parallel grid function corresponding to U. This is the
    //     local finite element solution on each processor.
    a->RecoverFEMSolution(U, b, u);
 
@@ -503,7 +495,7 @@ int main(int argc, char *argv[])
       }
    }
 
-   // 15. Save the refined mesh and the solution in parallel. This output can be
+   // 14. Save the refined mesh and the solution in parallel. This output can be
    //     viewed later using GLVis: "glvis -np <np> -m mesh -g sol".
    {
       ostringstream mesh_name, sol_r_name, sol_i_name;
@@ -523,7 +515,7 @@ int main(int argc, char *argv[])
       u.imag().Save(sol_i_ofs);
    }
 
-   // 16. Send the solution by socket to a GLVis server.
+   // 15. Send the solution by socket to a GLVis server.
    if (visualization)
    {
       char vishost[] = "localhost";
@@ -588,7 +580,7 @@ int main(int argc, char *argv[])
       }
    }
 
-   // 17. Free the used memory.
+   // 16. Free the used memory.
    delete a;
    delete u_exact;
    delete pcOp;

examples/ex24.cpp

@@ -70,7 +70,7 @@ int main(int argc, char *argv[])
    args.AddOption(&order, "-o", "--order",
                   "Finite element order (polynomial degree).");
    args.AddOption(&prob, "-p", "--problem-type",
-                  "Choose between 0: grad, 1: curl, 2: div");
+                  "Choose between 0: H(Curl) or 1: H(Div)");
    args.AddOption(&static_cond, "-sc", "--static-condensation", "-no-sc",
                   "--no-static-condensation", "Enable static condensation.");
    args.AddOption(&pa, "-pa", "--partial-assembly", "-no-pa",

examples/ex24p.cpp

@@ -76,7 +76,7 @@ int main(int argc, char *argv[])
    args.AddOption(&order, "-o", "--order",
                   "Finite element order (polynomial degree).");
    args.AddOption(&prob, "-p", "--problem-type",
-                  "Choose between 0: grad, 1: curl, 2: div");
+                  "Choose between 0: H(Curl) or 1: H(Div)");
    args.AddOption(&static_cond, "-sc", "--static-condensation", "-no-sc",
                   "--no-static-condensation", "Enable static condensation.");
    args.AddOption(&pa, "-pa", "--partial-assembly", "-no-pa",

examples/ex25.cpp

@@ -82,24 +82,24 @@ public:
 };
 
 // Class for returning the PML coefficients of the bilinear form
-class PMLDiagMatrixCoefficient : public VectorCoefficient
+class PMLMatrixCoefficient : public MatrixCoefficient
 {
 private:
    CartesianPML * pml = nullptr;
-   void (*Function)(const Vector &, CartesianPML * , Vector &);
+   void (*Function)(const Vector &, CartesianPML * , DenseMatrix &);
 public:
-   PMLDiagMatrixCoefficient(int dim, void(*F)(const Vector &, CartesianPML *,
-                                              Vector &),
-                            CartesianPML * pml_)
-      : VectorCoefficient(dim), pml(pml_), Function(F)
+   PMLMatrixCoefficient(int dim, void(*F)(const Vector &, CartesianPML *,
+                                          DenseMatrix &),
+                        CartesianPML * pml_)
+      : MatrixCoefficient(dim), pml(pml_), Function(F)
    {}
-   virtual void Eval(Vector &K, ElementTransformation &T,
+   virtual void Eval(DenseMatrix &K, ElementTransformation &T,
                      const IntegrationPoint &ip)
    {
       double x[3];
       Vector transip(x, 3);
       T.Transform(ip, transip);
-      K.SetSize(vdim);
+      K.SetSize(height, width);
       (*Function)(transip, pml, K);
    }
 };
@@ -116,13 +116,13 @@ void source(const Vector &x, Vector & f);
 
 // Functions for computing the necessary coefficients after PML stretching.
 // J is the Jacobian matrix of the stretching function
-void detJ_JT_J_inv_Re(const Vector &x, CartesianPML * pml, Vector &D);
-void detJ_JT_J_inv_Im(const Vector &x, CartesianPML * pml, Vector &D);
-void detJ_JT_J_inv_abs(const Vector &x, CartesianPML * pml, Vector &D);
+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, Vector &D);
-void detJ_inv_JT_J_Im(const Vector &x, CartesianPML * pml, Vector &D);
-void detJ_inv_JT_J_abs(const Vector &x, CartesianPML * pml, Vector &D);
+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);
+void detJ_inv_JT_J_abs(const Vector &x, CartesianPML * pml, DenseMatrix &M);
 
 Array2D<double> comp_domain_bdr;
 Array2D<double> domain_bdr;
@@ -365,19 +365,19 @@ int main(int argc, char *argv[])
    a.AddDomainIntegrator(new VectorFEMassIntegrator(restr_omeg),NULL);
 
    int cdim = (dim == 2) ? 1 : dim;
-   PMLDiagMatrixCoefficient pml_c1_Re(cdim,detJ_inv_JT_J_Re, pml);
-   PMLDiagMatrixCoefficient pml_c1_Im(cdim,detJ_inv_JT_J_Im, pml);
-   ScalarVectorProductCoefficient c1_Re(muinv,pml_c1_Re);
-   ScalarVectorProductCoefficient c1_Im(muinv,pml_c1_Im);
-   VectorRestrictedCoefficient restr_c1_Re(c1_Re,attrPML);
-   VectorRestrictedCoefficient restr_c1_Im(c1_Im,attrPML);
+   PMLMatrixCoefficient pml_c1_Re(cdim,detJ_inv_JT_J_Re, pml);
+   PMLMatrixCoefficient pml_c1_Im(cdim,detJ_inv_JT_J_Im, pml);
+   ScalarMatrixProductCoefficient c1_Re(muinv,pml_c1_Re);
+   ScalarMatrixProductCoefficient c1_Im(muinv,pml_c1_Im);
+   MatrixRestrictedCoefficient restr_c1_Re(c1_Re,attrPML);
+   MatrixRestrictedCoefficient restr_c1_Im(c1_Im,attrPML);
 
-   PMLDiagMatrixCoefficient pml_c2_Re(dim, detJ_JT_J_inv_Re,pml);
-   PMLDiagMatrixCoefficient pml_c2_Im(dim, detJ_JT_J_inv_Im,pml);
-   ScalarVectorProductCoefficient c2_Re(omeg,pml_c2_Re);
-   ScalarVectorProductCoefficient c2_Im(omeg,pml_c2_Im);
-   VectorRestrictedCoefficient restr_c2_Re(c2_Re,attrPML);
-   VectorRestrictedCoefficient restr_c2_Im(c2_Im,attrPML);
+   PMLMatrixCoefficient pml_c2_Re(dim, detJ_JT_J_inv_Re,pml);
+   PMLMatrixCoefficient pml_c2_Im(dim, detJ_JT_J_inv_Im,pml);
+   ScalarMatrixProductCoefficient c2_Re(omeg,pml_c2_Re);
+   ScalarMatrixProductCoefficient c2_Im(omeg,pml_c2_Im);
+   MatrixRestrictedCoefficient restr_c2_Re(c2_Re,attrPML);
+   MatrixRestrictedCoefficient restr_c2_Im(c2_Im,attrPML);
 
    // Integrators inside the PML region
    a.AddDomainIntegrator(new CurlCurlIntegrator(restr_c1_Re),
@@ -419,13 +419,13 @@ int main(int argc, char *argv[])
       prec.AddDomainIntegrator(new CurlCurlIntegrator(restr_muinv));
       prec.AddDomainIntegrator(new VectorFEMassIntegrator(restr_absomeg));
 
-      PMLDiagMatrixCoefficient pml_c1_abs(cdim,detJ_inv_JT_J_abs, pml);
-      ScalarVectorProductCoefficient c1_abs(muinv,pml_c1_abs);
-      VectorRestrictedCoefficient restr_c1_abs(c1_abs,attrPML);
+      PMLMatrixCoefficient pml_c1_abs(cdim,detJ_inv_JT_J_abs, pml);
+      ScalarMatrixProductCoefficient c1_abs(muinv,pml_c1_abs);
+      MatrixRestrictedCoefficient restr_c1_abs(c1_abs,attrPML);
 
-      PMLDiagMatrixCoefficient pml_c2_abs(dim, detJ_JT_J_inv_abs,pml);
-      ScalarVectorProductCoefficient c2_abs(absomeg,pml_c2_abs);
-      VectorRestrictedCoefficient restr_c2_abs(c2_abs,attrPML);
+      PMLMatrixCoefficient pml_c2_abs(dim, detJ_JT_J_inv_abs,pml);
+      ScalarMatrixProductCoefficient c2_abs(absomeg,pml_c2_abs);
+      MatrixRestrictedCoefficient restr_c2_abs(c2_abs,attrPML);
 
       prec.AddDomainIntegrator(new CurlCurlIntegrator(restr_c1_abs));
       prec.AddDomainIntegrator(new VectorFEMassIntegrator(restr_c2_abs));
@@ -763,7 +763,7 @@ void E_bdr_data_Im(const Vector &x, Vector &E)
    }
 }
 
-void detJ_JT_J_inv_Re(const Vector &x, CartesianPML * pml, Vector &D)
+void detJ_JT_J_inv_Re(const Vector &x, CartesianPML * pml, DenseMatrix &M)
 {
    vector<complex<double>> dxs(dim);
    complex<double> det(1.0, 0.0);
@@ -774,13 +774,14 @@ void detJ_JT_J_inv_Re(const Vector &x, CartesianPML * pml, Vector &D)
       det *= dxs[i];
    }
 
+   M = 0.0;
    for (int i = 0; i < dim; ++i)
    {
-      D(i) = (det / pow(dxs[i], 2)).real();
+      M(i, i) = (det / pow(dxs[i], 2)).real();
    }
 }
 
-void detJ_JT_J_inv_Im(const Vector &x, CartesianPML * pml, Vector &D)
+void detJ_JT_J_inv_Im(const Vector &x, CartesianPML * pml, DenseMatrix &M)
 {
    vector<complex<double>> dxs(dim);
    complex<double> det = 1.0;
@@ -791,13 +792,14 @@ void detJ_JT_J_inv_Im(const Vector &x, CartesianPML * pml, Vector &D)
       det *= dxs[i];
    }
 
+   M = 0.0;
    for (int i = 0; i < dim; ++i)
    {
-      D(i) = (det / pow(dxs[i], 2)).imag();
+      M(i, i) = (det / pow(dxs[i], 2)).imag();
    }
 }
 
-void detJ_JT_J_inv_abs(const Vector &x, CartesianPML * pml, Vector &D)
+void detJ_JT_J_inv_abs(const Vector &x, CartesianPML * pml, DenseMatrix &M)
 {
    vector<complex<double>> dxs(dim);
    complex<double> det = 1.0;
@@ -808,13 +810,14 @@ void detJ_JT_J_inv_abs(const Vector &x, CartesianPML * pml, Vector &D)
       det *= dxs[i];
    }
 
+   M = 0.0;
    for (int i = 0; i < dim; ++i)
    {
-      D(i) = abs(det / pow(dxs[i], 2));
+      M(i, i) = abs(det / pow(dxs[i], 2));
    }
 }
 
-void detJ_inv_JT_J_Re(const Vector &x, CartesianPML * pml, Vector &D)
+void detJ_inv_JT_J_Re(const Vector &x, CartesianPML * pml, DenseMatrix &M)
 {
    vector<complex<double>> dxs(dim);
    complex<double> det(1.0, 0.0);
@@ -828,18 +831,19 @@ void detJ_inv_JT_J_Re(const Vector &x, CartesianPML * pml, Vector &D)
    // in the 2D case the coefficient is scalar 1/det(J)
    if (dim == 2)
    {
-      D = (1.0 / det).real();
+      M = (1.0 / det).real();
    }
    else
    {
+      M = 0.0;
       for (int i = 0; i < dim; ++i)
       {
-         D(i) = (pow(dxs[i], 2) / det).real();
+         M(i, i) = (pow(dxs[i], 2) / det).real();
       }
    }
 }
 
-void detJ_inv_JT_J_Im(const Vector &x, CartesianPML * pml, Vector &D)
+void detJ_inv_JT_J_Im(const Vector &x, CartesianPML * pml, DenseMatrix &M)
 {
    vector<complex<double>> dxs(dim);
    complex<double> det = 1.0;
@@ -852,18 +856,19 @@ void detJ_inv_JT_J_Im(const Vector &x, CartesianPML * pml, Vector &D)
 
    if (dim == 2)
    {
-      D = (1.0 / det).imag();
+      M = (1.0 / det).imag();
    }
    else
    {
+      M = 0.0;
       for (int i = 0; i < dim; ++i)
       {
-         D(i) = (pow(dxs[i], 2) / det).imag();
+         M(i, i) = (pow(dxs[i], 2) / det).imag();
       }
    }
 }
 
-void detJ_inv_JT_J_abs(const Vector &x, CartesianPML * pml, Vector &D)
+void detJ_inv_JT_J_abs(const Vector &x, CartesianPML * pml, DenseMatrix &M)
 {
    vector<complex<double>> dxs(dim);
    complex<double> det = 1.0;
@@ -876,13 +881,14 @@ void detJ_inv_JT_J_abs(const Vector &x, CartesianPML * pml, Vector &D)
 
    if (dim == 2)
    {
-      D = abs(1.0 / det);
+      M = abs(1.0 / det);
    }
    else
    {
+      M = 0.0;
       for (int i = 0; i < dim; ++i)
       {
-         D(i) = abs(pow(dxs[i], 2) / det);
+         M(i, i) = abs(pow(dxs[i], 2) / det);
       }
    }
 }

examples/ex25p.cpp

@@ -82,24 +82,24 @@ public:
 };
 
 // Class for returning the PML coefficients of the bilinear form
-class PMLDiagMatrixCoefficient : public VectorCoefficient
+class PMLMatrixCoefficient : public MatrixCoefficient
 {
 private:
    CartesianPML * pml = nullptr;
-   void (*Function)(const Vector &, CartesianPML * , Vector &);
+   void (*Function)(const Vector &, CartesianPML * , DenseMatrix &);
 public:
-   PMLDiagMatrixCoefficient(int dim, void(*F)(const Vector &, CartesianPML *,
-                                              Vector &),
-                            CartesianPML * pml_)
-      : VectorCoefficient(dim), pml(pml_), Function(F)
+   PMLMatrixCoefficient(int dim, void(*F)(const Vector &, CartesianPML *,
+                                          DenseMatrix &),
+                        CartesianPML * pml_)
+      : MatrixCoefficient(dim), pml(pml_), Function(F)
    {}
-   virtual void Eval(Vector &K, ElementTransformation &T,
+   virtual void Eval(DenseMatrix &K, ElementTransformation &T,
                      const IntegrationPoint &ip)
    {
       double x[3];
       Vector transip(x, 3);
       T.Transform(ip, transip);
-      K.SetSize(vdim);
+      K.SetSize(height, width);
       (*Function)(transip, pml, K);
    }
 };
@@ -116,13 +116,13 @@ void source(const Vector &x, Vector & f);
 
 // Functions for computing the necessary coefficients after PML stretching.
 // J is the Jacobian matrix of the stretching function
-void detJ_JT_J_inv_Re(const Vector &x, CartesianPML * pml, Vector & D);
-void detJ_JT_J_inv_Im(const Vector &x, CartesianPML * pml, Vector & D);
-void detJ_JT_J_inv_abs(const Vector &x, CartesianPML * pml, Vector & D);
+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, Vector & D);
-void detJ_inv_JT_J_Im(const Vector &x, CartesianPML * pml, Vector & D);
-void detJ_inv_JT_J_abs(const Vector &x, CartesianPML * pml, Vector & D);
+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);
+void detJ_inv_JT_J_abs(const Vector &x, CartesianPML * pml, DenseMatrix &M);
 
 Array2D<double> comp_domain_bdr;
 Array2D<double> domain_bdr;
@@ -393,19 +393,19 @@ int main(int argc, char *argv[])
    a.AddDomainIntegrator(new VectorFEMassIntegrator(restr_omeg),NULL);
 
    int cdim = (dim == 2) ? 1 : dim;
-   PMLDiagMatrixCoefficient pml_c1_Re(cdim,detJ_inv_JT_J_Re, pml);
-   PMLDiagMatrixCoefficient pml_c1_Im(cdim,detJ_inv_JT_J_Im, pml);
-   ScalarVectorProductCoefficient c1_Re(muinv,pml_c1_Re);
-   ScalarVectorProductCoefficient c1_Im(muinv,pml_c1_Im);
-   VectorRestrictedCoefficient restr_c1_Re(c1_Re,attrPML);
-   VectorRestrictedCoefficient restr_c1_Im(c1_Im,attrPML);
+   PMLMatrixCoefficient pml_c1_Re(cdim,detJ_inv_JT_J_Re, pml);
+   PMLMatrixCoefficient pml_c1_Im(cdim,detJ_inv_JT_J_Im, pml);
+   ScalarMatrixProductCoefficient c1_Re(muinv,pml_c1_Re);
+   ScalarMatrixProductCoefficient c1_Im(muinv,pml_c1_Im);
+   MatrixRestrictedCoefficient restr_c1_Re(c1_Re,attrPML);
+   MatrixRestrictedCoefficient restr_c1_Im(c1_Im,attrPML);
 
-   PMLDiagMatrixCoefficient pml_c2_Re(dim, detJ_JT_J_inv_Re,pml);
-   PMLDiagMatrixCoefficient pml_c2_Im(dim, detJ_JT_J_inv_Im,pml);
-   ScalarVectorProductCoefficient c2_Re(omeg,pml_c2_Re);
-   ScalarVectorProductCoefficient c2_Im(omeg,pml_c2_Im);
-   VectorRestrictedCoefficient restr_c2_Re(c2_Re,attrPML);
-   VectorRestrictedCoefficient restr_c2_Im(c2_Im,attrPML);
+   PMLMatrixCoefficient pml_c2_Re(dim, detJ_JT_J_inv_Re,pml);
+   PMLMatrixCoefficient pml_c2_Im(dim, detJ_JT_J_inv_Im,pml);
+   ScalarMatrixProductCoefficient c2_Re(omeg,pml_c2_Re);
+   ScalarMatrixProductCoefficient c2_Im(omeg,pml_c2_Im);
+   MatrixRestrictedCoefficient restr_c2_Re(c2_Re,attrPML);
+   MatrixRestrictedCoefficient restr_c2_Im(c2_Im,attrPML);
 
    // Integrators inside the PML region
    a.AddDomainIntegrator(new CurlCurlIntegrator(restr_c1_Re),
@@ -453,13 +453,13 @@ int main(int argc, char *argv[])
       prec.AddDomainIntegrator(new CurlCurlIntegrator(restr_muinv));
       prec.AddDomainIntegrator(new VectorFEMassIntegrator(restr_absomeg));
 
-      PMLDiagMatrixCoefficient pml_c1_abs(cdim,detJ_inv_JT_J_abs, pml);
-      ScalarVectorProductCoefficient c1_abs(muinv,pml_c1_abs);
-      VectorRestrictedCoefficient restr_c1_abs(c1_abs,attrPML);
+      PMLMatrixCoefficient pml_c1_abs(cdim,detJ_inv_JT_J_abs, pml);
+      ScalarMatrixProductCoefficient c1_abs(muinv,pml_c1_abs);
+      MatrixRestrictedCoefficient restr_c1_abs(c1_abs,attrPML);
 
-      PMLDiagMatrixCoefficient pml_c2_abs(dim, detJ_JT_J_inv_abs,pml);
-      ScalarVectorProductCoefficient c2_abs(absomeg,pml_c2_abs);
-      VectorRestrictedCoefficient restr_c2_abs(c2_abs,attrPML);
+      PMLMatrixCoefficient pml_c2_abs(dim, detJ_JT_J_inv_abs,pml);
+      ScalarMatrixProductCoefficient c2_abs(absomeg,pml_c2_abs);
+      MatrixRestrictedCoefficient restr_c2_abs(c2_abs,attrPML);
 
       prec.AddDomainIntegrator(new CurlCurlIntegrator(restr_c1_abs));
       prec.AddDomainIntegrator(new VectorFEMassIntegrator(restr_c2_abs));
@@ -819,7 +819,7 @@ void E_bdr_data_Im(const Vector &x, Vector &E)
    }
 }
 
-void detJ_JT_J_inv_Re(const Vector &x, CartesianPML * pml, Vector & D)
+void detJ_JT_J_inv_Re(const Vector &x, CartesianPML * pml, DenseMatrix &M)
 {
    vector<complex<double>> dxs(dim);
    complex<double> det(1.0, 0.0);
@@ -830,13 +830,14 @@ void detJ_JT_J_inv_Re(const Vector &x, CartesianPML * pml, Vector & D)
       det *= dxs[i];
    }
 
+   M = 0.0;
    for (int i = 0; i < dim; ++i)
    {
-      D(i) = (det / pow(dxs[i], 2)).real();
+      M(i, i) = (det / pow(dxs[i], 2)).real();
    }
 }
 
-void detJ_JT_J_inv_Im(const Vector &x, CartesianPML * pml, Vector & D)
+void detJ_JT_J_inv_Im(const Vector &x, CartesianPML * pml, DenseMatrix &M)
 {
    vector<complex<double>> dxs(dim);
    complex<double> det = 1.0;
@@ -847,13 +848,14 @@ void detJ_JT_J_inv_Im(const Vector &x, CartesianPML * pml, Vector & D)
       det *= dxs[i];
    }
 
+   M = 0.0;
    for (int i = 0; i < dim; ++i)
    {
-      D(i) = (det / pow(dxs[i], 2)).imag();
+      M(i, i) = (det / pow(dxs[i], 2)).imag();
    }
 }
 
-void detJ_JT_J_inv_abs(const Vector &x, CartesianPML * pml, Vector & D)
+void detJ_JT_J_inv_abs(const Vector &x, CartesianPML * pml, DenseMatrix &M)
 {
    vector<complex<double>> dxs(dim);
    complex<double> det = 1.0;
@@ -864,13 +866,14 @@ void detJ_JT_J_inv_abs(const Vector &x, CartesianPML * pml, Vector & D)
       det *= dxs[i];
    }
 
+   M = 0.0;
    for (int i = 0; i < dim; ++i)
    {
-      D(i) = abs(det / pow(dxs[i], 2));
+      M(i, i) = abs(det / pow(dxs[i], 2));
    }
 }
 
-void detJ_inv_JT_J_Re(const Vector &x, CartesianPML * pml, Vector & D)
+void detJ_inv_JT_J_Re(const Vector &x, CartesianPML * pml, DenseMatrix &M)
 {
    vector<complex<double>> dxs(dim);
    complex<double> det(1.0, 0.0);
@@ -884,18 +887,19 @@ void detJ_inv_JT_J_Re(const Vector &x, CartesianPML * pml, Vector & D)
    // in the 2D case the coefficient is scalar 1/det(J)
    if (dim == 2)
    {
-      D = (1.0 / det).real();
+      M = (1.0 / det).real();
    }
    else
    {
+      M = 0.0;
       for (int i = 0; i < dim; ++i)
       {
-         D(i) = (pow(dxs[i], 2) / det).real();
+         M(i, i) = (pow(dxs[i], 2) / det).real();
       }
    }
 }
 
-void detJ_inv_JT_J_Im(const Vector &x, CartesianPML * pml, Vector & D)
+void detJ_inv_JT_J_Im(const Vector &x, CartesianPML * pml, DenseMatrix &M)
 {
    vector<complex<double>> dxs(dim);
    complex<double> det = 1.0;
@@ -908,18 +912,19 @@ void detJ_inv_JT_J_Im(const Vector &x, CartesianPML * pml, Vector & D)
 
    if (dim == 2)
    {
-      D = (1.0 / det).imag();
+      M = (1.0 / det).imag();
    }
    else
    {
+      M = 0.0;
       for (int i = 0; i < dim; ++i)
       {
-         D(i) = (pow(dxs[i], 2) / det).imag();
+         M(i, i) = (pow(dxs[i], 2) / det).imag();
       }
    }
 }
 
-void detJ_inv_JT_J_abs(const Vector &x, CartesianPML * pml, Vector & D)
+void detJ_inv_JT_J_abs(const Vector &x, CartesianPML * pml, DenseMatrix &M)
 {
    vector<complex<double>> dxs(dim);
    complex<double> det = 1.0;
@@ -932,13 +937,14 @@ void detJ_inv_JT_J_abs(const Vector &x, CartesianPML * pml, Vector & D)
 
    if (dim == 2)
    {
-      D = abs(1.0 / det);
+      M = abs(1.0 / det);
    }
    else
    {
+      M = 0.0;
       for (int i = 0; i < dim; ++i)
       {
-         D(i) = abs(pow(dxs[i], 2) / det);
+         M(i, i) = abs(pow(dxs[i], 2) / det);
       }
    }
 }

examples/ex2p.cpp

@@ -60,7 +60,6 @@ int main(int argc, char *argv[])
    bool static_cond = false;
    bool visualization = 1;
    bool amg_elast = 0;
-   bool reorder_space = false;
 
    OptionsParser args(argc, argv);
    args.AddOption(&mesh_file, "-m", "--mesh",
@@ -76,8 +75,6 @@ int main(int argc, char *argv[])
    args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
                   "--no-visualization",
                   "Enable or disable GLVis visualization.");
-   args.AddOption(&reorder_space, "-nodes", "--by-nodes", "-vdim", "--by-vdim",
-                  "Use byNODES ordering of vector space instead of byVDIM");
    args.Parse();
    if (!args.Good())
    {
@@ -159,14 +156,7 @@ int main(int argc, char *argv[])
    else
    {
       fec = new H1_FECollection(order, dim);
-      if (reorder_space)
-      {
-         fespace = new ParFiniteElementSpace(pmesh, fec, dim, Ordering::byNODES);
-      }
-      else
-      {
-         fespace = new ParFiniteElementSpace(pmesh, fec, dim, Ordering::byVDIM);
-      }
+      fespace = new ParFiniteElementSpace(pmesh, fec, dim, Ordering::byVDIM);
    }
    HYPRE_Int size = fespace->GlobalTrueVSize();
    if (myid == 0)
@@ -259,7 +249,7 @@ int main(int argc, char *argv[])
    }
    else
    {
-      amg->SetSystemsOptions(dim, reorder_space);
+      amg->SetSystemsOptions(dim);
    }
    HyprePCG *pcg = new HyprePCG(A);
    pcg->SetTol(1e-8);

examples/ex6.cpp

@@ -108,11 +108,7 @@ int main(int argc, char *argv[])
    //    the Laplace problem -\Delta u = 1. We don't assemble the discrete
    //    problem yet, this will be done in the main loop.
    BilinearForm a(&fespace);
-   if (pa)
-   {
-      a.SetAssemblyLevel(AssemblyLevel::PARTIAL);
-      a.SetDiagonalPolicy(Operator::DIAG_ONE);
-   }
+   if (pa) { a.SetAssemblyLevel(AssemblyLevel::PARTIAL); }
    LinearForm b(&fespace);
 
    ConstantCoefficient one(1.0);
@@ -203,10 +199,9 @@ int main(int argc, char *argv[])
          umf_solver.Mult(B, X);
 #endif
       }
-      else // Diagonal preconditioning in partial assembly mode.
+      else // No preconditioning for now in partial assembly mode.
       {
-         OperatorJacobiSmoother M(a, ess_tdof_list);
-         PCG(*A, M, B, X, 3, 2000, 1e-12, 0.0);
+         CG(*A, B, X, 3, 2000, 1e-12, 0.0);
       }
 
       // 18. After solving the linear system, reconstruct the solution as a

examples/ex6p.cpp

@@ -129,11 +129,7 @@ int main(int argc, char *argv[])
    //    the Laplace problem -\Delta u = 1. We don't assemble the discrete
    //    problem yet, this will be done in the main loop.
    ParBilinearForm a(&fespace);
-   if (pa)
-   {
-      a.SetAssemblyLevel(AssemblyLevel::PARTIAL);
-      a.SetDiagonalPolicy(Operator::DIAG_ONE);
-   }
+   if (pa) { a.SetAssemblyLevel(AssemblyLevel::PARTIAL); }
    ParLinearForm b(&fespace);
 
    ConstantCoefficient one(1.0);
@@ -224,26 +220,17 @@ int main(int argc, char *argv[])
 
       // 17. Solve the linear system A X = B.
       //     * With full assembly, use the BoomerAMG preconditioner from hypre.
-      //     * With partial assembly, use a diagonal preconditioner.
-      Solver *M = NULL;
-      if (pa)
-      {
-         M = new OperatorJacobiSmoother(a, ess_tdof_list);
-      }
-      else
-      {
-         HypreBoomerAMG *amg = new HypreBoomerAMG;
-         amg->SetPrintLevel(0);
-         M = amg;
-      }
+      //     * With partial assembly, use no preconditioner, for now.
+      HypreBoomerAMG *amg = NULL;
+      if (!pa) { amg = new HypreBoomerAMG; amg->SetPrintLevel(0); }
       CGSolver cg(MPI_COMM_WORLD);
       cg.SetRelTol(1e-6);
       cg.SetMaxIter(2000);
       cg.SetPrintLevel(3); // print the first and the last iterations only
-      cg.SetPreconditioner(*M);
+      if (amg) { cg.SetPreconditioner(*amg); }
       cg.SetOperator(*A);
       cg.Mult(B, X);
-      delete M;
+      delete amg;
 
       // 18. Switch back to the host and extract the parallel grid function
       //     corresponding to the finite element approximation X. This is the

examples/makefile

@@ -119,11 +119,6 @@ ex11p-test-superlu: ex11p
 	@$(call mfem-test,$<, $(RUN_MPI), SuperLU_DIST example,--superlu)
 test-par-YES: ex11p-test-superlu
 endif
-ifeq ($(MFEM_USE_MKL_CPARDISO),YES)
-ex11p-test-superlu: ex11p
-	@$(call mfem-test,$<, $(RUN_MPI), MKL_CPARDISO example,--pardiso)
-test-par-YES: ex11p-test-pardiso
-endif
 
 # Testing: "test" target and mfem-test* variables are defined in config/test.mk
 

fem/CMakeLists.txt

@@ -31,7 +31,6 @@ set(SRCS
   bilininteg_vecmass.cpp
   coefficient.cpp
   complex_fem.cpp
-  convergence.cpp
   datacollection.cpp
   eltrans.cpp
   estimators.cpp
@@ -51,10 +50,43 @@ set(SRCS
   fespacehierarchy.cpp
   nonlininteg_vectorconvection.cpp
   quadinterpolator.cpp
+  quadinterpolator_det.cpp
+  quadinterpolator_eval_by_nodes.cpp
+  quadinterpolator_eval_by_vdim.cpp
+  quadinterpolator_grad_by_nodes.cpp
+  quadinterpolator_grad_by_vdim.cpp
+  quadinterpolator_grad_phys_by_nodes.cpp
+  quadinterpolator_grad_phys_by_vdim.cpp
   quadinterpolator_face.cpp
   restriction.cpp
   staticcond.cpp
   tmop.cpp
+  tmop_pa.cpp
+  tmop_pa_h2d.cpp
+  tmop_pa_h2d_c0.cpp
+  tmop_pa_h2m.cpp
+  tmop_pa_h2m_c0.cpp
+  tmop_pa_h2s.cpp
+  tmop_pa_h2s_c0.cpp
+  tmop_pa_h3d.cpp
+  tmop_pa_h3d_c0.cpp
+  tmop_pa_h3m.cpp
+  tmop_pa_h3m_c0.cpp
+  tmop_pa_h3s.cpp
+  tmop_pa_h3s_c0.cpp
+  tmop_pa_jp2.cpp
+  tmop_pa_jp3.cpp
+  tmop_pa_jt2_tc.cpp
+  tmop_pa_jt3_datc.cpp
+  tmop_pa_jt3_tc.cpp
+  tmop_pa_p2.cpp
+  tmop_pa_p2_c0.cpp
+  tmop_pa_p3.cpp
+  tmop_pa_p3_c0.cpp
+  tmop_pa_w2.cpp
+  tmop_pa_w2_c0.cpp
+  tmop_pa_w3.cpp
+  tmop_pa_w3_c0.cpp
   tmop_tools.cpp
   gslib.cpp
   transfer.cpp
@@ -66,7 +98,6 @@ set(HDRS
   bilininteg.hpp
   coefficient.hpp
   complex_fem.hpp
-  convergence.hpp
   datacollection.hpp
   eltrans.hpp
   estimators.hpp
@@ -85,7 +116,10 @@ set(HDRS
   nonlinearform_ext.hpp
   nonlininteg.hpp
   quadinterpolator.hpp
+  quadinterpolator_eval.hpp
   quadinterpolator_face.hpp
+  quadinterpolator_grad.hpp
+  quadinterpolator_grad_phys.hpp
   restriction.hpp
   fespacehierarchy.hpp
   staticcond.hpp
@@ -98,6 +132,7 @@ set(HDRS
   tfespace.hpp
   tintrules.hpp
   tmop.hpp
+  tmop_pa.hpp
   tmop_tools.hpp
   gslib.hpp
   transfer.hpp

fem/bilinearform.cpp

@@ -627,33 +627,6 @@ void BilinearForm::AssembleDiagonal(Vector &diag) const
       MFEM_ASSERT(diag.Size() == fes->GetTrueVSize(),
                   "Vector for holding diagonal has wrong size!");
       const Operator *P = fes->GetProlongationMatrix();
-      // For an AMR mesh, a convergent diagonal is assembled with |P^T| d_e,
-      // where |P^T| has the entry-wise absolute values of the conforming
-      // prolongation transpose operator.
-      if (P && !fes->Conforming())
-      {
-         Vector local_diag(P->Height());
-         ext->AssembleDiagonal(local_diag);
-         const SparseMatrix *SP = dynamic_cast<const SparseMatrix*>(P);
-#ifdef MFEM_USE_MPI
-         const HypreParMatrix *HP = dynamic_cast<const HypreParMatrix*>(P);
-#endif
-         if (SP)
-         {
-            SP->AbsMultTranspose(local_diag, diag);
-         }
-#ifdef MFEM_USE_MPI
-         else if (HP)
-         {
-            HP->AbsMultTranspose(1.0, local_diag, 0.0, diag);
-         }
-#endif
-         else
-         {
-            MFEM_ABORT("Prolongation matrix has unexpected type.");
-         }
-         return;
-      }
       if (!IsIdentityProlongation(P))
       {
          Vector local_diag(P->Height());

fem/bilinearform_ext.cpp

@@ -96,9 +96,6 @@ void PABilinearFormExtension::Assemble()
       integrators[i]->AssemblePA(*a->FESpace());
    }
 
-   MFEM_VERIFY(a->GetBBFI()->Size() == 0,
-               "Partial assembly does not support AddBoundaryIntegrator yet.");
-
    Array<BilinearFormIntegrator*> &intFaceIntegrators = *a->GetFBFI();
    const int intFaceIntegratorCount = intFaceIntegrators.Size();
    for (int i = 0; i < intFaceIntegratorCount; ++i)
@@ -310,21 +307,19 @@ void EABilinearFormExtension::Assemble()
 
    ea_data.SetSize(ne*elemDofs*elemDofs, Device::GetMemoryType());
    ea_data.UseDevice(true);
+   ea_data = 0.0;
 
    Array<BilinearFormIntegrator*> &integrators = *a->GetDBFI();
    const int integratorCount = integrators.Size();
    for (int i = 0; i < integratorCount; ++i)
    {
-      integrators[i]->AssembleEA(*a->FESpace(), ea_data, i);
+      integrators[i]->AssembleEA(*a->FESpace(), ea_data);
    }
 
    faceDofs = trialFes ->
               GetTraceElement(0, trialFes->GetMesh()->GetFaceBaseGeometry(0)) ->
               GetDof();
 
-   MFEM_VERIFY(a->GetBBFI()->Size() == 0,
-               "Element assembly does not support AddBoundaryIntegrator yet.");
-
    Array<BilinearFormIntegrator*> &intFaceIntegrators = *a->GetFBFI();
    const int intFaceIntegratorCount = intFaceIntegrators.Size();
    if (intFaceIntegratorCount>0)
@@ -332,13 +327,14 @@ void EABilinearFormExtension::Assemble()
       nf_int = trialFes->GetNFbyType(FaceType::Interior);
       ea_data_int.SetSize(2*nf_int*faceDofs*faceDofs, Device::GetMemoryType());
       ea_data_ext.SetSize(2*nf_int*faceDofs*faceDofs, Device::GetMemoryType());
+      ea_data_int = 0.0;
+      ea_data_ext = 0.0;
    }
    for (int i = 0; i < intFaceIntegratorCount; ++i)
    {
       intFaceIntegrators[i]->AssembleEAInteriorFaces(*a->FESpace(),
                                                      ea_data_int,
-                                                     ea_data_ext,
-                                                     i);
+                                                     ea_data_ext);
    }
 
    Array<BilinearFormIntegrator*> &bdrFaceIntegrators = *a->GetBFBFI();
@@ -351,7 +347,7 @@ void EABilinearFormExtension::Assemble()
    }
    for (int i = 0; i < boundFaceIntegratorCount; ++i)
    {
-      bdrFaceIntegrators[i]->AssembleEABoundaryFaces(*a->FESpace(),ea_data_bdr,i);
+      bdrFaceIntegrators[i]->AssembleEABoundaryFaces(*a->FESpace(),ea_data_bdr);
    }
 
    if (factorize_face_terms && int_face_restrict_lex)
@@ -798,12 +794,6 @@ void PAMixedBilinearFormExtension::Assemble()
    {
       integrators[i]->AssemblePA(*trialFes, *testFes);
    }
-   MFEM_VERIFY(a->GetBBFI()->Size() == 0,
-               "Partial assembly does not support AddBoundaryIntegrator yet.");
-   MFEM_VERIFY(a->GetTFBFI()->Size() == 0,
-               "Partial assembly does not support AddTraceFaceIntegrator yet.");
-   MFEM_VERIFY(a->GetBTFBFI()->Size() == 0,
-               "Partial assembly does not support AddBdrTraceFaceIntegrator yet.");
 }
 
 void PAMixedBilinearFormExtension::Update()

fem/bilininteg.cpp

@@ -52,8 +52,7 @@ void BilinearFormIntegrator::AssembleDiagonalPA(Vector &)
 }
 
 void BilinearFormIntegrator::AssembleEA(const FiniteElementSpace &fes,
-                                        Vector &emat,
-                                        const bool add)
+                                        Vector &emat)
 {
    mfem_error ("BilinearFormIntegrator::AssembleEA(...)\n"
                "   is not implemented for this class.");
@@ -62,8 +61,7 @@ void BilinearFormIntegrator::AssembleEA(const FiniteElementSpace &fes,
 void BilinearFormIntegrator::AssembleEAInteriorFaces(const FiniteElementSpace
                                                      &fes,
                                                      Vector &ea_data_int,
-                                                     Vector &ea_data_ext,
-                                                     const bool add)
+                                                     Vector &ea_data_ext)
 {
    mfem_error ("BilinearFormIntegrator::AssembleEAInteriorFaces(...)\n"
                "   is not implemented for this class.");
@@ -71,8 +69,7 @@ void BilinearFormIntegrator::AssembleEAInteriorFaces(const FiniteElementSpace
 
 void BilinearFormIntegrator::AssembleEABoundaryFaces(const FiniteElementSpace
                                                      &fes,
-                                                     Vector &ea_data_bdr,
-                                                     const bool add)
+                                                     Vector &ea_data_bdr)
 {
    mfem_error ("BilinearFormIntegrator::AssembleEABoundaryFaces(...)\n"
                "   is not implemented for this class.");
@@ -1525,7 +1522,6 @@ void CurlCurlIntegrator::AssembleElementMatrix
    double w;
 
 #ifdef MFEM_THREAD_SAFE
-   Vector D;
    DenseMatrix curlshape(nd,dimc), curlshape_dFt(nd,dimc), M;
 #else
    curlshape.SetSize(nd,dimc);
@@ -1533,7 +1529,6 @@ void CurlCurlIntegrator::AssembleElementMatrix
 #endif
    elmat.SetSize(nd);
    if (MQ) { M.SetSize(dimc); }
-   if (DQ) { D.SetSize(dimc); }
 
    const IntegrationRule *ir = IntRule;
    if (ir == NULL)
@@ -1577,12 +1572,6 @@ void CurlCurlIntegrator::AssembleElementMatrix
          Mult(curlshape_dFt, M, curlshape);
          AddMultABt(curlshape, curlshape_dFt, elmat);
       }
-      else if (DQ)
-      {
-         DQ->Eval(D, Trans, ip);
-         D *= w;
-         AddMultADAt(curlshape_dFt, D, elmat);
-      }
       else if (Q)
       {
          w *= Q->Eval(Trans, ip);

fem/bilininteg.hpp

@@ -20,13 +20,6 @@
 namespace mfem
 {
 
-// Local maximum size of dofs and quads in 1D
-constexpr int HCURL_MAX_D1D = 5;
-constexpr int HCURL_MAX_Q1D = 6;
-
-constexpr int HDIV_MAX_D1D = 5;
-constexpr int HDIV_MAX_Q1D = 6;
-
 /// Abstract base class BilinearFormIntegrator
 class BilinearFormIntegrator : public NonlinearFormIntegrator
 {
@@ -86,10 +79,9 @@ public:
    virtual void AddMultTransposePA(const Vector &x, Vector &y) const;
 
    /// Method defining element assembly.
-   /** The result of the element assembly is added to the @a emat Vector if
-       @a add is true. Otherwise, if @a add is false, we set @a emat. */
-   virtual void AssembleEA(const FiniteElementSpace &fes, Vector &emat,
-                           const bool add = true);
+   /** The result of the element assembly is added and stored in the @a emat
+       Vector. */
+   virtual void AssembleEA(const FiniteElementSpace &fes, Vector &emat);
    /** Used with BilinearFormIntegrators that have different spaces. */
    // virtual void AssembleEA(const FiniteElementSpace &trial_fes,
    //                         const FiniteElementSpace &test_fes,
@@ -97,12 +89,10 @@ public:
 
    virtual void AssembleEAInteriorFaces(const FiniteElementSpace &fes,
                                         Vector &ea_data_int,
-                                        Vector &ea_data_ext,
-                                        const bool add = true);
+                                        Vector &ea_data_ext);
 
    virtual void AssembleEABoundaryFaces(const FiniteElementSpace &fes,
-                                        Vector &ea_data_bdr,
-                                        const bool add = true);
+                                        Vector &ea_data_bdr);
 
    /// Given a particular Finite Element computes the element matrix elmat.
    virtual void AssembleElementMatrix(const FiniteElement &el,
@@ -265,17 +255,14 @@ public:
       bfi->AddMultTransposePA(x, y);
    }
 
-   virtual void AssembleEA(const FiniteElementSpace &fes, Vector &emat,
-                           const bool add);
+   virtual void AssembleEA(const FiniteElementSpace &fes, Vector &emat);
 
    virtual void AssembleEAInteriorFaces(const FiniteElementSpace &fes,
                                         Vector &ea_data_int,
-                                        Vector &ea_data_ext,
-                                        const bool add);
+                                        Vector &ea_data_ext);
 
    virtual void AssembleEABoundaryFaces(const FiniteElementSpace &fes,
-                                        Vector &ea_data_bdr,
-                                        const bool add);
+                                        Vector &ea_data_bdr);
 
    virtual ~TransposeIntegrator() { if (own_bfi) { delete bfi; } }
 };
@@ -1958,8 +1945,7 @@ public:
 
    virtual void AssemblePA(const FiniteElementSpace &fes);
 
-   virtual void AssembleEA(const FiniteElementSpace &fes, Vector &emat,
-                           const bool add);
+   virtual void AssembleEA(const FiniteElementSpace &fes, Vector &emat);
 
    virtual void AssembleDiagonalPA(Vector &diag);
 
@@ -2034,8 +2020,7 @@ public:
 
    virtual void AssemblePA(const FiniteElementSpace &fes);
 
-   virtual void AssembleEA(const FiniteElementSpace &fes, Vector &emat,
-                           const bool add);
+   virtual void AssembleEA(const FiniteElementSpace &fes, Vector &emat);
 
    virtual void AssembleDiagonalPA(Vector &diag);
 
@@ -2091,8 +2076,7 @@ public:
 
    virtual void AssemblePA(const FiniteElementSpace&);
 
-   virtual void AssembleEA(const FiniteElementSpace &fes, Vector &emat,
-                           const bool add);
+   virtual void AssembleEA(const FiniteElementSpace &fes, Vector &emat);
 
    virtual void AddMultPA(const Vector&, Vector&) const;
 
@@ -2309,14 +2293,12 @@ class CurlCurlIntegrator: public BilinearFormIntegrator
 private:
    Vector vec, pointflux;
 #ifndef MFEM_THREAD_SAFE
-   Vector D;
    DenseMatrix curlshape, curlshape_dFt, M;
    DenseMatrix vshape, projcurl;
 #endif
 
 protected:
    Coefficient *Q;
-   VectorCoefficient *DQ;
    MatrixCoefficient *MQ;
 
    // PA extension
@@ -2325,17 +2307,12 @@ protected:
    const DofToQuad *mapsC;         ///< Not owned. DOF-to-quad map, closed.
    const GeometricFactors *geom;   ///< Not owned
    int dim, ne, nq, dofs1D, quad1D;
-   bool symmetric = true; ///< False if using a nonsymmetric matrix coefficient
 
 public:
-   CurlCurlIntegrator() { Q = NULL; DQ = NULL; MQ = NULL; }
+   CurlCurlIntegrator() { Q = NULL; MQ = NULL; }
    /// Construct a bilinear form integrator for Nedelec elements
-   CurlCurlIntegrator(Coefficient &q, const IntegrationRule *ir = NULL) :
-      BilinearFormIntegrator(ir), Q(&q) { DQ = NULL; MQ = NULL; }
-   CurlCurlIntegrator(VectorCoefficient &dq, const IntegrationRule *ir = NULL) :
-      BilinearFormIntegrator(ir), DQ(&dq) { Q = NULL; MQ = NULL; }
-   CurlCurlIntegrator(MatrixCoefficient &mq, const IntegrationRule *ir = NULL) :
-      BilinearFormIntegrator(ir), MQ(&mq) { Q = NULL; DQ = NULL; }
+   CurlCurlIntegrator(Coefficient &q) : Q(&q) { MQ = NULL; }
+   CurlCurlIntegrator(MatrixCoefficient &m) : MQ(&m) { Q = NULL; }
 
    /* Given a particular Finite Element, compute the
       element curl-curl matrix elmat */
@@ -2413,11 +2390,8 @@ protected:
    Vector pa_data;
    const DofToQuad *mapsO;         ///< Not owned. DOF-to-quad map, open.
    const DofToQuad *mapsC;         ///< Not owned. DOF-to-quad map, closed.
-   const DofToQuad *mapsOtest;     ///< Not owned. DOF-to-quad map, open.
-   const DofToQuad *mapsCtest;     ///< Not owned. DOF-to-quad map, closed.
    const GeometricFactors *geom;   ///< Not owned
-   int dim, ne, nq, dofs1D, dofs1Dtest, quad1D, trial_fetype, test_fetype;
-   bool symmetric = true; ///< False if using a nonsymmetric matrix coefficient
+   int dim, ne, nq, dofs1D, quad1D, fetype;
 
 public:
    VectorFEMassIntegrator() { Init(NULL, NULL, NULL); }
@@ -2438,8 +2412,6 @@ public:
 
    using BilinearFormIntegrator::AssemblePA;
    virtual void AssemblePA(const FiniteElementSpace &fes);
-   virtual void AssemblePA(const FiniteElementSpace &trial_fes,
-                           const FiniteElementSpace &test_fes);
    virtual void AddMultPA(const Vector &x, Vector &y) const;
    virtual void AssembleDiagonalPA(Vector& diag);
 };
@@ -2669,12 +2641,10 @@ public:
 
    virtual void AssembleEAInteriorFaces(const FiniteElementSpace& fes,
                                         Vector &ea_data_int,
-                                        Vector &ea_data_ext,
-                                        const bool add);
+                                        Vector &ea_data_ext);
 
    virtual void AssembleEABoundaryFaces(const FiniteElementSpace& fes,
-                                        Vector &ea_data_bdr,
-                                        const bool add);
+                                        Vector &ea_data_bdr);
 
    static const IntegrationRule &GetRule(Geometry::Type geom, int order,
                                          FaceElementTransformations &T);

fem/bilininteg_convection_ea.cpp

@@ -22,7 +22,6 @@ static void EAConvectionAssemble1D(const int NE,
                                    const Array<double> &g,
                                    const Vector &padata,
                                    Vector &eadata,
-                                   const bool add,
                                    const int d1d = 0,
                                    const int q1d = 0)
 {
@@ -55,14 +54,7 @@ static void EAConvectionAssemble1D(const int NE,
             {
                val += r_Bj[k1] * D(k1, e) * r_Gi[k1];
             }
-            if (add)
-            {
-               A(i1, j1, e) += val;
-            }
-            else
-            {
-               A(i1, j1, e) = val;
-            }
+            A(i1, j1, e) += val;
          }
       }
    });
@@ -74,7 +66,6 @@ static void EAConvectionAssemble2D(const int NE,
                                    const Array<double> &g,
                                    const Vector &padata,
                                    Vector &eadata,
-                                   const bool add,
                                    const int d1d = 0,
                                    const int q1d = 0)
 {
@@ -130,14 +121,7 @@ static void EAConvectionAssemble2D(const int NE,
                                * r_B[k1][j1]* r_B[k2][j2];
                      }
                   }
-                  if (add)
-                  {
-                     A(i1, i2, j1, j2, e) += val;
-                  }
-                  else
-                  {
-                     A(i1, i2, j1, j2, e) = val;
-                  }
+                  A(i1, i2, j1, j2, e) += val;
                }
             }
          }
@@ -151,7 +135,6 @@ static void EAConvectionAssemble3D(const int NE,
                                    const Array<double> &g,
                                    const Vector &padata,
                                    Vector &eadata,
-                                   const bool add,
                                    const int d1d = 0,
                                    const int q1d = 0)
 {
@@ -208,14 +191,7 @@ static void EAConvectionAssemble3D(const int NE,
                               }
                            }
                         }
-                        if (add)
-                        {
-                           A(i1, i2, i3, j1, j2, j3, e) += val;
-                        }
-                        else
-                        {
-                           A(i1, i2, i3, j1, j2, j3, e) = val;
-                        }
+                        A(i1, i2, i3, j1, j2, j3, e) += val;
                      }
                   }
                }
@@ -226,8 +202,7 @@ static void EAConvectionAssemble3D(const int NE,
 }
 
 void ConvectionIntegrator::AssembleEA(const FiniteElementSpace &fes,
-                                      Vector &ea_data,
-                                      const bool add)
+                                      Vector &ea_data)
 {
    AssemblePA(fes);
    const int ne = fes.GetMesh()->GetNE();
@@ -237,47 +212,44 @@ void ConvectionIntegrator::AssembleEA(const FiniteElementSpace &fes,
    {
       switch ((dofs1D << 4 ) | quad1D)
       {
-         case 0x22: return EAConvectionAssemble1D<2,2>(ne,B,G,pa_data,ea_data,add);
-         case 0x33: return EAConvectionAssemble1D<3,3>(ne,B,G,pa_data,ea_data,add);
-         case 0x44: return EAConvectionAssemble1D<4,4>(ne,B,G,pa_data,ea_data,add);
-         case 0x55: return EAConvectionAssemble1D<5,5>(ne,B,G,pa_data,ea_data,add);
-         case 0x66: return EAConvectionAssemble1D<6,6>(ne,B,G,pa_data,ea_data,add);
-         case 0x77: return EAConvectionAssemble1D<7,7>(ne,B,G,pa_data,ea_data,add);
-         case 0x88: return EAConvectionAssemble1D<8,8>(ne,B,G,pa_data,ea_data,add);
-         case 0x99: return EAConvectionAssemble1D<9,9>(ne,B,G,pa_data,ea_data,add);
-         default:   return EAConvectionAssemble1D(ne,B,G,pa_data,ea_data,add,
-                                                     dofs1D,quad1D);
+         case 0x22: return EAConvectionAssemble1D<2,2>(ne,B,G,pa_data,ea_data);
+         case 0x33: return EAConvectionAssemble1D<3,3>(ne,B,G,pa_data,ea_data);
+         case 0x44: return EAConvectionAssemble1D<4,4>(ne,B,G,pa_data,ea_data);
+         case 0x55: return EAConvectionAssemble1D<5,5>(ne,B,G,pa_data,ea_data);
+         case 0x66: return EAConvectionAssemble1D<6,6>(ne,B,G,pa_data,ea_data);
+         case 0x77: return EAConvectionAssemble1D<7,7>(ne,B,G,pa_data,ea_data);
+         case 0x88: return EAConvectionAssemble1D<8,8>(ne,B,G,pa_data,ea_data);
+         case 0x99: return EAConvectionAssemble1D<9,9>(ne,B,G,pa_data,ea_data);
+         default:   return EAConvectionAssemble1D(ne,B,G,pa_data,ea_data,dofs1D,quad1D);
       }
    }
    else if (dim == 2)
    {
       switch ((dofs1D << 4 ) | quad1D)
       {
-         case 0x22: return EAConvectionAssemble2D<2,2>(ne,B,G,pa_data,ea_data,add);
-         case 0x33: return EAConvectionAssemble2D<3,3>(ne,B,G,pa_data,ea_data,add);
-         case 0x44: return EAConvectionAssemble2D<4,4>(ne,B,G,pa_data,ea_data,add);
-         case 0x55: return EAConvectionAssemble2D<5,5>(ne,B,G,pa_data,ea_data,add);
-         case 0x66: return EAConvectionAssemble2D<6,6>(ne,B,G,pa_data,ea_data,add);
-         case 0x77: return EAConvectionAssemble2D<7,7>(ne,B,G,pa_data,ea_data,add);
-         case 0x88: return EAConvectionAssemble2D<8,8>(ne,B,G,pa_data,ea_data,add);
-         case 0x99: return EAConvectionAssemble2D<9,9>(ne,B,G,pa_data,ea_data,add);
-         default:   return EAConvectionAssemble2D(ne,B,G,pa_data,ea_data,add,
-                                                     dofs1D,quad1D);
+         case 0x22: return EAConvectionAssemble2D<2,2>(ne,B,G,pa_data,ea_data);
+         case 0x33: return EAConvectionAssemble2D<3,3>(ne,B,G,pa_data,ea_data);
+         case 0x44: return EAConvectionAssemble2D<4,4>(ne,B,G,pa_data,ea_data);
+         case 0x55: return EAConvectionAssemble2D<5,5>(ne,B,G,pa_data,ea_data);
+         case 0x66: return EAConvectionAssemble2D<6,6>(ne,B,G,pa_data,ea_data);
+         case 0x77: return EAConvectionAssemble2D<7,7>(ne,B,G,pa_data,ea_data);
+         case 0x88: return EAConvectionAssemble2D<8,8>(ne,B,G,pa_data,ea_data);
+         case 0x99: return EAConvectionAssemble2D<9,9>(ne,B,G,pa_data,ea_data);
+         default:   return EAConvectionAssemble2D(ne,B,G,pa_data,ea_data,dofs1D,quad1D);
       }
    }
    else if (dim == 3)
    {
       switch ((dofs1D << 4 ) | quad1D)
       {
-         case 0x23: return EAConvectionAssemble3D<2,3>(ne,B,G,pa_data,ea_data,add);
-         case 0x34: return EAConvectionAssemble3D<3,4>(ne,B,G,pa_data,ea_data,add);
-         case 0x45: return EAConvectionAssemble3D<4,5>(ne,B,G,pa_data,ea_data,add);
-         case 0x56: return EAConvectionAssemble3D<5,6>(ne,B,G,pa_data,ea_data,add);
-         case 0x67: return EAConvectionAssemble3D<6,7>(ne,B,G,pa_data,ea_data,add);
-         case 0x78: return EAConvectionAssemble3D<7,8>(ne,B,G,pa_data,ea_data,add);
-         case 0x89: return EAConvectionAssemble3D<8,9>(ne,B,G,pa_data,ea_data,add);
-         default:   return EAConvectionAssemble3D(ne,B,G,pa_data,ea_data,add,
-                                                     dofs1D,quad1D);
+         case 0x23: return EAConvectionAssemble3D<2,3>(ne,B,G,pa_data,ea_data);
+         case 0x34: return EAConvectionAssemble3D<3,4>(ne,B,G,pa_data,ea_data);
+         case 0x45: return EAConvectionAssemble3D<4,5>(ne,B,G,pa_data,ea_data);
+         case 0x56: return EAConvectionAssemble3D<5,6>(ne,B,G,pa_data,ea_data);
+         case 0x67: return EAConvectionAssemble3D<6,7>(ne,B,G,pa_data,ea_data);
+         case 0x78: return EAConvectionAssemble3D<7,8>(ne,B,G,pa_data,ea_data);
+         case 0x89: return EAConvectionAssemble3D<8,9>(ne,B,G,pa_data,ea_data);
+         default:   return EAConvectionAssemble3D(ne,B,G,pa_data,ea_data,dofs1D,quad1D);
       }
    }
    MFEM_ABORT("Unknown kernel.");

fem/bilininteg_convection_pa.cpp

@@ -13,6 +13,10 @@
 #include "bilininteg.hpp"
 #include "gridfunc.hpp"
 
+#include "restriction.hpp"
+#include "tmop_pa.hpp"
+#include "../linalg/kernels.hpp"
+
 using namespace std;
 
 namespace mfem
@@ -68,47 +72,53 @@ static void PAConvectionSetup3D(const int Q1D,
                                 const double alpha,
                                 Vector &op)
 {
-   const int NQ = Q1D*Q1D*Q1D;
-   auto W = w.Read();
-   auto J = Reshape(j.Read(), NQ, 3, 3, NE);
+   const auto W = Reshape(w.Read(), Q1D,Q1D,Q1D);
+   const auto J = Reshape(j.Read(), Q1D,Q1D,Q1D,3,3,NE);
    const bool const_v = vel.Size() == 3;
-   auto V =
-      const_v ? Reshape(vel.Read(), 3,1,1) : Reshape(vel.Read(), 3,NQ,NE);
-   auto y = Reshape(op.Write(), NQ, 3, NE);
-   MFEM_FORALL(e, NE,
+   const auto V = const_v ?
+                  Reshape(vel.Read(), 3,1,1,1,1) :
+                  Reshape(vel.Read(), 3,Q1D,Q1D,Q1D,NE);
+   auto y = Reshape(op.Write(), Q1D,Q1D,Q1D,3,NE);
+   MFEM_FORALL_3D(e, NE, Q1D, Q1D, Q1D,
    {
-      for (int q = 0; q < NQ; ++q)
+      MFEM_FOREACH_THREAD(qx,x,Q1D)
       {
-         const double J11 = J(q,0,0,e);
-         const double J21 = J(q,1,0,e);
-         const double J31 = J(q,2,0,e);
-         const double J12 = J(q,0,1,e);
-         const double J22 = J(q,1,1,e);
-         const double J32 = J(q,2,1,e);
-         const double J13 = J(q,0,2,e);
-         const double J23 = J(q,1,2,e);
-         const double J33 = J(q,2,2,e);
-         const double w = alpha * W[q];
-         const double v0 = const_v ? V(0,0,0) : V(0,q,e);
-         const double v1 = const_v ? V(1,0,0) : V(1,q,e);
-         const double v2 = const_v ? V(2,0,0) : V(2,q,e);
-         const double wx = w * v0;
-         const double wy = w * v1;
-         const double wz = w * v2;
-         // adj(J)
-         const double A11 = (J22 * J33) - (J23 * J32);
-         const double A12 = (J32 * J13) - (J12 * J33);
-         const double A13 = (J12 * J23) - (J22 * J13);
-         const double A21 = (J31 * J23) - (J21 * J33);
-         const double A22 = (J11 * J33) - (J13 * J31);
-         const double A23 = (J21 * J13) - (J11 * J23);
-         const double A31 = (J21 * J32) - (J31 * J22);
-         const double A32 = (J31 * J12) - (J11 * J32);
-         const double A33 = (J11 * J22) - (J12 * J21);
-         // q . J^{-1} = q . adj(J)
-         y(q,0,e) =  wx * A11 + wy * A12 + wz * A13;
-         y(q,1,e) =  wx * A21 + wy * A22 + wz * A23;
-         y(q,2,e) =  wx * A31 + wy * A32 + wz * A33;
+         MFEM_FOREACH_THREAD(qy,y,Q1D)
+         {
+            MFEM_FOREACH_THREAD(qz,z,Q1D)
+            {
+               const double J11 = J(qx,qy,qz,0,0,e);
+               const double J12 = J(qx,qy,qz,0,1,e);
+               const double J13 = J(qx,qy,qz,0,2,e);
+               const double J21 = J(qx,qy,qz,1,0,e);
+               const double J22 = J(qx,qy,qz,1,1,e);
+               const double J23 = J(qx,qy,qz,1,2,e);
+               const double J31 = J(qx,qy,qz,2,0,e);
+               const double J32 = J(qx,qy,qz,2,1,e);
+               const double J33 = J(qx,qy,qz,2,2,e);
+               const double w = alpha * W(qx,qy,qz);
+               const double v0 = const_v ? V(0,0,0,0,0) : V(0,qx,qy,qz,e);
+               const double v1 = const_v ? V(1,0,0,0,0) : V(1,qx,qy,qz,e);
+               const double v2 = const_v ? V(2,0,0,0,0) : V(2,qx,qy,qz,e);
+               const double wx = w * v0;
+               const double wy = w * v1;
+               const double wz = w * v2;
+               // adj(J)
+               const double A11 = (J22 * J33) - (J23 * J32);
+               const double A12 = (J32 * J13) - (J12 * J33);
+               const double A13 = (J12 * J23) - (J22 * J13);
+               const double A21 = (J31 * J23) - (J21 * J33);
+               const double A22 = (J11 * J33) - (J13 * J31);
+               const double A23 = (J21 * J13) - (J11 * J23);
+               const double A31 = (J21 * J32) - (J31 * J22);
+               const double A32 = (J31 * J12) - (J11 * J32);
+               const double A33 = (J11 * J22) - (J12 * J21);
+               // q . J^{-1} = q . adj(J)
+               y(qx,qy,qz,0,e) = wx * A11 + wy * A12 + wz * A13;
+               y(qx,qy,qz,1,e) = wx * A21 + wy * A22 + wz * A23;
+               y(qx,qy,qz,2,e) = wx * A31 + wy * A32 + wz * A33;
+            }
+         }
       }
    });
 }
@@ -184,8 +194,8 @@ void PAConvectionApply2D(const int ne,
             Gu[dy][qx] = 0.0;
             for (int dx = 0; dx < D1D; ++dx)
             {
-               const double bx  = B(qx,dx);
-               const double gx  = G(qx,dx);
+               const double bx = B(qx,dx);
+               const double gx = G(qx,dx);
                const double x = u[dy][dx];
                Bu[dy][qx] += bx * x;
                Gu[dy][qx] += gx * x;
@@ -202,8 +212,8 @@ void PAConvectionApply2D(const int ne,
             BGu[qy][qx] = 0.0;
             for (int dy = 0; dy < D1D; ++dy)
             {
-               const double bx  = B(qy,dy);
-               const double gx  = G(qy,dy);
+               const double bx = B(qy,dy);
+               const double gx = G(qy,dy);
                GBu[qy][qx] += gx * Bu[dy][qx];
                BGu[qy][qx] += bx * Gu[dy][qx];
             }
@@ -232,7 +242,7 @@ void PAConvectionApply2D(const int ne,
             BDGu[dy][qx] = 0.0;
             for (int qy = 0; qy < Q1D; ++qy)
             {
-               const double w  = Bt(dy,qy);
+               const double w = Bt(dy,qy);
                BDGu[dy][qx] += w * DGu[qy][qx];
             }
          }
@@ -244,7 +254,7 @@ void PAConvectionApply2D(const int ne,
             double BBDGu = 0.0;
             for (int qx = 0; qx < Q1D; ++qx)
             {
-               const double w  = Bt(dx,qx);
+               const double w = Bt(dx,qx);
                BBDGu += w * BDGu[dy][qx];
             }
             y(dx,dy,e) += BBDGu;
@@ -310,7 +320,7 @@ void SmemPAConvectionApply2D(const int ne,
             {
                const double bx = B(qx,dx);
                const double gx = G(qx,dx);
-               const double x  = u[tidz][dy][dx];
+               const double x = u[tidz][dy][dx];
                Bu[tidz][dy][qx] += bx * x;
                Gu[tidz][dy][qx] += gx * x;
             }
@@ -327,8 +337,8 @@ void SmemPAConvectionApply2D(const int ne,
             BGu[tidz][qy][qx] = 0.0;
             for (int dy = 0; dy < D1D; ++dy)
             {
-               const double bx  = B(qy,dy);
-               const double gx  = G(qy,dy);
+               const double bx = B(qy,dy);
+               const double gx = G(qy,dy);
                GBu[tidz][qy][qx] += gx * Bu[tidz][dy][qx];
                BGu[tidz][qy][qx] += bx * Gu[tidz][dy][qx];
             }
@@ -359,7 +369,7 @@ void SmemPAConvectionApply2D(const int ne,
             BDGu[tidz][dy][qx] = 0.0;
             for (int qy = 0; qy < Q1D; ++qy)
             {
-               const double w  = Bt(dy,qy);
+               const double w = Bt(dy,qy);
                BDGu[tidz][dy][qx] += w * DGu[tidz][qy][qx];
             }
          }
@@ -372,7 +382,7 @@ void SmemPAConvectionApply2D(const int ne,
             double BBDGu = 0.0;
             for (int qx = 0; qx < Q1D; ++qx)
             {
-               const double w  = Bt(dx,qx);
+               const double w = Bt(dx,qx);
                BBDGu += w * BDGu[tidz][dy][qx];
             }
             y(dx,dy,e) += BBDGu;
@@ -436,8 +446,8 @@ void PAConvectionApply3D(const int ne,
                Gu[dz][dy][qx] = 0.0;
                for (int dx = 0; dx < D1D; ++dx)
                {
-                  const double bx  = B(qx,dx);
-                  const double gx  = G(qx,dx);
+                  const double bx = B(qx,dx);
+                  const double gx = G(qx,dx);
                   const double x = u[dz][dy][dx];
                   Bu[dz][dy][qx] += bx * x;
                   Gu[dz][dy][qx] += gx * x;
@@ -459,8 +469,8 @@ void PAConvectionApply3D(const int ne,
                BGu[dz][qy][qx] = 0.0;
                for (int dy = 0; dy < D1D; ++dy)
                {
-                  const double bx  = B(qy,dy);
-                  const double gx  = G(qy,dy);
+                  const double bx = B(qy,dy);
+                  const double gx = G(qy,dy);
                   BBu[dz][qy][qx] += bx * Bu[dz][dy][qx];
                   GBu[dz][qy][qx] += gx * Bu[dz][dy][qx];
                   BGu[dz][qy][qx] += bx * Gu[dz][dy][qx];
@@ -482,8 +492,8 @@ void PAConvectionApply3D(const int ne,
                BBGu[qz][qy][qx] = 0.0;
                for (int dz = 0; dz < D1D; ++dz)
                {
-                  const double bx  = B(qz,dz);
-                  const double gx  = G(qz,dz);
+                  const double bx = B(qz,dz);
+                  const double gx = G(qz,dz);
                   GBBu[qz][qy][qx] += gx * BBu[dz][qy][qx];
                   BGBu[qz][qy][qx] += bx * GBu[dz][qy][qx];
                   BBGu[qz][qy][qx] += bx * BGu[dz][qy][qx];
@@ -521,7 +531,7 @@ void PAConvectionApply3D(const int ne,
                BDGu[dz][qy][qx] = 0.0;
                for (int qz = 0; qz < Q1D; ++qz)
                {
-                  const double w  = Bt(dz,qz);
+                  const double w = Bt(dz,qz);
                   BDGu[dz][qy][qx] += w * DGu[qz][qy][qx];
                }
             }
@@ -537,7 +547,7 @@ void PAConvectionApply3D(const int ne,
                BBDGu[dz][dy][qx] = 0.0;
                for (int qy = 0; qy < Q1D; ++qy)
                {
-                  const double w  = Bt(dy,qy);
+                  const double w = Bt(dy,qy);
                   BBDGu[dz][dy][qx] += w * BDGu[dz][qy][qx];
                }
             }
@@ -552,7 +562,7 @@ void PAConvectionApply3D(const int ne,
                double BBBDGu = 0.0;
                for (int qx = 0; qx < Q1D; ++qx)
                {
-                  const double w  = Bt(dx,qx);
+                  const double w = Bt(dx,qx);
                   BBBDGu += w * BBDGu[dz][dy][qx];
                }
                y(dx,dy,dz,e) += BBBDGu;
@@ -625,8 +635,8 @@ void SmemPAConvectionApply3D(const int ne,
                double Gu_ = 0.0;
                for (int dx = 0; dx < D1D; ++dx)
                {
-                  const double bx  = B(qx,dx);
-                  const double gx  = G(qx,dx);
+                  const double bx = B(qx,dx);
+                  const double gx = G(qx,dx);
                   const double x = u[dz][dy][dx];
                   Bu_ += bx * x;
                   Gu_ += gx * x;
@@ -651,8 +661,8 @@ void SmemPAConvectionApply3D(const int ne,
                double BGu_ = 0.0;
                for (int dy = 0; dy < D1D; ++dy)
                {
-                  const double bx  = B(qy,dy);
-                  const double gx  = G(qy,dy);
+                  const double bx = B(qy,dy);
+                  const double gx = G(qy,dy);
                   BBu_ += bx * Bu[dz][dy][qx];
                   GBu_ += gx * Bu[dz][dy][qx];
                   BGu_ += bx * Gu[dz][dy][qx];
@@ -678,8 +688,8 @@ void SmemPAConvectionApply3D(const int ne,
                double BBGu_ = 0.0;
                for (int dz = 0; dz < D1D; ++dz)
                {
-                  const double bx  = B(qz,dz);
-                  const double gx  = G(qz,dz);
+                  const double bx = B(qz,dz);
+                  const double gx = G(qz,dz);
                   GBBu_ += gx * BBu[dz][qy][qx];
                   BGBu_ += bx * GBu[dz][qy][qx];
                   BBGu_ += bx * BGu[dz][qy][qx];
@@ -721,7 +731,7 @@ void SmemPAConvectionApply3D(const int ne,
                double BDGu_ = 0.0;
                for (int qz = 0; qz < Q1D; ++qz)
                {
-                  const double w  = Bt(dz,qz);
+                  const double w = Bt(dz,qz);
                   BDGu_ += w * DGu[qz][qy][qx];
                }
                BDGu[dz][qy][qx] = BDGu_;
@@ -739,7 +749,7 @@ void SmemPAConvectionApply3D(const int ne,
                double BBDGu_ = 0.0;
                for (int qy = 0; qy < Q1D; ++qy)
                {
-                  const double w  = Bt(dy,qy);
+                  const double w = Bt(dy,qy);
                   BBDGu_ += w * BDGu[dz][qy][qx];
                }
                BBDGu[dz][dy][qx] = BBDGu_;
@@ -756,7 +766,7 @@ void SmemPAConvectionApply3D(const int ne,
                double BBBDGu = 0.0;
                for (int qx = 0; qx < Q1D; ++qx)
                {
-                  const double w  = Bt(dx,qx);
+                  const double w = Bt(dx,qx);
                   BBBDGu += w * BBDGu[dz][dy][qx];
                }
                y(dx,dy,dz,e) = BBBDGu;
@@ -766,6 +776,117 @@ void SmemPAConvectionApply3D(const int ne,
    });
 }
 
+template<int T_VDIM = 0, int T_D1D = 0, int T_Q1D = 0, int T_MAX = 0>
+static void QEvalVGF2D(const int NE,
+                       const double *b_,
+                       const double *x_,
+                       double *y_,
+                       const int vdim = 1,
+                       const int d1d = 0,
+                       const int q1d = 0)
+{
+   const int D1D = T_D1D ? T_D1D : d1d;
+   const int Q1D = T_Q1D ? T_Q1D : q1d;
+   const int VDIM = T_VDIM ? T_VDIM : vdim;
+
+   const auto b = Reshape(b_, Q1D, D1D);
+   const auto X = Reshape(x_, D1D, D1D, VDIM, NE);
+   auto C = Reshape(y_, VDIM, Q1D, Q1D, NE);
+
+   MFEM_FORALL_2D(e, NE, Q1D, Q1D, 1,
+   {
+      constexpr int NBZ = 1;
+      constexpr int MQ1 = T_Q1D ? T_Q1D : T_MAX;
+      constexpr int MD1 = T_D1D ? T_D1D : T_MAX;
+
+      const int D1D = T_D1D ? T_D1D : d1d;
+      const int Q1D = T_Q1D ? T_Q1D : q1d;
+      const int VDIM = T_VDIM ? T_VDIM : vdim;
+
+      MFEM_SHARED double B[MQ1*MD1];
+      mfem::kernels::LoadB<MD1,MQ1>(D1D,Q1D,b,B);
+
+      MFEM_SHARED double DD[NBZ][MD1*MD1];
+      MFEM_SHARED double DQ[NBZ][MD1*MQ1];
+      MFEM_SHARED double QQ[NBZ][MQ1*MQ1];
+
+      for (int c = 0; c < VDIM; c++)
+      {
+         mfem::kernels::LoadX<MD1,NBZ>(e,D1D,c,X,DD);
+         mfem::kernels::EvalX<MD1,MQ1,NBZ>(D1D,Q1D,B,DD,DQ);
+         mfem::kernels::EvalY<MD1,MQ1,NBZ>(D1D,Q1D,B,DQ,QQ);
+
+         MFEM_FOREACH_THREAD(qx,x,Q1D)
+         {
+            MFEM_FOREACH_THREAD(qy,y,Q1D)
+            {
+               double G;
+               mfem::kernels::PullEval<MQ1,NBZ>(qx,qy,QQ,G);
+               C(c,qx,qy,e) = G;
+            }
+         }
+         MFEM_SYNC_THREAD;
+      }
+   });
+}
+
+template<int T_VDIM = 0, int T_D1D = 0, int T_Q1D = 0, int T_MAX = 0>
+static void QEvalVGF3D(const int NE,
+                       const double *b_,
+                       const double *x_,
+                       double *y_,
+                       const int vdim = 1,
+                       const int d1d = 0,
+                       const int q1d = 0)
+{
+   const int D1D = T_D1D ? T_D1D : d1d;
+   const int Q1D = T_Q1D ? T_Q1D : q1d;
+   const int VDIM = T_VDIM ? T_VDIM : vdim;
+
+   const auto b = Reshape(b_, Q1D, D1D);
+   const auto X = Reshape(x_, D1D, D1D, D1D, VDIM, NE);
+   auto C = Reshape(y_, VDIM, Q1D, Q1D, Q1D, NE);
+
+   MFEM_FORALL_3D(e, NE, Q1D, Q1D, Q1D,
+   {
+      const int D1D = T_D1D ? T_D1D : d1d;
+      const int Q1D = T_Q1D ? T_Q1D : q1d;
+      const int VDIM = T_VDIM ? T_VDIM : vdim;
+      constexpr int MQ1 = T_Q1D ? T_Q1D : T_MAX;
+      constexpr int MD1 = T_D1D ? T_D1D : T_MAX;
+
+      MFEM_SHARED double B[MQ1*MD1];
+      mfem::kernels::LoadB<MD1,MQ1>(D1D,Q1D,b,B);
+
+      MFEM_SHARED double DDD[MD1*MD1*MD1];
+      MFEM_SHARED double DDQ[MD1*MD1*MQ1];
+      MFEM_SHARED double DQQ[MD1*MQ1*MQ1];
+      MFEM_SHARED double QQQ[MQ1*MQ1*MQ1];
+
+      for (int c = 0; c < VDIM; c++)
+      {
+         mfem::kernels::LoadX<MD1>(e,D1D,c,X,DDD);
+         mfem::kernels::EvalX<MD1,MQ1>(D1D,Q1D,B,DDD,DDQ);
+         mfem::kernels::EvalY<MD1,MQ1>(D1D,Q1D,B,DDQ,DQQ);
+         mfem::kernels::EvalZ<MD1,MQ1>(D1D,Q1D,B,DQQ,QQQ);
+
+         MFEM_FOREACH_THREAD(qx,x,Q1D)
+         {
+            MFEM_FOREACH_THREAD(qy,y,Q1D)
+            {
+               MFEM_FOREACH_THREAD(qz,z,Q1D)
+               {
+                  double G;
+                  mfem::kernels::PullEval<MQ1>(qx,qy,qz,QQQ,G);
+                  C(c,qx,qy,qz,e) = G;
+               }
+            }
+         }
+         MFEM_SYNC_THREAD;
+      }
+   });
+}
+
 void ConvectionIntegrator::AssemblePA(const FiniteElementSpace &fes)
 {
    // Assumes tensor-product elements
@@ -778,16 +899,90 @@ void ConvectionIntegrator::AssemblePA(const FiniteElementSpace &fes)
    const int nq = ir->GetNPoints();
    dim = mesh->Dimension();
    ne = fes.GetNE();
-   geom = mesh->GetGeometricFactors(*ir, GeometricFactors::JACOBIANS);
-   maps = &el.GetDofToQuad(*ir, DofToQuad::TENSOR);
+   const DofToQuad::Mode mode = DofToQuad::TENSOR;
+#ifdef MFEM_USE_UMPIRE
+   const MemoryType temp_type = Device::GetDeviceMemoryType() == MemoryType::DEVICE_UMPIRE
+      ? MemoryType::DEVICE_UMPIRE_2 : Device::GetDeviceMemoryType();
+#else
+   const MemoryType temp_type = Device::GetDeviceMemoryType();
+#endif
+
+   geom = mesh->GetGeometricFactors(*ir, GeometricFactors::JACOBIANS, mode, temp_type);
+   maps = &el.GetDofToQuad(*ir, mode);
    dofs1D = maps->ndof;
    quad1D = maps->nqpt;
-   pa_data.SetSize(symmDims * nq * ne, Device::GetMemoryType());
+   pa_data.SetSize(symmDims * nq * ne, temp_type);
    Vector vel;
-   if (VectorConstantCoefficient *cQ = dynamic_cast<VectorConstantCoefficient*>(Q))
+   if (VectorConstantCoefficient *cQ =
+          dynamic_cast<VectorConstantCoefficient*>(Q))
    {
       vel = cQ->GetVec();
    }
+   else if (VectorGridFunctionCoefficient *vgfQ =
+               dynamic_cast<VectorGridFunctionCoefficient*>(Q))
+   {
+      Vector xe;
+      vel.SetSize(dim * nq * ne, temp_type);
+
+      const GridFunction *gf = vgfQ->GetGridFunction();
+      const ElementDofOrdering ordering = ElementDofOrdering::LEXICOGRAPHIC;
+      const FiniteElementSpace &gf_fes = *gf->FESpace();
+
+      const int vdim = gf_fes.GetVDim();
+      const Operator *R = gf_fes.GetElementRestriction(ordering);
+      const FiniteElement &el_gf = *gf_fes.GetFE(0);
+      const DofToQuad *maps_gf = &el_gf.GetDofToQuad(*ir, mode);
+      const int D1D = maps_gf->ndof;
+      const int Q1D = maps_gf->nqpt;
+
+      MFEM_VERIFY(R,"");
+      MFEM_VERIFY(vdim == dim, "");
+      MFEM_VERIFY(dim==2 || dim==3,"");
+
+      xe.SetSize(R->Height(), Device::GetMemoryType());
+      xe.UseDevice(true);
+      R->Mult(*gf, xe);
+
+      const auto B = maps_gf->B.Read();
+      const auto x = xe.Read();
+      auto y = vel.Write();
+
+      const int id = (D1D << 4 ) | Q1D;
+      if (dim == 2)
+      {
+         switch (id)
+         {
+            case 0x22: QEvalVGF2D<2,2,2>(ne,B,x,y); break;
+            case 0x33: QEvalVGF2D<2,3,3>(ne,B,x,y); break;
+            case 0x34: QEvalVGF2D<2,3,4>(ne,B,x,y); break;
+            default:
+            {
+               constexpr int MAX_DQ = 8;
+               MFEM_VERIFY(D1D <= MAX_DQ, "");
+               MFEM_VERIFY(Q1D <= MAX_DQ, "");
+               QEvalVGF2D<0,0,0,MAX_DQ>(ne,B,x,y,vdim,D1D,Q1D);
+            }
+         }
+      }
+      if (dim == 3)
+      {
+         switch (id)
+         {
+            case 0x23: QEvalVGF3D<3,2,3>(ne,B,x,y); break;
+            case 0x34: QEvalVGF3D<3,3,4>(ne,B,x,y); break;
+            case 0x35: QEvalVGF3D<3,3,5>(ne,B,x,y); break;
+            case 0x46: QEvalVGF3D<3,4,6>(ne,B,x,y); break;
+            case 0x48: QEvalVGF3D<3,4,8>(ne,B,x,y); break;
+            default:
+            {
+               constexpr int MAX_DQ = 6;
+               MFEM_VERIFY(D1D <= MAX_DQ, "");
+               MFEM_VERIFY(Q1D <= MAX_DQ, "");
+               QEvalVGF3D<0,0,0,MAX_DQ>(ne,B,x,y,vdim,D1D,Q1D);
+            }
+         }
+      }
+   }
    else if (VectorQuadratureFunctionCoefficient* cQ =
                dynamic_cast<VectorQuadratureFunctionCoefficient*>(Q))
    {
@@ -841,9 +1036,12 @@ static void PAConvectionApply(const int dim,
       switch ((D1D << 4 ) | Q1D)
       {
          case 0x22: return SmemPAConvectionApply2D<2,2,8>(NE,B,G,Bt,Gt,op,x,y);
-         case 0x33: return SmemPAConvectionApply2D<3,3,3>(NE,B,G,Bt,Gt,op,x,y);
-         case 0x44: return SmemPAConvectionApply2D<4,4,2>(NE,B,G,Bt,Gt,op,x,y);
+         case 0x33: return SmemPAConvectionApply2D<3,3,4>(NE,B,G,Bt,Gt,op,x,y);
+         case 0x34: return SmemPAConvectionApply2D<3,4,4>(NE,B,G,Bt,Gt,op,x,y);
+         case 0x44: return SmemPAConvectionApply2D<4,4,4>(NE,B,G,Bt,Gt,op,x,y);
+         case 0x46: return SmemPAConvectionApply2D<4,6,4>(NE,B,G,Bt,Gt,op,x,y);
          case 0x55: return SmemPAConvectionApply2D<5,5,2>(NE,B,G,Bt,Gt,op,x,y);
+         case 0x58: return SmemPAConvectionApply2D<5,8,2>(NE,B,G,Bt,Gt,op,x,y);
          case 0x66: return SmemPAConvectionApply2D<6,6,1>(NE,B,G,Bt,Gt,op,x,y);
          case 0x77: return SmemPAConvectionApply2D<7,7,1>(NE,B,G,Bt,Gt,op,x,y);
          case 0x88: return SmemPAConvectionApply2D<8,8,1>(NE,B,G,Bt,Gt,op,x,y);
@@ -856,8 +1054,12 @@ static void PAConvectionApply(const int dim,
       switch ((D1D << 4 ) | Q1D)
       {
          case 0x23: return SmemPAConvectionApply3D<2,3>(NE,B,G,Bt,Gt,op,x,y);
+         case 0x24: return SmemPAConvectionApply3D<2,4>(NE,B,G,Bt,Gt,op,x,y);
+         case 0x26: return SmemPAConvectionApply3D<2,6>(NE,B,G,Bt,Gt,op,x,y);
          case 0x34: return SmemPAConvectionApply3D<3,4>(NE,B,G,Bt,Gt,op,x,y);
+         case 0x35: return SmemPAConvectionApply3D<3,5>(NE,B,G,Bt,Gt,op,x,y);
          case 0x45: return SmemPAConvectionApply3D<4,5>(NE,B,G,Bt,Gt,op,x,y);
+         case 0x48: return SmemPAConvectionApply3D<4,8>(NE,B,G,Bt,Gt,op,x,y);
          case 0x56: return SmemPAConvectionApply3D<5,6>(NE,B,G,Bt,Gt,op,x,y);
          case 0x67: return SmemPAConvectionApply3D<6,7>(NE,B,G,Bt,Gt,op,x,y);
          case 0x78: return SmemPAConvectionApply3D<7,8>(NE,B,G,Bt,Gt,op,x,y);

fem/bilininteg_dgtrace_ea.cpp

@@ -20,8 +20,7 @@ static void EADGTraceAssemble1DInt(const int NF,
                                    const Array<double> &basis,
                                    const Vector &padata,
                                    Vector &eadata_int,
-                                   Vector &eadata_ext,
-                                   const bool add)
+                                   Vector &eadata_ext)
 {
    auto D = Reshape(padata.Read(), 2, 2, NF);
    auto A_int = Reshape(eadata_int.ReadWrite(), 2, NF);
@@ -33,41 +32,23 @@ static void EADGTraceAssemble1DInt(const int NF,
       val_ext10 = D(1, 0, f);
       val_ext01 = D(0, 1, f);
       val_int1  = D(1, 1, f);
-      if (add)
-      {
-         A_int(0, f) += val_int0;
-         A_int(1, f) += val_int1;
-         A_ext(0, f) += val_ext01;
-         A_ext(1, f) += val_ext10;
-      }
-      else
-      {
-         A_int(0, f) = val_int0;
-         A_int(1, f) = val_int1;
-         A_ext(0, f) = val_ext01;
-         A_ext(1, f) = val_ext10;
-      }
+      A_int(0, f) += val_int0;
+      A_int(1, f) += val_int1;
+      A_ext(0, f) += val_ext01;
+      A_ext(1, f) += val_ext10;
    });
 }
 
 static void EADGTraceAssemble1DBdr(const int NF,
                                    const Array<double> &basis,
                                    const Vector &padata,
-                                   Vector &eadata_bdr,
-                                   const bool add)
+                                   Vector &eadata_bdr)
 {
    auto D = Reshape(padata.Read(), 2, 2, NF);
    auto A_bdr = Reshape(eadata_bdr.ReadWrite(), NF);
    MFEM_FORALL(f, NF,
    {
-      if (add)
-      {
-         A_bdr(f) += D(0, 0, f);
-      }
-      else
-      {
-         A_bdr(f) = D(0, 0, f);
-      }
+      A_bdr(f) += D(0, 0, f);
    });
 }
 
@@ -77,7 +58,6 @@ static void EADGTraceAssemble2DInt(const int NF,
                                    const Vector &padata,
                                    Vector &eadata_int,
                                    Vector &eadata_ext,
-                                   const bool add,
                                    const int d1d = 0,
                                    const int q1d = 0)
 {
@@ -108,20 +88,10 @@ static void EADGTraceAssemble2DInt(const int NF,
                val_ext10 += B(k1,i1) * B(k1,j1) * D(k1, 1, 0, f);
                val_int1  += B(k1,i1) * B(k1,j1) * D(k1, 1, 1, f);
             }
-            if (add)
-            {
-               A_int(i1, j1, 0, f) += val_int0;
-               A_int(i1, j1, 1, f) += val_int1;
-               A_ext(i1, j1, 0, f) += val_ext01;
-               A_ext(i1, j1, 1, f) += val_ext10;
-            }
-            else
-            {
-               A_int(i1, j1, 0, f) = val_int0;
-               A_int(i1, j1, 1, f) = val_int1;
-               A_ext(i1, j1, 0, f) = val_ext01;
-               A_ext(i1, j1, 1, f) = val_ext10;
-            }
+            A_int(i1, j1, 0, f) += val_int0;
+            A_int(i1, j1, 1, f) += val_int1;
+            A_ext(i1, j1, 0, f) += val_ext01;
+            A_ext(i1, j1, 1, f) += val_ext10;
          }
       }
    });
@@ -132,7 +102,6 @@ static void EADGTraceAssemble2DBdr(const int NF,
                                    const Array<double> &basis,
                                    const Vector &padata,
                                    Vector &eadata_bdr,
-                                   const bool add,
                                    const int d1d = 0,
                                    const int q1d = 0)
 {
@@ -156,14 +125,7 @@ static void EADGTraceAssemble2DBdr(const int NF,
             {
                val_bdr  += B(k1,i1) * B(k1,j1) * D(k1, 0, 0, f);
             }
-            if (add)
-            {
-               A_bdr(i1, j1, f) += val_bdr;
-            }
-            else
-            {
-               A_bdr(i1, j1, f) = val_bdr;
-            }
+            A_bdr(i1, j1, f) += val_bdr;
          }
       }
    });
@@ -175,7 +137,6 @@ static void EADGTraceAssemble3DInt(const int NF,
                                    const Vector &padata,
                                    Vector &eadata_int,
                                    Vector &eadata_ext,
-                                   const bool add,
                                    const int d1d = 0,
                                    const int q1d = 0)
 {
@@ -246,20 +207,10 @@ static void EADGTraceAssemble3DInt(const int NF,
                                     * s_D[k1][k2][1][0];
                      }
                   }
-                  if (add)
-                  {
-                     A_int(i1, i2, j1, j2, 0, f) += val_int0;
-                     A_int(i1, i2, j1, j2, 1, f) += val_int1;
-                     A_ext(i1, i2, j1, j2, 0, f) += val_ext01;
-                     A_ext(i1, i2, j1, j2, 1, f) += val_ext10;
-                  }
-                  else
-                  {
-                     A_int(i1, i2, j1, j2, 0, f) = val_int0;
-                     A_int(i1, i2, j1, j2, 1, f) = val_int1;
-                     A_ext(i1, i2, j1, j2, 0, f) = val_ext01;
-                     A_ext(i1, i2, j1, j2, 1, f) = val_ext10;
-                  }
+                  A_int(i1, i2, j1, j2, 0, f) += val_int0;
+                  A_int(i1, i2, j1, j2, 1, f) += val_int1;
+                  A_ext(i1, i2, j1, j2, 0, f) += val_ext01;
+                  A_ext(i1, i2, j1, j2, 1, f) += val_ext10;
                }
             }
          }
@@ -272,7 +223,6 @@ static void EADGTraceAssemble3DBdr(const int NF,
                                    const Array<double> &basis,
                                    const Vector &padata,
                                    Vector &eadata_bdr,
-                                   const bool add,
                                    const int d1d = 0,
                                    const int q1d = 0)
 {
@@ -330,14 +280,7 @@ static void EADGTraceAssemble3DBdr(const int NF,
                                    * s_D[k1][k2][0][0];
                      }
                   }
-                  if (add)
-                  {
-                     A_bdr(i1, i2, j1, j2, f) += val_bdr;
-                  }
-                  else
-                  {
-                     A_bdr(i1, i2, j1, j2, f) = val_bdr;
-                  }
+                  A_bdr(i1, i2, j1, j2, f) += val_bdr;
                }
             }
          }
@@ -347,8 +290,7 @@ static void EADGTraceAssemble3DBdr(const int NF,
 
 void DGTraceIntegrator::AssembleEAInteriorFaces(const FiniteElementSpace& fes,
                                                 Vector &ea_data_int,
-                                                Vector &ea_data_ext,
-                                                const bool add)
+                                                Vector &ea_data_ext)
 {
    SetupPA(fes, FaceType::Interior);
    nf = fes.GetNFbyType(FaceType::Interior);
@@ -356,7 +298,7 @@ void DGTraceIntegrator::AssembleEAInteriorFaces(const FiniteElementSpace& fes,
    const Array<double> &B = maps->B;
    if (dim == 1)
    {
-      return EADGTraceAssemble1DInt(nf,B,pa_data,ea_data_int,ea_data_ext,add);
+      return EADGTraceAssemble1DInt(nf,B,pa_data,ea_data_int,ea_data_ext);
    }
    else if (dim == 2)
    {
@@ -364,31 +306,31 @@ void DGTraceIntegrator::AssembleEAInteriorFaces(const FiniteElementSpace& fes,
       {
          case 0x22:
             return EADGTraceAssemble2DInt<2,2>(nf,B,pa_data,ea_data_int,
-                                               ea_data_ext,add);
+                                               ea_data_ext);
          case 0x33:
             return EADGTraceAssemble2DInt<3,3>(nf,B,pa_data,ea_data_int,
-                                               ea_data_ext,add);
+                                               ea_data_ext);
          case 0x44:
             return EADGTraceAssemble2DInt<4,4>(nf,B,pa_data,ea_data_int,
-                                               ea_data_ext,add);
+                                               ea_data_ext);
          case 0x55:
             return EADGTraceAssemble2DInt<5,5>(nf,B,pa_data,ea_data_int,
-                                               ea_data_ext,add);
+                                               ea_data_ext);
          case 0x66:
             return EADGTraceAssemble2DInt<6,6>(nf,B,pa_data,ea_data_int,
-                                               ea_data_ext,add);
+                                               ea_data_ext);
          case 0x77:
             return EADGTraceAssemble2DInt<7,7>(nf,B,pa_data,ea_data_int,
-                                               ea_data_ext,add);
+                                               ea_data_ext);
          case 0x88:
             return EADGTraceAssemble2DInt<8,8>(nf,B,pa_data,ea_data_int,
-                                               ea_data_ext,add);
+                                               ea_data_ext);
          case 0x99:
             return EADGTraceAssemble2DInt<9,9>(nf,B,pa_data,ea_data_int,
-                                               ea_data_ext,add);
+                                               ea_data_ext);
          default:
             return EADGTraceAssemble2DInt(nf,B,pa_data,ea_data_int,
-                                          ea_data_ext,add,dofs1D,quad1D);
+                                          ea_data_ext,dofs1D,quad1D);
       }
    }
    else if (dim == 3)
@@ -397,36 +339,35 @@ void DGTraceIntegrator::AssembleEAInteriorFaces(const FiniteElementSpace& fes,
       {
          case 0x23:
             return EADGTraceAssemble3DInt<2,3>(nf,B,pa_data,ea_data_int,
-                                               ea_data_ext,add);
+                                               ea_data_ext);
          case 0x34:
             return EADGTraceAssemble3DInt<3,4>(nf,B,pa_data,ea_data_int,
-                                               ea_data_ext,add);
+                                               ea_data_ext);
          case 0x45:
             return EADGTraceAssemble3DInt<4,5>(nf,B,pa_data,ea_data_int,
-                                               ea_data_ext,add);
+                                               ea_data_ext);
          case 0x56:
             return EADGTraceAssemble3DInt<5,6>(nf,B,pa_data,ea_data_int,
-                                               ea_data_ext,add);
+                                               ea_data_ext);
          case 0x67:
             return EADGTraceAssemble3DInt<6,7>(nf,B,pa_data,ea_data_int,
-                                               ea_data_ext,add);
+                                               ea_data_ext);
          case 0x78:
             return EADGTraceAssemble3DInt<7,8>(nf,B,pa_data,ea_data_int,
-                                               ea_data_ext,add);
+                                               ea_data_ext);
          case 0x89:
             return EADGTraceAssemble3DInt<8,9>(nf,B,pa_data,ea_data_int,
-                                               ea_data_ext,add);
+                                               ea_data_ext);
          default:
             return EADGTraceAssemble3DInt(nf,B,pa_data,ea_data_int,
-                                          ea_data_ext,add,dofs1D,quad1D);
+                                          ea_data_ext,dofs1D,quad1D);
       }
    }
    MFEM_ABORT("Unknown kernel.");
 }
 
 void DGTraceIntegrator::AssembleEABoundaryFaces(const FiniteElementSpace& fes,
-                                                Vector &ea_data_bdr,
-                                                const bool add)
+                                                Vector &ea_data_bdr)
 {
    SetupPA(fes, FaceType::Boundary);
    nf = fes.GetNFbyType(FaceType::Boundary);
@@ -434,37 +375,37 @@ void DGTraceIntegrator::AssembleEABoundaryFaces(const FiniteElementSpace& fes,
    const Array<double> &B = maps->B;
    if (dim == 1)
    {
-      return EADGTraceAssemble1DBdr(nf,B,pa_data,ea_data_bdr,add);
+      return EADGTraceAssemble1DBdr(nf,B,pa_data,ea_data_bdr);
    }
    else if (dim == 2)
    {
       switch ((dofs1D << 4 ) | quad1D)
       {
-         case 0x22: return EADGTraceAssemble2DBdr<2,2>(nf,B,pa_data,ea_data_bdr,add);
-         case 0x33: return EADGTraceAssemble2DBdr<3,3>(nf,B,pa_data,ea_data_bdr,add);
-         case 0x44: return EADGTraceAssemble2DBdr<4,4>(nf,B,pa_data,ea_data_bdr,add);
-         case 0x55: return EADGTraceAssemble2DBdr<5,5>(nf,B,pa_data,ea_data_bdr,add);
-         case 0x66: return EADGTraceAssemble2DBdr<6,6>(nf,B,pa_data,ea_data_bdr,add);
-         case 0x77: return EADGTraceAssemble2DBdr<7,7>(nf,B,pa_data,ea_data_bdr,add);
-         case 0x88: return EADGTraceAssemble2DBdr<8,8>(nf,B,pa_data,ea_data_bdr,add);
-         case 0x99: return EADGTraceAssemble2DBdr<9,9>(nf,B,pa_data,ea_data_bdr,add);
+         case 0x22: return EADGTraceAssemble2DBdr<2,2>(nf,B,pa_data,ea_data_bdr);
+         case 0x33: return EADGTraceAssemble2DBdr<3,3>(nf,B,pa_data,ea_data_bdr);
+         case 0x44: return EADGTraceAssemble2DBdr<4,4>(nf,B,pa_data,ea_data_bdr);
+         case 0x55: return EADGTraceAssemble2DBdr<5,5>(nf,B,pa_data,ea_data_bdr);
+         case 0x66: return EADGTraceAssemble2DBdr<6,6>(nf,B,pa_data,ea_data_bdr);
+         case 0x77: return EADGTraceAssemble2DBdr<7,7>(nf,B,pa_data,ea_data_bdr);
+         case 0x88: return EADGTraceAssemble2DBdr<8,8>(nf,B,pa_data,ea_data_bdr);
+         case 0x99: return EADGTraceAssemble2DBdr<9,9>(nf,B,pa_data,ea_data_bdr);
          default:
-            return EADGTraceAssemble2DBdr(nf,B,pa_data,ea_data_bdr,add,dofs1D,quad1D);
+            return EADGTraceAssemble2DBdr(nf,B,pa_data,ea_data_bdr,dofs1D,quad1D);
       }
    }
    else if (dim == 3)
    {
       switch ((dofs1D << 4 ) | quad1D)
       {
-         case 0x23: return EADGTraceAssemble3DBdr<2,3>(nf,B,pa_data,ea_data_bdr,add);
-         case 0x34: return EADGTraceAssemble3DBdr<3,4>(nf,B,pa_data,ea_data_bdr,add);
-         case 0x45: return EADGTraceAssemble3DBdr<4,5>(nf,B,pa_data,ea_data_bdr,add);
-         case 0x56: return EADGTraceAssemble3DBdr<5,6>(nf,B,pa_data,ea_data_bdr,add);
-         case 0x67: return EADGTraceAssemble3DBdr<6,7>(nf,B,pa_data,ea_data_bdr,add);
-         case 0x78: return EADGTraceAssemble3DBdr<7,8>(nf,B,pa_data,ea_data_bdr,add);
-         case 0x89: return EADGTraceAssemble3DBdr<8,9>(nf,B,pa_data,ea_data_bdr,add);
+         case 0x23: return EADGTraceAssemble3DBdr<2,3>(nf,B,pa_data,ea_data_bdr);
+         case 0x34: return EADGTraceAssemble3DBdr<3,4>(nf,B,pa_data,ea_data_bdr);
+         case 0x45: return EADGTraceAssemble3DBdr<4,5>(nf,B,pa_data,ea_data_bdr);
+         case 0x56: return EADGTraceAssemble3DBdr<5,6>(nf,B,pa_data,ea_data_bdr);
+         case 0x67: return EADGTraceAssemble3DBdr<6,7>(nf,B,pa_data,ea_data_bdr);
+         case 0x78: return EADGTraceAssemble3DBdr<7,8>(nf,B,pa_data,ea_data_bdr);
+         case 0x89: return EADGTraceAssemble3DBdr<8,9>(nf,B,pa_data,ea_data_bdr);
          default:
-            return EADGTraceAssemble3DBdr(nf,B,pa_data,ea_data_bdr,add,dofs1D,quad1D);
+            return EADGTraceAssemble3DBdr(nf,B,pa_data,ea_data_bdr,dofs1D,quad1D);
       }
    }
    MFEM_ABORT("Unknown kernel.");

fem/bilininteg_diffusion_ea.cpp

@@ -22,7 +22,6 @@ static void EADiffusionAssemble1D(const int NE,
                                   const Array<double> &g,
                                   const Vector &padata,
                                   Vector &eadata,
-                                  const bool add,
                                   const int d1d = 0,
                                   const int q1d = 0)
 {
@@ -54,14 +53,7 @@ static void EADiffusionAssemble1D(const int NE,
             {
                val += r_Gj[k1] * D(k1, e) * r_Gi[k1];
             }
-            if (add)
-            {
-               A(i1, j1, e) += val;
-            }
-            else
-            {
-               A(i1, j1, e) = val;
-            }
+            A(i1, j1, e) += val;
          }
       }
    });
@@ -73,7 +65,6 @@ static void EADiffusionAssemble2D(const int NE,
                                   const Array<double> &g,
                                   const Vector &padata,
                                   Vector &eadata,
-                                  const bool add,
                                   const int d1d = 0,
                                   const int q1d = 0)
 {
@@ -129,14 +120,7 @@ static void EADiffusionAssemble2D(const int NE,
                                + gbi * D11 * gbj;
                      }
                   }
-                  if (add)
-                  {
-                     A(i1, i2, j1, j2, e) += val;
-                  }
-                  else
-                  {
-                     A(i1, i2, j1, j2, e) = val;
-                  }
+                  A(i1, i2, j1, j2, e) += val;
                }
             }
          }
@@ -146,11 +130,10 @@ static void EADiffusionAssemble2D(const int NE,
 
 template<int T_D1D = 0, int T_Q1D = 0>
 static void EADiffusionAssemble3D(const int NE,
-                                  const Array<double> &b,
                                   const Array<double> &g,
+                                  const Array<double> &b,
                                   const Vector &padata,
                                   Vector &eadata,
-                                  const bool add,
                                   const int d1d = 0,
                                   const int q1d = 0)
 {
@@ -225,14 +208,7 @@ static void EADiffusionAssemble3D(const int NE,
                               }
                            }
                         }
-                        if (add)
-                        {
-                           A(i1, i2, i3, j1, j2, j3, e) += val;
-                        }
-                        else
-                        {
-                           A(i1, i2, i3, j1, j2, j3, e) = val;
-                        }
+                        A(i1, i2, i3, j1, j2, j3, e) += val;
                      }
                   }
                }
@@ -243,8 +219,7 @@ static void EADiffusionAssemble3D(const int NE,
 }
 
 void DiffusionIntegrator::AssembleEA(const FiniteElementSpace &fes,
-                                     Vector &ea_data,
-                                     const bool add)
+                                     Vector &ea_data)
 {
    AssemblePA(fes);
    const int ne = fes.GetMesh()->GetNE();
@@ -254,47 +229,44 @@ void DiffusionIntegrator::AssembleEA(const FiniteElementSpace &fes,
    {
       switch ((dofs1D << 4 ) | quad1D)
       {
-         case 0x22: return EADiffusionAssemble1D<2,2>(ne,B,G,pa_data,ea_data,add);
-         case 0x33: return EADiffusionAssemble1D<3,3>(ne,B,G,pa_data,ea_data,add);
-         case 0x44: return EADiffusionAssemble1D<4,4>(ne,B,G,pa_data,ea_data,add);
-         case 0x55: return EADiffusionAssemble1D<5,5>(ne,B,G,pa_data,ea_data,add);
-         case 0x66: return EADiffusionAssemble1D<6,6>(ne,B,G,pa_data,ea_data,add);
-         case 0x77: return EADiffusionAssemble1D<7,7>(ne,B,G,pa_data,ea_data,add);
-         case 0x88: return EADiffusionAssemble1D<8,8>(ne,B,G,pa_data,ea_data,add);
-         case 0x99: return EADiffusionAssemble1D<9,9>(ne,B,G,pa_data,ea_data,add);
-         default:   return EADiffusionAssemble1D(ne,B,G,pa_data,ea_data,add,
-                                                    dofs1D,quad1D);
+         case 0x22: return EADiffusionAssemble1D<2,2>(ne,B,G,pa_data,ea_data);
+         case 0x33: return EADiffusionAssemble1D<3,3>(ne,B,G,pa_data,ea_data);
+         case 0x44: return EADiffusionAssemble1D<4,4>(ne,B,G,pa_data,ea_data);
+         case 0x55: return EADiffusionAssemble1D<5,5>(ne,B,G,pa_data,ea_data);
+         case 0x66: return EADiffusionAssemble1D<6,6>(ne,B,G,pa_data,ea_data);
+         case 0x77: return EADiffusionAssemble1D<7,7>(ne,B,G,pa_data,ea_data);
+         case 0x88: return EADiffusionAssemble1D<8,8>(ne,B,G,pa_data,ea_data);
+         case 0x99: return EADiffusionAssemble1D<9,9>(ne,B,G,pa_data,ea_data);
+         default:   return EADiffusionAssemble1D(ne,B,G,pa_data,ea_data,dofs1D,quad1D);
       }
    }
    else if (dim == 2)
    {
       switch ((dofs1D << 4 ) | quad1D)
       {
-         case 0x22: return EADiffusionAssemble2D<2,2>(ne,B,G,pa_data,ea_data,add);
-         case 0x33: return EADiffusionAssemble2D<3,3>(ne,B,G,pa_data,ea_data,add);
-         case 0x44: return EADiffusionAssemble2D<4,4>(ne,B,G,pa_data,ea_data,add);
-         case 0x55: return EADiffusionAssemble2D<5,5>(ne,B,G,pa_data,ea_data,add);
-         case 0x66: return EADiffusionAssemble2D<6,6>(ne,B,G,pa_data,ea_data,add);
-         case 0x77: return EADiffusionAssemble2D<7,7>(ne,B,G,pa_data,ea_data,add);
-         case 0x88: return EADiffusionAssemble2D<8,8>(ne,B,G,pa_data,ea_data,add);
-         case 0x99: return EADiffusionAssemble2D<9,9>(ne,B,G,pa_data,ea_data,add);
-         default:   return EADiffusionAssemble2D(ne,B,G,pa_data,ea_data,add,
-                                                    dofs1D,quad1D);
+         case 0x22: return EADiffusionAssemble2D<2,2>(ne,B,G,pa_data,ea_data);
+         case 0x33: return EADiffusionAssemble2D<3,3>(ne,B,G,pa_data,ea_data);
+         case 0x44: return EADiffusionAssemble2D<4,4>(ne,B,G,pa_data,ea_data);
+         case 0x55: return EADiffusionAssemble2D<5,5>(ne,B,G,pa_data,ea_data);
+         case 0x66: return EADiffusionAssemble2D<6,6>(ne,B,G,pa_data,ea_data);
+         case 0x77: return EADiffusionAssemble2D<7,7>(ne,B,G,pa_data,ea_data);
+         case 0x88: return EADiffusionAssemble2D<8,8>(ne,B,G,pa_data,ea_data);
+         case 0x99: return EADiffusionAssemble2D<9,9>(ne,B,G,pa_data,ea_data);
+         default:   return EADiffusionAssemble2D(ne,B,G,pa_data,ea_data,dofs1D,quad1D);
       }
    }
    else if (dim == 3)
    {
       switch ((dofs1D << 4 ) | quad1D)
       {
-         case 0x23: return EADiffusionAssemble3D<2,3>(ne,B,G,pa_data,ea_data,add);
-         case 0x34: return EADiffusionAssemble3D<3,4>(ne,B,G,pa_data,ea_data,add);
-         case 0x45: return EADiffusionAssemble3D<4,5>(ne,B,G,pa_data,ea_data,add);
-         case 0x56: return EADiffusionAssemble3D<5,6>(ne,B,G,pa_data,ea_data,add);
-         case 0x67: return EADiffusionAssemble3D<6,7>(ne,B,G,pa_data,ea_data,add);
-         case 0x78: return EADiffusionAssemble3D<7,8>(ne,B,G,pa_data,ea_data,add);
-         case 0x89: return EADiffusionAssemble3D<8,9>(ne,B,G,pa_data,ea_data,add);
-         default:   return EADiffusionAssemble3D(ne,B,G,pa_data,ea_data,add,
-                                                    dofs1D,quad1D);
+         case 0x23: return EADiffusionAssemble3D<2,3>(ne,B,G,pa_data,ea_data);
+         case 0x34: return EADiffusionAssemble3D<3,4>(ne,B,G,pa_data,ea_data);
+         case 0x45: return EADiffusionAssemble3D<4,5>(ne,B,G,pa_data,ea_data);
+         case 0x56: return EADiffusionAssemble3D<5,6>(ne,B,G,pa_data,ea_data);
+         case 0x67: return EADiffusionAssemble3D<6,7>(ne,B,G,pa_data,ea_data);
+         case 0x78: return EADiffusionAssemble3D<7,8>(ne,B,G,pa_data,ea_data);
+         case 0x89: return EADiffusionAssemble3D<8,9>(ne,B,G,pa_data,ea_data);
+         default:   return EADiffusionAssemble3D(ne,B,G,pa_data,ea_data,dofs1D,quad1D);
       }
    }
    MFEM_ABORT("Unknown kernel.");

fem/bilininteg_diffusion_pa.cpp

@@ -96,28 +96,26 @@ void PADiffusionSetup2D<2>(const int Q1D,
                            const Vector &c,
                            Vector &d)
 {
+   const int NQ = Q1D*Q1D;
    const bool const_c = c.Size() == 1;
-   const auto W = Reshape(w.Read(), Q1D,Q1D);
-   const auto J = Reshape(j.Read(), Q1D,Q1D,2,2,NE);
-   const auto C = const_c ? Reshape(c.Read(), 1,1,1) :
-                  Reshape(c.Read(), Q1D,Q1D,NE);
-   auto D = Reshape(d.Write(), Q1D,Q1D, 3, NE);
-   MFEM_FORALL_2D(e, NE, Q1D,Q1D,1,
+   auto W = w.Read();
+   auto J = Reshape(j.Read(), NQ, 2, 2, NE);
+   auto C = const_c ? Reshape(c.Read(), 1, 1) : Reshape(c.Read(), NQ, NE);
+   auto D = Reshape(d.Write(), NQ, 3, NE);
+
+   MFEM_FORALL(e, NE,
    {
-      MFEM_FOREACH_THREAD(qx,x,Q1D)
+      for (int q = 0; q < NQ; ++q)
       {
-         MFEM_FOREACH_THREAD(qy,y,Q1D)
-         {
-            const double J11 = J(qx,qy,0,0,e);
-            const double J21 = J(qx,qy,1,0,e);
-            const double J12 = J(qx,qy,0,1,e);
-            const double J22 = J(qx,qy,1,1,e);
-            const double coeff = const_c ? C(0,0,0) : C(qx,qy,e);
-            const double c_detJ = W(qx,qy) * coeff / ((J11*J22)-(J21*J12));
-            D(qx,qy,0,e) =  c_detJ * (J12*J12 + J22*J22); // 1,1
-            D(qx,qy,1,e) = -c_detJ * (J12*J11 + J22*J21); // 1,2
-            D(qx,qy,2,e) =  c_detJ * (J11*J11 + J21*J21); // 2,2
-         }
+         const double J11 = J(q,0,0,e);
+         const double J21 = J(q,1,0,e);
+         const double J12 = J(q,0,1,e);
+         const double J22 = J(q,1,1,e);
+         const double coeff = const_c ? C(0,0) : C(q,e);
+         const double c_detJ = W[q] * coeff / ((J11*J22)-(J21*J12));
+         D(q,0,e) =  c_detJ * (J12*J12 + J22*J22); // 1,1
+         D(q,1,e) = -c_detJ * (J12*J11 + J22*J21); // 1,2
+         D(q,2,e) =  c_detJ * (J11*J11 + J21*J21); // 2,2
       }
    });
 }
@@ -133,35 +131,33 @@ void PADiffusionSetup2D<3>(const int Q1D,
 {
    constexpr int DIM = 2;
    constexpr int SDIM = 3;
+   const int NQ = Q1D*Q1D;
    const bool const_c = c.Size() == 1;
-   const auto W = Reshape(w.Read(), Q1D,Q1D);
-   const auto J = Reshape(j.Read(), Q1D,Q1D,SDIM,DIM,NE);
-   const auto C = const_c ? Reshape(c.Read(), 1,1,1) :
-                  Reshape(c.Read(), Q1D,Q1D,NE);
-   auto D = Reshape(d.Write(), Q1D,Q1D, 3, NE);
-   MFEM_FORALL_2D(e, NE, Q1D,Q1D,1,
+
+   auto W = w.Read();
+   auto J = Reshape(j.Read(), NQ, SDIM, DIM, NE);
+   auto C = const_c ? Reshape(c.Read(), 1, 1) : Reshape(c.Read(), NQ, NE);
+   auto D = Reshape(d.Write(), NQ, 3, NE);
+   MFEM_FORALL(e, NE,
    {
-      MFEM_FOREACH_THREAD(qx,x,Q1D)
+      for (int q = 0; q < NQ; ++q)
       {
-         MFEM_FOREACH_THREAD(qy,y,Q1D)
-         {
-            const double wq = W(qx,qy);
-            const double J11 = J(qx,qy,0,0,e);
-            const double J21 = J(qx,qy,1,0,e);
-            const double J31 = J(qx,qy,2,0,e);
-            const double J12 = J(qx,qy,0,1,e);
-            const double J22 = J(qx,qy,1,1,e);
-            const double J32 = J(qx,qy,2,1,e);
-            const double E = J11*J11 + J21*J21 + J31*J31;
-            const double G = J12*J12 + J22*J22 + J32*J32;
-            const double F = J11*J12 + J21*J22 + J31*J32;
-            const double iw = 1.0 / sqrt(E*G - F*F);
-            const double coeff = const_c ? C(0,0,0) : C(qx,qy,e);
-            const double alpha = wq * coeff * iw;
-            D(qx,qy,0,e) =  alpha * G; // 1,1
-            D(qx,qy,1,e) = -alpha * F; // 1,2
-            D(qx,qy,2,e) =  alpha * E; // 2,2
-         }
+         const double wq = W[q];
+         const double J11 = J(q,0,0,e);
+         const double J21 = J(q,1,0,e);
+         const double J31 = J(q,2,0,e);
+         const double J12 = J(q,0,1,e);
+         const double J22 = J(q,1,1,e);
+         const double J32 = J(q,2,1,e);
+         const double E = J11*J11 + J21*J21 + J31*J31;
+         const double G = J12*J12 + J22*J22 + J32*J32;
+         const double F = J11*J12 + J21*J22 + J31*J32;
+         const double iw = 1.0 / sqrt(E*G - F*F);
+         const double coeff = const_c ? C(0,0) : C(q,e);
+         const double alpha = wq * coeff * iw;
+         D(q,0,e) =  alpha * G; // 1,1
+         D(q,1,e) = -alpha * F; // 1,2
+         D(q,2,e) =  alpha * E; // 2,2
       }
    });
 }
@@ -286,9 +282,10 @@ void DiffusionIntegrator::SetupPA(const FiniteElementSpace &fes)
    const int nq = ir->GetNPoints();
    dim = mesh->Dimension();
    ne = fes.GetNE();
-   geom = mesh->GetGeometricFactors(*ir, GeometricFactors::JACOBIANS);
+   const DofToQuad::Mode mode = DofToQuad::TENSOR;
+   geom = mesh->GetGeometricFactors(*ir, GeometricFactors::JACOBIANS, mode);
    const int sdim = mesh->SpaceDimension();
-   maps = &el.GetDofToQuad(*ir, DofToQuad::TENSOR);
+   maps = &el.GetDofToQuad(*ir, mode);
    dofs1D = maps->ndof;
    quad1D = maps->nqpt;
    pa_data.SetSize(symmDims * nq * ne, Device::GetDeviceMemoryType());
@@ -743,6 +740,7 @@ static void PADiffusionAssembleDiagonal(const int dim,
          case 0x23: return SmemPADiffusionDiagonal3D<2,3>(NE,B,G,D,Y);
          case 0x34: return SmemPADiffusionDiagonal3D<3,4>(NE,B,G,D,Y);
          case 0x45: return SmemPADiffusionDiagonal3D<4,5>(NE,B,G,D,Y);
+         case 0x46: return SmemPADiffusionDiagonal3D<4,6>(NE,B,G,D,Y);
          case 0x56: return SmemPADiffusionDiagonal3D<5,6>(NE,B,G,D,Y);
          case 0x67: return SmemPADiffusionDiagonal3D<6,7>(NE,B,G,D,Y);
          case 0x78: return SmemPADiffusionDiagonal3D<7,8>(NE,B,G,D,Y);
@@ -1680,7 +1678,7 @@ static void PADiffusionApply(const int dim,
       MFEM_ABORT("OCCA PADiffusionApply unknown kernel!");
    }
 #endif // MFEM_USE_OCCA
-   const int ID = (D1D << 4) | Q1D;
+   const int ID = (D1D << 4 ) | Q1D;
 
    if (dim == 2)
    {
@@ -1703,6 +1701,7 @@ static void PADiffusionApply(const int dim,
       switch (ID)
       {
          case 0x23: return SmemPADiffusionApply3D<2,3>(NE,B,G,D,X,Y);
+         case 0x24: return SmemPADiffusionApply3D<2,4>(NE,B,G,D,X,Y);
          case 0x34: return SmemPADiffusionApply3D<3,4>(NE,B,G,D,X,Y);
          case 0x45: return SmemPADiffusionApply3D<4,5>(NE,B,G,D,X,Y);
          case 0x46: return SmemPADiffusionApply3D<4,6>(NE,B,G,D,X,Y);

fem/bilininteg_hdiv.cpp

@@ -23,6 +23,11 @@ using namespace std;
 namespace mfem
 {
 
+// Local maximum size of dofs and quads in 1D
+constexpr int HDIV_MAX_D1D = 5;
+constexpr int HDIV_MAX_Q1D = 6;
+
+
 // PA H(div) Mass Assemble 2D kernel
 void PAHdivSetup2D(const int Q1D,
                    const int NE,

fem/bilininteg_mass_ea.cpp

@@ -21,7 +21,6 @@ static void EAMassAssemble1D(const int NE,
                              const Array<double> &basis,
                              const Vector &padata,
                              Vector &eadata,
-                             const bool add,
                              const int d1d = 0,
                              const int q1d = 0)
 {
@@ -53,14 +52,7 @@ static void EAMassAssemble1D(const int NE,
             {
                val += r_Bi[k1] * r_Bj[k1] * D(k1, e);
             }
-            if (add)
-            {
-               M(i1, j1, e) += val;
-            }
-            else
-            {
-               M(i1, j1, e) = val;
-            }
+            M(i1, j1, e) += val;
          }
       }
    });
@@ -71,7 +63,6 @@ static void EAMassAssemble2D(const int NE,
                              const Array<double> &basis,
                              const Vector &padata,
                              Vector &eadata,
-                             const bool add,
                              const int d1d = 0,
                              const int q1d = 0)
 {
@@ -123,14 +114,7 @@ static void EAMassAssemble2D(const int NE,
                                * s_D[k1][k2];
                      }
                   }
-                  if (add)
-                  {
-                     M(i1, i2, j1, j2, e) += val;
-                  }
-                  else
-                  {
-                     M(i1, i2, j1, j2, e) = val;
-                  }
+                  M(i1, i2, j1, j2, e) += val;
                }
             }
          }
@@ -143,7 +127,6 @@ static void EAMassAssemble3D(const int NE,
                              const Array<double> &basis,
                              const Vector &padata,
                              Vector &eadata,
-                             const bool add,
                              const int d1d = 0,
                              const int q1d = 0)
 {
@@ -206,14 +189,7 @@ static void EAMassAssemble3D(const int NE,
                               }
                            }
                         }
-                        if (add)
-                        {
-                           M(i1, i2, i3, j1, j2, j3, e) += val;
-                        }
-                        else
-                        {
-                           M(i1, i2, i3, j1, j2, j3, e) = val;
-                        }
+                        M(i1, i2, i3, j1, j2, j3, e) += val;
                      }
                   }
                }
@@ -224,8 +200,7 @@ static void EAMassAssemble3D(const int NE,
 }
 
 void MassIntegrator::AssembleEA(const FiniteElementSpace &fes,
-                                Vector &ea_data,
-                                const bool add)
+                                Vector &ea_data)
 {
    AssemblePA(fes);
    const int ne = fes.GetMesh()->GetNE();
@@ -234,47 +209,44 @@ void MassIntegrator::AssembleEA(const FiniteElementSpace &fes,
    {
       switch ((dofs1D << 4 ) | quad1D)
       {
-         case 0x22: return EAMassAssemble1D<2,2>(ne,B,pa_data,ea_data,add);
-         case 0x33: return EAMassAssemble1D<3,3>(ne,B,pa_data,ea_data,add);
-         case 0x44: return EAMassAssemble1D<4,4>(ne,B,pa_data,ea_data,add);
-         case 0x55: return EAMassAssemble1D<5,5>(ne,B,pa_data,ea_data,add);
-         case 0x66: return EAMassAssemble1D<6,6>(ne,B,pa_data,ea_data,add);
-         case 0x77: return EAMassAssemble1D<7,7>(ne,B,pa_data,ea_data,add);
-         case 0x88: return EAMassAssemble1D<8,8>(ne,B,pa_data,ea_data,add);
-         case 0x99: return EAMassAssemble1D<9,9>(ne,B,pa_data,ea_data,add);
-         default:   return EAMassAssemble1D(ne,B,pa_data,ea_data,add,
-                                               dofs1D,quad1D);
+         case 0x22: return EAMassAssemble1D<2,2>(ne,B,pa_data,ea_data);
+         case 0x33: return EAMassAssemble1D<3,3>(ne,B,pa_data,ea_data);
+         case 0x44: return EAMassAssemble1D<4,4>(ne,B,pa_data,ea_data);
+         case 0x55: return EAMassAssemble1D<5,5>(ne,B,pa_data,ea_data);
+         case 0x66: return EAMassAssemble1D<6,6>(ne,B,pa_data,ea_data);
+         case 0x77: return EAMassAssemble1D<7,7>(ne,B,pa_data,ea_data);
+         case 0x88: return EAMassAssemble1D<8,8>(ne,B,pa_data,ea_data);
+         case 0x99: return EAMassAssemble1D<9,9>(ne,B,pa_data,ea_data);
+         default:   return EAMassAssemble1D(ne,B,pa_data,ea_data,dofs1D,quad1D);
       }
    }
    else if (dim == 2)
    {
       switch ((dofs1D << 4 ) | quad1D)
       {
-         case 0x22: return EAMassAssemble2D<2,2>(ne,B,pa_data,ea_data,add);
-         case 0x33: return EAMassAssemble2D<3,3>(ne,B,pa_data,ea_data,add);
-         case 0x44: return EAMassAssemble2D<4,4>(ne,B,pa_data,ea_data,add);
-         case 0x55: return EAMassAssemble2D<5,5>(ne,B,pa_data,ea_data,add);
-         case 0x66: return EAMassAssemble2D<6,6>(ne,B,pa_data,ea_data,add);
-         case 0x77: return EAMassAssemble2D<7,7>(ne,B,pa_data,ea_data,add);
-         case 0x88: return EAMassAssemble2D<8,8>(ne,B,pa_data,ea_data,add);
-         case 0x99: return EAMassAssemble2D<9,9>(ne,B,pa_data,ea_data,add);
-         default:   return EAMassAssemble2D(ne,B,pa_data,ea_data,add,
-                                               dofs1D,quad1D);
+         case 0x22: return EAMassAssemble2D<2,2>(ne,B,pa_data,ea_data);
+         case 0x33: return EAMassAssemble2D<3,3>(ne,B,pa_data,ea_data);
+         case 0x44: return EAMassAssemble2D<4,4>(ne,B,pa_data,ea_data);
+         case 0x55: return EAMassAssemble2D<5,5>(ne,B,pa_data,ea_data);
+         case 0x66: return EAMassAssemble2D<6,6>(ne,B,pa_data,ea_data);
+         case 0x77: return EAMassAssemble2D<7,7>(ne,B,pa_data,ea_data);
+         case 0x88: return EAMassAssemble2D<8,8>(ne,B,pa_data,ea_data);
+         case 0x99: return EAMassAssemble2D<9,9>(ne,B,pa_data,ea_data);
+         default:   return EAMassAssemble2D(ne,B,pa_data,ea_data,dofs1D,quad1D);
       }
    }
    else if (dim == 3)
    {
       switch ((dofs1D << 4 ) | quad1D)
       {
-         case 0x23: return EAMassAssemble3D<2,3>(ne,B,pa_data,ea_data,add);
-         case 0x34: return EAMassAssemble3D<3,4>(ne,B,pa_data,ea_data,add);
-         case 0x45: return EAMassAssemble3D<4,5>(ne,B,pa_data,ea_data,add);
-         case 0x56: return EAMassAssemble3D<5,6>(ne,B,pa_data,ea_data,add);
-         case 0x67: return EAMassAssemble3D<6,7>(ne,B,pa_data,ea_data,add);
-         case 0x78: return EAMassAssemble3D<7,8>(ne,B,pa_data,ea_data,add);
-         case 0x89: return EAMassAssemble3D<8,9>(ne,B,pa_data,ea_data,add);
-         default:   return EAMassAssemble3D(ne,B,pa_data,ea_data,add,
-                                               dofs1D,quad1D);
+         case 0x23: return EAMassAssemble3D<2,3>(ne,B,pa_data,ea_data);
+         case 0x34: return EAMassAssemble3D<3,4>(ne,B,pa_data,ea_data);
+         case 0x45: return EAMassAssemble3D<4,5>(ne,B,pa_data,ea_data);
+         case 0x56: return EAMassAssemble3D<5,6>(ne,B,pa_data,ea_data);
+         case 0x67: return EAMassAssemble3D<6,7>(ne,B,pa_data,ea_data);
+         case 0x78: return EAMassAssemble3D<7,8>(ne,B,pa_data,ea_data);
+         case 0x89: return EAMassAssemble3D<8,9>(ne,B,pa_data,ea_data);
+         default:   return EAMassAssemble3D(ne,B,pa_data,ea_data,dofs1D,quad1D);
       }
    }
    MFEM_ABORT("Unknown kernel.");

fem/bilininteg_mass_pa.cpp

@@ -25,6 +25,7 @@ namespace mfem
 
 void MassIntegrator::SetupPA(const FiniteElementSpace &fes)
 {
+
    // Assuming the same element type
    fespace = &fes;
    Mesh *mesh = fes.GetMesh();
@@ -45,22 +46,38 @@ void MassIntegrator::SetupPA(const FiniteElementSpace &fes)
    dim = mesh->Dimension();
    ne = fes.GetMesh()->GetNE();
    nq = ir->GetNPoints();
-   geom = mesh->GetGeometricFactors(*ir, GeometricFactors::COORDINATES |
-                                    GeometricFactors::JACOBIANS);
-   maps = &el.GetDofToQuad(*ir, DofToQuad::TENSOR);
+   const DofToQuad::Mode mode = DofToQuad::TENSOR;
+   const int flags = GeometricFactors::JACOBIANS |
+                     GeometricFactors::COORDINATES;
+#ifdef MFEM_USE_UMPIRE
+   const MemoryType temp_type = Device::GetDeviceMemoryType() == MemoryType::DEVICE_UMPIRE
+      ? MemoryType::DEVICE_UMPIRE_2 : Device::GetDeviceMemoryType();
+#else
+   const MemoryType temp_type = Device::GetDeviceMemoryType();
+#endif
+   geom = mesh->GetGeometricFactors(*ir, flags, mode, temp_type);
+   maps = &el.GetDofToQuad(*ir, mode);
    dofs1D = maps->ndof;
    quad1D = maps->nqpt;
    pa_data.SetSize(ne*nq, Device::GetDeviceMemoryType());
-   Vector coeff;
+   Vector *coeff{nullptr};
+   bool own_coeff{true};
    if (Q == nullptr)
    {
-      coeff.SetSize(1);
-      coeff(0) = 1.0;
+      coeff = new Vector;
+      coeff->SetSize(1);
+      (*coeff)(0) = 1.0;
    }
    else if (ConstantCoefficient* cQ = dynamic_cast<ConstantCoefficient*>(Q))
    {
-      coeff.SetSize(1);
-      coeff(0) = cQ->constant;
+      coeff = new Vector;
+      coeff->SetSize(1);
+      (*coeff)(0) = cQ->constant;
+   }
+   else if (QuadratureCoefficient* cQ = dynamic_cast<QuadratureCoefficient*>(Q))
+   {
+      coeff = cQ->Data();
+      own_coeff = false;
    }
    else if (QuadratureFunctionCoefficient* cQ =
                dynamic_cast<QuadratureFunctionCoefficient*>(Q))
@@ -73,12 +90,13 @@ void MassIntegrator::SetupPA(const FiniteElementSpace &fes)
                   "IntegrationRule used within integrator and in"
                   " QuadratureFunction appear to be different");
       qFun.Read();
-      coeff.MakeRef(const_cast<QuadratureFunction &>(qFun),0);
+      coeff->MakeRef(const_cast<QuadratureFunction &>(qFun),0);
    }
    else
    {
-      coeff.SetSize(nq * ne);
-      auto C = Reshape(coeff.HostWrite(), nq, ne);
+      coeff = new Vector;
+      coeff->SetSize(nq * ne);
+      auto C = Reshape(coeff->HostWrite(), nq, ne);
       for (int e = 0; e < ne; ++e)
       {
          ElementTransformation& T = *fes.GetElementTransformation(e);
@@ -92,27 +110,24 @@ void MassIntegrator::SetupPA(const FiniteElementSpace &fes)
    if (dim==2)
    {
       const int NE = ne;
-      const int Q1D = quad1D;
-      const bool const_c = coeff.Size() == 1;
-      const auto W = Reshape(ir->GetWeights().Read(), Q1D,Q1D);
-      const auto J = Reshape(geom->J.Read(), Q1D,Q1D,2,2,NE);
-      const auto C = const_c ? Reshape(coeff.Read(), 1,1,1) :
-                     Reshape(coeff.Read(), Q1D,Q1D,NE);
-      auto v = Reshape(pa_data.Write(), Q1D,Q1D, NE);
-      MFEM_FORALL_2D(e, NE, Q1D,Q1D,1,
+      const int NQ = nq;
+      const bool const_c = coeff->Size() == 1;
+      auto w = ir->GetWeights().Read();
+      auto J = Reshape(geom->J.Read(), NQ,2,2,NE);
+      auto C =
+         const_c ? Reshape(coeff->Read(), 1,1) : Reshape(coeff->Read(), NQ,NE);
+      auto v = Reshape(pa_data.Write(), NQ, NE);
+      MFEM_FORALL(e, NE,
       {
-         MFEM_FOREACH_THREAD(qx,x,Q1D)
+         for (int q = 0; q < NQ; ++q)
          {
-            MFEM_FOREACH_THREAD(qy,y,Q1D)
-            {
-               const double J11 = J(qx,qy,0,0,e);
-               const double J12 = J(qx,qy,1,0,e);
-               const double J21 = J(qx,qy,0,1,e);
-               const double J22 = J(qx,qy,1,1,e);
-               const double detJ = (J11*J22)-(J21*J12);
-               const double coeff = const_c ? C(0,0,0) : C(qx,qy,e);
-               v(qx,qy,e) =  W(qx,qy) * coeff * detJ;
-            }
+            const double J11 = J(q,0,0,e);
+            const double J12 = J(q,1,0,e);
+            const double J21 = J(q,0,1,e);
+            const double J22 = J(q,1,1,e);
+            const double detJ = (J11*J22)-(J21*J12);
+            const double coeff = const_c ? C(0,0) : C(q,e);
+            v(q,e) =  w[q] * coeff * detJ;
          }
       });
    }
@@ -120,12 +135,13 @@ void MassIntegrator::SetupPA(const FiniteElementSpace &fes)
    {
       const int NE = ne;
       const int Q1D = quad1D;
-      const bool const_c = coeff.Size() == 1;
-      const auto W = Reshape(ir->GetWeights().Read(), Q1D,Q1D,Q1D);
+      const bool const_c = coeff->Size() == 1;
+      const auto W = Reshape(ir->GetWeights().Read(),Q1D,Q1D,Q1D);
       const auto J = Reshape(geom->J.Read(), Q1D,Q1D,Q1D,3,3,NE);
-      const auto C = const_c ? Reshape(coeff.Read(), 1,1,1,1) :
-                     Reshape(coeff.Read(), Q1D,Q1D,Q1D,NE);
-      auto v = Reshape(pa_data.Write(), Q1D,Q1D,Q1D,NE);
+      const auto C = const_c ?
+                     Reshape(coeff->Read(), 1,1,1,1) :
+                     Reshape(coeff->Read(), Q1D,Q1D,Q1D,NE);
+      auto V = Reshape(pa_data.Write(), Q1D,Q1D,Q1D,NE);
       MFEM_FORALL_3D(e, NE, Q1D, Q1D, Q1D,
       {
          MFEM_FOREACH_THREAD(qx,x,Q1D)
@@ -135,24 +151,26 @@ void MassIntegrator::SetupPA(const FiniteElementSpace &fes)
                MFEM_FOREACH_THREAD(qz,z,Q1D)
                {
                   const double J11 = J(qx,qy,qz,0,0,e);
-                  const double J21 = J(qx,qy,qz,1,0,e);
-                  const double J31 = J(qx,qy,qz,2,0,e);
                   const double J12 = J(qx,qy,qz,0,1,e);
-                  const double J22 = J(qx,qy,qz,1,1,e);
-                  const double J32 = J(qx,qy,qz,2,1,e);
                   const double J13 = J(qx,qy,qz,0,2,e);
+                  const double J21 = J(qx,qy,qz,1,0,e);
+                  const double J22 = J(qx,qy,qz,1,1,e);
                   const double J23 = J(qx,qy,qz,1,2,e);
+                  const double J31 = J(qx,qy,qz,2,0,e);
+                  const double J32 = J(qx,qy,qz,2,1,e);
                   const double J33 = J(qx,qy,qz,2,2,e);
                   const double detJ = J11 * (J22 * J33 - J32 * J23) -
                   /* */               J21 * (J12 * J33 - J32 * J13) +
                   /* */               J31 * (J12 * J23 - J22 * J13);
                   const double coeff = const_c ? C(0,0,0,0) : C(qx,qy,qz,e);
-                  v(qx,qy,qz,e) = W(qx,qy,qz) * coeff * detJ;
+                  V(qx,qy,qz,e) = W(qx,qy,qz) * coeff * detJ;
                }
             }
          }
       });
    }
+
+   if (own_coeff) { delete coeff; }
 }
 
 void MassIntegrator::AssemblePA(const FiniteElementSpace &fes)
@@ -454,8 +472,12 @@ static void PAMassAssembleDiagonal(const int dim, const int D1D,
       switch ((D1D << 4 ) | Q1D)
       {
          case 0x23: return SmemPAMassAssembleDiagonal3D<2,3>(NE,B,D,Y);
+         case 0x24: return SmemPAMassAssembleDiagonal3D<2,4>(NE,B,D,Y);
+         case 0x26: return SmemPAMassAssembleDiagonal3D<2,6>(NE,B,D,Y);
          case 0x34: return SmemPAMassAssembleDiagonal3D<3,4>(NE,B,D,Y);
+         case 0x35: return SmemPAMassAssembleDiagonal3D<3,5>(NE,B,D,Y);
          case 0x45: return SmemPAMassAssembleDiagonal3D<4,5>(NE,B,D,Y);
+         case 0x48: return SmemPAMassAssembleDiagonal3D<4,8>(NE,B,D,Y);
          case 0x56: return SmemPAMassAssembleDiagonal3D<5,6>(NE,B,D,Y);
          case 0x67: return SmemPAMassAssembleDiagonal3D<6,7>(NE,B,D,Y);
          case 0x78: return SmemPAMassAssembleDiagonal3D<7,8>(NE,B,D,Y);
@@ -1189,10 +1211,13 @@ static void PAMassApply(const int dim,
          case 0x24: return SmemPAMassApply2D<2,4,16>(NE,B,Bt,D,X,Y);
          case 0x33: return SmemPAMassApply2D<3,3,16>(NE,B,Bt,D,X,Y);
          case 0x34: return SmemPAMassApply2D<3,4,16>(NE,B,Bt,D,X,Y);
+         case 0x35: return SmemPAMassApply2D<3,5,16>(NE,B,Bt,D,X,Y);
          case 0x36: return SmemPAMassApply2D<3,6,16>(NE,B,Bt,D,X,Y);
          case 0x44: return SmemPAMassApply2D<4,4,8>(NE,B,Bt,D,X,Y);
+         case 0x46: return SmemPAMassApply2D<4,6,8>(NE,B,Bt,D,X,Y);
          case 0x48: return SmemPAMassApply2D<4,8,4>(NE,B,Bt,D,X,Y);
          case 0x55: return SmemPAMassApply2D<5,5,8>(NE,B,Bt,D,X,Y);
+         case 0x57: return SmemPAMassApply2D<5,7,8>(NE,B,Bt,D,X,Y);
          case 0x58: return SmemPAMassApply2D<5,8,2>(NE,B,Bt,D,X,Y);
          case 0x66: return SmemPAMassApply2D<6,6,4>(NE,B,Bt,D,X,Y);
          case 0x77: return SmemPAMassApply2D<7,7,4>(NE,B,Bt,D,X,Y);
@@ -1200,6 +1225,7 @@ static void PAMassApply(const int dim,
          case 0x99: return SmemPAMassApply2D<9,9,2>(NE,B,Bt,D,X,Y);
          default:   return PAMassApply2D(NE,B,Bt,D,X,Y,D1D,Q1D);
       }
+      mfem::out << "Unknown 2D kernel 0x" << std::hex << id << std::endl;
    }
    else if (dim == 3)
    {
@@ -1208,7 +1234,9 @@ static void PAMassApply(const int dim,
          case 0x23: return SmemPAMassApply3D<2,3>(NE,B,Bt,D,X,Y);
          case 0x24: return SmemPAMassApply3D<2,4>(NE,B,Bt,D,X,Y);
          case 0x34: return SmemPAMassApply3D<3,4>(NE,B,Bt,D,X,Y);
+         case 0x35: return SmemPAMassApply3D<3,5>(NE,B,Bt,D,X,Y);
          case 0x36: return SmemPAMassApply3D<3,6>(NE,B,Bt,D,X,Y);
+         case 0x37: return SmemPAMassApply3D<3,7>(NE,B,Bt,D,X,Y);
          case 0x45: return SmemPAMassApply3D<4,5>(NE,B,Bt,D,X,Y);
          case 0x46: return SmemPAMassApply3D<4,6>(NE,B,Bt,D,X,Y);
          case 0x48: return SmemPAMassApply3D<4,8>(NE,B,Bt,D,X,Y);
@@ -1220,8 +1248,8 @@ static void PAMassApply(const int dim,
          case 0x9A: return SmemPAMassApply3D<9,10>(NE,B,Bt,D,X,Y);
          default:   return PAMassApply3D(NE,B,Bt,D,X,Y,D1D,Q1D);
       }
+      mfem::out << "Unknown 3D kernel 0x" << std::hex << id << std::endl;
    }
-   mfem::out << "Unknown kernel 0x" << std::hex << id << std::endl;
    MFEM_ABORT("Unknown kernel.");
 }
 

fem/bilininteg_transpose_ea.cpp

@@ -16,171 +16,88 @@ namespace mfem
 {
 
 void TransposeIntegrator::AssembleEA(const FiniteElementSpace &fes,
-                                     Vector &ea_data, const bool add)
+                                     Vector &ea_data)
 {
-   if (add)
+   Vector ea_data_tmp(ea_data.Size());
+   ea_data_tmp = 0.0;
+   bfi->AssembleEA(fes, ea_data_tmp);
+   const int ne = fes.GetNE();
+   if (ne == 0) { return; }
+   const int dofs = fes.GetFE(0)->GetDof();
+   auto A = Reshape(ea_data_tmp.Write(), dofs, dofs, ne);
+   auto AT = Reshape(ea_data.ReadWrite(), dofs, dofs, ne);
+   MFEM_FORALL(e, ne,
    {
-      Vector ea_data_tmp(ea_data.Size());
-      bfi->AssembleEA(fes, ea_data_tmp, false);
-      const int ne = fes.GetNE();
-      if (ne == 0) { return; }
-      const int dofs = fes.GetFE(0)->GetDof();
-      auto A = Reshape(ea_data_tmp.Read(), dofs, dofs, ne);
-      auto AT = Reshape(ea_data.ReadWrite(), dofs, dofs, ne);
-      MFEM_FORALL(e, ne,
+      for (int i = 0; i < dofs; i++)
       {
-         for (int i = 0; i < dofs; i++)
+         for (int j = 0; j < dofs; j++)
          {
-            for (int j = 0; j < dofs; j++)
-            {
-               const double a = A(i, j, e);
-               AT(j, i, e) += a;
-            }
+            const double a = A(i, j, e);
+            AT(j, i, e) += a;
          }
-      });
-   }
-   else
-   {
-      bfi->AssembleEA(fes, ea_data, false);
-      const int ne = fes.GetNE();
-      if (ne == 0) { return; }
-      const int dofs = fes.GetFE(0)->GetDof();
-      auto A = Reshape(ea_data.ReadWrite(), dofs, dofs, ne);
-      MFEM_FORALL(e, ne,
-      {
-         for (int i = 0; i < dofs; i++)
-         {
-            for (int j = i+1; j < dofs; j++)
-            {
-               const double aij = A(i, j, e);
-               const double aji = A(j, i, e);
-               A(j, i, e) = aij;
-               A(i, j, e) = aji;
-            }
-         }
-      });
-   }
+      }
+   });
 }
 
 void TransposeIntegrator::AssembleEAInteriorFaces(const FiniteElementSpace& fes,
                                                   Vector &ea_data_int,
-                                                  Vector &ea_data_ext,
-                                                  const bool add)
+                                                  Vector &ea_data_ext)
 {
    const int nf = fes.GetNFbyType(FaceType::Interior);
    if (nf == 0) { return; }
-   if (add)
+   Vector ea_data_int_tmp(ea_data_int.Size());
+   Vector ea_data_ext_tmp(ea_data_ext.Size());
+   ea_data_int_tmp = 0.0;
+   ea_data_ext_tmp = 0.0;
+   bfi->AssembleEAInteriorFaces(fes, ea_data_int_tmp, ea_data_ext_tmp);
+   const int faceDofs = fes.GetTraceElement(0,
+                                            fes.GetMesh()->GetFaceBaseGeometry(0))->GetDof();
+   auto A_int = Reshape(ea_data_int_tmp.Read(), faceDofs, faceDofs, 2, nf);
+   auto A_ext = Reshape(ea_data_ext_tmp.Read(), faceDofs, faceDofs, 2, nf);
+   auto AT_int = Reshape(ea_data_int.ReadWrite(), faceDofs, faceDofs, 2, nf);
+   auto AT_ext = Reshape(ea_data_ext.ReadWrite(), faceDofs, faceDofs, 2, nf);
+   MFEM_FORALL(f, nf,
    {
-      Vector ea_data_int_tmp(ea_data_int.Size());
-      Vector ea_data_ext_tmp(ea_data_ext.Size());
-      bfi->AssembleEAInteriorFaces(fes, ea_data_int_tmp, ea_data_ext_tmp, false);
-      const int faceDofs = fes.GetTraceElement(0,
-                                               fes.GetMesh()->GetFaceBaseGeometry(0))->GetDof();
-      auto A_int = Reshape(ea_data_int_tmp.Read(), faceDofs, faceDofs, 2, nf);
-      auto A_ext = Reshape(ea_data_ext_tmp.Read(), faceDofs, faceDofs, 2, nf);
-      auto AT_int = Reshape(ea_data_int.ReadWrite(), faceDofs, faceDofs, 2, nf);
-      auto AT_ext = Reshape(ea_data_ext.ReadWrite(), faceDofs, faceDofs, 2, nf);
-      MFEM_FORALL(f, nf,
+      for (int i = 0; i < faceDofs; i++)
       {
-         for (int i = 0; i < faceDofs; i++)
+         for (int j = 0; j < faceDofs; j++)
          {
-            for (int j = 0; j < faceDofs; j++)
-            {
-               const double a_int0 = A_int(i, j, 0, f);
-               const double a_int1 = A_int(i, j, 1, f);
-               const double a_ext0 = A_ext(i, j, 0, f);
-               const double a_ext1 = A_ext(i, j, 1, f);
-               AT_int(j, i, 0, f) += a_int0;
-               AT_int(j, i, 1, f) += a_int1;
-               AT_ext(j, i, 0, f) += a_ext1;
-               AT_ext(j, i, 1, f) += a_ext0;
-            }
+            const double a_int0 = A_int(i, j, 0, f);
+            const double a_int1 = A_int(i, j, 1, f);
+            const double a_ext0 = A_ext(i, j, 0, f);
+            const double a_ext1 = A_ext(i, j, 1, f);
+            AT_int(j, i, 0, f) += a_int0;
+            AT_int(j, i, 1, f) += a_int1;
+            AT_ext(j, i, 0, f) += a_ext1;
+            AT_ext(j, i, 1, f) += a_ext0;
          }
-      });
-   }
-   else
-   {
-      bfi->AssembleEAInteriorFaces(fes, ea_data_int, ea_data_ext, false);
-      const int faceDofs = fes.GetTraceElement(0,
-                                               fes.GetMesh()->GetFaceBaseGeometry(0))->GetDof();
-      auto A_int = Reshape(ea_data_int.ReadWrite(), faceDofs, faceDofs, 2, nf);
-      auto A_ext = Reshape(ea_data_ext.ReadWrite(), faceDofs, faceDofs, 2, nf);
-      MFEM_FORALL(f, nf,
-      {
-         for (int i = 0; i < faceDofs; i++)
-         {
-            for (int j = i+1; j < faceDofs; j++)
-            {
-               const double aij_int0 = A_int(i, j, 0, f);
-               const double aij_int1 = A_int(i, j, 1, f);
-               const double aji_int0 = A_int(j, i, 0, f);
-               const double aji_int1 = A_int(j, i, 1, f);
-               A_int(j, i, 0, f) = aij_int0;
-               A_int(j, i, 1, f) = aij_int1;
-               A_int(i, j, 0, f) = aji_int0;
-               A_int(i, j, 1, f) = aji_int1;
-            }
-         }
-         for (int i = 0; i < faceDofs; i++)
-         {
-            for (int j = 0; j < faceDofs; j++)
-            {
-               const double aij_ext0 = A_ext(i, j, 0, f);
-               const double aji_ext1 = A_ext(j, i, 1, f);
-               A_ext(j, i, 1, f) = aij_ext0;
-               A_ext(i, j, 0, f) = aji_ext1;
-            }
-         }
-      });
-   }
+      }
+   });
 }
 
 void TransposeIntegrator::AssembleEABoundaryFaces(const FiniteElementSpace& fes,
-                                                  Vector &ea_data_bdr,
-                                                  const bool add)
+                                                  Vector &ea_data_bdr)
 {
    const int nf = fes.GetNFbyType(FaceType::Boundary);
    if (nf == 0) { return; }
-   if (add)
+   Vector ea_data_bdr_tmp(ea_data_bdr.Size());
+   ea_data_bdr_tmp = 0.0;
+   bfi->AssembleEABoundaryFaces(fes, ea_data_bdr_tmp);
+   const int faceDofs = fes.GetTraceElement(0,
+                                            fes.GetMesh()->GetFaceBaseGeometry(0))->GetDof();
+   auto A_bdr = Reshape(ea_data_bdr_tmp.Read(), faceDofs, faceDofs, nf);
+   auto AT_bdr = Reshape(ea_data_bdr.ReadWrite(), faceDofs, faceDofs, nf);
+   MFEM_FORALL(f, nf,
    {
-      Vector ea_data_bdr_tmp(ea_data_bdr.Size());
-      bfi->AssembleEABoundaryFaces(fes, ea_data_bdr_tmp, false);
-      const int faceDofs = fes.GetTraceElement(0,
-                                               fes.GetMesh()->GetFaceBaseGeometry(0))->GetDof();
-      auto A_bdr = Reshape(ea_data_bdr_tmp.Read(), faceDofs, faceDofs, nf);
-      auto AT_bdr = Reshape(ea_data_bdr.ReadWrite(), faceDofs, faceDofs, nf);
-      MFEM_FORALL(f, nf,
+      for (int i = 0; i < faceDofs; i++)
       {
-         for (int i = 0; i < faceDofs; i++)
+         for (int j = 0; j < faceDofs; j++)
          {
-            for (int j = 0; j < faceDofs; j++)
-            {
-               const double a_bdr = A_bdr(i, j, f);
-               AT_bdr(j, i, f) += a_bdr;
-            }
+            const double a_bdr = A_bdr(i, j, f);
+            AT_bdr(j, i, f) += a_bdr;
          }
-      });
-   }
-   else
-   {
-      bfi->AssembleEABoundaryFaces(fes, ea_data_bdr, false);
-      const int faceDofs = fes.GetTraceElement(0,
-                                               fes.GetMesh()->GetFaceBaseGeometry(0))->GetDof();
-      auto A_bdr = Reshape(ea_data_bdr.ReadWrite(), faceDofs, faceDofs, nf);
-      MFEM_FORALL(f, nf,
-      {
-         for (int i = 0; i < faceDofs; i++)
-         {
-            for (int j = i+1; j < faceDofs; j++)
-            {
-               const double aij_bdr = A_bdr(i, j, f);
-               const double aji_bdr = A_bdr(j, i, f);
-               A_bdr(j, i, f) = aij_bdr;
-               A_bdr(i, j, f) = aji_bdr;
-            }
-         }
-      });
-   }
+      }
+   });
 }
 
 }

fem/coefficient.cpp

@@ -12,6 +12,7 @@
 // Implementation of Coefficient class
 
 #include "fem.hpp"
+#include "../linalg/dtensor.hpp"
 
 #include <cmath>
 #include <limits>
@@ -21,6 +22,13 @@ namespace mfem
 
 using namespace std;
 
+double QuadratureCoefficient::Eval(ElementTransformation & T,
+                                   const IntegrationPoint & ip)
+{
+   auto coeff = mfem::Reshape(qData->HostRead(), nip, NE);
+   return coeff(ip.index, T.ElementNo);
+}
+
 double PWConstCoefficient::Eval(ElementTransformation & T,
                                 const IntegrationPoint & ip)
 {
@@ -319,31 +327,6 @@ void MatrixFunctionCoefficient::Eval(DenseMatrix &K, ElementTransformation &T,
    }
 }
 
-void MatrixFunctionCoefficient::EvalSymmetric(Vector &K,
-                                              ElementTransformation &T,
-                                              const IntegrationPoint &ip)
-{
-   MFEM_VERIFY(symmetric && height == width && height < 4 && SymmFunction,
-               "MatrixFunctionCoefficient is not symmetric");
-
-   double x[3];
-   Vector transip(x, 3);
-
-   T.Transform(ip, transip);
-
-   K.SetSize((width * (width + 1)) / 2); // 1x1: 1, 2x2: 3, 3x3: 6
-
-   if (SymmFunction)
-   {
-      (*SymmFunction)(transip, K);
-   }
-
-   if (Q)
-   {
-      K *= Q->Eval(T, ip, GetTime());
-   }
-}
-
 MatrixArrayCoefficient::MatrixArrayCoefficient (int dim)
    : MatrixCoefficient (dim)
 {

fem/coefficient.hpp

@@ -87,6 +87,33 @@ public:
    { return (constant); }
 };
 
+
+/// class for quadrature coefficient
+class QuadratureCoefficient : public Coefficient
+{
+
+private:
+   const int nip;
+   const int NE;
+public:
+   Vector *qData{nullptr};
+
+   //Set external data
+   QuadratureCoefficient(Vector *Data, int in_nip, int in_NE)
+      : qData(Data), nip(in_nip), NE(in_NE)
+   { }
+
+   virtual double Eval(ElementTransformation &T,
+                       const IntegrationPoint &ip);
+
+   Vector *Data()
+   {
+      return qData;
+   }
+
+};
+
+/// class for piecewise constant coefficient
 /** @brief A piecewise constant coefficient with the constants keyed
     off the element attribute numbers. */
 class PWConstCoefficient : public Coefficient
@@ -695,16 +722,13 @@ class MatrixCoefficient
 protected:
    int height, width;
    double time;
-   bool symmetric;
 
 public:
    /// Construct a dim x dim matrix coefficient.
-   explicit MatrixCoefficient(int dim, bool symm=false)
-   { height = width = dim; time = 0.; symmetric = symm; }
+   explicit MatrixCoefficient(int dim) { height = width = dim; time = 0.; }
 
    /// Construct a h x w matrix coefficient.
-   MatrixCoefficient(int h, int w, bool symm=false) :
-      height(h), width(w), time(0.), symmetric(symm) { }
+   MatrixCoefficient(int h, int w) : height(h), width(w), time(0.) { }
 
    /// Set the time for time dependent coefficients
    void SetTime(double t) { time = t; }
@@ -721,9 +745,6 @@ public:
    /// For backward compatibility get the width of the matrix.
    int GetVDim() const { return width; }
 
-   void SetSymmetric(bool s) { symmetric = s; }
-   bool IsSymmetric() const { return symmetric; }
-
    /** @brief Evaluate the matrix coefficient in the element described by @a T
        at the point @a ip, storing the result in @a K. */
    /** @note When this method is called, the caller must make sure that the
@@ -732,15 +753,6 @@ public:
    virtual void Eval(DenseMatrix &K, ElementTransformation &T,
                      const IntegrationPoint &ip) = 0;
 
-   /** @brief Evaluate the upper triangular entries of the matrix coefficient
-       in the symmetric case, similarly to Eval. Matrix entry (i,j) is stored
-       in K[j - i + os_i] for 0 <= i <= j < width, os_0 = 0,
-       os_{i+1} = os_i + width - i. That is, K = {M(0,0), ..., M(0,w-1),
-       M(1,1), ..., M(1,w-1), ..., M(w-1,w-1) with w = width. */
-   virtual void EvalSymmetric(Vector &K, ElementTransformation &T,
-                              const IntegrationPoint &ip)
-   { mfem_error("MatrixCoefficient::EvalSymmetric"); }
-
    virtual ~MatrixCoefficient() { }
 };
 
@@ -768,7 +780,6 @@ class MatrixFunctionCoefficient : public MatrixCoefficient
 {
 private:
    void (*Function)(const Vector &, DenseMatrix &);
-   void (*SymmFunction)(const Vector &, Vector &);
    void (*TDFunction)(const Vector &, double, DenseMatrix &);
    Coefficient *Q;
    DenseMatrix mat;
@@ -806,26 +817,10 @@ public:
       mat.SetSize(0);
    }
 
-   /// Construct a symmetric square matrix coefficient from a C-function
-   /// defining a vector function used by EvalSymmetric
-   MatrixFunctionCoefficient(int dim, void (*F)(const Vector &, Vector &),
-                             Coefficient *q = NULL)
-      : MatrixCoefficient(dim, true), Q(q)
-   {
-      SymmFunction = F;
-      Function = NULL;
-      TDFunction = NULL;
-      mat.SetSize(0);
-   }
-
    /// Evaluate the matrix coefficient at @a ip.
    virtual void Eval(DenseMatrix &K, ElementTransformation &T,
                      const IntegrationPoint &ip);
 
-   /// Evaluate the symmetric matrix coefficient at @a ip.
-   virtual void EvalSymmetric(Vector &K, ElementTransformation &T,
-                              const IntegrationPoint &ip);
-
    virtual ~MatrixFunctionCoefficient() { }
 };
 

fem/complex_fem.cpp

@@ -10,7 +10,6 @@
 // CONTRIBUTING.md for details.
 
 #include "complex_fem.hpp"
-#include "../general/forall.hpp"
 
 using namespace std;
 
@@ -20,21 +19,16 @@ namespace mfem
 ComplexGridFunction::ComplexGridFunction(FiniteElementSpace *fes)
    : Vector(2*(fes->GetVSize()))
 {
-   UseDevice(true);
-   this->Vector::operator=(0.0);
-
-   gfr = new GridFunction();
-   gfr->MakeRef(fes, *this, 0);
-
-   gfi = new GridFunction();
-   gfi->MakeRef(fes, *this, fes->GetVSize());
+   gfr = new GridFunction(fes, data);
+   gfi = new GridFunction(fes, &data[fes->GetVSize()]);
 }
 
 void
 ComplexGridFunction::Update()
 {
-   FiniteElementSpace *fes = gfr->FESpace();
-   const int vsize = fes->GetVSize();
+   FiniteElementSpace * fes = gfr->FESpace();
+
+   int vsize = fes->GetVSize();
 
    const Operator *T = fes->GetUpdateOperator();
    if (T)
@@ -46,36 +40,30 @@ ComplexGridFunction::Update()
 
       // Our data array now contains old data as well as being the wrong size so
       // reallocate it.
-      UseDevice(true);
       this->SetSize(2 * vsize);
-      this->Vector::operator=(0.0);
 
       // Create temporary vectors which point to the new data array
-      Vector gf_r; gf_r.MakeRef(*this, 0, vsize);
-      Vector gf_i; gf_i.MakeRef(*this, vsize, vsize);
+      Vector gf_r(data, vsize);
+      Vector gf_i((data) ? &data[vsize] : data, vsize);
 
       // Copy the updated GridFunctions into the new data array
       gf_r = *gfr;
       gf_i = *gfi;
-      gf_r.SyncAliasMemory(*this);
-      gf_i.SyncAliasMemory(*this);
 
       // Replace the individual data arrays with pointers into the new data
       // array
-      gfr->MakeRef(*this, 0, vsize);
-      gfi->MakeRef(*this, vsize, vsize);
+      gfr->NewDataAndSize(data, vsize);
+      gfi->NewDataAndSize((data) ? &data[vsize] : data, vsize);
    }
    else
    {
       // The existing data will not be transferred to the new GridFunctions so
-      // delete it and allocate a new array
-      UseDevice(true);
+      // delete it a allocate a new array
       this->SetSize(2 * vsize);
-      this->Vector::operator=(0.0);
 
       // Point the individual GridFunctions to the new data array
-      gfr->MakeRef(*this, 0, vsize);
-      gfi->MakeRef(*this, vsize, vsize);
+      gfr->NewDataAndSize(data, vsize);
+      gfi->NewDataAndSize((data) ? &data[vsize] : data, vsize);
 
       // These updates will only set the proper 'sequence' value within the
       // individual GridFunction objects because their sizes are already correct
@@ -88,24 +76,16 @@ void
 ComplexGridFunction::ProjectCoefficient(Coefficient &real_coeff,
                                         Coefficient &imag_coeff)
 {
-   gfr->SyncMemory(*this);
-   gfi->SyncMemory(*this);
    gfr->ProjectCoefficient(real_coeff);
    gfi->ProjectCoefficient(imag_coeff);
-   gfr->SyncAliasMemory(*this);
-   gfi->SyncAliasMemory(*this);
 }
 
 void
 ComplexGridFunction::ProjectCoefficient(VectorCoefficient &real_vcoeff,
                                         VectorCoefficient &imag_vcoeff)
 {
-   gfr->SyncMemory(*this);
-   gfi->SyncMemory(*this);
    gfr->ProjectCoefficient(real_vcoeff);
    gfi->ProjectCoefficient(imag_vcoeff);
-   gfr->SyncAliasMemory(*this);
-   gfi->SyncAliasMemory(*this);
 }
 
 void
@@ -113,12 +93,8 @@ ComplexGridFunction::ProjectBdrCoefficient(Coefficient &real_coeff,
                                            Coefficient &imag_coeff,
                                            Array<int> &attr)
 {
-   gfr->SyncMemory(*this);
-   gfi->SyncMemory(*this);
    gfr->ProjectBdrCoefficient(real_coeff, attr);
    gfi->ProjectBdrCoefficient(imag_coeff, attr);
-   gfr->SyncAliasMemory(*this);
-   gfi->SyncAliasMemory(*this);
 }
 
 void
@@ -126,12 +102,8 @@ ComplexGridFunction::ProjectBdrCoefficientNormal(VectorCoefficient &real_vcoeff,
                                                  VectorCoefficient &imag_vcoeff,
                                                  Array<int> &attr)
 {
-   gfr->SyncMemory(*this);
-   gfi->SyncMemory(*this);
    gfr->ProjectBdrCoefficientNormal(real_vcoeff, attr);
    gfi->ProjectBdrCoefficientNormal(imag_vcoeff, attr);
-   gfr->SyncAliasMemory(*this);
-   gfi->SyncAliasMemory(*this);
 }
 
 void
@@ -141,28 +113,18 @@ ComplexGridFunction::ProjectBdrCoefficientTangent(VectorCoefficient
                                                   &imag_vcoeff,
                                                   Array<int> &attr)
 {
-   gfr->SyncMemory(*this);
-   gfi->SyncMemory(*this);
    gfr->ProjectBdrCoefficientTangent(real_vcoeff, attr);
    gfi->ProjectBdrCoefficientTangent(imag_vcoeff, attr);
-   gfr->SyncAliasMemory(*this);
-   gfi->SyncAliasMemory(*this);
 }
 
 
-ComplexLinearForm::ComplexLinearForm(FiniteElementSpace *fes,
+ComplexLinearForm::ComplexLinearForm(FiniteElementSpace *f,
                                      ComplexOperator::Convention convention)
-   : Vector(2*(fes->GetVSize())),
+   : Vector(2*(f->GetVSize())),
      conv(convention)
 {
-   UseDevice(true);
-   this->Vector::operator=(0.0);
-
-   lfr = new LinearForm();
-   lfr->MakeRef(fes, *this, 0);
-
-   lfi = new LinearForm();
-   lfi->MakeRef(fes, *this, fes->GetVSize());
+   lfr = new LinearForm(f, data);
+   lfi = new LinearForm(f, &data[f->GetVSize()]);
 }
 
 ComplexLinearForm::ComplexLinearForm(FiniteElementSpace *fes,
@@ -171,14 +133,8 @@ ComplexLinearForm::ComplexLinearForm(FiniteElementSpace *fes,
    : Vector(2*(fes->GetVSize())),
      conv(convention)
 {
-   UseDevice(true);
-   this->Vector::operator=(0.0);
-
-   lfr = new LinearForm(fes, lf_r);
-   lfi = new LinearForm(fes, lf_i);
-
-   lfr->MakeRef(fes, *this, 0);
-   lfi->MakeRef(fes, *this, fes->GetVSize());
+   lfr = new LinearForm(fes, lf_r);  lfr->SetData(data);
+   lfi = new LinearForm(fes, lf_i);  lfi->SetData(&data[fes->GetVSize()]);
 }
 
 ComplexLinearForm::~ComplexLinearForm()
@@ -233,43 +189,42 @@ void
 ComplexLinearForm::Update()
 {
    FiniteElementSpace *fes = lfr->FESpace();
+
    this->Update(fes);
 }
 
 void
 ComplexLinearForm::Update(FiniteElementSpace *fes)
 {
-   UseDevice(true);
-   SetSize(2 * fes->GetVSize());
-   this->Vector::operator=(0.0);
+   int vsize = fes->GetVSize();
+   SetSize(2 * vsize);
 
-   lfr->MakeRef(fes, *this, 0);
-   lfi->MakeRef(fes, *this, fes->GetVSize());
+   Vector vlfr(data, vsize);
+   Vector vlfi((data) ? &data[vsize] : data, vsize);
+
+   lfr->Update(fes, vlfr, 0);
+   lfi->Update(fes, vlfi, 0);
 }
 
 void
 ComplexLinearForm::Assemble()
 {
-   lfr->SyncMemory(*this);
-   lfi->SyncMemory(*this);
    lfr->Assemble();
    lfi->Assemble();
-   if (conv == ComplexOperator::BLOCK_SYMMETRIC) { *lfi *= -1.0; }
-   lfr->SyncAliasMemory(*this);
-   lfi->SyncAliasMemory(*this);
+   if (conv == ComplexOperator::BLOCK_SYMMETRIC)
+   {
+      *lfi *= -1.0;
+   }
 }
 
 complex<double>
 ComplexLinearForm::operator()(const ComplexGridFunction &gf) const
 {
-   double s = (conv == ComplexOperator::HERMITIAN) ? 1.0 : -1.0;
-   lfr->SyncMemory(*this);
-   lfi->SyncMemory(*this);
+   double s = (conv == ComplexOperator::HERMITIAN)?1.0:-1.0;
    return complex<double>((*lfr)(gf.real()) - s * (*lfi)(gf.imag()),
                           (*lfr)(gf.imag()) + s * (*lfi)(gf.real()));
 }
 
-
 bool SesquilinearForm::RealInteg()
 {
    int nint = blfr->GetFBFI()->Size() + blfr->GetDBFI()->Size() +
@@ -386,45 +341,34 @@ SesquilinearForm::FormLinearSystem(const Array<int> &ess_tdof_list,
                                    Vector &X, Vector &B,
                                    int ci)
 {
-   FiniteElementSpace *fes = blfr->FESpace();
-   const int vsize = fes->GetVSize();
+   FiniteElementSpace * fes = blfr->FESpace();
+   int vsize  = fes->GetVSize();
 
-   // Allocate temporary vector
-   Vector b_0;
-   b_0.UseDevice(true);
-   b_0.SetSize(vsize);
-   b_0 = 0.0;
+   // Allocate temporary vectors
+   Vector b_0(vsize); b_0 = 0.0;
 
    // Extract the real and imaginary parts of the input vectors
    MFEM_ASSERT(x.Size() == 2 * vsize, "Input GridFunction of incorrect size!");
-   x.Read();
-   Vector x_r; x_r.MakeRef(x, 0, vsize);
-   Vector x_i; x_i.MakeRef(x, vsize, vsize);
+   Vector x_r(x.GetData(), vsize);
+   Vector x_i(&(x.GetData())[vsize], vsize);
 
    MFEM_ASSERT(b.Size() == 2 * vsize, "Input LinearForm of incorrect size!");
-   b.Read();
-   Vector b_r; b_r.MakeRef(b, 0, vsize);
-   Vector b_i; b_i.MakeRef(b, vsize, vsize);
+   Vector b_r(b.GetData(), vsize);
+   Vector b_i(&(b.GetData())[vsize], vsize);
 
    if (conv == ComplexOperator::BLOCK_SYMMETRIC) { b_i *= -1.0; }
 
-   const int tvsize = fes->GetTrueVSize();
+   int tvsize = fes->GetTrueVSize();
    OperatorHandle A_r, A_i;
 
-   X.UseDevice(true);
    X.SetSize(2 * tvsize);
-   X = 0.0;
-
-   B.UseDevice(true);
    B.SetSize(2 * tvsize);
-   B = 0.0;
 
-   Vector X_r; X_r.MakeRef(X, 0, tvsize);
-   Vector X_i; X_i.MakeRef(X, tvsize, tvsize);
-   Vector B_r; B_r.MakeRef(B, 0, tvsize);
-   Vector B_i; B_i.MakeRef(B, tvsize, tvsize);
-
-   Vector X_0, B_0;
+   Vector X_0(tvsize), B_0(tvsize);
+   Vector X_r(X.GetData(),tvsize);
+   Vector X_i(&(X.GetData())[tvsize], tvsize);
+   Vector B_r(B.GetData(), tvsize);
+   Vector B_i(&(B.GetData())[tvsize], tvsize);
 
    if (RealInteg())
    {
@@ -474,18 +418,13 @@ SesquilinearForm::FormLinearSystem(const Array<int> &ess_tdof_list,
       // conform with standard essential BC treatment
       if (A_i.Is<ConstrainedOperator>())
       {
-         const int n = ess_tdof_list.Size();
-         auto d_B_r = B_r.Write();
-         auto d_B_i = B_i.Write();
-         auto d_X_r = X_r.Read();
-         auto d_X_i = X_i.Read();
-         auto d_idx = ess_tdof_list.Read();
-         MFEM_FORALL(i, n,
+         int n = ess_tdof_list.Size();
+         for (int k = 0; k < n; k++)
          {
-            const int j = d_idx[i];
-            d_B_r[j] = d_X_r[j];
-            d_B_i[j] = d_X_i[j];
-         });
+            int j = ess_tdof_list[k];
+            B_r(j) = X_r(j);
+            B_i(j) = X_i(j);
+         }
          A_i.As<ConstrainedOperator>()->SetDiagonalPolicy
          (mfem::Operator::DiagonalPolicy::DIAG_ZERO);
       }
@@ -497,16 +436,6 @@ SesquilinearForm::FormLinearSystem(const Array<int> &ess_tdof_list,
       b_i *= -1.0;
    }
 
-   x_r.SyncAliasMemory(x);
-   x_i.SyncAliasMemory(x);
-   b_r.SyncAliasMemory(b);
-   b_i.SyncAliasMemory(b);
-
-   X_r.SyncAliasMemory(X);
-   X_i.SyncAliasMemory(X);
-   B_r.SyncAliasMemory(B);
-   B_i.SyncAliasMemory(B);
-
    // A = A_r + i A_i
    A.Clear();
    if ( A_r.Type() == Operator::MFEM_SPARSEMAT ||
@@ -599,32 +528,29 @@ void
 SesquilinearForm::RecoverFEMSolution(const Vector &X, const Vector &b,
                                      Vector &x)
 {
-   FiniteElementSpace *fes = blfr->FESpace();
+   FiniteElementSpace * fes = blfr->FESpace();
 
    const SparseMatrix *P = fes->GetConformingProlongation();
+
+   int vsize  = fes->GetVSize();
+   int tvsize = X.Size() / 2;
+
+   Vector X_r(X.GetData(), tvsize);
+   Vector X_i(&(X.GetData())[tvsize], tvsize);
+
+   Vector x_r(x.GetData(), vsize);
+   Vector x_i(&(x.GetData())[vsize], vsize);
+
    if (!P)
    {
       x = X;
-      return;
    }
-
-   const int vsize  = fes->GetVSize();
-   const int tvsize = X.Size() / 2;
-
-   X.Read();
-   Vector X_r; X_r.MakeRef(const_cast<Vector&>(X), 0, tvsize);
-   Vector X_i; X_i.MakeRef(const_cast<Vector&>(X), tvsize, tvsize);
-
-   x.Write();
-   Vector x_r; x_r.MakeRef(x, 0, vsize);
-   Vector x_i; x_i.MakeRef(x, vsize, vsize);
-
-   // Apply conforming prolongation
-   P->Mult(X_r, x_r);
-   P->Mult(X_i, x_i);
-
-   x_r.SyncAliasMemory(x);
-   x_i.SyncAliasMemory(x);
+   else
+   {
+      // Apply conforming prolongation
+      P->Mult(X_r, x_r);
+      P->Mult(X_i, x_i);
+   }
 }
 
 void
@@ -640,21 +566,16 @@ SesquilinearForm::Update(FiniteElementSpace *nfes)
 ParComplexGridFunction::ParComplexGridFunction(ParFiniteElementSpace *pfes)
    : Vector(2*(pfes->GetVSize()))
 {
-   UseDevice(true);
-   this->Vector::operator=(0.0);
-
-   pgfr = new ParGridFunction();
-   pgfr->MakeRef(pfes, *this, 0);
-
-   pgfi = new ParGridFunction();
-   pgfi->MakeRef(pfes, *this, pfes->GetVSize());
+   pgfr = new ParGridFunction(pfes, data);
+   pgfi = new ParGridFunction(pfes, (data) ? &data[pfes->GetVSize()]:data);
 }
 
 void
 ParComplexGridFunction::Update()
 {
-   ParFiniteElementSpace *pfes = pgfr->ParFESpace();
-   const int vsize = pfes->GetVSize();
+   ParFiniteElementSpace * pfes = pgfr->ParFESpace();
+
+   int vsize = pfes->GetVSize();
 
    const Operator *T = pfes->GetUpdateOperator();
    if (T)
@@ -666,34 +587,30 @@ ParComplexGridFunction::Update()
 
       // Our data array now contains old data as well as being the wrong size so
       // reallocate it.
-      UseDevice(true);
       this->SetSize(2 * vsize);
-      this->Vector::operator=(0.0);
 
       // Create temporary vectors which point to the new data array
-      Vector gf_r; gf_r.MakeRef(*this, 0, vsize);
-      Vector gf_i; gf_i.MakeRef(*this, vsize, vsize);
+      Vector gf_r(data, vsize);
+      Vector gf_i((data) ? &data[vsize] : data, vsize);
 
       // Copy the updated GridFunctions into the new data array
-      gf_r = *pgfr; gf_r.SyncAliasMemory(*this);
-      gf_i = *pgfi; gf_i.SyncAliasMemory(*this);
+      gf_r = *pgfr;
+      gf_i = *pgfi;
 
       // Replace the individual data arrays with pointers into the new data
       // array
-      pgfr->MakeRef(*this, 0, vsize);
-      pgfi->MakeRef(*this, vsize, vsize);
+      pgfr->NewDataAndSize(data, vsize);
+      pgfi->NewDataAndSize((data) ? &data[vsize] : data, vsize);
    }
    else
    {
       // The existing data will not be transferred to the new GridFunctions so
-      // delete it and allocate a new array
-      UseDevice(true);
+      // delete it a allocate a new array
       this->SetSize(2 * vsize);
-      this->Vector::operator=(0.0);
 
       // Point the individual GridFunctions to the new data array
-      pgfr->MakeRef(*this, 0, vsize);
-      pgfi->MakeRef(*this, vsize, vsize);
+      pgfr->NewDataAndSize(data, vsize);
+      pgfi->NewDataAndSize((data) ? &data[vsize] : data, vsize);
 
       // These updates will only set the proper 'sequence' value within the
       // individual GridFunction objects because their sizes are already correct
@@ -706,24 +623,16 @@ void
 ParComplexGridFunction::ProjectCoefficient(Coefficient &real_coeff,
                                            Coefficient &imag_coeff)
 {
-   pgfr->SyncMemory(*this);
-   pgfi->SyncMemory(*this);
    pgfr->ProjectCoefficient(real_coeff);
    pgfi->ProjectCoefficient(imag_coeff);
-   pgfr->SyncAliasMemory(*this);
-   pgfi->SyncAliasMemory(*this);
 }
 
 void
 ParComplexGridFunction::ProjectCoefficient(VectorCoefficient &real_vcoeff,
                                            VectorCoefficient &imag_vcoeff)
 {
-   pgfr->SyncMemory(*this);
-   pgfi->SyncMemory(*this);
    pgfr->ProjectCoefficient(real_vcoeff);
    pgfi->ProjectCoefficient(imag_vcoeff);
-   pgfr->SyncAliasMemory(*this);
-   pgfi->SyncAliasMemory(*this);
 }
 
 void
@@ -731,12 +640,8 @@ ParComplexGridFunction::ProjectBdrCoefficient(Coefficient &real_coeff,
                                               Coefficient &imag_coeff,
                                               Array<int> &attr)
 {
-   pgfr->SyncMemory(*this);
-   pgfi->SyncMemory(*this);
    pgfr->ProjectBdrCoefficient(real_coeff, attr);
    pgfi->ProjectBdrCoefficient(imag_coeff, attr);
-   pgfr->SyncAliasMemory(*this);
-   pgfi->SyncAliasMemory(*this);
 }
 
 void
@@ -746,12 +651,8 @@ ParComplexGridFunction::ProjectBdrCoefficientNormal(VectorCoefficient
                                                     &imag_vcoeff,
                                                     Array<int> &attr)
 {
-   pgfr->SyncMemory(*this);
-   pgfi->SyncMemory(*this);
    pgfr->ProjectBdrCoefficientNormal(real_vcoeff, attr);
    pgfi->ProjectBdrCoefficientNormal(imag_vcoeff, attr);
-   pgfr->SyncAliasMemory(*this);
-   pgfi->SyncAliasMemory(*this);
 }
 
 void
@@ -761,51 +662,36 @@ ParComplexGridFunction::ProjectBdrCoefficientTangent(VectorCoefficient
                                                      &imag_vcoeff,
                                                      Array<int> &attr)
 {
-   pgfr->SyncMemory(*this);
-   pgfi->SyncMemory(*this);
    pgfr->ProjectBdrCoefficientTangent(real_vcoeff, attr);
    pgfi->ProjectBdrCoefficientTangent(imag_vcoeff, attr);
-   pgfr->SyncAliasMemory(*this);
-   pgfi->SyncAliasMemory(*this);
 }
 
 void
 ParComplexGridFunction::Distribute(const Vector *tv)
 {
-   ParFiniteElementSpace *pfes = pgfr->ParFESpace();
-   const int tvsize = pfes->GetTrueVSize();
+   ParFiniteElementSpace * pfes = pgfr->ParFESpace();
+   HYPRE_Int size = pfes->GetTrueVSize();
 
-   tv->Read();
-   Vector tvr; tvr.MakeRef(const_cast<Vector&>(*tv), 0, tvsize);
-   Vector tvi; tvi.MakeRef(const_cast<Vector&>(*tv), tvsize, tvsize);
+   double * tvd = tv->GetData();
+   Vector tvr(tvd, size);
+   Vector tvi((tvd) ? &tvd[size] : tvd, size);
 
-   pgfr->SyncMemory(*this);
-   pgfi->SyncMemory(*this);
    pgfr->Distribute(tvr);
    pgfi->Distribute(tvi);
-   pgfr->SyncAliasMemory(*this);
-   pgfi->SyncAliasMemory(*this);
 }
 
 void
 ParComplexGridFunction::ParallelProject(Vector &tv) const
 {
-   ParFiniteElementSpace *pfes = pgfr->ParFESpace();
-   const int tvsize = pfes->GetTrueVSize();
+   ParFiniteElementSpace * pfes = pgfr->ParFESpace();
+   HYPRE_Int size = pfes->GetTrueVSize();
 
-   tv.Write();
-   Vector tvr; tvr.MakeRef(tv, 0, tvsize);
-   Vector tvi; tvi.MakeRef(tv, tvsize, tvsize);
+   double * tvd = tv.GetData();
+   Vector tvr(tvd, size);
+   Vector tvi((tvd) ? &tvd[size] : tvd, size);
 
-   pgfr->SyncMemory(*this);
-   pgfi->SyncMemory(*this);
    pgfr->ParallelProject(tvr);
    pgfi->ParallelProject(tvi);
-   pgfr->SyncAliasMemory(*this);
-   pgfi->SyncAliasMemory(*this);
-
-   tvr.SyncAliasMemory(tv);
-   tvi.SyncAliasMemory(tv);
 }
 
 
@@ -815,16 +701,10 @@ ParComplexLinearForm::ParComplexLinearForm(ParFiniteElementSpace *pfes,
    : Vector(2*(pfes->GetVSize())),
      conv(convention)
 {
-   UseDevice(true);
-   this->Vector::operator=(0.0);
+   plfr = new ParLinearForm(pfes, data);
+   plfi = new ParLinearForm(pfes, (data) ? &data[pfes->GetVSize()]:data);
 
-   plfr = new ParLinearForm();
-   plfr->MakeRef(pfes, *this, 0);
-
-   plfi = new ParLinearForm();
-   plfi->MakeRef(pfes, *this, pfes->GetVSize());
-
-   HYPRE_Int *tdof_offsets_fes = pfes->GetTrueDofOffsets();
+   HYPRE_Int * tdof_offsets_fes = pfes->GetTrueDofOffsets();
 
    int n = (HYPRE_AssumedPartitionCheck()) ? 2 : pfes->GetNRanks();
    tdof_offsets = new HYPRE_Int[n+1];
@@ -844,16 +724,12 @@ ParComplexLinearForm::ParComplexLinearForm(ParFiniteElementSpace *pfes,
    : Vector(2*(pfes->GetVSize())),
      conv(convention)
 {
-   UseDevice(true);
-   this->Vector::operator=(0.0);
-
    plfr = new ParLinearForm(pfes, plf_r);
+   plfr->SetData(data);
    plfi = new ParLinearForm(pfes, plf_i);
+   plfi->SetData((data) ? &data[pfes->GetVSize()]:data);
 
-   plfr->MakeRef(pfes, *this, 0);
-   plfi->MakeRef(pfes, *this, pfes->GetVSize());
-
-   HYPRE_Int *tdof_offsets_fes = pfes->GetTrueDofOffsets();
+   HYPRE_Int * tdof_offsets_fes = pfes->GetTrueDofOffsets();
 
    int n = (HYPRE_AssumedPartitionCheck()) ? 2 : pfes->GetNRanks();
    tdof_offsets = new HYPRE_Int[n+1];
@@ -916,71 +792,58 @@ ParComplexLinearForm::AddBdrFaceIntegrator(LinearFormIntegrator *lfi_real,
 void
 ParComplexLinearForm::Update(ParFiniteElementSpace *pf)
 {
-   ParFiniteElementSpace *pfes = (pf != NULL) ? pf : plfr->ParFESpace();
+   ParFiniteElementSpace *pfes = (pf!=NULL)?pf:plfr->ParFESpace();
+   int vsize = pfes->GetVSize();
+   SetSize(2 * vsize);
 
-   UseDevice(true);
-   SetSize(2 * pfes->GetVSize());
-   this->Vector::operator=(0.0);
+   Vector vplfr(data, vsize);
+   Vector vplfi((data) ? &data[vsize] : data, vsize);
 
-   plfr->MakeRef(pfes, *this, 0);
-   plfi->MakeRef(pfes, *this, pfes->GetVSize());
+   plfr->Update(pfes, vplfr, 0);
+   plfi->Update(pfes, vplfi, 0);
 }
 
 void
 ParComplexLinearForm::Assemble()
 {
-   plfr->SyncMemory(*this);
-   plfi->SyncMemory(*this);
    plfr->Assemble();
    plfi->Assemble();
-   if (conv == ComplexOperator::BLOCK_SYMMETRIC) { *plfi *= -1.0; }
-   plfr->SyncAliasMemory(*this);
-   plfi->SyncAliasMemory(*this);
+   if (conv == ComplexOperator::BLOCK_SYMMETRIC)
+   {
+      *plfi *= -1.0;
+   }
 }
 
 void
 ParComplexLinearForm::ParallelAssemble(Vector &tv)
 {
-   const int tvsize = plfr->ParFESpace()->GetTrueVSize();
+   HYPRE_Int size = plfr->ParFESpace()->GetTrueVSize();
 
-   tv.Write();
-   Vector tvr; tvr.MakeRef(tv, 0, tvsize);
-   Vector tvi; tvi.MakeRef(tv, tvsize, tvsize);
+   double * tvd = tv.GetData();
+   Vector tvr(tvd, size);
+   Vector tvi((tvd) ? &tvd[size] : tvd, size);
 
-   plfr->SyncMemory(*this);
-   plfi->SyncMemory(*this);
    plfr->ParallelAssemble(tvr);
    plfi->ParallelAssemble(tvi);
-   plfr->SyncAliasMemory(*this);
-   plfi->SyncAliasMemory(*this);
-
-   tvr.SyncAliasMemory(tv);
-   tvi.SyncAliasMemory(tv);
 }
 
 HypreParVector *
 ParComplexLinearForm::ParallelAssemble()
 {
-   const ParFiniteElementSpace *pfes = plfr->ParFESpace();
-   const int tvsize = pfes->GetTrueVSize();
+   const ParFiniteElementSpace * pfes = plfr->ParFESpace();
 
-   HypreParVector *tv = new HypreParVector(pfes->GetComm(),
-                                           2*(pfes->GlobalTrueVSize()),
-                                           tdof_offsets);
+   HypreParVector * tv = new HypreParVector(pfes->GetComm(),
+                                            2*(pfes->GlobalTrueVSize()),
+                                            tdof_offsets);
 
-   tv->Write();
-   Vector tvr; tvr.MakeRef(*tv, 0, tvsize);
-   Vector tvi; tvi.MakeRef(*tv, tvsize, tvsize);
+   HYPRE_Int size = pfes->GetTrueVSize();
+
+   double * tvd = tv->GetData();
+   Vector tvr(tvd, size);
+   Vector tvi((tvd) ? &tvd[size] : tvd, size);
 
-   plfr->SyncMemory(*this);
-   plfi->SyncMemory(*this);
    plfr->ParallelAssemble(tvr);
    plfi->ParallelAssemble(tvi);
-   plfr->SyncAliasMemory(*this);
-   plfi->SyncAliasMemory(*this);
-
-   tvr.SyncAliasMemory(*tv);
-   tvi.SyncAliasMemory(*tv);
 
    return tv;
 }
@@ -988,14 +851,13 @@ ParComplexLinearForm::ParallelAssemble()
 complex<double>
 ParComplexLinearForm::operator()(const ParComplexGridFunction &gf) const
 {
-   plfr->SyncMemory(*this);
-   plfi->SyncMemory(*this);
-   double s = (conv == ComplexOperator::HERMITIAN) ? 1.0 : -1.0;
+   double s = (conv == ComplexOperator::HERMITIAN)?1.0:-1.0;
    return complex<double>((*plfr)(gf.real()) - s * (*plfi)(gf.imag()),
                           (*plfr)(gf.imag()) + s * (*plfi)(gf.real()));
 }
 
 
+
 bool ParSesquilinearForm::RealInteg()
 {
    int nint = pblfr->GetFBFI()->Size() + pblfr->GetDBFI()->Size() +
@@ -1102,6 +964,7 @@ ParSesquilinearForm::ParallelAssemble()
    return new ComplexHypreParMatrix(pblfr->ParallelAssemble(),
                                     pblfi->ParallelAssemble(),
                                     true, true, conv);
+
 }
 
 void
@@ -1111,45 +974,35 @@ ParSesquilinearForm::FormLinearSystem(const Array<int> &ess_tdof_list,
                                       Vector &X, Vector &B,
                                       int ci)
 {
-   ParFiniteElementSpace *pfes = pblfr->ParFESpace();
-   const int vsize = pfes->GetVSize();
+   ParFiniteElementSpace * pfes = pblfr->ParFESpace();
+   int vsize = pfes->GetVSize();
 
-   // Allocate temporary vector
-   Vector b_0;
-   b_0.UseDevice(true);
-   b_0.SetSize(vsize);
-   b_0 = 0.0;
+   // Allocate temporary vectors
+   Vector b_0(vsize); b_0 = 0.0;
 
    // Extract the real and imaginary parts of the input vectors
    MFEM_ASSERT(x.Size() == 2 * vsize, "Input GridFunction of incorrect size!");
-   x.Read();
-   Vector x_r; x_r.MakeRef(x, 0, vsize);
-   Vector x_i; x_i.MakeRef(x, vsize, vsize);
+   Vector x_r(x.GetData(), vsize);
+   Vector x_i(&(x.GetData())[vsize], vsize);
 
    MFEM_ASSERT(b.Size() == 2 * vsize, "Input LinearForm of incorrect size!");
-   b.Read();
-   Vector b_r; b_r.MakeRef(b, 0, vsize);
-   Vector b_i; b_i.MakeRef(b, vsize, vsize);
+   Vector b_r(b.GetData(), vsize);
+   Vector b_i(&(b.GetData())[vsize], vsize);
 
    if (conv == ComplexOperator::BLOCK_SYMMETRIC) { b_i *= -1.0; }
 
-   const int tvsize = pfes->GetTrueVSize();
+   int tvsize = pfes->GetTrueVSize();
+
    OperatorHandle A_r, A_i;
 
-   X.UseDevice(true);
    X.SetSize(2 * tvsize);
-   X = 0.0;
-
-   B.UseDevice(true);
    B.SetSize(2 * tvsize);
-   B = 0.0;
 
-   Vector X_r; X_r.MakeRef(X, 0, tvsize);
-   Vector X_i; X_i.MakeRef(X, tvsize, tvsize);
-   Vector B_r; B_r.MakeRef(B, 0, tvsize);
-   Vector B_i; B_i.MakeRef(B, tvsize, tvsize);
-
-   Vector X_0, B_0;
+   Vector X_0(tvsize), B_0(tvsize);
+   Vector X_r(X.GetData(),tvsize);
+   Vector X_i(&(X.GetData())[tvsize], tvsize);
+   Vector B_r(B.GetData(), tvsize);
+   Vector B_i(&(B.GetData())[tvsize], tvsize);
 
    if (RealInteg())
    {
@@ -1189,29 +1042,24 @@ ParSesquilinearForm::FormLinearSystem(const Array<int> &ess_tdof_list,
 
    if (RealInteg() && ImagInteg())
    {
+      int n = ess_tdof_list.Size();
       // Modify RHS to conform with standard essential BC treatment
-      const int n = ess_tdof_list.Size();
-      auto d_B_r = B_r.Write();
-      auto d_B_i = B_i.Write();
-      auto d_X_r = X_r.Read();
-      auto d_X_i = X_i.Read();
-      auto d_idx = ess_tdof_list.Read();
-      MFEM_FORALL(i, n,
+      for (int k = 0; k < n; k++)
       {
-         const int j = d_idx[i];
-         d_B_r[j] = d_X_r[j];
-         d_B_i[j] = d_X_i[j];
-      });
+         int j=ess_tdof_list[k];
+         B_r(j) = X_r(j);
+         B_i(j) = X_i(j);
+      }
       // Modify offdiagonal blocks (imaginary parts of the matrix) to conform
       // with standard essential BC treatment
-      if (A_i.Type() == Operator::Hypre_ParCSR)
+      if ( A_i.Type() == Operator::Hypre_ParCSR )
       {
          HypreParMatrix * Ah;
          A_i.Get(Ah);
          hypre_ParCSRMatrix *Aih = *Ah;
          for (int k = 0; k < n; k++)
          {
-            const int j = ess_tdof_list[k];
+            int j = ess_tdof_list[k];
             Aih->diag->data[Aih->diag->i[j]] = 0.0;
          }
       }
@@ -1228,16 +1076,6 @@ ParSesquilinearForm::FormLinearSystem(const Array<int> &ess_tdof_list,
       b_i *= -1.0;
    }
 
-   x_r.SyncAliasMemory(x);
-   x_i.SyncAliasMemory(x);
-   b_r.SyncAliasMemory(b);
-   b_i.SyncAliasMemory(b);
-
-   X_r.SyncAliasMemory(X);
-   X_i.SyncAliasMemory(X);
-   B_r.SyncAliasMemory(B);
-   B_i.SyncAliasMemory(B);
-
    // A = A_r + i A_i
    A.Clear();
    if ( A_r.Type() == Operator::Hypre_ParCSR ||
@@ -1337,27 +1175,22 @@ void
 ParSesquilinearForm::RecoverFEMSolution(const Vector &X, const Vector &b,
                                         Vector &x)
 {
-   ParFiniteElementSpace *pfes = pblfr->ParFESpace();
+   ParFiniteElementSpace * pfes = pblfr->ParFESpace();
 
    const Operator &P = *pfes->GetProlongationMatrix();
 
-   const int vsize  = pfes->GetVSize();
-   const int tvsize = X.Size() / 2;
+   int vsize  = pfes->GetVSize();
+   int tvsize = X.Size() / 2;
 
-   X.Read();
-   Vector X_r; X_r.MakeRef(const_cast<Vector&>(X), 0, tvsize);
-   Vector X_i; X_i.MakeRef(const_cast<Vector&>(X), tvsize, tvsize);
+   Vector X_r(X.GetData(), tvsize);
+   Vector X_i(&(X.GetData())[tvsize], tvsize);
 
-   x.Write();
-   Vector x_r; x_r.MakeRef(x, 0, vsize);
-   Vector x_i; x_i.MakeRef(x, vsize, vsize);
+   Vector x_r(x.GetData(), vsize);
+   Vector x_i(&(x.GetData())[vsize], vsize);
 
    // Apply conforming prolongation
    P.Mult(X_r, x_r);
    P.Mult(X_i, x_i);
-
-   x_r.SyncAliasMemory(x);
-   x_i.SyncAliasMemory(x);
 }
 
 void

fem/complex_fem.hpp

@@ -38,8 +38,8 @@ protected:
    void Destroy() { delete gfr; delete gfi; }
 
 public:
-   /** @brief Construct a ComplexGridFunction associated with the
-       FiniteElementSpace @a *f. */
+   /* @brief Construct a ComplexGridFunction associated with the
+      FiniteElementSpace @a *f. */
    ComplexGridFunction(FiniteElementSpace *f);
 
    void Update();
@@ -71,14 +71,6 @@ public:
    const GridFunction & real() const { return *gfr; }
    const GridFunction & imag() const { return *gfi; }
 
-   /// Update the memory location of the real and imaginary GridFunction @a gfr
-   /// and @a gfi to match the ComplexGridFunction.
-   void Sync() { gfr->SyncMemory(*this); gfi->SyncMemory(*this); }
-
-   /// Update the alias memory location of the real and imaginary GridFunction
-   /// @a gfr and @a gfi to match the ComplexGridFunction.
-   void SyncAlias() { gfr->SyncAliasMemory(*this); gfi->SyncAliasMemory(*this); }
-
    /// Destroys the grid function.
    virtual ~ComplexGridFunction() { Destroy(); }
 
@@ -107,8 +99,8 @@ public:
                      ComplexOperator::Convention
                      convention = ComplexOperator::HERMITIAN);
 
-   /** @brief Create a ComplexLinearForm on the FiniteElementSpace @a fes, using
-       the same integrators as the LinearForms @a lf_r (real) and @a lf_i (imag).
+   /** @brief Create a ComplexLinearForm on the FiniteElementSpace @a f, using
+       the same integrators as the LinearForms @a lfr (real) and @a lfi (imag) .
 
        The pointer @a fes is not owned by the newly constructed object.
 
@@ -165,14 +157,6 @@ public:
    const LinearForm & real() const { return *lfr; }
    const LinearForm & imag() const { return *lfi; }
 
-   /// Update the memory location of the real and imaginary LinearForm @a lfr
-   /// and @a lfi to match the ComplexLinearForm.
-   void Sync() { lfr->SyncMemory(*this); lfi->SyncMemory(*this); }
-
-   /// Update the alias memory location of the real and imaginary LinearForm @a
-   /// lfr and @a lfi to match the ComplexLinearForm.
-   void SyncAlias() { lfr->SyncAliasMemory(*this); lfi->SyncAliasMemory(*this); }
-
    void Update();
    void Update(FiniteElementSpace *f);
 
@@ -211,8 +195,8 @@ private:
    BilinearForm *blfr;
    BilinearForm *blfi;
 
-   /* These methods check if the real/imag parts of the sesquilinear form are
-      not empty */
+   /* These methods check if the real/imag parts of the sesqulinear form are not
+      empty */
    bool RealInteg();
    bool ImagInteg();
 
@@ -220,7 +204,7 @@ public:
    SesquilinearForm(FiniteElementSpace *fes,
                     ComplexOperator::Convention
                     convention = ComplexOperator::HERMITIAN);
-   /** @brief Create a SesquilinearForm on the FiniteElementSpace @a fes, using
+   /** @brief Create a SesquilinearForm on the FiniteElementSpace @a f, using
        the same integrators as the BilinearForms @a bfr and @a bfi .
 
        The pointer @a fes is not owned by the newly constructed object.
@@ -339,8 +323,8 @@ protected:
 
 public:
 
-   /** @brief Construct a ParComplexGridFunction associated with the
-       ParFiniteElementSpace @a *pf. */
+   /* @brief Construct a ParComplexGridFunction associated with the
+      ParFiniteElementSpace @a *f. */
    ParComplexGridFunction(ParFiniteElementSpace *pf);
 
    void Update();
@@ -381,15 +365,6 @@ public:
    const ParGridFunction & real() const { return *pgfr; }
    const ParGridFunction & imag() const { return *pgfi; }
 
-   /// Update the memory location of the real and imaginary ParGridFunction @a
-   /// pgfr and @a pgfi to match the ParComplexGridFunction.
-   void Sync() { pgfr->SyncMemory(*this); pgfi->SyncMemory(*this); }
-
-   /// Update the alias memory location of the real and imaginary
-   /// ParGridFunction @a pgfr and @a pgfi to match the ParComplexGridFunction.
-   void SyncAlias() { pgfr->SyncAliasMemory(*this); pgfi->SyncAliasMemory(*this); }
-
-
    virtual double ComputeL2Error(Coefficient &exsolr, Coefficient &exsoli,
                                  const IntegrationRule *irs[] = NULL) const
    {
@@ -441,8 +416,8 @@ public:
                         convention = ComplexOperator::HERMITIAN);
 
    /** @brief Create a ParComplexLinearForm on the ParFiniteElementSpace @a pf,
-       using the same integrators as the LinearForms @a plf_r (real) and
-       @a plf_i (imag).
+       using the same integrators as the LinearForms @a plfr (real) and @a plfi
+       (imag) .
 
       The pointer @a fes is not owned by the newly constructed object.
 
@@ -500,14 +475,6 @@ public:
    const ParLinearForm & real() const { return *plfr; }
    const ParLinearForm & imag() const { return *plfi; }
 
-   /// Update the memory location of the real and imaginary ParLinearForm @a lfr
-   /// and @a lfi to match the ParComplexLinearForm.
-   void Sync() { plfr->SyncMemory(*this); plfi->SyncMemory(*this); }
-
-   /// Update the alias memory location of the real and imaginary ParLinearForm
-   /// @a plfr and @a plfi to match the ParComplexLinearForm.
-   void SyncAlias() { plfr->SyncAliasMemory(*this); plfi->SyncAliasMemory(*this); }
-
    void Update(ParFiniteElementSpace *pf = NULL);
 
    /// Assembles the linear form i.e. sums over all domain/bdr integrators.

fem/convergence.cpp

@@ -1,297 +0,0 @@
-#include "convergence.hpp"
-
-using namespace std;
-
-
-namespace mfem
-{
-
-void ConvergenceStudy::Reset()
-{
-   counter=0;
-   dcounter=0;
-   fcounter=0;
-   cont_type=-1;
-   print_flag=1;
-   L2Errors.SetSize(0);
-   L2Rates.SetSize(0);
-   DErrors.SetSize(0);
-   DRates.SetSize(0);
-   EnErrors.SetSize(0);
-   EnRates.SetSize(0);
-   DGFaceErrors.SetSize(0);
-   DGFaceRates.SetSize(0);
-   ndofs.SetSize(0);
-}
-
-double ConvergenceStudy::GetNorm(GridFunction *gf, Coefficient *scalar_u,
-                                 VectorCoefficient *vector_u)
-{
-   bool norm_set = false;
-   double norm=0.0;
-   int order = gf->FESpace()->GetOrder(0);
-   int order_quad = std::max(2, 2*order+1);
-   const IntegrationRule *irs[Geometry::NumGeom];
-   for (int i=0; i < Geometry::NumGeom; ++i)
-   {
-      irs[i] = &(IntRules.Get(i, order_quad));
-   }
-#ifdef MFEM_USE_MPI
-   ParGridFunction *pgf = dynamic_cast<ParGridFunction *>(gf);
-   if (pgf)
-   {
-      ParMesh *pmesh = pgf->ParFESpace()->GetParMesh();
-      if (scalar_u)
-      {
-         norm = ComputeGlobalLpNorm(2.0,*scalar_u,*pmesh,irs);
-      }
-      else if (vector_u)
-      {
-         norm = ComputeGlobalLpNorm(2.0,*vector_u,*pmesh,irs);
-      }
-      norm_set = true;
-   }
-#endif
-   if (!norm_set)
-   {
-      Mesh *mesh = gf->FESpace()->GetMesh();
-      if (scalar_u)
-      {
-         norm = ComputeLpNorm(2.0,*scalar_u,*mesh,irs);
-      }
-      else if (vector_u)
-      {
-         norm = ComputeLpNorm(2.0,*vector_u,*mesh,irs);
-      }
-   }
-   return norm;
-}
-
-void ConvergenceStudy::AddL2Error(GridFunction *gf,
-                                  Coefficient *scalar_u, VectorCoefficient *vector_u)
-{
-   int tdofs=0;
-#ifdef MFEM_USE_MPI
-   ParGridFunction *pgf = dynamic_cast<ParGridFunction *>(gf);
-   if (pgf)
-   {
-      MPI_Comm comm = pgf->ParFESpace()->GetComm();
-      int rank;
-      MPI_Comm_rank(comm, &rank);
-      print_flag = 0;
-      if (rank==0) { print_flag = 1; }
-      tdofs = pgf->ParFESpace()->GlobalTrueVSize();
-   }
-#endif
-   if (!tdofs) { tdofs = gf->FESpace()->GetTrueVSize(); }
-   ndofs.Append(tdofs);
-   double L2Err;
-   if (scalar_u)
-   {
-      L2Err = gf->ComputeL2Error(*scalar_u);
-      CoeffNorm = GetNorm(gf,scalar_u,nullptr);
-   }
-   else if (vector_u)
-   {
-      L2Err = gf->ComputeL2Error(*vector_u);
-      CoeffNorm = GetNorm(gf,nullptr,vector_u);
-   }
-   else
-   {
-      MFEM_ABORT("Exact Solution Coefficient pointer is NULL");
-   }
-   L2Errors.Append(L2Err);
-   // Compute the rate of convergence by:
-   // rate = log (||u - u_h|| / ||u - u_{h/2}||)/log(2)
-   double val = (counter) ? log(L2Errors[counter-1]/L2Err)/log(2.0) : 0.0;
-   L2Rates.Append(val);
-   counter++;
-}
-
-void ConvergenceStudy::AddGf(GridFunction *gf, Coefficient *scalar_u,
-                             VectorCoefficient *grad,
-                             Coefficient *ell_coeff, double Nu)
-{
-   cont_type = gf->FESpace()->FEColl()->GetContType();
-
-   MFEM_VERIFY((cont_type == mfem::FiniteElementCollection::CONTINUOUS) ||
-               (cont_type == mfem::FiniteElementCollection::DISCONTINUOUS),
-               "This constructor is intended for H1 or L2 Elements")
-
-   AddL2Error(gf,scalar_u, nullptr);
-
-   if (grad)
-   {
-      double GradErr = gf->ComputeGradError(grad);
-      DErrors.Append(GradErr);
-      double err = sqrt(L2Errors[counter-1]*L2Errors[counter-1]+GradErr*GradErr);
-      EnErrors.Append(err);
-      // Compute the rate of convergence by:
-      // rate = log (||u - u_h|| / ||u - u_{h/2}||)/log(2)
-      double val = (dcounter) ? log(DErrors[dcounter-1]/GradErr)/log(2.0) : 0.0;
-      double eval = (dcounter) ? log(EnErrors[dcounter-1]/err)/log(2.0) : 0.0;
-      DRates.Append(val);
-      EnRates.Append(eval);
-      CoeffDNorm = GetNorm(gf,nullptr,grad);
-      dcounter++;
-      MFEM_VERIFY(counter == dcounter,
-                  "Number of added solutions and derivatives do not match")
-   }
-
-   if (cont_type == mfem::FiniteElementCollection::DISCONTINUOUS && ell_coeff)
-   {
-      double DGErr = gf->ComputeDGFaceJumpError(scalar_u,ell_coeff,Nu);
-      DGFaceErrors.Append(DGErr);
-      // Compute the rate of convergence by:
-      // rate = log (||u - u_h|| / ||u - u_{h/2}||)/log(2)
-      double val=(fcounter) ? log(DGFaceErrors[fcounter-1]/DGErr)/log(2.0):0.;
-      DGFaceRates.Append(val);
-      fcounter++;
-      MFEM_VERIFY(fcounter == counter, "Number of added solutions mismatch");
-   }
-}
-
-void ConvergenceStudy::AddGf(GridFunction *gf, VectorCoefficient *vector_u,
-                             VectorCoefficient *curl, Coefficient *div)
-{
-   cont_type = gf->FESpace()->FEColl()->GetContType();
-
-   AddL2Error(gf,nullptr,vector_u);
-   double DErr = 0.0;
-   bool derivative = false;
-   if (curl)
-   {
-      DErr = gf->ComputeCurlError(curl);
-      CoeffDNorm = GetNorm(gf,nullptr,curl);
-      derivative = true;
-   }
-   else if (div)
-   {
-      DErr = gf->ComputeDivError(div);
-      // update coefficient norm
-      CoeffDNorm = GetNorm(gf,div,nullptr);
-      derivative = true;
-   }
-   if (derivative)
-   {
-      double err = sqrt(L2Errors[counter-1]*L2Errors[counter-1] + DErr*DErr);
-      DErrors.Append(DErr);
-      EnErrors.Append(err);
-      // Compute the rate of convergence by:
-      // rate = log (||u - u_h|| / ||u - u_{h/2}||)/log(2)
-      double val = (dcounter) ? log(DErrors[dcounter-1]/DErr)/log(2.0) : 0.0;
-      double eval = (dcounter) ? log(EnErrors[dcounter-1]/err)/log(2.0) : 0.0;
-      DRates.Append(val);
-      EnRates.Append(eval);
-      dcounter++;
-      MFEM_VERIFY(counter == dcounter,
-                  "Number of added solutions and derivatives do not match")
-   }
-}
-
-void ConvergenceStudy::Print(bool relative, std::ostream &out)
-{
-   if (print_flag)
-   {
-      std::string title = (relative) ? "Relative " : "Absolute ";
-      out << "\n";
-      out << " -------------------------------------------" << "\n";
-      out <<  std::setw(21) << title << "L2 Error " << "\n";
-      out << " -------------------------------------------"
-          << "\n";
-      out << std::right<< std::setw(11)<< "DOFs "<< std::setw(13) << "Error ";
-      out <<  std::setw(15) << "Rate " << "\n";
-      out << " -------------------------------------------"
-          << "\n";
-      out << std::setprecision(4);
-      double d = (relative) ? CoeffNorm : 1.0;
-      for (int i =0; i<counter; i++)
-      {
-         out << std::right << std::setw(10)<< ndofs[i] << std::setw(16)
-             << std::scientific << L2Errors[i]/d << std::setw(13)
-             << std::fixed << L2Rates[i] << "\n";
-      }
-      out << "\n";
-      if (dcounter == counter)
-      {
-         std::string dname;
-         switch (cont_type)
-         {
-            case 0: dname = "Grad"; break;
-            case 1: dname = "Curl"; break;
-            case 2: dname = "Div";  break;
-            case 3: dname = "DG Grad";  break;
-            default: break;
-         }
-         out << " -------------------------------------------" << "\n";
-         out <<  std::setw(21) << title << dname << " Error  " << "\n";
-         out << " -------------------------------------------" << "\n";
-         out << std::right<<std::setw(11)<< "DOFs "<< std::setw(13) << "Error";
-         out << std::setw(15) << "Rate " << "\n";
-         out << " -------------------------------------------"
-             << "\n";
-         out << std::setprecision(4);
-         d = (relative) ? CoeffDNorm : 1.0;
-         for (int i =0; i<dcounter; i++)
-         {
-            out << std::right << std::setw(10)<< ndofs[i] << std::setw(16)
-                << std::scientific << DErrors[i]/d << std::setw(13)
-                << std::fixed << DRates[i] << "\n";
-         }
-         out << "\n";
-         switch (cont_type)
-         {
-            case 0: dname = "H1"; break;
-            case 1: dname = "H(Curl)"; break;
-            case 2: dname = "H(Div)";  break;
-            case 3: dname = "DG H1";  break;
-            default: break;
-         }
-
-         if (dcounter)
-         {
-            d = (relative) ?
-                sqrt(CoeffNorm*CoeffNorm + CoeffDNorm*CoeffDNorm):1.0;
-
-            out << " -------------------------------------------" << "\n";
-            out << std::setw(21) << title << dname << " Error   " << "\n";
-            out << " -------------------------------------------" << "\n";
-            out << std::right<< std::setw(11)<< "DOFs "<< std::setw(13);
-            out << "Error ";
-            out << std::setw(15) << "Rate " << "\n";
-            out << " -------------------------------------------"
-                << "\n";
-            out << std::setprecision(4);
-            for (int i =0; i<dcounter; i++)
-            {
-               out << std::right << std::setw(10)<< ndofs[i] << std::setw(16)
-                   << std::scientific << EnErrors[i]/d << std::setw(13)
-                   << std::fixed << EnRates[i] << "\n";
-            }
-            out << "\n";
-         }
-         if (cont_type == 3 && fcounter)
-         {
-            out << " -------------------------------------------" << "\n";
-            out << "            DG Face Jump Error          " << "\n";
-            out << " -------------------------------------------"
-                << "\n";
-            out << std::right<< std::setw(11)<< "DOFs "<< std::setw(13);
-            out << "Error ";
-            out << std::setw(15) << "Rate " << "\n";
-            out << " -------------------------------------------"
-                << "\n";
-            out << std::setprecision(4);
-            for (int i =0; i<fcounter; i++)
-            {
-               out << std::right << std::setw(10)<< ndofs[i] << std::setw(16)
-                   << std::scientific << DGFaceErrors[i] << std::setw(13)
-                   << std::fixed << DGFaceRates[i] << "\n";
-            }
-            out << "\n";
-         }
-      }
-   }
-}
-
-} // namespace mfem

fem/convergence.hpp

@@ -1,149 +0,0 @@
-// Copyright (c) 2010-2020, 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_CONVERGENCE
-#define MFEM_CONVERGENCE
-
-#include "../linalg/linalg.hpp"
-#include "gridfunc.hpp"
-#ifdef MFEM_USE_MPI
-#include "pgridfunc.hpp"
-#endif
-
-namespace mfem
-{
-
-/** @brief Class to compute error and convergence rates.
-    It supports H1, H(curl) (ND elements), H(div) (RT elements) and L2 (DG).
-
-    For "smooth enough" solutions the Galerkin error measured in the appropriate
-    norm satisfies || u - u_h || ~ h^k
-
-    Here, k is called the asymptotic rate of convergence
-
-    For successive uniform h-refinements the rate can be estimated by
-    k = log(||u - u_h|| / ||u - u_{h/2}||)/log(2)
-*/
-class ConvergenceStudy
-{
-private:
-   // counters for solutions/derivatives
-   int counter=0;
-   int dcounter=0;
-   int fcounter=0;
-
-   // space continuity type
-   int cont_type=-1;
-
-   // printing flag for helpful for MPI calls
-   int print_flag=1;
-
-   // exact solution and derivatives
-   double CoeffNorm;
-   double CoeffDNorm;
-
-   // Arrays to store error/rates
-   Array<double> L2Errors, DGFaceErrors, DErrors, EnErrors;
-   Array<double> L2Rates, DGFaceRates, DRates, EnRates;
-   Array<int> ndofs;
-
-   void AddL2Error(GridFunction *gf, Coefficient *scalar_u,
-                   VectorCoefficient *vector_u);
-   void AddGf(GridFunction *gf, Coefficient *scalar_u,
-              VectorCoefficient *grad=nullptr,
-              Coefficient *ell_coeff=nullptr, double Nu=1.0);
-   void AddGf(GridFunction *gf, VectorCoefficient *vector_u,
-              VectorCoefficient *curl, Coefficient *div);
-   // returns the L2-norm of scalar_u or vector_u
-   double GetNorm(GridFunction *gf, Coefficient *scalar_u,
-                  VectorCoefficient *vector_u);
-
-public:
-
-   /// Clear any internal data
-   void Reset();
-
-   /// Add L2 GridFunction, the exact solution and possibly its gradient and/or
-   /// DG face jumps parameters
-   void AddL2GridFunction(GridFunction *gf, Coefficient *scalar_u,
-                          VectorCoefficient *grad=nullptr,
-                          Coefficient *ell_coeff=nullptr, double Nu=1.0)
-   {
-      AddGf(gf, scalar_u, grad, ell_coeff, Nu);
-   }
-
-   /// Add H1 GridFunction, the exact solution and possibly its gradient
-   void AddH1GridFunction(GridFunction *gf, Coefficient *scalar_u,
-                          VectorCoefficient *grad=nullptr)
-   {
-      AddGf(gf, scalar_u, grad);
-   }
-
-   /// Add H(curl) GridFunction, the exact solution and possibly its curl
-   void AddHcurlGridFunction(GridFunction *gf, VectorCoefficient *vector_u,
-                             VectorCoefficient *curl=nullptr)
-   {
-      AddGf(gf, vector_u, curl, nullptr);
-   }
-
-   /// Add H(div) GridFunction, the exact solution and possibly its div
-   void AddHdivGridFunction(GridFunction *gf, VectorCoefficient *vector_u,
-                            Coefficient *div=nullptr)
-   {
-      AddGf(gf,vector_u, nullptr, div);
-   }
-
-   /// Get the L2 error at step n
-   double GetL2Error(int n)
-   {
-      MFEM_VERIFY( n <= counter,"Step out of bounds")
-      return L2Errors[n];
-   }
-
-   /// Get all L2 errors
-   void GetL2Errors(Array<double> & L2Errors_)
-   {
-      L2Errors_ = L2Errors;
-   }
-
-   /// Get the Grad/Curl/Div error at step n
-   double GetDError(int n)
-   {
-      MFEM_VERIFY(n <= dcounter,"Step out of bounds")
-      return DErrors[n];
-   }
-
-   /// Get all Grad/Curl/Div errors
-   void GetDErrors(Array<double> & DErrors_)
-   {
-      DErrors_ = DErrors;
-   }
-
-   /// Get the DGFaceJumps error at step n
-   double GetDGFaceJumpsError(int n)
-   {
-      MFEM_VERIFY(n<= fcounter,"Step out of bounds")
-      return DGFaceErrors[n];
-   }
-
-   /// Get all DGFaceJumps errors
-   void GetDGFaceJumpsErrors(Array<double> & DGFaceErrors_)
-   {
-      DGFaceErrors_ = DGFaceErrors;
-   }
-
-   /// Print rates and errors
-   void Print(bool relative = false, std::ostream &out = mfem::out);
-};
-
-} // namespace mfem
-
-#endif // MFEM_CONVERGENCE

fem/datacollection.cpp

@@ -563,8 +563,6 @@ void VisItDataCollection::LoadVisItRootFile(const std::string& root_name)
 
 void VisItDataCollection::LoadMesh()
 {
-   // GetMeshFileName() uses 'serial', so we need to set it in advance.
-   serial = (format == SERIAL_FORMAT);
    std::string mesh_fname = GetMeshFileName();
    named_ifgzstream file(mesh_fname);
    // TODO: in parallel, check for errors on all processors

fem/eltrans.hpp

@@ -77,9 +77,6 @@ public:
 
    ElementTransformation();
 
-   /** @brief Force the reevaluation of the Jacobian in the next call. */
-   void Reset() { EvalState = 0; }
-
    /** @brief Set the integration point @a ip that weights and Jacobians will
        be evaluated at. */
    void SetIntPoint(const IntegrationPoint *ip)
@@ -360,17 +357,9 @@ private:
    // Evaluate the Hessian of the transformation at the IntPoint and store it
    // in d2Fdx2.
    virtual const DenseMatrix &EvalHessian();
-
 public:
-   IsoparametricTransformation() : FElem(NULL) {}
-
    /// Set the element that will be used to compute the transformations
-   void SetFE(const FiniteElement *FE)
-   {
-      MFEM_ASSERT(FE != NULL, "Must provide a valid FiniteElement object!");
-      EvalState = (FE != FElem) ? 0 : EvalState;
-      FElem = FE; geom = FE->GetGeomType();
-   }
+   void SetFE(const FiniteElement *FE) { FElem = FE; geom = FE->GetGeomType(); }
 
    /// Get the current element used to compute the transformations
    const FiniteElement* GetFE() const { return FElem; }
@@ -385,15 +374,12 @@ public:
        the column-vector of all basis functions evaluated at \f$ \hat x \f$ .
        The columns of @a P represent the control points in physical space
        defining the transformation. */
-   void SetPointMat(const DenseMatrix &pm) { PointMat = pm; EvalState = 0; }
+   void SetPointMat(const DenseMatrix &pm) { PointMat = pm; }
 
    /// Return the stored point matrix.
    const DenseMatrix &GetPointMat() const { return PointMat; }
 
-   /// @brief Write access to the stored point matrix. Use with caution.
-   /** If the point matrix is altered using this member function the Reset
-       function should also be called to force the reevaluation of the
-       Jacobian, etc.. */
+   /// Write access to the stored point matrix. Use with caution.
    DenseMatrix &GetPointMat() { return PointMat; }
 
    /// Set the FiniteElement Geometry for the reference elements being used.

fem/fe.cpp

@@ -139,12 +139,6 @@ void FiniteElement::Project (
    mfem_error ("FiniteElement::Project (...) (vector) is not overloaded !");
 }
 
-void FiniteElement::ProjectFromNodes(Vector &vc, ElementTransformation &Trans,
-                                     Vector &dofs) const
-{
-   mfem_error ("FiniteElement::ProjectFromNodes() (vector) is not overloaded!");
-}
-
 void FiniteElement::ProjectMatrixCoefficient(
    MatrixCoefficient &mc, ElementTransformation &T, Vector &dofs) const
 {
@@ -931,23 +925,6 @@ void VectorFiniteElement::Project_RT(
    }
 }
 
-void VectorFiniteElement::Project_RT(
-   const double *nk, const Array<int> &d2n,
-   Vector &vc, ElementTransformation &Trans, Vector &dofs) const
-{
-   const int sdim = Trans.GetSpaceDim();
-   const bool square_J = (dim == sdim);
-
-   for (int k = 0; k < dof; k++)
-   {
-      Trans.SetIntPoint(&Nodes.IntPoint(k));
-      // dof_k = nk^t adj(J) xk
-      Vector vk(vc.GetData()+k*sdim, sdim);
-      dofs(k) = Trans.AdjugateJacobian().InnerProduct(vk, nk + d2n[k]*dim);
-      if (!square_J) { dofs(k) /= Trans.Weight(); }
-   }
-}
-
 void VectorFiniteElement::ProjectMatrixCoefficient_RT(
    const double *nk, const Array<int> &d2n,
    MatrixCoefficient &mc, ElementTransformation &T, Vector &dofs) const
@@ -1124,19 +1101,6 @@ void VectorFiniteElement::Project_ND(
    }
 }
 
-void VectorFiniteElement::Project_ND(
-   const double *tk, const Array<int> &d2t,
-   Vector &vc, ElementTransformation &Trans, Vector &dofs) const
-{
-   for (int k = 0; k < dof; k++)
-   {
-      Trans.SetIntPoint(&Nodes.IntPoint(k));
-      Vector vk(vc.GetData()+k*dim, dim);
-      // dof_k = xk^t J tk
-      dofs(k) = Trans.Jacobian().InnerProduct(tk + d2t[k]*dim, vk);
-   }
-}
-
 void VectorFiniteElement::ProjectMatrixCoefficient_ND(
    const double *tk, const Array<int> &d2t,
    MatrixCoefficient &mc, ElementTransformation &T, Vector &dofs) const
@@ -7070,95 +7034,6 @@ void Poly_1D::Basis::Eval(const double y, Vector &u, Vector &d) const
    }
 }
 
-void Poly_1D::Basis::Eval(const double y, Vector &u, Vector &d,
-                          Vector &d2) const
-{
-   MFEM_VERIFY(etype == Barycentric,
-               "Basis::Eval with second order derivatives not implemented for"
-               " etype = " << etype);
-   switch (etype)
-   {
-      case ChangeOfBasis:
-      {
-         CalcBasis(Ai.Width() - 1, y, x, w);
-         Ai.Mult(x, u);
-         Ai.Mult(w, d);
-         // set d2 (not implemented yet)
-         break;
-      }
-      case Barycentric:
-      {
-         int i, k, p = x.Size() - 1;
-         double l, lp, lp2, lk, sk, si, sk2;
-
-         if (p == 0)
-         {
-            u(0) = 1.0;
-            d(0) = 0.0;
-            d2(0) = 0.0;
-            return;
-         }
-
-         lk = 1.0;
-         for (k = 0; k < p; k++)
-         {
-            if (y >= (x(k) + x(k+1))/2)
-            {
-               lk *= y - x(k);
-            }
-            else
-            {
-               for (i = k+1; i <= p; i++)
-               {
-                  lk *= y - x(i);
-               }
-               break;
-            }
-         }
-         l = lk * (y - x(k));
-
-         sk = 0.0;
-         sk2 = 0.0;
-         for (i = 0; i < k; i++)
-         {
-            si = 1.0/(y - x(i));
-            sk += si;
-            sk2 -= si * si;
-            u(i) = l * si * w(i);
-         }
-         u(k) = lk * w(k);
-         for (i++; i <= p; i++)
-         {
-            si = 1.0/(y - x(i));
-            sk += si;
-            sk2 -= si * si;
-            u(i) = l * si * w(i);
-         }
-         lp = l * sk + lk;
-         lp2 = lp * sk + l * sk2 + sk * lk;
-
-         for (i = 0; i < k; i++)
-         {
-            d(i) = (lp * w(i) - u(i))/(y - x(i));
-            d2(i) = (lp2 * w(i) - 2 * d(i))/(y - x(i));
-         }
-         d(k) = sk * u(k);
-         d2(k) = sk2 * u(k) + sk * d(k);
-         for (i++; i <= p; i++)
-         {
-            d(i) = (lp * w(i) - u(i))/(y - x(i));
-            d2(i) = (lp2 * w(i) - 2 * d(i))/(y - x(i));
-         }
-         break;
-      }
-      case Positive:
-         CalcBernstein(x.Size() - 1, y, u, d);
-         break;
-
-      default: break;
-   }
-}
-
 const int *Poly_1D::Binom(const int p)
 {
    if (binom.NumCols() <= p)
@@ -7714,7 +7589,6 @@ H1_SegmentElement::H1_SegmentElement(const int p, const int btype)
 #ifndef MFEM_THREAD_SAFE
    shape_x.SetSize(p+1);
    dshape_x.SetSize(p+1);
-   d2shape_x.SetSize(p+1);
 #endif
 
    Nodes.IntPoint(0).x = cp[0];
@@ -7763,25 +7637,6 @@ void H1_SegmentElement::CalcDShape(const IntegrationPoint &ip,
    }
 }
 
-void H1_SegmentElement::CalcHessian(const IntegrationPoint &ip,
-                                    DenseMatrix &Hessian) const
-{
-   const int p = order;
-
-#ifdef MFEM_THREAD_SAFE
-   Vector shape_x(p+1), dshape_x(p+1), d2shape_x(p+1);
-#endif
-
-   basis1d.Eval(ip.x, shape_x, dshape_x, d2shape_x);
-
-   Hessian(0,0) = d2shape_x(0);
-   Hessian(1,0) = d2shape_x(p);
-   for (int i = 1; i < p; i++)
-   {
-      Hessian(i+1,0) = d2shape_x(i);
-   }
-}
-
 void H1_SegmentElement::ProjectDelta(int vertex, Vector &dofs) const
 {
    const int p = order;
@@ -7822,8 +7677,6 @@ H1_QuadrilateralElement::H1_QuadrilateralElement(const int p, const int btype)
    shape_y.SetSize(p1);
    dshape_x.SetSize(p1);
    dshape_y.SetSize(p1);
-   d2shape_x.SetSize(p1);
-   d2shape_y.SetSize(p1);
 #endif
 
    int o = 0;
@@ -7877,30 +7730,6 @@ void H1_QuadrilateralElement::CalcDShape(const IntegrationPoint &ip,
    }
 }
 
-void H1_QuadrilateralElement::CalcHessian(const IntegrationPoint &ip,
-                                          DenseMatrix &Hessian) const
-{
-   const int p = order;
-
-#ifdef MFEM_THREAD_SAFE
-   Vector shape_x(p+1), shape_y(p+1), dshape_x(p+1), dshape_y(p+1),
-          d2shape_x(p+1), d2shape_y(p+1);
-#endif
-
-   basis1d.Eval(ip.x, shape_x, dshape_x, d2shape_x);
-   basis1d.Eval(ip.y, shape_y, dshape_y, d2shape_y);
-
-   for (int o = 0, j = 0; j <= p; j++)
-   {
-      for (int i = 0; i <= p; i++)
-      {
-         Hessian(dof_map[o],0) = d2shape_x(i)*  shape_y(j);
-         Hessian(dof_map[o],1) =  dshape_x(i)* dshape_y(j);
-         Hessian(dof_map[o],2) =   shape_x(i)*d2shape_y(j);  o++;
-      }
-   }
-}
-
 void H1_QuadrilateralElement::ProjectDelta(int vertex, Vector &dofs) const
 {
    const int p = order;
@@ -7964,9 +7793,6 @@ H1_HexahedronElement::H1_HexahedronElement(const int p, const int btype)
    dshape_x.SetSize(p1);
    dshape_y.SetSize(p1);
    dshape_z.SetSize(p1);
-   d2shape_x.SetSize(p1);
-   d2shape_y.SetSize(p1);
-   d2shape_z.SetSize(p1);
 #endif
 
    int o = 0;
@@ -8023,35 +7849,6 @@ void H1_HexahedronElement::CalcDShape(const IntegrationPoint &ip,
          }
 }
 
-void H1_HexahedronElement::CalcHessian(const IntegrationPoint &ip,
-                                       DenseMatrix &Hessian) const
-{
-   const int p = order;
-
-#ifdef MFEM_THREAD_SAFE
-   Vector shape_x(p+1),  shape_y(p+1),  shape_z(p+1);
-   Vector dshape_x(p+1), dshape_y(p+1), dshape_z(p+1);
-   Vector d2shape_x(p+1), d2shape_y(p+1), ds2hape_z(p+1);
-#endif
-
-   basis1d.Eval(ip.x, shape_x, dshape_x, d2shape_x);
-   basis1d.Eval(ip.y, shape_y, dshape_y, d2shape_y);
-   basis1d.Eval(ip.z, shape_z, dshape_z, d2shape_z);
-
-   for (int o = 0, k = 0; k <= p; k++)
-      for (int j = 0; j <= p; j++)
-         for (int i = 0; i <= p; i++)
-         {
-            Hessian(dof_map[o],0) = d2shape_x(i)*  shape_y(j)*  shape_z(k);
-            Hessian(dof_map[o],1) =  dshape_x(i)* dshape_y(j)*  shape_z(k);
-            Hessian(dof_map[o],2) =  dshape_x(i)*  shape_y(j)* dshape_z(k);
-            Hessian(dof_map[o],3) =   shape_x(i)*d2shape_y(j)*  shape_z(k);
-            Hessian(dof_map[o],4) =   shape_x(i)* dshape_y(j)* dshape_z(k);
-            Hessian(dof_map[o],5) =   shape_x(i)*  shape_y(j)*d2shape_z(k);
-            o++;
-         }
-}
-
 void H1_HexahedronElement::ProjectDelta(int vertex, Vector &dofs) const
 {
    const int p = order;

fem/fe.hpp

@@ -446,7 +446,7 @@ public:
    /** Each row of the result DenseMatrix @a Hessian contains upper triangular
        part of the Hessian of one shape function.
        The order in 2D is {u_xx, u_xy, u_yy}.
-       The size (#dof x (#dim (#dim+1)/2) of @a Hessian must be set in advance.*/
+       The size (#dof x (#dim (#dim-1)/2) of @a Hessian must be set in advance.*/
    virtual void CalcHessian (const IntegrationPoint &ip,
                              DenseMatrix &Hessian) const;
 
@@ -504,21 +504,14 @@ public:
    /** @brief Given a coefficient and a transformation, compute its projection
        (approximation) in the local finite dimensional space in terms
        of the degrees of freedom. */
-   virtual void Project(Coefficient &coeff,
-                        ElementTransformation &Trans, Vector &dofs) const;
+   virtual void Project (Coefficient &coeff,
+                         ElementTransformation &Trans, Vector &dofs) const;
 
    /** @brief Given a vector coefficient and a transformation, compute its
        projection (approximation) in the local finite dimensional space
        in terms of the degrees of freedom. (VectorFiniteElements) */
-   virtual void Project(VectorCoefficient &vc,
-                        ElementTransformation &Trans, Vector &dofs) const;
-
-   /** @brief Given a vector of values at the finite element nodes and a
-       transformation, compute its projection (approximation) in the local
-       finite dimensional space in terms of the degrees of freedom. Valid for
-       VectorFiniteElements. */
-   virtual void ProjectFromNodes(Vector &vc, ElementTransformation &Trans,
-                                 Vector &dofs) const;
+   virtual void Project (VectorCoefficient &vc,
+                         ElementTransformation &Trans, Vector &dofs) const;
 
    /** @brief Given a matrix coefficient and a transformation, compute an
        approximation ("projection") in the local finite dimensional space in
@@ -804,12 +797,7 @@ protected:
                    VectorCoefficient &vc, ElementTransformation &Trans,
                    Vector &dofs) const;
 
-   /// Projects the vector of values given at FE nodes to RT space
-   void Project_RT(const double *nk, const Array<int> &d2n,
-                   Vector &vc, ElementTransformation &Trans,
-                   Vector &dofs) const;
-
-   /// Project the rows of the matrix coefficient in an RT space
+   // project the rows of the matrix coefficient in an RT space
    void ProjectMatrixCoefficient_RT(
       const double *nk, const Array<int> &d2n,
       MatrixCoefficient &mc, ElementTransformation &T, Vector &dofs) const;
@@ -837,12 +825,7 @@ protected:
                    VectorCoefficient &vc, ElementTransformation &Trans,
                    Vector &dofs) const;
 
-   /// Projects the vector of values given at FE nodes to ND space
-   void Project_ND(const double *tk, const Array<int> &d2t,
-                   Vector &vc, ElementTransformation &Trans,
-                   Vector &dofs) const;
-
-   /// Project the rows of the matrix coefficient in an ND space
+   /// project the rows of the matrix coefficient in an ND space
    void ProjectMatrixCoefficient_ND(
       const double *tk, const Array<int> &d2t,
       MatrixCoefficient &mc, ElementTransformation &T, Vector &dofs) const;
@@ -1867,7 +1850,6 @@ public:
       Basis(const int p, const double *nodes, EvalType etype = Barycentric);
       void Eval(const double x, Vector &u) const;
       void Eval(const double x, Vector &u, Vector &d) const;
-      void Eval(const double x, Vector &u, Vector &d, Vector &d2) const;
    };
 
 private:
@@ -2118,7 +2100,7 @@ class H1_SegmentElement : public NodalTensorFiniteElement
 {
 private:
 #ifndef MFEM_THREAD_SAFE
-   mutable Vector shape_x, dshape_x, d2shape_x;
+   mutable Vector shape_x, dshape_x;
 #endif
 
 public:
@@ -2127,8 +2109,6 @@ public:
    virtual void CalcShape(const IntegrationPoint &ip, Vector &shape) const;
    virtual void CalcDShape(const IntegrationPoint &ip,
                            DenseMatrix &dshape) const;
-   virtual void CalcHessian(const IntegrationPoint &ip,
-                            DenseMatrix &Hessian) const;
    virtual void ProjectDelta(int vertex, Vector &dofs) const;
 };
 
@@ -2138,7 +2118,7 @@ class H1_QuadrilateralElement : public NodalTensorFiniteElement
 {
 private:
 #ifndef MFEM_THREAD_SAFE
-   mutable Vector shape_x, shape_y, dshape_x, dshape_y, d2shape_x, d2shape_y;
+   mutable Vector shape_x, shape_y, dshape_x, dshape_y;
 #endif
 
 public:
@@ -2148,8 +2128,6 @@ public:
    virtual void CalcShape(const IntegrationPoint &ip, Vector &shape) const;
    virtual void CalcDShape(const IntegrationPoint &ip,
                            DenseMatrix &dshape) const;
-   virtual void CalcHessian(const IntegrationPoint &ip,
-                            DenseMatrix &Hessian) const;
    virtual void ProjectDelta(int vertex, Vector &dofs) const;
 };
 
@@ -2159,8 +2137,7 @@ class H1_HexahedronElement : public NodalTensorFiniteElement
 {
 private:
 #ifndef MFEM_THREAD_SAFE
-   mutable Vector shape_x, shape_y, shape_z, dshape_x, dshape_y, dshape_z,
-           d2shape_x, d2shape_y, d2shape_z;
+   mutable Vector shape_x, shape_y, shape_z, dshape_x, dshape_y, dshape_z;
 #endif
 
 public:
@@ -2169,8 +2146,6 @@ public:
    virtual void CalcShape(const IntegrationPoint &ip, Vector &shape) const;
    virtual void CalcDShape(const IntegrationPoint &ip,
                            DenseMatrix &dshape) const;
-   virtual void CalcHessian(const IntegrationPoint &ip,
-                            DenseMatrix &Hessian) const;
    virtual void ProjectDelta(int vertex, Vector &dofs) const;
 };
 
@@ -2706,9 +2681,6 @@ public:
    virtual void Project(VectorCoefficient &vc,
                         ElementTransformation &Trans, Vector &dofs) const
    { Project_RT(nk, dof2nk, vc, Trans, dofs); }
-   virtual void ProjectFromNodes(Vector &vc, ElementTransformation &Trans,
-                                 Vector &dofs) const
-   { Project_RT(nk, dof2nk, vc, Trans, dofs); }
    virtual void ProjectMatrixCoefficient(
       MatrixCoefficient &mc, ElementTransformation &T, Vector &dofs) const
    { ProjectMatrixCoefficient_RT(nk, dof2nk, mc, T, dofs); }
@@ -2767,9 +2739,6 @@ public:
    virtual void Project(VectorCoefficient &vc,
                         ElementTransformation &Trans, Vector &dofs) const
    { Project_RT(nk, dof2nk, vc, Trans, dofs); }
-   virtual void ProjectFromNodes(Vector &vc, ElementTransformation &Trans,
-                                 Vector &dofs) const
-   { Project_RT(nk, dof2nk, vc, Trans, dofs); }
    virtual void ProjectMatrixCoefficient(
       MatrixCoefficient &mc, ElementTransformation &T, Vector &dofs) const
    { ProjectMatrixCoefficient_RT(nk, dof2nk, mc, T, dofs); }
@@ -2821,9 +2790,6 @@ public:
    virtual void Project(VectorCoefficient &vc,
                         ElementTransformation &Trans, Vector &dofs) const
    { Project_RT(nk, dof2nk, vc, Trans, dofs); }
-   virtual void ProjectFromNodes(Vector &vc, ElementTransformation &Trans,
-                                 Vector &dofs) const
-   { Project_RT(nk, dof2nk, vc, Trans, dofs); }
    virtual void ProjectMatrixCoefficient(
       MatrixCoefficient &mc, ElementTransformation &T, Vector &dofs) const
    { ProjectMatrixCoefficient_RT(nk, dof2nk, mc, T, dofs); }
@@ -2881,9 +2847,6 @@ public:
    virtual void Project(VectorCoefficient &vc,
                         ElementTransformation &Trans, Vector &dofs) const
    { Project_RT(nk, dof2nk, vc, Trans, dofs); }
-   virtual void ProjectFromNodes(Vector &vc, ElementTransformation &Trans,
-                                 Vector &dofs) const
-   { Project_RT(nk, dof2nk, vc, Trans, dofs); }
    virtual void ProjectMatrixCoefficient(
       MatrixCoefficient &mc, ElementTransformation &T, Vector &dofs) const
    { ProjectMatrixCoefficient_RT(nk, dof2nk, mc, T, dofs); }
@@ -2943,10 +2906,6 @@ public:
                         ElementTransformation &Trans, Vector &dofs) const
    { Project_ND(tk, dof2tk, vc, Trans, dofs); }
 
-   virtual void ProjectFromNodes(Vector &vc, ElementTransformation &Trans,
-                                 Vector &dofs) const
-   { Project_ND(tk, dof2tk, vc, Trans, dofs); }
-
    virtual void ProjectMatrixCoefficient(
       MatrixCoefficient &mc, ElementTransformation &T, Vector &dofs) const
    { ProjectMatrixCoefficient_ND(tk, dof2tk, mc, T, dofs); }
@@ -3006,9 +2965,6 @@ public:
    virtual void Project(VectorCoefficient &vc,
                         ElementTransformation &Trans, Vector &dofs) const
    { Project_ND(tk, dof2tk, vc, Trans, dofs); }
-   virtual void ProjectFromNodes(Vector &vc, ElementTransformation &Trans,
-                                 Vector &dofs) const
-   { Project_ND(tk, dof2tk, vc, Trans, dofs); }
    virtual void ProjectMatrixCoefficient(
       MatrixCoefficient &mc, ElementTransformation &T, Vector &dofs) const
    { ProjectMatrixCoefficient_ND(tk, dof2tk, mc, T, dofs); }
@@ -3060,9 +3016,6 @@ public:
    virtual void Project(VectorCoefficient &vc,
                         ElementTransformation &Trans, Vector &dofs) const
    { Project_ND(tk, dof2tk, vc, Trans, dofs); }
-   virtual void ProjectFromNodes(Vector &vc, ElementTransformation &Trans,
-                                 Vector &dofs) const
-   { Project_ND(tk, dof2tk, vc, Trans, dofs); }
    virtual void ProjectMatrixCoefficient(
       MatrixCoefficient &mc, ElementTransformation &T, Vector &dofs) const
    { ProjectMatrixCoefficient_ND(tk, dof2tk, mc, T, dofs); }
@@ -3119,9 +3072,6 @@ public:
    virtual void Project(VectorCoefficient &vc,
                         ElementTransformation &Trans, Vector &dofs) const
    { Project_ND(tk, dof2tk, vc, Trans, dofs); }
-   virtual void ProjectFromNodes(Vector &vc, ElementTransformation &Trans,
-                                 Vector &dofs) const
-   { Project_ND(tk, dof2tk, vc, Trans, dofs); }
    virtual void ProjectMatrixCoefficient(
       MatrixCoefficient &mc, ElementTransformation &T, Vector &dofs) const
    { ProjectMatrixCoefficient_ND(tk, dof2tk, mc, T, dofs); }

fem/fem.hpp

@@ -19,7 +19,6 @@
 #include "eltrans.hpp"
 #include "coefficient.hpp"
 #include "complex_fem.hpp"
-#include "convergence.hpp"
 #include "lininteg.hpp"
 #include "nonlininteg.hpp"
 #include "bilininteg.hpp"

fem/fespace.cpp

@@ -440,7 +440,6 @@ void FiniteElementSpace::MarkerToList(const Array<int> &marker,
       if (marker[i]) { num_marked++; }
    }
    list.SetSize(0);
-   list.HostWrite();
    list.Reserve(num_marked);
    for (int i = 0; i < marker.Size(); i++)
    {
@@ -452,9 +451,7 @@ void FiniteElementSpace::MarkerToList(const Array<int> &marker,
 void FiniteElementSpace::ListToMarker(const Array<int> &list, int marker_size,
                                       Array<int> &marker, int mark_val)
 {
-   list.HostRead(); // make sure we can read the array on host
    marker.SetSize(marker_size);
-   marker.HostWrite();
    marker = 0;
    for (int i = 0; i < list.Size(); i++)
    {
@@ -947,7 +944,7 @@ const Operator *FiniteElementSpace::GetFaceRestriction(
 }
 
 const QuadratureInterpolator *FiniteElementSpace::GetQuadratureInterpolator(
-   const IntegrationRule &ir) const
+   const IntegrationRule &ir, const DofToQuad::Mode mode) const
 {
    for (int i = 0; i < E2Q_array.Size(); i++)
    {
@@ -955,13 +952,13 @@ const QuadratureInterpolator *FiniteElementSpace::GetQuadratureInterpolator(
       if (qi->IntRule == &ir) { return qi; }
    }
 
-   QuadratureInterpolator *qi = new QuadratureInterpolator(*this, ir);
+   QuadratureInterpolator *qi = new QuadratureInterpolator(*this, ir, mode);
    E2Q_array.Append(qi);
    return qi;
 }
 
 const QuadratureInterpolator *FiniteElementSpace::GetQuadratureInterpolator(
-   const QuadratureSpace &qs) const
+   const QuadratureSpace &qs, const DofToQuad::Mode mode) const
 {
    for (int i = 0; i < E2Q_array.Size(); i++)
    {
@@ -969,7 +966,7 @@ const QuadratureInterpolator *FiniteElementSpace::GetQuadratureInterpolator(
       if (qi->qspace == &qs) { return qi; }
    }
 
-   QuadratureInterpolator *qi = new QuadratureInterpolator(*this, qs);
+   QuadratureInterpolator *qi = new QuadratureInterpolator(*this, qs, mode);
    E2Q_array.Append(qi);
    return qi;
 }
@@ -986,8 +983,8 @@ const FaceQuadratureInterpolator
          if (qi->IntRule == &ir) { return qi; }
       }
 
-      FaceQuadratureInterpolator *qi = new FaceQuadratureInterpolator(*this, ir,
-                                                                      type);
+      FaceQuadratureInterpolator *qi =
+         new FaceQuadratureInterpolator(*this, ir, type);
       E2IFQ_array.Append(qi);
       return qi;
    }

fem/fespace.hpp

@@ -367,7 +367,7 @@ public:
        All elements will use the same IntegrationRule, @a ir as the target
        quadrature points. */
    const QuadratureInterpolator *GetQuadratureInterpolator(
-      const IntegrationRule &ir) const;
+      const IntegrationRule &ir, const DofToQuad::Mode = DofToQuad::FULL) const;
 
    /** @brief Return a QuadratureInterpolator that interpolates E-vectors to
        quadrature point values and/or derivatives (Q-vectors). */
@@ -378,7 +378,7 @@ public:
        The target quadrature points in the elements are described by the given
        QuadratureSpace, @a qs. */
    const QuadratureInterpolator *GetQuadratureInterpolator(
-      const QuadratureSpace &qs) const;
+      const QuadratureSpace &qs, const DofToQuad::Mode = DofToQuad::FULL) const;
 
    /** @brief Return a FaceQuadratureInterpolator that interpolates E-vectors to
        quadrature point values and/or derivatives (Q-vectors). */

fem/gridfunc.cpp

@@ -199,7 +199,8 @@ void GridFunction::MakeRef(FiniteElementSpace *f, Vector &v, int v_offset)
    if (f != fes) { Destroy(); }
    fes = f;
    v.UseDevice(true);
-   this->Vector::MakeRef(v, v_offset, fes->GetVSize());
+   NewMemoryAndSize(Memory<double>(v.GetMemory(), v_offset, fes->GetVSize()),
+                    fes->GetVSize(), true);
    sequence = fes->GetSequence();
 }
 
@@ -1833,19 +1834,6 @@ void GridFunction::ImposeBounds(int i, const Vector &weights,
    ImposeBounds(i, weights, minv, maxv);
 }
 
-void GridFunction::RestrictConforming()
-{
-   const SparseMatrix *R = fes->GetRestrictionMatrix();
-   const Operator *P = fes->GetProlongationMatrix();
-
-   if (P && R)
-   {
-      Vector tmp(R->Height());
-      R->Mult(*this, tmp);
-      P->Mult(tmp, *this);
-   }
-}
-
 void GridFunction::GetNodalValues(Vector &nval, int vdim) const
 {
    int i, j;
@@ -2614,7 +2602,11 @@ double GridFunction::ComputeL2Error(
       }
    }
 
-   return (error < 0.0) ? -sqrt(-error) : sqrt(error);
+   if (error < 0.0)
+   {
+      return -sqrt(-error);
+   }
+   return sqrt(error);
 }
 
 double GridFunction::ComputeL2Error(
@@ -2655,199 +2647,94 @@ double GridFunction::ComputeL2Error(
       }
    }
 
-   return (error < 0.0) ? -sqrt(-error) : sqrt(error);
-}
-
-double GridFunction::ComputeGradError(VectorCoefficient *exgrad,
-                                      const IntegrationRule *irs[]) const
-{
-   double error = 0.0;
-   const FiniteElement *fe;
-   ElementTransformation *Tr;
-   Array<int> dofs;
-   Vector grad;
-   int intorder;
-   int dim = fes->GetMesh()->SpaceDimension();
-   Vector vec(dim);
-
-   for (int i = 0; i < fes->GetNE(); i++)
+   if (error < 0.0)
    {
-      fe = fes->GetFE(i);
-      Tr = fes->GetElementTransformation(i);
-      intorder = 2*fe->GetOrder() + 3; // <--------
-      const IntegrationRule *ir;
-      if (irs)
-      {
-         ir = irs[fe->GetGeomType()];
-      }
-      else
-      {
-         ir = &(IntRules.Get(fe->GetGeomType(), intorder));
-      }
-      fes->GetElementDofs(i, dofs);
-      for (int j = 0; j < ir->GetNPoints(); j++)
-      {
-         const IntegrationPoint &ip = ir->IntPoint(j);
-         Tr->SetIntPoint(&ip);
-         GetGradient(*Tr,grad);
-         exgrad->Eval(vec,*Tr,ip);
-         vec-=grad;
-         error += ip.weight * Tr->Weight() * (vec * vec);
-      }
+      return -sqrt(-error);
    }
-   return (error < 0.0) ? -sqrt(-error) : sqrt(error);
+   return sqrt(error);
 }
 
-double GridFunction::ComputeCurlError(VectorCoefficient *excurl,
-                                      const IntegrationRule *irs[]) const
+double GridFunction::ComputeH1Error(
+   Coefficient *exsol, VectorCoefficient *exgrad,
+   Coefficient *ell_coeff, double Nu, int norm_type) const
 {
-   double error = 0.0;
-   const FiniteElement *fe;
-   ElementTransformation *Tr;
-   Array<int> dofs;
-   Vector curl;
-   int intorder;
-   int dim = fes->GetMesh()->SpaceDimension();
-   int n = (dim == 3) ? dim : 1;
-   Vector vec(n);
-
-   for (int i = 0; i < fes->GetNE(); i++)
-   {
-      fe = fes->GetFE(i);
-      Tr = fes->GetElementTransformation(i);
-      intorder = 2*fe->GetOrder() + 3;
-      const IntegrationRule *ir;
-      if (irs)
-      {
-         ir = irs[fe->GetGeomType()];
-      }
-      else
-      {
-         ir = &(IntRules.Get(fe->GetGeomType(), intorder));
-      }
-      fes->GetElementDofs(i, dofs);
-      for (int j = 0; j < ir->GetNPoints(); j++)
-      {
-         const IntegrationPoint &ip = ir->IntPoint(j);
-         Tr->SetIntPoint(&ip);
-         GetCurl(*Tr,curl);
-         excurl->Eval(vec,*Tr,ip);
-         vec-=curl;
-         error += ip.weight * Tr->Weight() * ( vec * vec );
-      }
-   }
-
-   return (error < 0.0) ? -sqrt(-error) : sqrt(error);
-}
-
-double GridFunction::ComputeDivError(
-   Coefficient *exdiv, const IntegrationRule *irs[]) const
-{
-   double error = 0.0, a;
-   const FiniteElement *fe;
-   ElementTransformation *Tr;
-   Array<int> dofs;
-   int intorder;
-
-   for (int i = 0; i < fes->GetNE(); i++)
-   {
-      fe = fes->GetFE(i);
-      Tr = fes->GetElementTransformation(i);
-      intorder = 2*fe->GetOrder() + 3;
-      const IntegrationRule *ir;
-      if (irs)
-      {
-         ir = irs[fe->GetGeomType()];
-      }
-      else
-      {
-         ir = &(IntRules.Get(fe->GetGeomType(), intorder));
-      }
-      fes->GetElementDofs(i, dofs);
-      for (int j = 0; j < ir->GetNPoints(); j++)
-      {
-         const IntegrationPoint &ip = ir->IntPoint(j);
-         Tr->SetIntPoint (&ip);
-         a = GetDivergence(*Tr) - exdiv->Eval(*Tr, ip);
-         error += ip.weight * Tr->Weight() * a * a;
-      }
-   }
-
-   return (error < 0.0) ? -sqrt(-error) : sqrt(error);
-}
-
-double GridFunction::ComputeDGFaceJumpError(Coefficient *exsol,
-                                            Coefficient *ell_coeff, double Nu,
-                                            const IntegrationRule *irs[])  const
-{
-   int fdof, dim, intorder, k;
+   // assuming vdim is 1
+   int i, fdof, dim, intorder, j, k;
    Mesh *mesh;
    const FiniteElement *fe;
    ElementTransformation *transf;
    FaceElementTransformations *face_elem_transf;
-   Vector shape, el_dofs, err_val, ell_coeff_val;
+   Vector e_grad, a_grad, shape, el_dofs, err_val, ell_coeff_val;
+   DenseMatrix dshape, dshapet, Jinv;
    Array<int> vdofs;
    IntegrationPoint eip;
    double error = 0.0;
 
    mesh = fes->GetMesh();
    dim = mesh->Dimension();
+   e_grad.SetSize(dim);
+   a_grad.SetSize(dim);
+   Jinv.SetSize(dim);
 
-   for (int i = 0; i < mesh->GetNumFaces(); i++)
-   {
-      face_elem_transf = mesh->GetFaceElementTransformations(i, 5);
-      int i1 = face_elem_transf->Elem1No;
-      int i2 = face_elem_transf->Elem2No;
-      intorder = fes->GetFE(i1)->GetOrder();
-      if (i2 >= 0)
-         if ( (k = fes->GetFE(i2)->GetOrder()) > intorder )
-         {
-            intorder = k;
-         }
-      intorder = 2 * intorder;  // <-------------
-      const IntegrationRule *ir;
-      if (irs)
+   if (norm_type & 1)
+      for (i = 0; i < mesh->GetNE(); i++)
       {
-         ir = irs[face_elem_transf->GetGeometryType()];
-      }
-      else
-      {
-         ir = &(IntRules.Get(face_elem_transf->GetGeometryType(), intorder));
-      }
-      err_val.SetSize(ir->GetNPoints());
-      ell_coeff_val.SetSize(ir->GetNPoints());
-      // side 1
-      transf = face_elem_transf->Elem1;
-      fe = fes->GetFE(i1);
-      fdof = fe->GetDof();
-      fes->GetElementVDofs(i1, vdofs);
-      shape.SetSize(fdof);
-      el_dofs.SetSize(fdof);
-      for (k = 0; k < fdof; k++)
-         if (vdofs[k] >= 0)
-         {
-            el_dofs(k) =   (*this)(vdofs[k]);
-         }
-         else
-         {
-            el_dofs(k) = - (*this)(-1-vdofs[k]);
-         }
-      for (int j = 0; j < ir->GetNPoints(); j++)
-      {
-         face_elem_transf->Loc1.Transform(ir->IntPoint(j), eip);
-         fe->CalcShape(eip, shape);
-         transf->SetIntPoint(&eip);
-         ell_coeff_val(j) = ell_coeff->Eval(*transf, eip);
-         err_val(j) = exsol->Eval(*transf, eip) - (shape * el_dofs);
-      }
-      if (i2 >= 0)
-      {
-         // side 2
-         face_elem_transf = mesh->GetFaceElementTransformations(i, 10);
-         transf = face_elem_transf->Elem2;
-         fe = fes->GetFE(i2);
+         fe = fes->GetFE(i);
          fdof = fe->GetDof();
-         fes->GetElementVDofs(i2, vdofs);
+         transf = mesh->GetElementTransformation(i);
+         el_dofs.SetSize(fdof);
+         dshape.SetSize(fdof, dim);
+         dshapet.SetSize(fdof, dim);
+         intorder = 2 * fe->GetOrder(); // <----------
+         const IntegrationRule &ir = IntRules.Get(fe->GetGeomType(), intorder);
+         fes->GetElementVDofs(i, vdofs);
+         for (k = 0; k < fdof; k++)
+            if (vdofs[k] >= 0)
+            {
+               el_dofs(k) =   (*this)(vdofs[k]);
+            }
+            else
+            {
+               el_dofs(k) = - (*this)(-1-vdofs[k]);
+            }
+         for (j = 0; j < ir.GetNPoints(); j++)
+         {
+            const IntegrationPoint &ip = ir.IntPoint(j);
+            fe->CalcDShape(ip, dshape);
+            transf->SetIntPoint(&ip);
+            exgrad->Eval(e_grad, *transf, ip);
+            CalcInverse(transf->Jacobian(), Jinv);
+            Mult(dshape, Jinv, dshapet);
+            dshapet.MultTranspose(el_dofs, a_grad);
+            e_grad -= a_grad;
+            error += (ip.weight * transf->Weight() *
+                      ell_coeff->Eval(*transf, ip) *
+                      (e_grad * e_grad));
+         }
+      }
+
+   if (norm_type & 2)
+      for (i = 0; i < mesh->GetNFaces(); i++)
+      {
+         face_elem_transf = mesh->GetFaceElementTransformations(i, 5);
+         int i1 = face_elem_transf->Elem1No;
+         int i2 = face_elem_transf->Elem2No;
+         intorder = fes->GetFE(i1)->GetOrder();
+         if (i2 >= 0)
+            if ( (k = fes->GetFE(i2)->GetOrder()) > intorder )
+            {
+               intorder = k;
+            }
+         intorder = 2 * intorder;  // <-------------
+         const IntegrationRule &ir =
+            IntRules.Get(face_elem_transf->GetGeometryType(), intorder);
+         err_val.SetSize(ir.GetNPoints());
+         ell_coeff_val.SetSize(ir.GetNPoints());
+         // side 1
+         transf = face_elem_transf->Elem1;
+         fe = fes->GetFE(i1);
+         fdof = fe->GetDof();
+         fes->GetElementVDofs(i1, vdofs);
          shape.SetSize(fdof);
          el_dofs.SetSize(fdof);
          for (k = 0; k < fdof; k++)
@@ -2859,69 +2746,60 @@ double GridFunction::ComputeDGFaceJumpError(Coefficient *exsol,
             {
                el_dofs(k) = - (*this)(-1-vdofs[k]);
             }
-         for (int j = 0; j < ir->GetNPoints(); j++)
+         for (j = 0; j < ir.GetNPoints(); j++)
          {
-            face_elem_transf->Loc2.Transform(ir->IntPoint(j), eip);
+            face_elem_transf->Loc1.Transform(ir.IntPoint(j), eip);
             fe->CalcShape(eip, shape);
             transf->SetIntPoint(&eip);
-            ell_coeff_val(j) += ell_coeff->Eval(*transf, eip);
-            ell_coeff_val(j) *= 0.5;
-            err_val(j) -= (exsol->Eval(*transf, eip) - (shape * el_dofs));
+            ell_coeff_val(j) = ell_coeff->Eval(*transf, eip);
+            err_val(j) = exsol->Eval(*transf, eip) - (shape * el_dofs);
+         }
+         if (i2 >= 0)
+         {
+            // side 2
+            face_elem_transf = mesh->GetFaceElementTransformations(i, 10);
+            transf = face_elem_transf->Elem2;
+            fe = fes->GetFE(i2);
+            fdof = fe->GetDof();
+            fes->GetElementVDofs(i2, vdofs);
+            shape.SetSize(fdof);
+            el_dofs.SetSize(fdof);
+            for (k = 0; k < fdof; k++)
+               if (vdofs[k] >= 0)
+               {
+                  el_dofs(k) =   (*this)(vdofs[k]);
+               }
+               else
+               {
+                  el_dofs(k) = - (*this)(-1-vdofs[k]);
+               }
+            for (j = 0; j < ir.GetNPoints(); j++)
+            {
+               face_elem_transf->Loc2.Transform(ir.IntPoint(j), eip);
+               fe->CalcShape(eip, shape);
+               transf->SetIntPoint(&eip);
+               ell_coeff_val(j) += ell_coeff->Eval(*transf, eip);
+               ell_coeff_val(j) *= 0.5;
+               err_val(j) -= (exsol->Eval(*transf, eip) - (shape * el_dofs));
+            }
+         }
+         face_elem_transf = mesh->GetFaceElementTransformations(i, 16);
+         transf = face_elem_transf;
+         for (j = 0; j < ir.GetNPoints(); j++)
+         {
+            const IntegrationPoint &ip = ir.IntPoint(j);
+            transf->SetIntPoint(&ip);
+            error += (ip.weight * Nu * ell_coeff_val(j) *
+                      pow(transf->Weight(), 1.0-1.0/(dim-1)) *
+                      err_val(j) * err_val(j));
          }
       }
-      face_elem_transf = mesh->GetFaceElementTransformations(i, 16);
-      transf = face_elem_transf;
-      for (int j = 0; j < ir->GetNPoints(); j++)
-      {
-         const IntegrationPoint &ip = ir->IntPoint(j);
-         transf->SetIntPoint(&ip);
-         error += (ip.weight * Nu * ell_coeff_val(j) *
-                   pow(transf->Weight(), 1.0-1.0/(dim-1)) *
-                   err_val(j) * err_val(j));
-      }
+
+   if (error < 0.0)
+   {
+      return -sqrt(-error);
    }
-
-   return (error < 0.0) ? -sqrt(-error) : sqrt(error);
-}
-
-double GridFunction::ComputeH1Error(Coefficient *exsol,
-                                    VectorCoefficient *exgrad,
-                                    Coefficient *ell_coef, double Nu,
-                                    int norm_type) const
-{
-   double error1 = 0.0;
-   double error2 = 0.0;
-   if (norm_type & 1) { error1 = GridFunction::ComputeGradError(exgrad); }
-   if (norm_type & 2) { error2 = GridFunction::ComputeDGFaceJumpError(exsol,ell_coef,Nu); }
-
-   return sqrt(error1 * error1 + error2 * error2);
-}
-
-double GridFunction::ComputeH1Error(Coefficient *exsol,
-                                    VectorCoefficient *exgrad,
-                                    const IntegrationRule *irs[]) const
-{
-   double L2error = GridFunction::ComputeLpError(2.0,*exsol,NULL,irs);
-   double GradError = ComputeGradError(exgrad,irs);
-   return sqrt(L2error*L2error + GradError*GradError);
-}
-
-double GridFunction::ComputeHDivError(VectorCoefficient *exsol,
-                                      Coefficient *exdiv,
-                                      const IntegrationRule *irs[]) const
-{
-   double L2error = GridFunction::ComputeLpError(2.0,*exsol,NULL,NULL,irs);
-   double DivError = ComputeDivError(exdiv,irs);
-   return sqrt(L2error*L2error + DivError*DivError);
-}
-
-double GridFunction::ComputeHCurlError(VectorCoefficient *exsol,
-                                       VectorCoefficient *excurl,
-                                       const IntegrationRule *irs[]) const
-{
-   double L2error = GridFunction::ComputeLpError(2.0,*exsol,NULL,NULL,irs);
-   double CurlError = ComputeCurlError(excurl,irs);
-   return sqrt(L2error*L2error + CurlError*CurlError);
+   return sqrt(error);
 }
 
 double GridFunction::ComputeMaxError(
@@ -2977,6 +2855,7 @@ double GridFunction::ComputeMaxError(
          }
       }
    }
+
    return error;
 }
 

fem/gridfunc.hpp

@@ -334,11 +334,6 @@ public:
    void ImposeBounds(int i, const Vector &weights,
                      double _min = 0.0, double _max = infinity());
 
-   /** On a non-conforming mesh, make sure the function lies in the conforming
-       space by multiplying with R and then with P, the conforming restriction
-       and prolongation matrices of the space, respectively. */
-   void RestrictConforming();
-
    /** @brief Project the @a src GridFunction to @a this GridFunction, both of
        which must be on the same mesh. */
    /** The current implementation assumes that all elements use the same
@@ -427,7 +422,6 @@ public:
    virtual void ProjectBdrCoefficientTangent(VectorCoefficient &vcoeff,
                                              Array<int> &bdr_attr);
 
-
    virtual double ComputeL2Error(Coefficient &exsol,
                                  const IntegrationRule *irs[] = NULL) const
    { return ComputeLpError(2.0, exsol, NULL, irs); }
@@ -439,50 +433,10 @@ public:
                                  const IntegrationRule *irs[] = NULL,
                                  Array<int> *elems = NULL) const;
 
-   /// Returns ||grad u_ex - grad u_h||_L2 for H1 or L2 elements
-   virtual double ComputeGradError(VectorCoefficient *exgrad,
-                                   const IntegrationRule *irs[] = NULL) const;
-
-   /// Returns ||curl u_ex - curl u_h||_L2 for ND elements
-   virtual double ComputeCurlError(VectorCoefficient *excurl,
-                                   const IntegrationRule *irs[] = NULL) const;
-
-   /// Returns ||div u_ex - div u_h||_L2 for RT elements
-   virtual double ComputeDivError(Coefficient *exdiv,
-                                  const IntegrationRule *irs[] = NULL) const;
-
-   /// Returns the Face Jumps error for L2 elements
-   virtual double ComputeDGFaceJumpError(Coefficient *exsol,
-                                         Coefficient *ell_coeff,
-                                         double Nu,
-                                         const IntegrationRule *irs[] = NULL)
-   const;
-
-   /** This method is kept for backward compatibility.
-
-       Returns either the H1-seminorm, or the DG face jumps error, or both
-       depending on norm_type = 1, 2, 3. Additional arguments for the DG face
-       jumps norm: ell_coeff: mesh-depended coefficient (weight) Nu: scalar
-       constant weight */
    virtual double ComputeH1Error(Coefficient *exsol, VectorCoefficient *exgrad,
                                  Coefficient *ell_coef, double Nu,
                                  int norm_type) const;
 
-   /// Returns the error measured in H1-norm for H1 elements or in "broken"
-   /// H1-norm for L2 elements
-   virtual double ComputeH1Error(Coefficient *exsol, VectorCoefficient *exgrad,
-                                 const IntegrationRule *irs[] = NULL) const;
-
-   /// Returns the error measured in H(div)-norm for RT elements
-   virtual double ComputeHDivError(VectorCoefficient *exsol,
-                                   Coefficient *exdiv,
-                                   const IntegrationRule *irs[] = NULL) const;
-
-   /// Returns the error measured in H(curl)-norm for ND elements
-   virtual double ComputeHCurlError(VectorCoefficient *exsol,
-                                    VectorCoefficient *excurl,
-                                    const IntegrationRule *irs[] = NULL) const;
-
    virtual double ComputeMaxError(Coefficient &exsol,
                                   const IntegrationRule *irs[] = NULL) const
    {

fem/gslib.cpp

@@ -29,13 +29,10 @@ namespace mfem
 {
 
 FindPointsGSLIB::FindPointsGSLIB()
-   : mesh(NULL), meshsplit(NULL), ir_simplex(NULL),
-     fdata2D(NULL), fdata3D(NULL), cr(NULL), gsl_comm(NULL),
-     dim(-1), points_cnt(0), setupflag(false), default_interp_value(0),
-     avgtype(AvgType::ARITHMETIC)
+   : mesh(NULL), ir_simplex(NULL), fdata2D(NULL), fdata3D(NULL),
+     dim(-1), gsl_mesh(), gsl_ref(), gsl_dist(), setupflag(false)
 {
    gsl_comm = new comm;
-   cr       = new crystal;
 #ifdef MFEM_USE_MPI
    int initialized;
    MPI_Initialized(&initialized);
@@ -50,20 +47,15 @@ FindPointsGSLIB::FindPointsGSLIB()
 FindPointsGSLIB::~FindPointsGSLIB()
 {
    delete gsl_comm;
-   delete cr;
    delete ir_simplex;
-   delete meshsplit;
 }
 
 #ifdef MFEM_USE_MPI
 FindPointsGSLIB::FindPointsGSLIB(MPI_Comm _comm)
-   : mesh(NULL), meshsplit(NULL), ir_simplex(NULL),
-     fdata2D(NULL), fdata3D(NULL), cr(NULL), gsl_comm(NULL),
-     dim(-1), points_cnt(0), setupflag(false), default_interp_value(0),
-     avgtype(AvgType::ARITHMETIC)
+   : mesh(NULL), ir_simplex(NULL), fdata2D(NULL), fdata3D(NULL),
+     dim(-1), gsl_mesh(), gsl_ref(), gsl_dist(), setupflag(false)
 {
    gsl_comm = new comm;
-   cr      = new crystal;
    comm_init(gsl_comm, _comm);
 }
 #endif
@@ -78,7 +70,6 @@ void FindPointsGSLIB::Setup(Mesh &m, const double bb_t, const double newt_tol,
    // call FreeData if FindPointsGSLIB::Setup has been called already
    if (setupflag) { FreeData(); }
 
-   crystal_init(cr, gsl_comm);
    mesh = &m;
    dim  = mesh->Dimension();
    const FiniteElement *fe = mesh->GetNodalFESpace()->GetFE(0);
@@ -122,16 +113,14 @@ void FindPointsGSLIB::Setup(Mesh &m, const double bb_t, const double newt_tol,
    setupflag = true;
 }
 
-void FindPointsGSLIB::FindPoints(const Vector &point_pos)
+void FindPointsGSLIB::FindPoints(const Vector &point_pos,
+                                 Array<unsigned int> &codes,
+                                 Array<unsigned int> &proc_ids,
+                                 Array<unsigned int> &elem_ids,
+                                 Vector &ref_pos, Vector &dist)
 {
    MFEM_VERIFY(setupflag, "Use FindPointsGSLIB::Setup before finding points.");
-   points_cnt = point_pos.Size() / dim;
-   gsl_code.SetSize(points_cnt);
-   gsl_proc.SetSize(points_cnt);
-   gsl_elem.SetSize(points_cnt);
-   gsl_ref.SetSize(points_cnt * dim);
-   gsl_dist.SetSize(points_cnt);
-
+   const int points_cnt = point_pos.Size() / dim;
    if (dim == 2)
    {
       const double *xv_base[2];
@@ -140,11 +129,11 @@ void FindPointsGSLIB::FindPoints(const Vector &point_pos)
       unsigned xv_stride[2];
       xv_stride[0] = sizeof(double);
       xv_stride[1] = sizeof(double);
-      findpts_2(gsl_code.GetData(), sizeof(unsigned int),
-                gsl_proc.GetData(), sizeof(unsigned int),
-                gsl_elem.GetData(), sizeof(unsigned int),
-                gsl_ref.GetData(),  sizeof(double) * dim,
-                gsl_dist.GetData(), sizeof(double),
+      findpts_2(codes.GetData(), sizeof(unsigned int),
+                proc_ids.GetData(), sizeof(unsigned int),
+                elem_ids.GetData(), sizeof(unsigned int),
+                ref_pos.GetData(), sizeof(double) * dim,
+                dist.GetData(), sizeof(double),
                 xv_base, xv_stride, points_cnt, fdata2D);
    }
    else
@@ -157,27 +146,25 @@ void FindPointsGSLIB::FindPoints(const Vector &point_pos)
       xv_stride[0] = sizeof(double);
       xv_stride[1] = sizeof(double);
       xv_stride[2] = sizeof(double);
-      findpts_3(gsl_code.GetData(), sizeof(unsigned int),
-                gsl_proc.GetData(), sizeof(unsigned int),
-                gsl_elem.GetData(), sizeof(unsigned int),
-                gsl_ref.GetData(),  sizeof(double) * dim,
-                gsl_dist.GetData(), sizeof(double),
+      findpts_3(codes.GetData(), sizeof(unsigned int),
+                proc_ids.GetData(), sizeof(unsigned int),
+                elem_ids.GetData(), sizeof(unsigned int),
+                ref_pos.GetData(), sizeof(double) * dim,
+                dist.GetData(), sizeof(double),
                 xv_base, xv_stride, points_cnt, fdata3D);
    }
+}
 
-   // Set the element number and reference position to 0 for points not found
-   for (int i = 0; i < points_cnt; i++)
-   {
-      if (gsl_code[i] == 2)
-      {
-         gsl_elem[i] = 0;
-         for (int d = 0; d < dim; d++) { gsl_ref(i*dim + d) = -1.; }
-      }
-   }
+void FindPointsGSLIB::FindPoints(const Vector &point_pos)
+{
+   const int points_cnt = point_pos.Size() / dim;
+   gsl_code.SetSize(points_cnt);
+   gsl_proc.SetSize(points_cnt);
+   gsl_elem.SetSize(points_cnt);
+   gsl_ref.SetSize(points_cnt * dim);
+   gsl_dist.SetSize(points_cnt);
 
-   // Map element number for simplices, and ref_pos from [-1,1] to [0,1] for
-   // both simplices and quads.
-   MapRefPosAndElemIndices();
+   FindPoints(point_pos, gsl_code, gsl_proc, gsl_elem, gsl_ref, gsl_dist);
 }
 
 void FindPointsGSLIB::FindPoints(Mesh &m, const Vector &point_pos,
@@ -191,24 +178,72 @@ void FindPointsGSLIB::FindPoints(Mesh &m, const Vector &point_pos,
    FindPoints(point_pos);
 }
 
+void FindPointsGSLIB::Interpolate(Array<unsigned int> &codes,
+                                  Array<unsigned int> &proc_ids,
+                                  Array<unsigned int> &elem_ids,
+                                  Vector &ref_pos, const GridFunction &field_in,
+                                  Vector &field_out)
+{
+
+   FiniteElementSpace ind_fes(mesh, field_in.FESpace()->FEColl());
+   GridFunction field_in_scalar(&ind_fes);
+   Vector node_vals;
+
+   const int ncomp      = field_in.FESpace()->GetVDim(),
+             points_fld = field_in.Size() / ncomp,
+             points_cnt = codes.Size();
+   field_out.SetSize(points_cnt*ncomp);
+
+   for (int i = 0; i < ncomp; i++)
+   {
+      const int dataptrin  = i*points_fld,
+                dataptrout = i*points_cnt;
+      field_in_scalar.NewDataAndSize(field_in.GetData()+dataptrin, points_fld);
+      GetNodeValues(field_in_scalar, node_vals);
+
+      if (dim==2)
+      {
+         findpts_eval_2(field_out.GetData()+dataptrout, sizeof(double),
+                        codes.GetData(),       sizeof(unsigned int),
+                        proc_ids.GetData(),    sizeof(unsigned int),
+                        elem_ids.GetData(),    sizeof(unsigned int),
+                        ref_pos.GetData(),     sizeof(double) * dim,
+                        points_cnt, node_vals.GetData(), fdata2D);
+      }
+      else
+      {
+         findpts_eval_3(field_out.GetData()+dataptrout, sizeof(double),
+                        codes.GetData(),       sizeof(unsigned int),
+                        proc_ids.GetData(),    sizeof(unsigned int),
+                        elem_ids.GetData(),    sizeof(unsigned int),
+                        ref_pos.GetData(),     sizeof(double) * dim,
+                        points_cnt, node_vals.GetData(), fdata3D);
+      }
+   }
+}
+
+void FindPointsGSLIB::Interpolate(const GridFunction &field_in,
+                                  Vector &field_out)
+{
+   Interpolate(gsl_code, gsl_proc, gsl_elem, gsl_ref, field_in, field_out);
+}
+
 void FindPointsGSLIB::Interpolate(const Vector &point_pos,
                                   const GridFunction &field_in, Vector &field_out)
 {
    FindPoints(point_pos);
-   Interpolate(field_in, field_out);
+   Interpolate(gsl_code, gsl_proc, gsl_elem, gsl_ref, field_in, field_out);
 }
 
 void FindPointsGSLIB::Interpolate(Mesh &m, const Vector &point_pos,
                                   const GridFunction &field_in, Vector &field_out)
 {
    FindPoints(m, point_pos);
-   Interpolate(field_in, field_out);
+   Interpolate(gsl_code, gsl_proc, gsl_elem, gsl_ref, field_in, field_out);
 }
 
 void FindPointsGSLIB::FreeData()
 {
-   if (!setupflag) { return; }
-   crystal_free(cr);
    if (dim == 2)
    {
       findpts_free_2(fdata2D);
@@ -217,13 +252,13 @@ void FindPointsGSLIB::FreeData()
    {
       findpts_free_3(fdata3D);
    }
+   setupflag = false;
    gsl_code.DeleteAll();
    gsl_proc.DeleteAll();
    gsl_elem.DeleteAll();
    gsl_mesh.Destroy();
    gsl_ref.Destroy();
    gsl_dist.Destroy();
-   setupflag = false;
 }
 
 void FindPointsGSLIB::GetNodeValues(const GridFunction &gf_in,
@@ -323,8 +358,9 @@ void FindPointsGSLIB::GetSimplexNodalCoordinates()
    const FiniteElement *fe   = mesh->GetNodalFESpace()->GetFE(0);
    const Geometry::Type gt   = fe->GetGeomType();
    const GridFunction *nodes = mesh->GetNodes();
+   Mesh *meshsplit           = NULL;
    const int NE              = mesh->GetNE();
-   int NEsplit = 0;
+   int NEsplit = -1;
 
    // Split the reference element into a reference submesh of quads or hexes.
    if (gt == Geometry::TRIANGLE)
@@ -480,361 +516,8 @@ void FindPointsGSLIB::GetSimplexNodalCoordinates()
          pt_id++;
       }
    }
-}
 
-void FindPointsGSLIB::MapRefPosAndElemIndices()
-{
-   gsl_mfem_ref = gsl_ref;
-   gsl_mfem_elem = gsl_elem;
-   const FiniteElement *fe = mesh->GetNodalFESpace()->GetFE(0);
-   const Geometry::Type gt = fe->GetGeomType();
-   int NEsplit = 0;
-
-   gsl_mfem_ref -= -1.;  // map  [-1, 1] to
-   gsl_mfem_ref *= 0.5;  //      [0,  1]
-   if (gt == Geometry::SQUARE || gt == Geometry::CUBE) { return; }
-
-   H1_FECollection feclin(1, dim);
-   FiniteElementSpace nodal_fes_lin(meshsplit, &feclin, dim);
-   GridFunction gf_lin(&nodal_fes_lin);
-
-   if (gt == Geometry::TRIANGLE)
-   {
-      const double quad_v[7][2] =
-      {
-         {0, 0}, {0.5, 0}, {1, 0}, {0, 0.5},
-         {1./3., 1./3.}, {0.5, 0.5}, {0, 1}
-      };
-      for (int k = 0; k < dim; k++)
-      {
-         for (int j = 0; j < gf_lin.Size()/dim; j++)
-         {
-            gf_lin(j+k*gf_lin.Size()/dim) = quad_v[j][k];
-         }
-      }
-      NEsplit = 3;
-   }
-   else if (gt == Geometry::TETRAHEDRON)
-   {
-      const double hex_v[15][3] =
-      {
-         {0, 0, 0.}, {1, 0., 0.}, {0., 1., 0.}, {0, 0., 1.},
-         {0.5, 0., 0.}, {0.5, 0.5, 0.}, {0., 0.5, 0.},
-         {0., 0., 0.5}, {0.5, 0., 0.5}, {0., 0.5, 0.5},
-         {1./3., 0., 1./3.}, {1./3., 1./3., 1./3.}, {0, 1./3., 1./3.},
-         {1./3., 1./3., 0}, {0.25, 0.25, 0.25}
-      };
-      for (int k = 0; k < dim; k++)
-      {
-         for (int j = 0; j < gf_lin.Size()/dim; j++)
-         {
-            gf_lin(j+k*gf_lin.Size()/dim) = hex_v[j][k];
-         }
-      }
-      NEsplit = 4;
-   }
-   else if (gt == Geometry::PRISM)
-   {
-      const double hex_v[14][3] =
-      {
-         {0, 0, 0}, {0.5, 0, 0}, {1, 0, 0}, {0, 0.5, 0},
-         {1./3., 1./3., 0}, {0.5, 0.5, 0}, {0, 1, 0},
-         {0, 0, 1}, {0.5, 0, 1}, {1, 0, 1}, {0, 0.5, 1},
-         {1./3., 1./3., 1}, {0.5, 0.5, 1}, {0, 1, 1}
-      };
-      for (int k = 0; k < dim; k++)
-      {
-         for (int j = 0; j < gf_lin.Size()/dim; j++)
-         {
-            gf_lin(j+k*gf_lin.Size()/dim) = hex_v[j][k];
-         }
-      }
-      NEsplit = 3;
-   }
-   else
-   {
-      MFEM_ABORT("Element type not currently supported.");
-   }
-
-   // Simplices are split into quads/hexes for GSLIB. For MFEM, we need to find
-   // the original element number and map the rst from micro to macro element.
-   for (int i = 0; i < points_cnt; i++)
-   {
-      if (gsl_code[i] == 2) { continue; }
-      int local_elem   = gsl_elem[i]%NEsplit;
-      gsl_mfem_elem[i] = (gsl_elem[i] - local_elem)/NEsplit; // macro element number
-
-      IntegrationPoint ip;
-      Vector mfem_ref(gsl_mfem_ref.GetData()+i*dim, dim);
-      ip.Set2(mfem_ref.GetData());
-      if (dim == 3) { ip.z = mfem_ref(2); }
-      gf_lin.GetVectorValue(local_elem, ip, mfem_ref); // map to rst of macro element
-   }
-}
-
-void FindPointsGSLIB::Interpolate(const GridFunction &field_in,
-                                  Vector &field_out)
-{
-   const int  gf_order   = field_in.FESpace()->GetFE(0)->GetOrder(),
-              mesh_order = mesh->GetNodalFESpace()->GetFE(0)->GetOrder();
-
-   const FiniteElementCollection *fec_in =  field_in.FESpace()->FEColl();
-   const H1_FECollection *fec_h1 = dynamic_cast<const H1_FECollection *>(fec_in);
-   const L2_FECollection *fec_l2 = dynamic_cast<const L2_FECollection *>(fec_in);
-
-   if (fec_h1 && gf_order == mesh_order &&
-       fec_h1->GetBasisType() == BasisType::GaussLobatto)
-   {
-      InterpolateH1(field_in, field_out);
-      return;
-   }
-   else
-   {
-      InterpolateGeneral(field_in, field_out);
-      if (!fec_l2 || avgtype == AvgType::NONE) { return; }
-   }
-
-   // For points on element borders, project the L2 GridFunction to H1 and
-   // re-interpolate.
-   if (fec_l2)
-   {
-      Array<int> indl2;
-      for (int i = 0; i < points_cnt; i++)
-      {
-         if (gsl_code[i] == 1) { indl2.Append(i); }
-      }
-      if (indl2.Size() == 0) { return; } // no points on element borders
-
-      Vector field_out_l2(field_out.Size());
-      VectorGridFunctionCoefficient field_in_dg(&field_in);
-      int gf_order_h1 = std::max(gf_order, 1); // H1 should be at least order 1
-      H1_FECollection fec(gf_order_h1, dim);
-      const int ncomp = field_in.FESpace()->GetVDim();
-      FiniteElementSpace fes(mesh, &fec, ncomp);
-      GridFunction field_in_h1(&fes);
-
-      if (avgtype == AvgType::ARITHMETIC)
-      {
-         field_in_h1.ProjectDiscCoefficient(field_in_dg, GridFunction::ARITHMETIC);
-      }
-      else if (avgtype == AvgType::HARMONIC)
-      {
-         field_in_h1.ProjectDiscCoefficient(field_in_dg, GridFunction::HARMONIC);
-      }
-      else
-      {
-         MFEM_ABORT("Invalid averaging type.");
-      }
-
-      if (gf_order_h1 == mesh_order) // basis is GaussLobatto by default
-      {
-         InterpolateH1(field_in_h1, field_out_l2);
-      }
-      else
-      {
-         InterpolateGeneral(field_in_h1, field_out_l2);
-      }
-
-      // Copy interpolated values for the points on element border
-      for (int j = 0; j < ncomp; j++)
-      {
-         for (int i = 0; i < indl2.Size(); i++)
-         {
-            int idx = indl2[i] + j*points_cnt;
-            field_out(idx) = field_out_l2(idx);
-         }
-      }
-   }
-}
-
-void FindPointsGSLIB::InterpolateH1(const GridFunction &field_in,
-                                    Vector &field_out)
-{
-   FiniteElementSpace ind_fes(mesh, field_in.FESpace()->FEColl());
-   GridFunction field_in_scalar(&ind_fes);
-   Vector node_vals;
-
-   const int ncomp      = field_in.FESpace()->GetVDim(),
-             points_fld = field_in.Size() / ncomp,
-             points_cnt = gsl_code.Size();
-
-   field_out.SetSize(points_cnt*ncomp);
-   field_out = default_interp_value;
-
-   for (int i = 0; i < ncomp; i++)
-   {
-      const int dataptrin  = i*points_fld,
-                dataptrout = i*points_cnt;
-      field_in_scalar.NewDataAndSize(field_in.GetData()+dataptrin, points_fld);
-      GetNodeValues(field_in_scalar, node_vals);
-
-      if (dim==2)
-      {
-         findpts_eval_2(field_out.GetData()+dataptrout, sizeof(double),
-                        gsl_code.GetData(),       sizeof(unsigned int),
-                        gsl_proc.GetData(),    sizeof(unsigned int),
-                        gsl_elem.GetData(),    sizeof(unsigned int),
-                        gsl_ref.GetData(),     sizeof(double) * dim,
-                        points_cnt, node_vals.GetData(), fdata2D);
-      }
-      else
-      {
-         findpts_eval_3(field_out.GetData()+dataptrout, sizeof(double),
-                        gsl_code.GetData(),       sizeof(unsigned int),
-                        gsl_proc.GetData(),    sizeof(unsigned int),
-                        gsl_elem.GetData(),    sizeof(unsigned int),
-                        gsl_ref.GetData(),     sizeof(double) * dim,
-                        points_cnt, node_vals.GetData(), fdata3D);
-      }
-   }
-}
-
-void FindPointsGSLIB::InterpolateGeneral(const GridFunction &field_in,
-                                         Vector &field_out)
-{
-   int ncomp   = field_in.VectorDim(),
-       nptorig = points_cnt,
-       npt     = points_cnt;
-
-   field_out.SetSize(points_cnt*ncomp);
-   field_out = default_interp_value;
-
-   if (gsl_comm->np == 1) // serial
-   {
-      for (int index = 0; index < npt; index++)
-      {
-         if (gsl_code[index] == 2) { continue; }
-         IntegrationPoint ip;
-         ip.Set2(gsl_mfem_ref.GetData()+index*dim);
-         if (dim == 3) { ip.z = gsl_mfem_ref(index*dim + 2); }
-         Vector localval(ncomp);
-         field_in.GetVectorValue(gsl_mfem_elem[index], ip, localval);
-         for (int i = 0; i < ncomp; i++)
-         {
-            field_out(index + i*npt) = localval(i);
-         }
-      }
-   }
-   else // parallel
-   {
-      // Determine number of points to be sent
-      int nptsend = 0;
-      for (int index = 0; index < npt; index++)
-      {
-         if (gsl_code[index] != 2) { nptsend +=1; }
-      }
-
-      // Pack data to send via crystal router
-      struct array *outpt = new array;
-      struct out_pt { double r[3], ival; uint index, el, proc; };
-      struct out_pt *pt;
-      array_init(struct out_pt, outpt, nptsend);
-      outpt->n=nptsend;
-      pt = (struct out_pt *)outpt->ptr;
-      for (int index = 0; index < npt; index++)
-      {
-         if (gsl_code[index] == 2) { continue; }
-         for (int d = 0; d < dim; ++d) { pt->r[d]= gsl_mfem_ref(index*dim + d); }
-         pt->index = index;
-         pt->proc  = gsl_proc[index];
-         pt->el    = gsl_mfem_elem[index];
-         ++pt;
-      }
-
-      // Transfer data to target MPI ranks
-      sarray_transfer(struct out_pt, outpt, proc, 1, cr);
-
-      if (ncomp == 1)
-      {
-         // Interpolate the grid function
-         npt = outpt->n;
-         pt = (struct out_pt *)outpt->ptr;
-         for (int index = 0; index < npt; index++)
-         {
-            IntegrationPoint ip;
-            ip.Set3(&pt->r[0]);
-            pt->ival = field_in.GetValue(pt->el, ip, 1);
-            ++pt;
-         }
-
-         // Transfer data back to source MPI rank
-         sarray_transfer(struct out_pt, outpt, proc, 1, cr);
-         npt = outpt->n;
-         pt = (struct out_pt *)outpt->ptr;
-         for (int index = 0; index < npt; index++)
-         {
-            field_out(pt->index) = pt->ival;
-            ++pt;
-         }
-         array_free(outpt);
-         delete outpt;
-      }
-      else // ncomp > 1
-      {
-         // Interpolate data and store in a Vector
-         npt = outpt->n;
-         pt = (struct out_pt *)outpt->ptr;
-         Vector vec_int_vals(npt*ncomp);
-         for (int index = 0; index < npt; index++)
-         {
-            IntegrationPoint ip;
-            ip.Set3(&pt->r[0]);
-            Vector localval(vec_int_vals.GetData()+index*ncomp, ncomp);
-            field_in.GetVectorValue(pt->el, ip, localval);
-            ++pt;
-         }
-
-         // Save index and proc data in a struct
-         struct array *savpt = new array;
-         struct sav_pt { uint index, proc; };
-         struct sav_pt *spt;
-         array_init(struct sav_pt, savpt, npt);
-         savpt->n=npt;
-         spt = (struct sav_pt *)savpt->ptr;
-         pt  = (struct out_pt *)outpt->ptr;
-         for (int index = 0; index < npt; index++)
-         {
-            spt->index = pt->index;
-            spt->proc  = pt->proc;
-            ++pt; ++spt;
-         }
-
-         array_free(outpt);
-         delete outpt;
-
-         // Copy data from save struct to send struct and send component wise
-         struct array *sendpt = new array;
-         struct send_pt { double ival; uint index, proc; };
-         struct send_pt *sdpt;
-         for (int j = 0; j < ncomp; j++)
-         {
-            array_init(struct send_pt, sendpt, npt);
-            sendpt->n=npt;
-            spt  = (struct sav_pt *)savpt->ptr;
-            sdpt = (struct send_pt *)sendpt->ptr;
-            for (int index = 0; index < npt; index++)
-            {
-               sdpt->index = spt->index;
-               sdpt->proc  = spt->proc;
-               sdpt->ival  = vec_int_vals(j + index*ncomp);
-               ++sdpt; ++spt;
-            }
-
-            sarray_transfer(struct send_pt, sendpt, proc, 1, cr);
-            sdpt = (struct send_pt *)sendpt->ptr;
-            for (int index = 0; index < nptorig; index++)
-            {
-               int idx = sdpt->index + j*nptorig;
-               field_out(idx) = sdpt->ival;
-               ++sdpt;
-            }
-            array_free(sendpt);
-         }
-         array_free(savpt);
-         delete sendpt;
-         delete savpt;
-      } // ncomp > 1
-   } // parallel
+   delete meshsplit;
 }
 
 } // namespace mfem

fem/gslib.hpp

@@ -20,66 +20,28 @@
 struct comm;
 struct findpts_data_2;
 struct findpts_data_3;
-struct array;
-struct crystal;
 
 namespace mfem
 {
 
-/** \brief FindPointsGSLIB can robustly evaluate a GridFunction on an arbitrary
- *  collection of points. There are three key functions in FindPointsGSLIB:
- *
- *  1. Setup - constructs the internal data structures of gslib.
- *
- *  2. FindPoints - for any given arbitrary set of points in physical space,
- *     gslib finds the element number, MPI rank, and the reference space
- *     coordinates inside the element that each point is located in. gslib also
- *     returns a code that indicates whether the point was found inside an
- *     element, on element border, or not found in the domain.
- *
- *  3. Interpolate - Interpolates any grid function at the points found using 2.
- *
- *  FindPointsGSLIB provides interface to use these functions individually or
- *  using a single call.
- */
 class FindPointsGSLIB
 {
-public:
-   enum AvgType {NONE, ARITHMETIC, HARMONIC}; // Average type for L2 functions
-
 protected:
-   Mesh *mesh, *meshsplit;
-   IntegrationRule *ir_simplex;    // IntegrationRule to split quads/hex -> simplex
-   struct findpts_data_2 *fdata2D; // gslib's internal data
-   struct findpts_data_3 *fdata3D; // gslib's internal data
-   struct crystal *cr;             // gslib's internal data
-   struct comm *gsl_comm;          // gslib's internal data
-   int dim, points_cnt;
-   Array<unsigned int> gsl_code, gsl_proc, gsl_elem, gsl_mfem_elem;
-   Vector gsl_mesh, gsl_ref, gsl_dist, gsl_mfem_ref;
-   bool setupflag;              // flag to indicate whether gslib data has been setup
-   double default_interp_value; // used for points that are not found in the mesh
-   AvgType avgtype;             // average type used for L2 functions
+   Mesh *mesh;
+   IntegrationRule *ir_simplex;
+   struct findpts_data_2 *fdata2D;
+   struct findpts_data_3 *fdata3D;
+   int dim;
+   Array<unsigned int> gsl_code, gsl_proc, gsl_elem;
+   Vector gsl_mesh, gsl_ref, gsl_dist;
+   bool setupflag;
+
+   struct comm *gsl_comm;
 
-   /// Get GridFunction from MFEM format to GSLIB format
    void GetNodeValues(const GridFunction &gf_in, Vector &node_vals);
-   /// Get nodal coordinates from mesh to the format expected by GSLIB for quads
-   /// and hexes
    void GetQuadHexNodalCoordinates();
-   /// Convert simplices to quad/hexes and then get nodal coordinates for each
-   /// split element into format expected by GSLIB
    void GetSimplexNodalCoordinates();
 
-   /// Use GSLIB for communication and interpolation
-   void InterpolateH1(const GridFunction &field_in, Vector &field_out);
-   /// Uses GSLIB Crystal Router for communication followed by MFEM's
-   /// interpolation functions
-   void InterpolateGeneral(const GridFunction &field_in, Vector &field_out);
-   /// Map {r,s,t} coordinates from [-1,1] to [0,1] for MFEM. For simplices mesh
-   /// find the original element number (that was split into micro quads/hexes
-   /// by GetSimplexNodalCoordinates())
-   void MapRefPosAndElemIndices();
-
 public:
    FindPointsGSLIB();
 
@@ -102,37 +64,45 @@ public:
    void Setup(Mesh &m, const double bb_t = 0.1, const double newt_tol = 1.0e-12,
               const int npt_max = 256);
 
-   /** Searches positions given in physical space by @a point_pos. These positions
-       must by ordered by nodes: (XXX...,YYY...,ZZZ).
-       This function populates the following member variables:
-       #gsl_code        Return codes for each point: inside element (0),
-                        element boundary (1), not found (2).
-       #gsl_proc        MPI proc ids where the points were found.
-       #gsl_elem        Element ids where the points were found.
-                        Defaults to 0 for points that were not found.
-       #gsl_mfem_elem   Element ids corresponding to MFEM-mesh where the points
-                        were found. #gsl_mfem_elem != #gsl_elem for simplices
-                        Defaults to 0 for points that were not found.
-       #gsl_ref         Reference coordinates of the found point.
-                        Ordered by vdim (XYZ,XYZ,XYZ...). Defaults to -1 for
-                        points that were not found. Note: the gslib reference
-                        frame is [-1,1].
-       #gsl_mfem_ref    Reference coordinates #gsl_ref mapped to [0,1].
-                        Defaults to 0 for points that were not found.
-       #gsl_dist        Distance between the sought and the found point
-                        in physical space. */
+   /** Searches positions given in physical space by @a point_pos. All output
+       Arrays and Vectors are expected to have the correct size.
+
+       @param[in]  point_pos  Positions to be found. Must by ordered by nodes
+                              (XXX...,YYY...,ZZZ).
+       @param[out] codes      Return codes for each point: inside element (0),
+                              element boundary (1), not found (2).
+       @param[out] proc_ids   MPI proc ids where the points were found.
+       @param[out] elem_ids   Element ids where the points were found.
+       @param[out] ref_pos    Reference coordinates of the found point. Ordered
+                              by vdim (XYZ,XYZ,XYZ...).
+                              Note: the gslib reference frame is [-1,1].
+       @param[out] dist       Distance between the sought and the found point
+                              in physical space. */
+   void FindPoints(const Vector &point_pos, Array<unsigned int> &codes,
+                   Array<unsigned int> &proc_ids, Array<unsigned int> &elem_ids,
+                   Vector &ref_pos, Vector &dist);
    void FindPoints(const Vector &point_pos);
    /// Setup FindPoints and search positions
    void FindPoints(Mesh &m, const Vector &point_pos, const double bb_t = 0.1,
                    const double newt_tol = 1.0e-12,  const int npt_max = 256);
 
    /** Interpolation of field values at prescribed reference space positions.
+
+       @param[in] codes       Return codes for each point: inside element (0),
+                              element boundary (1), not found (2).
+       @param[in] proc_ids    MPI proc ids where the points were found.
+       @param[in] elem_ids    Element ids where the points were found.
+       @param[in] ref_pos     Reference coordinates of the found point. Ordered
+                              by vdim (XYZ,XYZ,XYZ...).
+                              Note: the gslib reference frame is [-1,1].
        @param[in] field_in    Function values that will be interpolated on the
                               reference positions. Note: it is assumed that
                               @a field_in is in H1 and in the same space as the
                               mesh that was given to Setup().
-       @param[out] field_out  Interpolated values. For points that are not found
-                              the value is set to #default_interp_value. */
+       @param[out] field_out  Interpolated values. */
+   void Interpolate(Array<unsigned int> &codes, Array<unsigned int> &proc_ids,
+                    Array<unsigned int> &elem_ids, Vector &ref_pos,
+                    const GridFunction &field_in, Vector &field_out);
    void Interpolate(const GridFunction &field_in, Vector &field_out);
    /** Search positions and interpolate */
    void Interpolate(const Vector &point_pos, const GridFunction &field_in,
@@ -141,45 +111,27 @@ public:
    void Interpolate(Mesh &m, const Vector &point_pos,
                     const GridFunction &field_in, Vector &field_out);
 
-   /// Average type to be used for L2 functions in-case a point is located at
-   /// an element boundary where the function might be multi-valued.
-   void SetL2AvgType(AvgType avgtype_) { avgtype = avgtype_; }
-
-   /// Set the default interpolation value for points that are not found in the
-   /// mesh.
-   void SetDefaultInterpolationValue(double interp_value_)
-   {
-      default_interp_value = interp_value_;
-   }
-
    /** Cleans up memory allocated internally by gslib.
-       Note that in parallel, this must be called before MPI_Finalize(), as it
-       calls MPI_Comm_free() for internal gslib communicators. */
+       Note that in parallel, this must be called before MPI_Finalize(), as
+       it calls MPI_Comm_free() for internal gslib communicators. */
    void FreeData();
 
    /// Return code for each point searched by FindPoints: inside element (0), on
    /// element boundary (1), or not found (2).
    const Array<unsigned int> &GetCode() const { return gsl_code; }
    /// Return element number for each point found by FindPoints.
-   const Array<unsigned int> &GetElem() const { return gsl_mfem_elem; }
+   const Array<unsigned int> &GetElem() const { return gsl_elem; }
    /// Return MPI rank on which each point was found by FindPoints.
    const Array<unsigned int> &GetProc() const { return gsl_proc; }
    /// Return reference coordinates for each point found by FindPoints.
-   const Vector &GetReferencePosition() const { return gsl_mfem_ref;  }
+   const Vector &GetReferencePosition() const { return gsl_ref;  }
    /// Return distance Distance between the sought and the found point
    /// in physical space, for each point found by FindPoints.
    const Vector &GetDist()              const { return gsl_dist; }
-
-   /// Return element number for each point found by FindPoints corresponding to
-   /// GSLIB mesh. gsl_mfem_elem != gsl_elem for mesh with simplices.
-   const Array<unsigned int> &GetGSLIBElem() const { return gsl_elem; }
-   /// Return reference coordinates in [-1,1] (internal range in GSLIB) for each
-   /// point found by FindPoints.
-   const Vector &GetGSLIBReferencePosition() const { return gsl_ref; }
 };
 
 } // namespace mfem
 
-#endif // MFEM_USE_GSLIB
+#endif //MFEM_USE_GSLIB
 
-#endif // MFEM_GSLIB
+#endif //MFEM_GSLIB guard

fem/libceed/ceed.cpp

@@ -35,9 +35,6 @@ extern Ceed ceed;
 
 std::string ceed_path;
 
-extern CeedBasisMap ceed_basis_map;
-extern CeedRestrMap ceed_restr_map;
-
 }
 
 void InitCeedCoeff(Coefficient* Q, CeedData* ptr)
@@ -84,9 +81,10 @@ static CeedElemTopology GetCeedTopology(Geometry::Type geom)
    }
 }
 
-static void InitCeedNonTensorBasis(const FiniteElementSpace &fes,
-                                   const IntegrationRule &ir,
-                                   Ceed ceed, CeedBasis *basis)
+static void InitCeedNonTensorBasisAndRestriction(const FiniteElementSpace &fes,
+                                                 const IntegrationRule &ir,
+                                                 Ceed ceed, CeedBasis *basis,
+                                                 CeedElemRestriction *restr)
 {
    Mesh *mesh = fes.GetMesh();
    const FiniteElement *fe = fes.GetFE(0);
@@ -99,73 +97,7 @@ static void InitCeedNonTensorBasis(const FiniteElementSpace &fes,
    Vector qweight(Q);
    Vector shape_i(P);
    DenseMatrix grad_i(P, dim);
-   const Table &el_dof = fes.GetElementToDofTable();
-   Array<int> tp_el_dof(el_dof.Size_of_connections());
-   const TensorBasisElement * tfe =
-      dynamic_cast<const TensorBasisElement *>(fe);
-   if (tfe) // Lexicographic ordering using dof_map
-   {
-      const Array<int>& dof_map = tfe->GetDofMap();
-      for (int i = 0; i < Q; i++)
-      {
-         const IntegrationPoint &ip = ir.IntPoint(i);
-         qref(0,i) = ip.x;
-         if (dim>1) { qref(1,i) = ip.y; }
-         if (dim>2) { qref(2,i) = ip.z; }
-         qweight(i) = ip.weight;
-         fe->CalcShape(ip, shape_i);
-         fe->CalcDShape(ip, grad_i);
-         for (int j = 0; j < P; j++)
-         {
-            shape(j, i) = shape_i(dof_map[j]);
-            for (int d = 0; d < dim; ++d)
-            {
-               grad(j+i*P+d*Q*P) = grad_i(dof_map[j], d);
-            }
-         }
-      }
-   }
-   else  // Native ordering
-   {
-      for (int i = 0; i < Q; i++)
-      {
-         const IntegrationPoint &ip = ir.IntPoint(i);
-         qref(0,i) = ip.x;
-         if (dim>1) { qref(1,i) = ip.y; }
-         if (dim>2) { qref(2,i) = ip.z; }
-         qweight(i) = ip.weight;
-         fe->CalcShape(ip, shape_i);
-         fe->CalcDShape(ip, grad_i);
-         for (int j = 0; j < P; j++)
-         {
-            shape(j, i) = shape_i(j);
-            for (int d = 0; d < dim; ++d)
-            {
-               grad(j+i*P+d*Q*P) = grad_i(j, d);
-            }
-         }
-      }
-   }
-   CeedBasisCreateH1(ceed, GetCeedTopology(fe->GetGeomType()), fes.GetVDim(),
-                     fe->GetDof(), ir.GetNPoints(), shape.GetData(),
-                     grad.GetData(), qref.GetData(), qweight.GetData(), basis);
-}
 
-static void InitCeedNonTensorRestriction(const FiniteElementSpace &fes,
-                                         const IntegrationRule &ir,
-                                         Ceed ceed, CeedElemRestriction *restr)
-{
-   Mesh *mesh = fes.GetMesh();
-   const FiniteElement *fe = fes.GetFE(0);
-   const int dim = mesh->Dimension();
-   const int P = fe->GetDof();
-   const int Q = ir.GetNPoints();
-   DenseMatrix shape(P, Q);
-   Vector grad(P*dim*Q);
-   DenseMatrix qref(dim, Q);
-   Vector qweight(Q);
-   Vector shape_i(P);
-   DenseMatrix grad_i(P, dim);
    CeedInt compstride = fes.GetOrdering()==Ordering::byVDIM ? 1 : fes.GetNDofs();
    const Table &el_dof = fes.GetElementToDofTable();
    Array<int> tp_el_dof(el_dof.Size_of_connections());
@@ -192,6 +124,7 @@ static void InitCeedNonTensorRestriction(const FiniteElementSpace &fes,
             }
          }
       }
+
       for (int i = 0; i < mesh->GetNE(); i++)
       {
          const int el_offset = fe->GetDof() * i;
@@ -229,6 +162,7 @@ static void InitCeedNonTensorRestriction(const FiniteElementSpace &fes,
             }
          }
       }
+
       for (int e = 0; e < mesh->GetNE(); e++)
       {
          for (int i = 0; i < P; i++)
@@ -244,15 +178,19 @@ static void InitCeedNonTensorRestriction(const FiniteElementSpace &fes,
          }
       }
    }
+   CeedBasisCreateH1(ceed, GetCeedTopology(fe->GetGeomType()), fes.GetVDim(),
+                     fe->GetDof(), ir.GetNPoints(), shape.GetData(),
+                     grad.GetData(), qref.GetData(), qweight.GetData(), basis);
    CeedElemRestrictionCreate(ceed, mesh->GetNE(), fe->GetDof(), fes.GetVDim(),
                              compstride, (fes.GetVDim())*(fes.GetNDofs()),
                              CEED_MEM_HOST, CEED_COPY_VALUES,
                              tp_el_dof.GetData(), restr);
 }
 
-static void InitCeedTensorBasis(const FiniteElementSpace &fes,
-                                const IntegrationRule &ir,
-                                Ceed ceed, CeedBasis *basis)
+static void InitCeedTensorBasisAndRestriction(const FiniteElementSpace &fes,
+                                              const IntegrationRule &ir,
+                                              Ceed ceed, CeedBasis *basis,
+                                              CeedElemRestriction *restr)
 {
    Mesh *mesh = fes.GetMesh();
    const FiniteElement *fe = fes.GetFE(0);
@@ -260,6 +198,7 @@ static void InitCeedTensorBasis(const FiniteElementSpace &fes,
    const TensorBasisElement * tfe =
       dynamic_cast<const TensorBasisElement *>(fe);
    MFEM_VERIFY(tfe, "invalid FE");
+   const Array<int>& dof_map = tfe->GetDofMap();
    const FiniteElement *fe1d =
       fes.FEColl()->FiniteElementForGeometry(Geometry::SEGMENT);
    DenseMatrix shape1d(fe1d->GetDof(), ir.GetNPoints());
@@ -288,28 +227,6 @@ static void InitCeedTensorBasis(const FiniteElementSpace &fes,
                            ir.GetNPoints(), shape1d.GetData(),
                            grad1d.GetData(), qref1d.GetData(),
                            qweight1d.GetData(), basis);
-}
-
-static void InitCeedTensorRestriction(const FiniteElementSpace &fes,
-                                      const IntegrationRule &ir,
-                                      Ceed ceed, CeedElemRestriction *restr)
-{
-   Mesh *mesh = fes.GetMesh();
-   const FiniteElement *fe = fes.GetFE(0);
-   const TensorBasisElement * tfe =
-      dynamic_cast<const TensorBasisElement *>(fe);
-   MFEM_VERIFY(tfe, "invalid FE");
-   const Array<int>& dof_map = tfe->GetDofMap();
-   const FiniteElement *fe1d =
-      fes.FEColl()->FiniteElementForGeometry(Geometry::SEGMENT);
-   DenseMatrix shape1d(fe1d->GetDof(), ir.GetNPoints());
-   DenseMatrix grad1d(fe1d->GetDof(), ir.GetNPoints());
-   Vector qref1d(ir.GetNPoints()), qweight1d(ir.GetNPoints());
-   Vector shape_i(shape1d.Height());
-   DenseMatrix grad_i(grad1d.Height(), 1);
-   const H1_SegmentElement *h1_fe1d =
-      dynamic_cast<const H1_SegmentElement *>(fe1d);
-   MFEM_VERIFY(h1_fe1d, "invalid FE");
 
    CeedInt compstride = fes.GetOrdering()==Ordering::byVDIM ? 1 : fes.GetNDofs();
    const Table &el_dof = fes.GetElementToDofTable();
@@ -341,52 +258,14 @@ void InitCeedBasisAndRestriction(const FiniteElementSpace &fes,
                                  Ceed ceed, CeedBasis *basis,
                                  CeedElemRestriction *restr)
 {
-   // Check for FES -> basis, restriction in hash tables
-   const Mesh *mesh = fes.GetMesh();
-   const FiniteElement *fe = fes.GetFE(0);
-   const int P = fe->GetDof();
-   const int Q = irm.GetNPoints();
-   const int nelem = mesh->GetNE();
-   const int ncomp = fes.GetVDim();
-   CeedBasisKey basis_key(&fes, &irm, ncomp, P, Q);
-   auto basis_itr = internal::ceed_basis_map.find(basis_key);
-   CeedRestrKey restr_key(&fes, nelem, P, ncomp);
-   auto restr_itr = internal::ceed_restr_map.find(restr_key);
-
-   // Init or retreive key values
-   if (basis_itr == internal::ceed_basis_map.end())
+   if (UsesTensorBasis(fes))
    {
-      if (UsesTensorBasis(fes))
-      {
-         const IntegrationRule &ir = IntRules.Get(Geometry::SEGMENT, irm.GetOrder());
-         InitCeedTensorBasis(fes, ir, ceed, basis);
-      }
-      else
-      {
-         InitCeedNonTensorBasis(fes, irm, ceed, basis);
-      }
-      internal::ceed_basis_map[basis_key] = *basis;
+      const IntegrationRule &ir = IntRules.Get(Geometry::SEGMENT, irm.GetOrder());
+      InitCeedTensorBasisAndRestriction(fes, ir, ceed, basis, restr);
    }
    else
    {
-      *basis = basis_itr->second;
-   }
-   if (restr_itr == internal::ceed_restr_map.end())
-   {
-      if (UsesTensorBasis(fes))
-      {
-         const IntegrationRule &ir = IntRules.Get(Geometry::SEGMENT, irm.GetOrder());
-         InitCeedTensorRestriction(fes, ir, ceed, restr);
-      }
-      else
-      {
-         InitCeedNonTensorRestriction(fes, irm, ceed, restr);
-      }
-      internal::ceed_restr_map[restr_key] = *restr;
-   }
-   else
-   {
-      *restr = restr_itr->second;
+      InitCeedNonTensorBasisAndRestriction(fes, irm, ceed, basis, restr);
    }
 }
 
@@ -448,8 +327,8 @@ void CeedPAAssemble(const CeedPAOperator& op,
    CeedVectorCreate(ceed, nelem * nqpts * qdatasize, &ceedData.rho);
 
    // Context data to be passed to the 'f_build_diff' Q-function.
-   ceedData.build_ctx_data.dim = mesh->Dimension();
-   ceedData.build_ctx_data.space_dim = mesh->SpaceDimension();
+   ceedData.build_ctx.dim = mesh->Dimension();
+   ceedData.build_ctx.space_dim = mesh->SpaceDimension();
 
    std::string qf_file = GetCeedPath() + op.header;
    std::string qf;
@@ -463,7 +342,7 @@ void CeedPAAssemble(const CeedPAOperator& op,
          CeedQFunctionCreateInterior(ceed, 1, op.const_qf,
                                      qf.c_str(),
                                      &ceedData.build_qfunc);
-         ceedData.build_ctx_data.coeff = ((CeedConstCoeff*)ceedData.coeff)->val;
+         ceedData.build_ctx.coeff = ((CeedConstCoeff*)ceedData.coeff)->val;
          break;
       case CeedCoeff::Grid:
          qf = qf_file + op.grid_func;
@@ -479,12 +358,8 @@ void CeedPAAssemble(const CeedPAOperator& op,
    CeedQFunctionAddInput(ceedData.build_qfunc, "weights", 1, CEED_EVAL_WEIGHT);
    CeedQFunctionAddOutput(ceedData.build_qfunc, "qdata", qdatasize,
                           CEED_EVAL_NONE);
-
-   CeedQFunctionContextCreate(ceed, &ceedData.build_ctx);
-   CeedQFunctionContextSetData(ceedData.build_ctx, CEED_MEM_HOST, CEED_USE_POINTER,
-                               sizeof(ceedData.build_ctx_data),
-                               &ceedData.build_ctx_data);
-   CeedQFunctionSetContext(ceedData.build_qfunc, ceedData.build_ctx);
+   CeedQFunctionSetContext(ceedData.build_qfunc, &ceedData.build_ctx,
+                           sizeof(ceedData.build_ctx));
 
    // Create the operator that builds the quadrature data for the operator.
    CeedOperatorCreate(ceed, ceedData.build_qfunc, NULL, NULL,
@@ -524,7 +399,8 @@ void CeedPAAssemble(const CeedPAOperator& op,
    CeedQFunctionAddInput(ceedData.apply_qfunc, "qdata", qdatasize,
                          CEED_EVAL_NONE);
    CeedQFunctionAddOutput(ceedData.apply_qfunc, "v", dimV, op.test_op);
-   CeedQFunctionSetContext(ceedData.apply_qfunc, ceedData.build_ctx);
+   CeedQFunctionSetContext(ceedData.apply_qfunc, &ceedData.build_ctx,
+                           sizeof(ceedData.build_ctx));
 
    // Create the diff operator.
    CeedOperatorCreate(ceed, ceedData.apply_qfunc, NULL, NULL, &ceedData.oper);

fem/libceed/ceed.hpp

@@ -18,9 +18,6 @@
 #include "../../general/device.hpp"
 #include "../../linalg/vector.hpp"
 #include <ceed.h>
-#include <ceed-hash.h>
-#include <tuple>
-#include <unordered_map>
 
 namespace mfem
 {
@@ -30,47 +27,7 @@ class GridFunction;
 class IntegrationRule;
 class Coefficient;
 
-// Hash table for CeedBasis
-using CeedBasisKey =
-   std::tuple<const FiniteElementSpace*, const IntegrationRule*, int, int, int>;
-struct CeedBasisHash
-{
-   std::size_t operator()(const CeedBasisKey& k) const
-   {
-      return CeedHashCombine(CeedHashCombine(CeedHashInt(
-                                                reinterpret_cast<CeedHash64_t>(std::get<0>(k))),
-                                             CeedHashInt(
-                                                reinterpret_cast<CeedHash64_t>(std::get<1>(k)))),
-                             CeedHashCombine(CeedHashCombine(CeedHashInt(std::get<2>(k)),
-                                                             CeedHashInt(std::get<3>(k))),
-                                             CeedHashInt(std::get<4>(k))));
-   }
-};
-using CeedBasisMap =
-   std::unordered_map<const CeedBasisKey, CeedBasis, CeedBasisHash>;
-
-// Hash table for CeedElemRestriction
-using CeedRestrKey = std::tuple<const FiniteElementSpace*, int, int, int>;
-struct CeedRestrHash
-{
-   std::size_t operator()(const CeedRestrKey& k) const
-   {
-      return CeedHashCombine(CeedHashCombine(CeedHashInt(
-                                                reinterpret_cast<CeedHash64_t>(std::get<0>(k))),
-                                             CeedHashInt(std::get<1>(k))),
-                             CeedHashCombine(CeedHashInt(std::get<2>(k)),
-                                             CeedHashInt(std::get<3>(k))));
-   }
-};
-using CeedRestrMap =
-   std::unordered_map<const CeedRestrKey, CeedElemRestriction, CeedRestrHash>;
-
-namespace internal
-{
-extern Ceed ceed; // defined in device.cpp
-extern CeedBasisMap basis_map;
-extern CeedRestrMap restr_map;
-}
+namespace internal { extern Ceed ceed; } // defined in device.cpp
 
 /// A structure used to pass additional data to f_build_diff and f_apply_diff
 struct BuildContext { CeedInt dim, space_dim; CeedScalar coeff; };
@@ -99,8 +56,7 @@ struct CeedData
    CeedVector node_coords, rho;
    CeedCoeff coeff_type;
    void* coeff;
-   CeedQFunctionContext build_ctx;
-   BuildContext build_ctx_data;
+   BuildContext build_ctx;
 
    CeedVector u, v;
 
@@ -108,6 +64,10 @@ struct CeedData
    {
       CeedOperatorDestroy(&build_oper);
       CeedOperatorDestroy(&oper);
+      CeedBasisDestroy(&basis);
+      CeedBasisDestroy(&mesh_basis);
+      CeedElemRestrictionDestroy(&restr);
+      CeedElemRestrictionDestroy(&mesh_restr);
       CeedElemRestrictionDestroy(&restr_i);
       CeedElemRestrictionDestroy(&mesh_restr_i);
       CeedQFunctionDestroy(&apply_qfunc);
@@ -117,6 +77,8 @@ struct CeedData
       if (coeff_type==CeedCoeff::Grid)
       {
          CeedGridCoeff* c = (CeedGridCoeff*)coeff;
+         CeedBasisDestroy(&c->basis);
+         CeedElemRestrictionDestroy(&c->restr);
          CeedVectorDestroy(&c->coeffVector);
          delete c;
       }

fem/linearform.cpp

@@ -204,14 +204,6 @@ void LinearForm::Update(FiniteElementSpace *f, Vector &v, int v_offset)
    ResetDeltaLocations();
 }
 
-void LinearForm::MakeRef(FiniteElementSpace *f, Vector &v, int v_offset)
-{
-   MFEM_ASSERT(v.Size() >= v_offset + f->GetVSize(), "");
-   fes = f;
-   v.UseDevice(true);
-   this->Vector::MakeRef(v, v_offset, fes->GetVSize());
-}
-
 void LinearForm::AssembleDelta()
 {
    if (dlfi_delta.Size() == 0) { return; }

fem/linearform.hpp

@@ -26,7 +26,7 @@ protected:
    /// FE space on which the LinearForm lives. Not owned.
    FiniteElementSpace *fes;
 
-   /** @brief Indicates the LinearFormIntegrator%s stored in #dlfi, #dlfi_delta,
+   /** @brief Indicates the LinerFormIntegrator%s stored in #dlfi, #dlfi_delta,
        #blfi, and #flfi are owned by another LinearForm. */
    int extern_lfs;
 
@@ -175,16 +175,6 @@ public:
        @note This method does not perform assembly. */
    void Update(FiniteElementSpace *f, Vector &v, int v_offset);
 
-   /** @brief Make the LinearForm reference external data on a new
-       FiniteElementSpace. */
-   /** This method changes the FiniteElementSpace associated with the LinearForm
-       @a *f and sets the data of the Vector @a v (plus the @a v_offset) as
-       external data in the LinearForm.
-
-       @note This version of the method will also perform bounds checks when the
-       build option MFEM_DEBUG is enabled. */
-   virtual void MakeRef(FiniteElementSpace *f, Vector &v, int v_offset);
-
    /// Return the action of the LinearForm as a linear mapping.
    /** Linear forms are linear functionals which map GridFunctions to
        the real numbers.  This method performs this mapping which in

fem/lininteg.cpp

@@ -457,8 +457,20 @@ void VectorFEDomainLFCurlIntegrator::AssembleRHSElementVect(
 
       Tr.SetIntPoint (&ip);
       el.CalcPhysCurlShape(Tr, curlshape);
-      QF->Eval(vec, Tr, ip);
 
+      switch (spaceDim)
+      {
+         case 3:
+            MFEM_VERIFY(QF, "VectorFunctionCoefficient not provided");
+            QF->Eval(vec, Tr, ip);
+            break;
+         case 2:
+            MFEM_VERIFY(Q, "FunctionCoefficient (Scalar) not provided");
+            vec[0] = Q->Eval(Tr, ip);
+            break;
+         default:
+            break; // This should be unreachable
+      }
       vec *= ip.weight * Tr.Weight();
       curlshape.AddMult (vec, elvect);
    }
@@ -468,17 +480,38 @@ void VectorFEDomainLFCurlIntegrator::AssembleDeltaElementVect(
    const FiniteElement &fe, ElementTransformation &Trans, Vector &elvect)
 {
    int spaceDim = Trans.GetSpaceDim();
-   MFEM_ASSERT(vec_delta != NULL,
-               "coefficient must be VectorDeltaCoefficient");
+   switch (spaceDim)
+   {
+      case 3:
+         MFEM_ASSERT(vec_delta != NULL,
+                     "coefficient must be VectorDeltaCoefficient");
+         break;
+      case 2:
+         MFEM_ASSERT(delta != NULL,
+                     "coefficient must be DeltaCoefficient");
+         break;
+      default:
+         break; // This should be unreachable
+   }
    int dof = fe.GetDof();
    int n=(spaceDim == 3)? spaceDim : 1;
-   vec.SetSize(n);
    curlshape.SetSize(dof, n);
    elvect.SetSize(dof);
    fe.CalcPhysCurlShape(Trans, curlshape);
 
-   vec_delta->EvalDelta(vec, Trans, Trans.GetIntPoint());
-   curlshape.Mult(vec, elvect);
+   switch (spaceDim)
+   {
+      case 3:
+         vec_delta->EvalDelta(vec, Trans, Trans.GetIntPoint());
+         curlshape.Mult(vec, elvect);
+         break;
+      case 2:
+         curlshape.GetColumn(0,elvect);
+         elvect *= delta->EvalDelta(Trans, Trans.GetIntPoint());
+         break;
+      default:
+         break; // This should be unreachable
+   }
 }
 
 void VectorFEDomainLFDivIntegrator::AssembleRHSElementVect(

fem/lininteg.hpp

@@ -284,6 +284,7 @@ class VectorFEDomainLFCurlIntegrator : public DeltaLFIntegrator
 {
 private:
    VectorCoefficient *QF=nullptr;
+   Coefficient *Q=nullptr;
    DenseMatrix curlshape;
    Vector vec;
 
@@ -291,6 +292,8 @@ public:
    /// Constructs the domain integrator (Q, curl v)
    VectorFEDomainLFCurlIntegrator(VectorCoefficient &F)
       : DeltaLFIntegrator(F), QF(&F) { }
+   VectorFEDomainLFCurlIntegrator(Coefficient &F)
+      : DeltaLFIntegrator(F), Q(&F) { }
 
    virtual void AssembleRHSElementVect(const FiniteElement &el,
                                        ElementTransformation &Tr,

fem/nonlinearform.cpp

@@ -10,6 +10,7 @@
 // CONTRIBUTING.md for details.
 
 #include "fem.hpp"
+#include "../general/forall.hpp"
 
 namespace mfem
 {
@@ -27,7 +28,7 @@ void NonlinearForm::SetAssemblyLevel(AssemblyLevel assembly_level)
          // This is the default behavior.
          break;
       case AssemblyLevel::PARTIAL:
-         ext = new PANonlinearFormExtension(this);
+         ext = new PANonlinearForm(this);
          break;
       default:
          mfem_error("Unknown assembly level for this form.");
@@ -80,6 +81,13 @@ void NonlinearForm::SetEssentialVDofs(const Array<int> &ess_vdofs_list)
 
 double NonlinearForm::GetGridFunctionEnergy(const Vector &x) const
 {
+   if (ext)
+   {
+      MFEM_VERIFY(!fnfi.Size(), "Interior faces terms not yet implemented!");
+      MFEM_VERIFY(!bfnfi.Size(), "Boundary face terms not yet implemented!");
+      return ext->GetGridFunctionEnergy(x);
+   }
+
    Array<int> vdofs;
    Vector el_x;
    const FiniteElement *fe;
@@ -138,6 +146,14 @@ void NonlinearForm::Mult(const Vector &x, Vector &y) const
    if (ext)
    {
       ext->Mult(px, py);
+      if (Serial())
+      {
+         if (cP) { cP->MultTranspose(py, y); }
+         const int N = ess_tdof_list.Size();
+         const auto tdof = ess_tdof_list.Read();
+         auto Y = y.ReadWrite();
+         MFEM_FORALL(i, N, Y[tdof[i]] = 0.0; );
+      }
       return;
    }
 
@@ -264,7 +280,16 @@ Operator &NonlinearForm::GetGradient(const Vector &x) const
 {
    if (ext)
    {
-      MFEM_ABORT("Not yet implemented!");
+      Operator &grad = ext->GetGradient(Prolongate(x));
+      hGrad.Reset(&grad, false);
+      if (Serial())
+      {
+         Operator *Gop;
+         if (cP) { hGrad.Reset(new RAPOperator(*cP, grad, *cP)); }
+         hGrad.Ptr()->Operator::FormSystemOperator(ess_tdof_list, Gop);
+         hGrad.Reset(Gop);
+      }
+      return *hGrad.Ptr();
    }
 
    const int skip_zeros = 0;
@@ -426,7 +451,31 @@ void NonlinearForm::Update()
 
 void NonlinearForm::Setup()
 {
-   if (ext) { return ext->AssemblePA(); }
+   if (ext) { return ext->Setup(); }
+}
+
+void NonlinearForm::AssembleGradientDiagonal(Vector &diag) const
+{
+   if (ext)
+   {
+      MFEM_ASSERT(diag.Size() == fes->GetTrueVSize(),
+                  "Vector for holding diagonal has wrong size!");
+      const Operator *P = fes->GetProlongationMatrix();
+      if (!IsIdentityProlongation(P))
+      {
+         Vector local_diag(P->Height());
+         ext->AssembleGradientDiagonal(local_diag);
+         P->MultTranspose(local_diag, diag);
+      }
+      else
+      {
+         ext->AssembleGradientDiagonal(diag);
+      }
+   }
+   else
+   {
+      MFEM_ABORT("Not implemented. Can be obtained through GetGradient().");
+   }
 }
 
 NonlinearForm::~NonlinearForm()

fem/nonlinearform.hpp

@@ -45,6 +45,7 @@ protected:
    Array<Array<int>*>              bfnfi_marker; // not owned
 
    mutable SparseMatrix *Grad, *cGrad; // owned
+   mutable OperatorHandle hGrad;
 
    /// A list of all essential true dofs
    Array<int> ess_tdof_list;
@@ -165,6 +166,15 @@ public:
    /// Setup the NonlinearForm
    virtual void Setup();
 
+   /** @brief Assemble the diagonal of the gradient into diag
+
+       For adaptively refined meshes, this returns P^T d_e, where d_e is the
+       locally assembled diagonal on each element and P^T is the transpose of
+       the conforming prolongation. In general this is not the correct diagonal
+       for an AMR mesh. */
+   void AssembleGradientDiagonal(Vector &diag) const;
+
+
    /// Get the finite element space prolongation matrix
    virtual const Operator *GetProlongation() const { return P; }
    /// Get the finite element space restriction matrix

fem/nonlinearform_ext.cpp

@@ -13,62 +13,101 @@
 // PABilinearFormExtension and MFBilinearFormExtension.
 
 #include "nonlinearform.hpp"
+#include "../general/forall.hpp"
 
 namespace mfem
 {
 
-NonlinearFormExtension::NonlinearFormExtension(NonlinearForm *form)
-   : Operator(form->FESpace()->GetTrueVSize()), n(form)
+NonlinearFormExtension::NonlinearFormExtension(const NonlinearForm *nlf)
+   : Operator(nlf->FESpace()->GetTrueVSize()), nlf(nlf) { }
+
+PANonlinearForm::PANonlinearForm(NonlinearForm *nlf):
+   NonlinearFormExtension(nlf),
+   x_grad(NULL),
+   fes(*nlf->FESpace()),
+   dnfi(*nlf->GetDNFI()),
+   R(fes.GetElementRestriction(ElementDofOrdering::LEXICOGRAPHIC))
 {
-   // empty
+   MFEM_VERIFY(R, "Not yet implemented!");
+   xe.SetSize(R->Height(), Device::GetMemoryType());
+   ye.SetSize(R->Height(), Device::GetMemoryType());
+   ye.UseDevice(true);
 }
 
-PANonlinearFormExtension::PANonlinearFormExtension(NonlinearForm *form):
-   NonlinearFormExtension(form), fes(*form->FESpace())
+double PANonlinearForm::GetGridFunctionEnergy(const Vector &x) const
 {
-   const ElementDofOrdering ordering = ElementDofOrdering::LEXICOGRAPHIC;
-   elem_restrict_lex = fes.GetElementRestriction(ordering);
-   if (elem_restrict_lex)
+   double energy = 0.0;
+
+   R->Mult(x, xe);
+   for (int i = 0; i < dnfi.Size(); i++)
    {
-      localX.SetSize(elem_restrict_lex->Height(), Device::GetMemoryType());
-      localY.SetSize(elem_restrict_lex->Height(), Device::GetMemoryType());
-      localY.UseDevice(true); // ensure 'localY = 0.0' is done on device
+      energy += dnfi[i]->GetGridFunctionEnergyPA(xe);
    }
+   return energy;
 }
 
-void PANonlinearFormExtension::AssemblePA()
+void PANonlinearForm::Setup()
 {
-   Array<NonlinearFormIntegrator*> &integrators = *n->GetDNFI();
-   const int Ni = integrators.Size();
-   for (int i = 0; i < Ni; ++i)
-   {
-      integrators[i]->AssemblePA(*n->FESpace());
-   }
+   for (int i = 0; i < dnfi.Size(); ++i) { dnfi[i]->AssemblePA(fes); }
 }
 
-void PANonlinearFormExtension::Mult(const Vector &x, Vector &y) const
+void PANonlinearForm::Mult(const Vector &x, Vector &y) const
 {
-   Array<NonlinearFormIntegrator*> &integrators = *n->GetDNFI();
-   const int iSz = integrators.Size();
-   if (elem_restrict_lex)
-   {
-      elem_restrict_lex->Mult(x, localX);
-      localY = 0.0;
-      for (int i = 0; i < iSz; ++i)
-      {
-         integrators[i]->AddMultPA(localX, localY);
-      }
-      elem_restrict_lex->MultTranspose(localY, y);
-   }
-   else
-   {
-      y.UseDevice(true); // typically this is a large vector, so store on device
-      y = 0.0;
-      for (int i = 0; i < iSz; ++i)
-      {
-         integrators[i]->AddMultPA(x, y);
-      }
-   }
+   ye = 0.0;
+   R->Mult(x, xe);
+   for (int i = 0; i < dnfi.Size(); ++i) { dnfi[i]->AddMultPA(xe, ye); }
+   R->MultTranspose(ye, y);
 }
 
+void PANonlinearForm::AssembleGradientDiagonal(Vector &diag) const
+{
+   MFEM_VERIFY(x_grad, "GetGradient() has not been called");
+   R->Mult(*x_grad, xe);
+
+   ye = 0.0;
+   for (int i = 0; i < dnfi.Size(); ++i)
+   {
+      dnfi[i]->AssembleGradientDiagonalPA(xe, ye);
+   }
+   R->MultTranspose(ye, diag);
 }
+
+Operator &PANonlinearForm::GetGradient(const Vector &x) const
+{
+   // Store the last x that was used to compute the gradient.
+   x_grad = &x;
+
+   Grad.Reset(new PANonlinearForm::Gradient(x, *this));
+   return *Grad.Ptr();
+}
+
+PANonlinearForm::Gradient::Gradient(const Vector &x, const PANonlinearForm &e):
+   Operator(e.fes.GetVSize()), R(e.R), dnfi(e.dnfi)
+{
+   ge.UseDevice(true);
+   ge.SetSize(R->Height(), Device::GetMemoryType());
+   R->Mult(x, ge);
+
+   xe.UseDevice(true);
+   xe.SetSize(R->Height(), Device::GetMemoryType());
+
+   ye.UseDevice(true);
+   ye.SetSize(R->Height(), Device::GetMemoryType());
+
+   ze.UseDevice(true);
+   ze.SetSize(R->Height(), Device::GetMemoryType());
+
+   // Do we still need to do this?
+   for (int i = 0; i < dnfi.Size(); ++i) { dnfi[i]->AssemblePA(e.fes); }
+}
+
+void PANonlinearForm::Gradient::Mult(const Vector &x, Vector &y) const
+{
+   ze = x;
+   ye = 0.0;
+   R->Mult(ze, xe);
+   for (int i = 0; i < dnfi.Size(); ++i) { dnfi[i]->AddMultGradPA(ge, xe, ye); }
+   R->MultTranspose(ye, y);
+}
+
+} // namespace mfem

fem/nonlinearform_ext.hpp

@@ -17,28 +17,60 @@
 
 namespace mfem
 {
-class NonlinearForm;
 
+class NonlinearForm;
+class NonlinearFormIntegrator;
+
+/** @brief Class extending the NonlinearForm class to support the different
+    AssemblyLevel%s. */
 class NonlinearFormExtension : public Operator
 {
 protected:
-   NonlinearForm *n; ///< Not owned
+   const NonlinearForm *nlf;
 public:
-   NonlinearFormExtension(NonlinearForm *form);
-   virtual void AssemblePA() = 0;
+   NonlinearFormExtension(const NonlinearForm*);
+   virtual void Setup() = 0;
+   virtual Operator &GetGradient(const Vector&) const = 0;
+   virtual double GetGridFunctionEnergy(const Vector &x) const = 0;
+   virtual void AssembleGradientDiagonal(Vector &diag) const
+   {
+      MFEM_ABORT("Not implemented for this assembly level!");
+   }
 };
 
+class PANonlinearForm;
+
+
 /// Data and methods for partially-assembled nonlinear forms
-class PANonlinearFormExtension : public NonlinearFormExtension
+class PANonlinearForm : public NonlinearFormExtension
 {
+private:
+   class Gradient : public Operator
+   {
+   protected:
+      const Operator *R;
+      mutable Vector ge, xe, ye, ze;
+      const Array<NonlinearFormIntegrator*> &dnfi;
+   public:
+      Gradient(const Vector &x, const PANonlinearForm &ext);
+      virtual void Mult(const Vector &x, Vector &y) const;
+   };
+
 protected:
-   const FiniteElementSpace &fes; // Not owned
-   mutable Vector localX, localY;
-   const Operator *elem_restrict_lex; // Not owned
+   mutable Vector xe, ye;
+   mutable const Vector *x_grad;
+   mutable OperatorHandle Grad;
+   const FiniteElementSpace &fes;
+   const Array<NonlinearFormIntegrator*> &dnfi;
+   const Operator *R;
+
 public:
-   PANonlinearFormExtension(NonlinearForm*);
-   void AssemblePA();
+   PANonlinearForm(NonlinearForm *nlf);
+   void Setup();
    void Mult(const Vector &x, Vector &y) const;
+   Operator &GetGradient(const Vector &x) const;
+   double GetGridFunctionEnergy(const Vector &x) const;
+   void AssembleGradientDiagonal(Vector &diag) const;
 };
 }
 #endif // NONLINEARFORM_EXT_HPP

fem/nonlininteg.cpp

@@ -15,6 +15,13 @@
 namespace mfem
 {
 
+double NonlinearFormIntegrator::GetGridFunctionEnergyPA(const Vector &x) const
+{
+   mfem_error ("NonlinearFormIntegrator::GetGridFunctionEnergyPA(...)\n"
+               "   is not implemented for this class.");
+   return 0.0;
+}
+
 void NonlinearFormIntegrator::AssemblePA(const FiniteElementSpace&)
 {
    mfem_error ("NonlinearFormIntegrator::AssemblePA(...)\n"
@@ -34,6 +41,20 @@ void NonlinearFormIntegrator::AddMultPA(const Vector &, Vector &) const
                "   is not implemented for this class.");
 }
 
+void NonlinearFormIntegrator::AddMultGradPA(const Vector&,
+                                            const Vector&, Vector&) const
+{
+   mfem_error ("NonlinearFormIntegrator::AddMultGradPA(...)\n"
+               "   is not implemented for this class.");
+}
+
+void NonlinearFormIntegrator::AssembleGradientDiagonalPA(const mfem::Vector &x,
+                                                         mfem::Vector &diag) const
+{
+   mfem_error ("NonlinearFormIntegrator::AssembleDiagonalPA(...)\n"
+               "   is not implemented for this class.");
+}
+
 void NonlinearFormIntegrator::AssembleElementVector(
    const FiniteElement &el, ElementTransformation &Tr,
    const Vector &elfun, Vector &elvect)

fem/nonlininteg.hpp

@@ -68,6 +68,9 @@ public:
                                    ElementTransformation &Tr,
                                    const Vector &elfun);
 
+   /// Compute the local energy with partial assembly.
+   virtual double GetGridFunctionEnergyPA(const Vector &x) const;
+
    /// Method defining partial assembly.
    /** The result of the partial assembly is stored internally so that it can be
        used later in the methods AddMultPA(). */
@@ -88,6 +91,12 @@ public:
        called. */
    virtual void AddMultPA(const Vector &x, Vector &y) const;
 
+   /// Method for partially assembled gradient action.
+   virtual void AddMultGradPA(const Vector &g,
+                              const Vector &x, Vector &y) const;
+
+   virtual void AssembleGradientDiagonalPA(const Vector &x, Vector &diag) const;
+
    virtual ~NonlinearFormIntegrator() { }
 };
 

fem/pfespace.cpp

@@ -3147,7 +3147,7 @@ static void SetSubVector(const int N,
                          const Array<int> &indices,
                          const Vector &in, Vector &out)
 {
-   auto y = out.Write();
+   auto y = out.ReadWrite();
    const auto x = in.Read();
    const auto I = indices.Read();
    MFEM_FORALL(i, N, y[I[i]] = x[i];);
@@ -3234,7 +3234,7 @@ static void AddSubVector(const int num_unique_dst_indices,
                          const Vector &src,
                          Vector &dst)
 {
-   auto y = dst.Write();
+   auto y = dst.ReadWrite();
    const auto x = src.Read();
    const auto DST_I = unique_dst_indices.Read();
    const auto SRC_O = unique_to_src_offsets.Read();

fem/pgridfunc.cpp

@@ -655,167 +655,6 @@ void ParGridFunction::ProjectBdrCoefficientTangent(VectorCoefficient &vcoeff,
 #endif
 }
 
-double ParGridFunction::ComputeDGFaceJumpError(Coefficient *exsol,
-                                               Coefficient *ell_coeff,
-                                               double Nu,
-                                               const IntegrationRule *irs[]) const
-{
-   const_cast<ParGridFunction *>(this)->ExchangeFaceNbrData();
-
-   int fdof, dim, intorder, k;
-   ElementTransformation *transf;
-   Vector shape, el_dofs, err_val, ell_coeff_val;
-   Array<int> vdofs;
-   IntegrationPoint eip;
-   double error = 0.0;
-
-   ParMesh *mesh = pfes->GetParMesh();
-   dim = mesh->Dimension();
-
-   std::map<int,int> local_to_shared;
-   for (int i = 0; i < mesh->GetNSharedFaces(); ++i)
-   {
-      int i_local = mesh->GetSharedFace(i);
-      local_to_shared[i_local] = i;
-   }
-
-   for (int i = 0; i < mesh->GetNumFaces(); i++)
-   {
-      double shared_face_factor = 1.0;
-      bool shared_face = false;
-      int iel1, iel2, info1, info2;
-      mesh->GetFaceElements(i, &iel1, &iel2);
-      mesh->GetFaceInfos(i, &info1, &info2);
-
-      intorder = fes->GetFE(iel1)->GetOrder();
-
-      FaceElementTransformations *face_elem_transf;
-      const FiniteElement *fe1, *fe2;
-      if (info2 >= 0 && iel2 < 0)
-      {
-         int ishared = local_to_shared[i];
-         face_elem_transf = mesh->GetSharedFaceTransformations(ishared);
-         iel2 = face_elem_transf->Elem2No - mesh->GetNE();
-         fe2 = pfes->GetFaceNbrFE(iel2);
-         if ( (k = fe2->GetOrder()) > intorder )
-         {
-            intorder = k;
-         }
-         shared_face = true;
-         shared_face_factor = 0.5;
-      }
-      else
-      {
-         face_elem_transf = mesh->GetFaceElementTransformations(i);
-
-         if (iel2 >= 0)
-         {
-            fe2 = pfes->GetFE(iel2);
-            if ( (k = fe2->GetOrder()) > intorder )
-            {
-               intorder = k;
-            }
-         }
-         else
-         {
-            fe2 = NULL;
-         }
-      }
-
-      intorder = 2 * intorder;  // <-------------
-      const IntegrationRule *ir;
-      if (irs)
-      {
-         ir = irs[face_elem_transf->GetGeometryType()];
-      }
-      else
-      {
-         ir = &(IntRules.Get(face_elem_transf->GetGeometryType(), intorder));
-      }
-      err_val.SetSize(ir->GetNPoints());
-      ell_coeff_val.SetSize(ir->GetNPoints());
-      // side 1
-      transf = face_elem_transf->Elem1;
-      fe1 = fes->GetFE(iel1);
-      fdof = fe1->GetDof();
-      fes->GetElementVDofs(iel1, vdofs);
-      shape.SetSize(fdof);
-      el_dofs.SetSize(fdof);
-      for (k = 0; k < fdof; k++)
-         if (vdofs[k] >= 0)
-         {
-            el_dofs(k) =   (*this)(vdofs[k]);
-         }
-         else
-         {
-            el_dofs(k) = - (*this)(-1-vdofs[k]);
-         }
-      for (int j = 0; j < ir->GetNPoints(); j++)
-      {
-         face_elem_transf->Loc1.Transform(ir->IntPoint(j), eip);
-         fe1->CalcShape(eip, shape);
-         transf->SetIntPoint(&eip);
-         ell_coeff_val(j) = ell_coeff->Eval(*transf, eip);
-         err_val(j) = exsol->Eval(*transf, eip) - (shape * el_dofs);
-      }
-      if (fe2 != NULL)
-      {
-         // side 2
-         transf = face_elem_transf->Elem2;
-         fdof = fe2->GetDof();
-         shape.SetSize(fdof);
-         el_dofs.SetSize(fdof);
-         if (shared_face)
-         {
-            pfes->GetFaceNbrElementVDofs(iel2, vdofs);
-            for (k = 0; k < fdof; k++)
-               if (vdofs[k] >= 0)
-               {
-                  el_dofs(k) = face_nbr_data[vdofs[k]];
-               }
-               else
-               {
-                  el_dofs(k) = - face_nbr_data[-1-vdofs[k]];
-               }
-         }
-         else
-         {
-            pfes->GetElementVDofs(iel2, vdofs);
-            for (k = 0; k < fdof; k++)
-               if (vdofs[k] >= 0)
-               {
-                  el_dofs(k) = (*this)(vdofs[k]);
-               }
-               else
-               {
-                  el_dofs(k) = - (*this)(-1 - vdofs[k]);
-               }
-         }
-         for (int j = 0; j < ir->GetNPoints(); j++)
-         {
-            face_elem_transf->Loc2.Transform(ir->IntPoint(j), eip);
-            fe2->CalcShape(eip, shape);
-            transf->SetIntPoint(&eip);
-            ell_coeff_val(j) += ell_coeff->Eval(*transf, eip);
-            ell_coeff_val(j) *= 0.5;
-            err_val(j) -= (exsol->Eval(*transf, eip) - (shape * el_dofs));
-         }
-      }
-      transf = face_elem_transf;
-      for (int j = 0; j < ir->GetNPoints(); j++)
-      {
-         const IntegrationPoint &ip = ir->IntPoint(j);
-         transf->SetIntPoint(&ip);
-         error += shared_face_factor*(ip.weight * Nu * ell_coeff_val(j) *
-                                      pow(transf->Weight(), 1.0-1.0/(dim-1)) *
-                                      err_val(j) * err_val(j));
-      }
-   }
-
-   error = (error < 0.0) ? -sqrt(-error) : sqrt(error);
-   return GlobalLpNorm(2.0, error, pfes->GetComm());
-}
-
 void ParGridFunction::Save(std::ostream &out) const
 {
    double *data_  = const_cast<double*>(HostRead());
@@ -872,9 +711,9 @@ void ParGridFunction::SaveAsOne(std::ostream &out)
    int *nfdofs = new int[NRanks];
    int *nrdofs = new int[NRanks];
 
-   double * h_data = const_cast<double *>(this->HostRead());
+   HostReadWrite();
+   values[0] = data;
 
-   values[0] = h_data;
    nv[0]     = pfes -> GetVSize();
    nvdofs[0] = pfes -> GetNVDofs();
    nedofs[0] = pfes -> GetNEDofs();
@@ -975,7 +814,7 @@ void ParGridFunction::SaveAsOne(std::ostream &out)
       MPI_Send(&nvdofs[0], 1, MPI_INT, 0, 456, MyComm);
       MPI_Send(&nedofs[0], 1, MPI_INT, 0, 457, MyComm);
       MPI_Send(&nfdofs[0], 1, MPI_INT, 0, 458, MyComm);
-      MPI_Send(h_data, nv[0], MPI_DOUBLE, 0, 460, MyComm);
+      MPI_Send(data, nv[0], MPI_DOUBLE, 0, 460, MyComm);
    }
 
    delete [] values;
@@ -1021,6 +860,7 @@ double GlobalLpNorm(const double p, double loc_norm, MPI_Comm comm)
    return glob_norm;
 }
 
+
 void ParGridFunction::ComputeFlux(
    BilinearFormIntegrator &blfi,
    GridFunction &flux, bool wcoef, int subdomain)
@@ -1161,6 +1001,6 @@ double L2ZZErrorEstimator(BilinearFormIntegrator &flux_integrator,
    return pow(glob_error, 1.0/norm_p);
 }
 
-} // namespace mfem
+}
 
 #endif // MFEM_USE_MPI

fem/pgridfunc.hpp

@@ -283,77 +283,6 @@ public:
                           pfes->GetComm());
    }
 
-   /// Returns ||grad u_ex - grad u_h||_L2 for H1 or L2 elements
-   virtual double ComputeGradError(VectorCoefficient *exgrad,
-                                   const IntegrationRule *irs[] = NULL) const
-   {
-      return GlobalLpNorm(2.0, GridFunction::ComputeGradError(exgrad,irs),
-                          pfes->GetComm());
-   }
-
-   /// Returns ||curl u_ex - curl u_h||_L2 for ND elements
-   virtual double ComputeCurlError(VectorCoefficient *excurl,
-                                   const IntegrationRule *irs[] = NULL) const
-   {
-      return GlobalLpNorm(2.0, GridFunction::ComputeCurlError(excurl,irs),
-                          pfes->GetComm());
-   }
-
-   /// Returns ||div u_ex - div u_h||_L2 for RT elements
-   virtual double ComputeDivError(Coefficient *exdiv,
-                                  const IntegrationRule *irs[] = NULL) const
-   {
-      return GlobalLpNorm(2.0, GridFunction::ComputeDivError(exdiv,irs),
-                          pfes->GetComm());
-   }
-
-   /// Returns the Face Jumps error for L2 elements
-   virtual double ComputeDGFaceJumpError(Coefficient *exsol,
-                                         Coefficient *ell_coeff,
-                                         double Nu,
-                                         const IntegrationRule *irs[]=NULL)
-   const;
-
-   /// Returns either the H1-seminorm or the DG Face Jumps error or both
-   /// depending on norm_type = 1, 2, 3
-   virtual double ComputeH1Error(Coefficient *exsol, VectorCoefficient *exgrad,
-                                 Coefficient *ell_coef, double Nu,
-                                 int norm_type) const
-   {
-      return GlobalLpNorm(2.0,
-                          GridFunction::ComputeH1Error(exsol,exgrad,ell_coef,
-                                                       Nu, norm_type),
-                          pfes->GetComm());
-   }
-
-   /// Returns the error measured in H1-norm for H1 elements or in "broken"
-   /// H1-norm for L2 elements
-   virtual double ComputeH1Error(Coefficient *exsol, VectorCoefficient *exgrad,
-                                 const IntegrationRule *irs[] = NULL) const
-   {
-      return GlobalLpNorm(2.0, GridFunction::ComputeH1Error(exsol,exgrad,irs),
-                          pfes->GetComm());
-   }
-
-   /// Returns the error measured H(div)-norm for RT elements
-   virtual double ComputeHDivError(VectorCoefficient *exsol,
-                                   Coefficient *exdiv,
-                                   const IntegrationRule *irs[] = NULL) const
-   {
-      return GlobalLpNorm(2.0, GridFunction::ComputeHDivError(exsol,exdiv,irs),
-                          pfes->GetComm());
-   }
-
-   /// Returns the error measured H(curl)-norm for ND elements
-   virtual double ComputeHCurlError(VectorCoefficient *exsol,
-                                    VectorCoefficient *excurl,
-                                    const IntegrationRule *irs[] = NULL) const
-   {
-      return GlobalLpNorm(2.0,
-                          GridFunction::ComputeHCurlError(exsol,excurl,irs),
-                          pfes->GetComm());
-   }
-
    virtual double ComputeMaxError(Coefficient *exsol[],
                                   const IntegrationRule *irs[] = NULL) const
    {

fem/plinearform.cpp

@@ -21,6 +21,7 @@ namespace mfem
 void ParLinearForm::Update(ParFiniteElementSpace *pf)
 {
    if (pf) { pfes = pf; }
+
    LinearForm::Update(pfes);
 }
 
@@ -30,19 +31,6 @@ void ParLinearForm::Update(ParFiniteElementSpace *pf, Vector &v, int v_offset)
    LinearForm::Update(pf,v,v_offset);
 }
 
-void ParLinearForm::MakeRef(FiniteElementSpace *f, Vector &v, int v_offset)
-{
-   LinearForm::MakeRef(f, v, v_offset);
-   pfes = dynamic_cast<ParFiniteElementSpace*>(f);
-   MFEM_ASSERT(pfes != NULL, "not a ParFiniteElementSpace");
-}
-
-void ParLinearForm::MakeRef(ParFiniteElementSpace *pf, Vector &v, int v_offset)
-{
-   LinearForm::MakeRef(pf, v, v_offset);
-   pfes = pf;
-}
-
 void ParLinearForm::ParallelAssemble(Vector &tv)
 {
    const Operator* prolong = pfes->GetProlongationMatrix();

fem/plinearform.hpp

@@ -92,27 +92,6 @@ public:
        @note This method does not perform assembly. */
    void Update(ParFiniteElementSpace *pf, Vector &v, int v_offset);
 
-
-   /** @brief Make the ParLinearForm reference external data on a new
-       FiniteElementSpace. */
-   /** This method changes the FiniteElementSpace associated with the
-       ParLinearForm to @a *f and sets the data of the Vector @a v (plus the @a
-       v_offset) as external data in the ParLinearForm.
-
-       @note This version of the method will also perform bounds checks when the
-       build option MFEM_DEBUG is enabled. */
-   virtual void MakeRef(FiniteElementSpace *f, Vector &v, int v_offset);
-
-   /** @brief Make the ParLinearForm reference external data on a new
-       ParFiniteElementSpace. */
-   /** This method changes the ParFiniteElementSpace associated with the
-       ParLinearForm to @a *pf and sets the data of the Vector @a v (plus the @a
-       v_offset) as external data in the ParLinearForm.
-
-       @note This version of the method will also perform bounds checks when the
-       build option MFEM_DEBUG is enabled. */
-   void MakeRef(ParFiniteElementSpace *pf, Vector &v, int v_offset);
-
    /// Assemble the vector on the true dofs, i.e. P^t v.
    void ParallelAssemble(Vector &tv);
 
@@ -120,10 +99,10 @@ public:
    HypreParVector *ParallelAssemble();
 
    /// Return the action of the ParLinearForm as a linear mapping.
-   /** Linear forms are linear functionals which map ParGridFunction%s to the
-       real numbers. This method performs this mapping which in this case is
-       equivalent as an inner product of the ParLinearForm and
-       ParGridFunction. */
+   /** Linear forms are linear functionals which map ParGridFunction%s to
+       the real numbers.  This method performs this mapping which in
+       this case is equivalent as an inner product of the ParLinearForm
+       and ParGridFunction. */
    double operator()(const ParGridFunction &gf) const
    {
       return InnerProduct(pfes->GetComm(), *this, gf);

fem/pnonlinearform.cpp

@@ -14,6 +14,7 @@
 #ifdef MFEM_USE_MPI
 
 #include "fem.hpp"
+#include "../general/forall.hpp"
 
 namespace mfem
 {
@@ -49,6 +50,7 @@ void ParNonlinearForm::Mult(const Vector &x, Vector &y) const
 
    if (fnfi.Size())
    {
+      MFEM_VERIFY(!NonlinearForm::ext,"");
       // Terms over shared interior faces in parallel.
       ParFiniteElementSpace *pfes = ParFESpace();
       ParMesh *pmesh = pfes->GetParMesh();
@@ -86,15 +88,16 @@ void ParNonlinearForm::Mult(const Vector &x, Vector &y) const
 
    P->MultTranspose(aux2, y);
 
-   y.HostReadWrite();
-   for (int i = 0; i < ess_tdof_list.Size(); i++)
-   {
-      y(ess_tdof_list[i]) = 0.0;
-   }
+   const int N = ess_tdof_list.Size();
+   const auto idx = ess_tdof_list.Read();
+   auto Y = y.ReadWrite();
+   MFEM_FORALL(i, N, Y[idx[i]] = 0.0; );
 }
 
 const SparseMatrix &ParNonlinearForm::GetLocalGradient(const Vector &x) const
 {
+   if (NonlinearForm::ext) { MFEM_ABORT("Not yet implemented!"); }
+
    NonlinearForm::GetGradient(x); // (re)assemble Grad, no b.c.
 
    return *Grad;
@@ -104,16 +107,20 @@ Operator &ParNonlinearForm::GetGradient(const Vector &x) const
 {
    ParFiniteElementSpace *pfes = ParFESpace();
 
-   pGrad.Clear();
+   Operator &grad = NonlinearForm::GetGradient(x); // (re)assemble Grad, no b.c.
 
-   NonlinearForm::GetGradient(x); // (re)assemble Grad, no b.c.
+   pGrad.Clear();
 
    OperatorHandle dA(pGrad.Type()), Ph(pGrad.Type());
 
    if (fnfi.Size() == 0)
    {
-      dA.MakeSquareBlockDiag(pfes->GetComm(), pfes->GlobalVSize(),
-                             pfes->GetDofOffsets(), Grad);
+      if (NonlinearForm::ext) { dA.Reset(&grad, false); }
+      else
+      {
+         dA.MakeSquareBlockDiag(pfes->GetComm(), pfes->GlobalVSize(),
+                                pfes->GetDofOffsets(), Grad);
+      }
    }
    else
    {

fem/quadinterpolator.hpp

@@ -41,10 +41,11 @@ protected:
    const FiniteElementSpace *fespace;  ///< Not owned
    const QuadratureSpace *qspace;      ///< Not owned
    const IntegrationRule *IntRule;     ///< Not owned
+
    mutable QVectorLayout q_layout;     ///< Output Q-vector layout
+   mutable bool use_tensor_products;   ///< Tensor product evaluation mmode
 
-   mutable bool use_tensor_products;
-
+public:
    static const int MAX_NQ2D = 100;
    static const int MAX_ND2D = 100;
    static const int MAX_VDIM2D = 3;
@@ -53,7 +54,6 @@ protected:
    static const int MAX_ND3D = 1000;
    static const int MAX_VDIM3D = 3;
 
-public:
    enum EvalFlags
    {
       VALUES       = 1 << 0,  ///< Evaluate the values at quadrature points
@@ -61,21 +61,28 @@ public:
       /** @brief Assuming the derivative at quadrature points form a matrix,
           this flag can be used to compute and store their determinants. This
           flag can only be used in Mult(). */
-      DETERMINANTS = 1 << 2
+      DETERMINANTS = 1 << 2,
+      PHYSICAL_DERIVATIVES = 1 << 3 ///< Evaluate the physical derivatives
    };
 
    QuadratureInterpolator(const FiniteElementSpace &fes,
-                          const IntegrationRule &ir);
+                          const IntegrationRule &ir,
+                          const bool use_tensor_products = false);
 
    QuadratureInterpolator(const FiniteElementSpace &fes,
-                          const QuadratureSpace &qs);
+                          const QuadratureSpace &qs,
+                          const bool use_tensor_products = false);
 
    /** @brief Disable the use of tensor product evaluations, for tensor-product
        elements, e.g. quads and hexes. */
-   /** Currently, tensor product evaluations are not implemented and this method
-       has no effect. */
-   void DisableTensorProducts(bool disable = true) const
-   { use_tensor_products = !disable; }
+   void DisableTensorProducts() const { use_tensor_products = false; }
+
+   /** @brief Enable the use of tensor product evaluations, for tensor-product
+       elements, e.g. quads and hexes. */
+   void EnableTensorProducts() const { use_tensor_products = true; }
+
+   /** @brief Query the current evaluation mode. */
+   bool UseTensorProducts() const { return use_tensor_products; }
 
    /** @brief Query the current output Q-vector layout. The default value is
        QVectorLayout::byNODES. */
@@ -83,8 +90,7 @@ public:
 
    /** @brief Set the desired output Q-vector layout. The default value is
        QVectorLayout::byNODES. */
-   void SetOutputLayout(QVectorLayout out_layout) const
-   { q_layout = out_layout; }
+   void SetOutputLayout(QVectorLayout layout) const { q_layout = layout; }
 
    /// Interpolate the E-vector @a e_vec to quadrature points.
    /** The @a eval_flags are a bitwise mask of constants from the EvalFlags
@@ -99,26 +105,36 @@ public:
              Vector &q_val, Vector &q_der, Vector &q_det) const;
 
    /// Interpolate the values of the E-vector @a e_vec at quadrature points.
+   template <QVectorLayout>
+   void Values(const Vector &e_vec, Vector &q_val) const;
    void Values(const Vector &e_vec, Vector &q_val) const;
 
    /** @brief Interpolate the derivatives of the E-vector @a e_vec at quadrature
        points. */
+   template <QVectorLayout>
+   void Derivatives(const Vector &e_vec, Vector &q_der) const;
    void Derivatives(const Vector &e_vec, Vector &q_der) const;
 
    /** @brief Interpolate the derivatives in physical space of the E-vector
        @a e_vec at quadrature points. */
+   template <QVectorLayout>
    void PhysDerivatives(const Vector &e_vec, Vector &q_der) const;
+   void PhysDerivatives(const Vector &e_vec, Vector &q_der) const;
+
+   /// Compute the determinant of the E-vector @a e_vec at quadrature points.
+   void Determinants(const Vector &e_vec, Vector &q_det) const;
 
    /// Perform the transpose operation of Mult(). (TODO)
    void MultTranspose(unsigned eval_flags, const Vector &q_val,
                       const Vector &q_der, Vector &e_vec) const;
 
    // Compute kernels follow (cannot be private or protected with nvcc)
-
    /// Template compute kernel for 2D.
    template<const int T_VDIM = 0, const int T_ND = 0, const int T_NQ = 0>
-   static void Eval2D(const int NE,
+   static void Mult2D(const int NE,
                       const int vdim,
+                      const QVectorLayout q_layout,
+                      const GeometricFactors *geom,
                       const DofToQuad &maps,
                       const Vector &e_vec,
                       Vector &q_val,
@@ -128,8 +144,10 @@ public:
 
    /// Template compute kernel for 3D.
    template<const int T_VDIM = 0, const int T_ND = 0, const int T_NQ = 0>
-   static void Eval3D(const int NE,
+   static void Mult3D(const int NE,
                       const int vdim,
+                      const QVectorLayout q_layout,
+                      const GeometricFactors *geom,
                       const DofToQuad &maps,
                       const Vector &e_vec,
                       Vector &q_val,

fem/quadinterpolator_det.cpp

@@ -0,0 +1,208 @@
+// Copyright (c) 2010-2020, 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 "tmop_pa.hpp"
+#include "quadinterpolator.hpp"
+#include "../general/forall.hpp"
+#include "../linalg/dtensor.hpp"
+#include "../fem/kernels.hpp"
+#include "../linalg/kernels.hpp"
+
+using namespace mfem;
+
+namespace mfem
+{
+
+template<int T_D1D = 0, int T_Q1D = 0, int MAX_D1D = 0, int MAX_Q1D = 0>
+static void Det2D(const int NE,
+                  const double *b,
+                  const double *g,
+                  const double *x,
+                  double *y,
+                  const int vdim = 1,
+                  const int d1d = 0,
+                  const int q1d = 0)
+{
+   constexpr int DIM = 2;
+   constexpr int NBZ = 1;
+
+   const int D1D = T_D1D ? T_D1D : d1d;
+   const int Q1D = T_Q1D ? T_Q1D : q1d;
+
+   const auto B = Reshape(b, Q1D, D1D);
+   const auto G = Reshape(g, Q1D, D1D);
+   const auto X = Reshape(x,  D1D, D1D, DIM, NE);
+   auto Y = Reshape(y, Q1D, Q1D, NE);
+
+   MFEM_FORALL_2D(e, NE, Q1D, Q1D, NBZ,
+   {
+      constexpr int NBZ = 1;
+      constexpr int MQ1 = T_Q1D ? T_Q1D : MAX_Q1D;
+      constexpr int MD1 = T_D1D ? T_D1D : MAX_D1D;
+      const int D1D = T_D1D ? T_D1D : d1d;
+      const int Q1D = T_Q1D ? T_Q1D : q1d;
+
+      MFEM_SHARED double BG[2][MQ1*MD1];
+      MFEM_SHARED double XY[2][NBZ][MD1*MD1];
+      MFEM_SHARED double DQ[4][NBZ][MD1*MQ1];
+      MFEM_SHARED double QQ[4][NBZ][MQ1*MQ1];
+
+      kernels::LoadX<MD1,NBZ>(e,D1D,X,XY);
+      kernels::LoadBG<MD1,MQ1>(D1D,Q1D,B,G,BG);
+
+      kernels::GradX<MD1,MQ1,NBZ>(D1D,Q1D,BG,XY,DQ);
+      kernels::GradY<MD1,MQ1,NBZ>(D1D,Q1D,BG,DQ,QQ);
+
+      MFEM_FOREACH_THREAD(qy,y,Q1D)
+      {
+         MFEM_FOREACH_THREAD(qx,x,Q1D)
+         {
+            double J[4];
+            kernels::PullGrad<MQ1,NBZ>(qx,qy,QQ,J);
+            Y(qx,qy,e) = kernels::Det<2>(J);
+         }
+      }
+   });
+}
+
+template<int T_D1D = 0, int T_Q1D = 0, int MAX_D1D = 0, int MAX_Q1D = 0>
+static void Det3D(const int NE,
+                  const double *b,
+                  const double *g,
+                  const double *x,
+                  double *y,
+                  const int vdim = 1,
+                  const int d1d = 0,
+                  const int q1d = 0)
+{
+   constexpr int DIM = 3;
+   const int D1D = T_D1D ? T_D1D : d1d;
+   const int Q1D = T_Q1D ? T_Q1D : q1d;
+
+   const auto B = Reshape(b, Q1D, D1D);
+   const auto G = Reshape(g, Q1D, D1D);
+   const auto X = Reshape(x, D1D, D1D, D1D, DIM, NE);
+   auto Y = Reshape(y, Q1D, Q1D, Q1D, NE);
+
+   MFEM_FORALL_3D(e, NE, Q1D, Q1D, Q1D,
+   {
+      const int D1D = T_D1D ? T_D1D : d1d;
+      const int Q1D = T_Q1D ? T_Q1D : q1d;
+      constexpr int MQ1 = T_Q1D ? T_Q1D : MAX_Q1D;
+      constexpr int MD1 = T_D1D ? T_D1D : MAX_D1D;
+      constexpr int MDQ = MQ1 > MD1 ? MQ1 : MD1;
+
+      MFEM_SHARED double BG[2][MQ1*MD1];
+      MFEM_SHARED double sm0[9][MDQ*MDQ*MDQ];
+      MFEM_SHARED double sm1[9][MDQ*MDQ*MDQ];
+
+      double (*DDD)[MD1*MD1*MD1] = (double (*)[MD1*MD1*MD1]) (sm0);
+      double (*DDQ)[MD1*MD1*MQ1] = (double (*)[MD1*MD1*MQ1]) (sm1);
+      double (*DQQ)[MD1*MQ1*MQ1] = (double (*)[MD1*MQ1*MQ1]) (sm0);
+      double (*QQQ)[MQ1*MQ1*MQ1] = (double (*)[MQ1*MQ1*MQ1]) (sm1);
+
+      kernels::LoadX<MD1>(e,D1D,X,DDD);
+      kernels::LoadBG<MD1,MQ1>(D1D,Q1D,B,G,BG);
+
+      kernels::GradX<MD1,MQ1>(D1D,Q1D,BG,DDD,DDQ);
+      kernels::GradY<MD1,MQ1>(D1D,Q1D,BG,DDQ,DQQ);
+      kernels::GradZ<MD1,MQ1>(D1D,Q1D,BG,DQQ,QQQ);
+
+      MFEM_FOREACH_THREAD(qz,z,Q1D)
+      {
+         MFEM_FOREACH_THREAD(qy,y,Q1D)
+         {
+            MFEM_FOREACH_THREAD(qx,x,Q1D)
+            {
+               double J[9];
+               kernels::PullGrad<MQ1>(qx,qy,qz, QQQ, J);
+               Y(qx,qy,qz,e) = kernels::Det<3>(J);
+            }
+         }
+      }
+   });
+}
+
+void QuadratureInterpolator::Determinants(const Vector &e_vec,
+                                          Vector &q_det) const
+{
+   if (use_tensor_products)
+   {
+      const int NE = fespace->GetNE();
+      if (NE == 0) { return; }
+      const int vdim = fespace->GetVDim();
+      const int dim = fespace->GetMesh()->Dimension();
+      const FiniteElement *fe = fespace->GetFE(0);
+      const IntegrationRule *ir =
+         IntRule ? IntRule : &qspace->GetElementIntRule(0);
+      const DofToQuad &maps = fe->GetDofToQuad(*ir, DofToQuad::TENSOR);
+      const int D1D = maps.ndof;
+      const int Q1D = maps.nqpt;
+      const double *B = maps.B.Read();
+      const double *G = maps.G.Read();
+      const double *X = e_vec.Read();
+      double *Y = q_det.Write();
+
+      const int id = (dim<<12) | (vdim<<8) | (D1D<<4) | Q1D;
+
+      switch (id)
+      {
+         case 0x2222: return Det2D<2,2>(NE,B,G,X,Y);
+         case 0x2223: return Det2D<2,3>(NE,B,G,X,Y);
+         case 0x2224: return Det2D<2,4>(NE,B,G,X,Y);
+         case 0x2226: return Det2D<2,6>(NE,B,G,X,Y);
+         case 0x2234: return Det2D<3,4>(NE,B,G,X,Y);
+         case 0x2236: return Det2D<3,6>(NE,B,G,X,Y);
+         case 0x2244: return Det2D<4,4>(NE,B,G,X,Y);
+         case 0x2246: return Det2D<4,6>(NE,B,G,X,Y);
+         case 0x2256: return Det2D<5,6>(NE,B,G,X,Y);
+
+         case 0x3324: return Det3D<2,4>(NE,B,G,X,Y);
+         case 0x3333: return Det3D<3,3>(NE,B,G,X,Y);
+         case 0x3335: return Det3D<3,5>(NE,B,G,X,Y);
+         case 0x3336: return Det3D<3,6>(NE,B,G,X,Y);
+         //case 0x3348: return Det3D<4,8>(NE,B,G,X,Y);
+
+         default:
+         {
+            if (dim == 2)
+            {
+               constexpr int MD1 = 8;
+               constexpr int MQ1 = 8;
+               MFEM_VERIFY(D1D <= MD1, "Orders higher than " << MD1-1
+                           << " are not supported!");
+               MFEM_VERIFY(Q1D <= MQ1, "Quadrature rules with more than "
+                           << MQ1 << " 1D points are not supported!");
+               return Det2D<0,0,MD1,MQ1>(NE,B,G,X,Y,vdim,D1D,Q1D);
+            }
+            if (dim == 3)
+            {
+               constexpr int MD1 = 6;
+               constexpr int MQ1 = 6;
+               MFEM_VERIFY(D1D <= MD1, "Orders higher than " << MD1-1
+                           << " are not supported!");
+               MFEM_VERIFY(Q1D <= MQ1, "Quadrature rules with more than "
+                           << MQ1 << " 1D points are not supported!");
+               return Det3D<0,0,MD1,MQ1>(NE,B,G,X,Y,vdim,D1D,Q1D);
+            }
+         }
+
+      }
+      MFEM_ABORT("Kernel " << std::hex << id << std::dec << " not supported yet");
+   }
+   else
+   {
+      Vector empty;
+      Mult(e_vec, DETERMINANTS, empty, empty, q_det);
+   }
+}
+
+} // namespace mfem

fem/quadinterpolator_eval.hpp

@@ -0,0 +1,233 @@
+// Copyright (c) 2010-2020, 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 "quadinterpolator.hpp"
+#include "../general/forall.hpp"
+#include "../linalg/dtensor.hpp"
+#include "../linalg/kernels.hpp"
+
+namespace mfem
+{
+
+template<QVectorLayout Q_LAYOUT,
+         int T_VDIM = 0, int T_D1D = 0, int T_Q1D = 0,
+         int T_NBZ = 1, int MAX_D1D = 0, int MAX_Q1D = 0>
+static void Eval2D(const int NE,
+                   const double *b_,
+                   const double *x_,
+                   double *y_,
+                   const int vdim = 0,
+                   const int d1d = 0,
+                   const int q1d = 0)
+{
+   constexpr int NBZ = T_NBZ ? T_NBZ : 1;
+
+   const int D1D = T_D1D ? T_D1D : d1d;
+   const int Q1D = T_Q1D ? T_Q1D : q1d;
+   const int VDIM = T_VDIM ? T_VDIM : vdim;
+
+   const auto b = Reshape(b_, Q1D, D1D);
+   const auto x = Reshape(x_, D1D, D1D, VDIM, NE);
+   auto y = Q_LAYOUT == QVectorLayout:: byNODES ?
+            Reshape(y_, Q1D, Q1D, VDIM, NE):
+            Reshape(y_, VDIM, Q1D, Q1D, NE);
+
+   MFEM_FORALL_2D(e, NE, Q1D, Q1D, NBZ,
+   {
+      const int D1D = T_D1D ? T_D1D : d1d;
+      const int Q1D = T_Q1D ? T_Q1D : q1d;
+      const int VDIM = T_VDIM ? T_VDIM : vdim;
+      constexpr int MQ1 = T_Q1D ? T_Q1D : MAX_Q1D;
+      constexpr int MD1 = T_D1D ? T_D1D : MAX_D1D;
+      constexpr int NBZ = T_NBZ ? T_NBZ : 1;
+      const int tidz = MFEM_THREAD_ID(z);
+
+      MFEM_SHARED double s_B[MQ1*MD1];
+      DeviceTensor<2,double> B(s_B, Q1D, D1D);
+
+      MFEM_SHARED double s_DD[NBZ][MD1*MD1];
+      DeviceTensor<2,double> DD((double*)(s_DD+tidz), MD1, MD1);
+
+      MFEM_SHARED double s_DQ[NBZ][MD1*MQ1];
+      DeviceTensor<2,double> DQ((double*)(s_DQ+tidz), MD1, MQ1);
+
+      if (tidz == 0)
+      {
+         MFEM_FOREACH_THREAD(d,y,D1D)
+         {
+            MFEM_FOREACH_THREAD(q,x,Q1D)
+            {
+               B(q,d) = b(q,d);
+            }
+         }
+      }
+      MFEM_SYNC_THREAD;
+
+      for (int c = 0; c < VDIM; c++)
+      {
+         MFEM_FOREACH_THREAD(dx,x,D1D)
+         {
+            MFEM_FOREACH_THREAD(dy,y,D1D)
+            {
+               DD(dx,dy) = x(dx,dy,c,e);
+            }
+         }
+         MFEM_SYNC_THREAD;
+         MFEM_FOREACH_THREAD(dy,y,D1D)
+         {
+            MFEM_FOREACH_THREAD(qx,x,Q1D)
+            {
+               double u = 0.0;
+               for (int dx = 0; dx < D1D; ++dx)
+               {
+                  u += B(qx,dx) *  DD(dx,dy);
+               }
+               DQ(dy,qx) = u;
+            }
+         }
+         MFEM_SYNC_THREAD;
+         MFEM_FOREACH_THREAD(qy,y,Q1D)
+         {
+            MFEM_FOREACH_THREAD(qx,x,Q1D)
+            {
+               double u = 0.0;
+               for (int dy = 0; dy < D1D; ++dy)
+               {
+                  u += DQ(dy,qx) * B(qy,dy);
+               }
+               if (Q_LAYOUT == QVectorLayout::byVDIM) { y(c,qx,qy,e) = u; }
+               if (Q_LAYOUT == QVectorLayout::byNODES) { y(qx,qy,c,e) = u; }
+            }
+         }
+         MFEM_SYNC_THREAD;
+      }
+   });
+}
+
+template<QVectorLayout Q_LAYOUT,
+         int T_VDIM = 0, int T_D1D = 0, int T_Q1D = 0,
+         int MAX_D1D = 0, int MAX_Q1D = 0>
+static void Eval3D(const int NE,
+                   const double *b_,
+                   const double *x_,
+                   double *y_,
+                   const int vdim = 0,
+                   const int d1d = 0,
+                   const int q1d = 0)
+{
+   const int D1D = T_D1D ? T_D1D : d1d;
+   const int Q1D = T_Q1D ? T_Q1D : q1d;
+   const int VDIM = T_VDIM ? T_VDIM : vdim;
+
+   const auto b = Reshape(b_, Q1D, D1D);
+   const auto x = Reshape(x_, D1D, D1D, D1D, VDIM, NE);
+   auto y = Q_LAYOUT == QVectorLayout:: byNODES ?
+            Reshape(y_, Q1D, Q1D, Q1D, VDIM, NE):
+            Reshape(y_, VDIM, Q1D, Q1D, Q1D, NE);
+
+   MFEM_FORALL_3D(e, NE, Q1D, Q1D, Q1D,
+   {
+      const int D1D = T_D1D ? T_D1D : d1d;
+      const int Q1D = T_Q1D ? T_Q1D : q1d;
+      const int VDIM = T_VDIM ? T_VDIM : vdim;
+      constexpr int MQ1 = T_Q1D ? T_Q1D : MAX_Q1D;
+      constexpr int MD1 = T_D1D ? T_D1D : MAX_D1D;
+      constexpr int MDQ = (MQ1 > MD1) ? MQ1 : MD1;
+      const int tidz = MFEM_THREAD_ID(z);
+
+      MFEM_SHARED double s_B[MQ1*MD1];
+      DeviceTensor<2,double> B(s_B, Q1D, D1D);
+
+      MFEM_SHARED double sm0[MDQ*MDQ*MDQ];
+      MFEM_SHARED double sm1[MDQ*MDQ*MDQ];
+      DeviceTensor<3,double> DDD(sm0, MD1, MD1, MD1);
+      DeviceTensor<3,double> DDQ(sm1, MD1, MD1, MQ1);
+      DeviceTensor<3,double> DQQ(sm0, MD1, MQ1, MQ1);
+
+      if (tidz == 0)
+      {
+         MFEM_FOREACH_THREAD(d,y,D1D)
+         {
+            MFEM_FOREACH_THREAD(q,x,Q1D)
+            {
+               B(q,d) = b(q,d);
+            }
+         }
+      }
+      MFEM_SYNC_THREAD;
+
+      for (int c = 0; c < VDIM; c++)
+      {
+         MFEM_FOREACH_THREAD(dz,z,D1D)
+         {
+            MFEM_FOREACH_THREAD(dy,y,D1D)
+            {
+               MFEM_FOREACH_THREAD(dx,x,D1D)
+               {
+                  DDD(dx,dy,dz) = x(dx,dy,dz,c,e);
+               }
+            }
+         }
+         MFEM_SYNC_THREAD;
+         MFEM_FOREACH_THREAD(dz,z,D1D)
+         {
+            MFEM_FOREACH_THREAD(dy,y,D1D)
+            {
+               MFEM_FOREACH_THREAD(qx,x,Q1D)
+               {
+                  double u = 0.0;
+                  for (int dx = 0; dx < D1D; ++dx)
+                  {
+                     u += B(qx,dx) * DDD(dx,dy,dz);
+                  }
+                  DDQ(dz,dy,qx) = u;
+               }
+            }
+         }
+         MFEM_SYNC_THREAD;
+         MFEM_FOREACH_THREAD(dz,z,D1D)
+         {
+            MFEM_FOREACH_THREAD(qy,y,Q1D)
+            {
+               MFEM_FOREACH_THREAD(qx,x,Q1D)
+               {
+                  double u = 0.0;
+                  for (int dy = 0; dy < D1D; ++dy)
+                  {
+                     u += DDQ(dz,dy,qx) * B(qy,dy);
+                  }
+                  DQQ(dz,qy,qx) = u;
+               }
+            }
+         }
+         MFEM_SYNC_THREAD;
+         MFEM_FOREACH_THREAD(qz,z,Q1D)
+         {
+            MFEM_FOREACH_THREAD(qy,y,Q1D)
+            {
+               MFEM_FOREACH_THREAD(qx,x,Q1D)
+               {
+                  double u = 0.0;
+                  for (int dz = 0; dz < D1D; ++dz)
+                  {
+                     u +=  DQQ(dz,qy,qx) * B(qz,dz);
+                  }
+                  if (Q_LAYOUT == QVectorLayout::byVDIM) { y(c,qx,qy,qz,e) = u; }
+                  if (Q_LAYOUT == QVectorLayout::byNODES) { y(qx,qy,qz,c,e) = u; }
+               }
+            }
+         }
+         MFEM_SYNC_THREAD;
+      }
+   });
+}
+
+} // namespace mfem

fem/quadinterpolator_eval_by_nodes.cpp

@@ -0,0 +1,110 @@
+// Copyright (c) 2010-2020, 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 "quadinterpolator.hpp"
+#include "quadinterpolator_eval.hpp"
+#include "../general/forall.hpp"
+#include "../linalg/dtensor.hpp"
+#include "../linalg/kernels.hpp"
+
+namespace mfem
+{
+
+template<>
+void QuadratureInterpolator::Values<QVectorLayout::byNODES>(
+   const Vector &e_vec, Vector &q_val) const
+{
+   const int NE = fespace->GetNE();
+   if (NE == 0) { return; }
+   const int vdim = fespace->GetVDim();
+   const int dim = fespace->GetMesh()->Dimension();
+   const FiniteElement *fe = fespace->GetFE(0);
+   const IntegrationRule *ir =
+      IntRule ? IntRule : &qspace->GetElementIntRule(0);
+   const DofToQuad &maps = fe->GetDofToQuad(*ir, DofToQuad::TENSOR);
+   const int D1D = maps.ndof;
+   const int Q1D = maps.nqpt;
+   const double *B = maps.B.Read();
+   const double *X = e_vec.Read();
+   double *Y = q_val.Write();
+
+   constexpr QVectorLayout L = QVectorLayout::byNODES;
+
+   const int id = (dim<<12) | (vdim<<8) | (D1D<<4) | Q1D;
+
+   switch (id)
+   {
+      case 0x2133: return Eval2D<L,1,3,3>(NE,B,X,Y);
+      case 0x2124: return Eval2D<L,1,2,4>(NE,B,X,Y);
+      case 0x2132: return Eval2D<L,1,3,2>(NE,B,X,Y);
+      case 0x2134: return Eval2D<L,1,3,4>(NE,B,X,Y);
+      case 0x2143: return Eval2D<L,1,4,3>(NE,B,X,Y);
+      case 0x2144: return Eval2D<L,1,4,4>(NE,B,X,Y);
+
+      case 0x2222: return Eval2D<L,2,2,2>(NE,B,X,Y);
+      case 0x2223: return Eval2D<L,2,2,3>(NE,B,X,Y);
+      case 0x2224: return Eval2D<L,2,2,4>(NE,B,X,Y);
+      case 0x2225: return Eval2D<L,2,2,5>(NE,B,X,Y);
+      case 0x2226: return Eval2D<L,2,2,6>(NE,B,X,Y);
+      case 0x2233: return Eval2D<L,2,3,3>(NE,B,X,Y);
+      case 0x2234: return Eval2D<L,2,3,4>(NE,B,X,Y);
+      case 0x2236: return Eval2D<L,2,3,6>(NE,B,X,Y);
+      case 0x2243: return Eval2D<L,2,4,3>(NE,B,X,Y);
+      case 0x2244: return Eval2D<L,2,4,4>(NE,B,X,Y);
+      case 0x2245: return Eval2D<L,2,4,5>(NE,B,X,Y);
+      case 0x2246: return Eval2D<L,2,4,6>(NE,B,X,Y);
+      case 0x2247: return Eval2D<L,2,4,7>(NE,B,X,Y);
+      case 0x2256: return Eval2D<L,2,5,6>(NE,B,X,Y);
+
+      case 0x3124: return Eval3D<L,1,2,4>(NE,B,X,Y);
+      case 0x3133: return Eval3D<L,1,3,3>(NE,B,X,Y);
+      case 0x3134: return Eval3D<L,1,3,4>(NE,B,X,Y);
+      case 0x3136: return Eval3D<L,1,3,6>(NE,B,X,Y);
+      case 0x3143: return Eval3D<L,1,4,3>(NE,B,X,Y);
+      case 0x3144: return Eval3D<L,1,4,4>(NE,B,X,Y);
+      case 0x3148: return Eval3D<L,1,4,8>(NE,B,X,Y);
+
+      case 0x3222: return Eval3D<L,2,2,2>(NE,B,X,Y);
+      case 0x3223: return Eval3D<L,2,2,3>(NE,B,X,Y);
+      case 0x3234: return Eval3D<L,2,3,4>(NE,B,X,Y);
+
+      case 0x3323: return Eval3D<L,3,2,3>(NE,B,X,Y);
+      case 0x3324: return Eval3D<L,3,2,4>(NE,B,X,Y);
+      case 0x3325: return Eval3D<L,3,2,5>(NE,B,X,Y);
+      case 0x3326: return Eval3D<L,3,2,6>(NE,B,X,Y);
+      case 0x3333: return Eval3D<L,3,3,3>(NE,B,X,Y);
+      case 0x3334: return Eval3D<L,3,3,4>(NE,B,X,Y);
+      case 0x3335: return Eval3D<L,3,3,5>(NE,B,X,Y);
+      case 0x3336: return Eval3D<L,3,3,6>(NE,B,X,Y);
+      case 0x3343: return Eval3D<L,3,4,3>(NE,B,X,Y);
+      case 0x3344: return Eval3D<L,3,4,4>(NE,B,X,Y);
+      case 0x3346: return Eval3D<L,3,4,6>(NE,B,X,Y);
+      case 0x3347: return Eval3D<L,3,4,7>(NE,B,X,Y);
+      case 0x3348: return Eval3D<L,3,4,8>(NE,B,X,Y);
+
+      default:
+      {
+         constexpr int MD1 = 8;
+         constexpr int MQ1 = 8;
+         MFEM_VERIFY(D1D <= MD1, "Orders higher than " << MD1-1
+                     << " are not supported!");
+         MFEM_VERIFY(Q1D <= MQ1, "Quadrature rules with more than "
+                     << MQ1 << " 1D points are not supported!");
+         if (dim == 2) { Eval2D<L,0,0,0,0,MD1,MQ1>(NE,B,X,Y,vdim,D1D,Q1D); }
+         if (dim == 3) { Eval3D<L,0,0,0,MD1,MQ1>(NE,B,X,Y,vdim,D1D,Q1D); }
+         return;
+      }
+   }
+   mfem::out << "Unknown kernel 0x" << std::hex << id << std::endl;
+   MFEM_ABORT("Kernel not supported yet");
+}
+
+} // namespace mfem

fem/quadinterpolator_eval_by_vdim.cpp

@@ -0,0 +1,79 @@
+// Copyright (c) 2010-2020, 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 "quadinterpolator.hpp"
+#include "quadinterpolator_eval.hpp"
+#include "../general/forall.hpp"
+#include "../linalg/dtensor.hpp"
+#include "../linalg/kernels.hpp"
+
+namespace mfem
+{
+
+template<>
+void QuadratureInterpolator::Values<QVectorLayout::byVDIM>(
+   const Vector &e_vec, Vector &q_val) const
+{
+   const int NE = fespace->GetNE();
+   if (NE == 0) { return; }
+   const int vdim = fespace->GetVDim();
+   const int dim = fespace->GetMesh()->Dimension();
+   const FiniteElement *fe = fespace->GetFE(0);
+   const IntegrationRule *ir =
+      IntRule ? IntRule : &qspace->GetElementIntRule(0);
+   const DofToQuad &maps = fe->GetDofToQuad(*ir, DofToQuad::TENSOR);
+   const int D1D = maps.ndof;
+   const int Q1D = maps.nqpt;
+   const double *B = maps.B.Read();
+   const double *X = e_vec.Read();
+   double *Y = q_val.Write();
+
+   constexpr QVectorLayout L = QVectorLayout::byVDIM;
+
+   const int id = (dim<<12) | (vdim<<8) | (D1D<<4) | Q1D;
+
+   switch (id)
+   {
+      case 0x2124: return Eval2D<L,1,2,4,8>(NE,B,X,Y);
+      case 0x2136: return Eval2D<L,1,3,6,4>(NE,B,X,Y);
+      case 0x2148: return Eval2D<L,1,4,8,2>(NE,B,X,Y);
+
+      case 0x2224: return Eval2D<L,2,2,4,8>(NE,B,X,Y);
+      case 0x2234: return Eval2D<L,2,3,4,8>(NE,B,X,Y);
+      case 0x2236: return Eval2D<L,2,3,6,4>(NE,B,X,Y);
+      case 0x2248: return Eval2D<L,2,4,8,2>(NE,B,X,Y);
+
+      case 0x3124: return Eval3D<L,1,2,4>(NE,B,X,Y);
+      case 0x3136: return Eval3D<L,1,3,6>(NE,B,X,Y);
+      case 0x3148: return Eval3D<L,1,4,8>(NE,B,X,Y);
+
+      case 0x3324: return Eval3D<L,3,2,4>(NE,B,X,Y);
+      case 0x3336: return Eval3D<L,3,3,6>(NE,B,X,Y);
+      case 0x3348: return Eval3D<L,3,4,8>(NE,B,X,Y);
+
+      default:
+      {
+         constexpr int MD1 = 8;
+         constexpr int MQ1 = 8;
+         MFEM_VERIFY(D1D <= MD1, "Orders higher than " << MD1-1
+                     << " are not supported!");
+         MFEM_VERIFY(Q1D <= MQ1, "Quadrature rules with more than "
+                     << MQ1 << " 1D points are not supported!");
+         if (dim == 2) { Eval2D<L,0,0,0,0,MD1,MQ1>(NE,B,X,Y,vdim,D1D,Q1D); }
+         if (dim == 3) { Eval3D<L,0,0,0,MD1,MQ1>(NE,B,X,Y,vdim,D1D,Q1D); }
+         return;
+      }
+   }
+   mfem::out << "Unknown kernel 0x" << std::hex << id << std::endl;
+   MFEM_ABORT("Kernel not supported yet");
+}
+
+} // namespace mfem

fem/quadinterpolator_face.cpp

@@ -495,8 +495,9 @@ void FaceQuadratureInterpolator::Mult(
    }
 }
 
-void FaceQuadratureInterpolator::Values(
-   const Vector &e_vec, Vector &q_val) const
+
+void FaceQuadratureInterpolator::Values(const Vector &e_vec,
+                                        Vector &q_val) const
 {
    Vector q_der, q_det, q_nor;
    Mult(e_vec, VALUES, q_val, q_der, q_det, q_nor);

fem/quadinterpolator_grad.hpp

@@ -0,0 +1,282 @@
+// Copyright (c) 2010-2020, 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 "quadinterpolator.hpp"
+#include "../general/forall.hpp"
+#include "../linalg/dtensor.hpp"
+#include "../linalg/kernels.hpp"
+
+namespace mfem
+{
+
+template<QVectorLayout Q_LAYOUT,
+         int T_VDIM = 0, int T_D1D = 0, int T_Q1D = 0,
+         int T_NBZ = 1, int MAX_D1D = 0, int MAX_Q1D = 0>
+static void Grad2D(const int NE,
+                   const double *b_,
+                   const double *g_,
+                   const double *x_,
+                   double *y_,
+                   const int vdim = 0,
+                   const int d1d = 0,
+                   const int q1d = 0)
+{
+   const int D1D = T_D1D ? T_D1D : d1d;
+   const int Q1D = T_Q1D ? T_Q1D : q1d;
+   constexpr int NBZ = T_NBZ ? T_NBZ : 1;
+   const int VDIM = T_VDIM ? T_VDIM : vdim;
+
+   const auto b = Reshape(b_, Q1D, D1D);
+   const auto g = Reshape(g_, Q1D, D1D);
+   const auto x = Reshape(x_, D1D, D1D, VDIM, NE);
+   auto y = Q_LAYOUT ==QVectorLayout:: byNODES ?
+            Reshape(y_, Q1D, Q1D, VDIM, 2, NE):
+            Reshape(y_, VDIM, 2, Q1D, Q1D, NE);
+
+   MFEM_FORALL_2D(e, NE, Q1D, Q1D, NBZ,
+   {
+      const int D1D = T_D1D ? T_D1D : d1d;
+      const int Q1D = T_Q1D ? T_Q1D : q1d;
+      const int VDIM = T_VDIM ? T_VDIM : vdim;
+      constexpr int MQ1 = T_Q1D ? T_Q1D : MAX_Q1D;
+      constexpr int MD1 = T_D1D ? T_D1D : MAX_D1D;
+      constexpr int NBZ = T_NBZ ? T_NBZ : 1;
+      const int tidz = MFEM_THREAD_ID(z);
+      MFEM_SHARED double s_B[MQ1*MD1];
+      MFEM_SHARED double s_G[MQ1*MD1];
+      DeviceTensor<2,double> B(s_B, Q1D, D1D);
+      DeviceTensor<2,double> G(s_G, Q1D, D1D);
+
+      MFEM_SHARED double s_X[NBZ][MD1*MD1];
+      DeviceTensor<2,double> X((double*)(s_X+tidz), MD1, MD1);
+
+      MFEM_SHARED double s_DQ[2][NBZ][MD1*MQ1];
+      DeviceTensor<2,double> DQ0((double*)(s_DQ[0]+tidz), MD1, MQ1);
+      DeviceTensor<2,double> DQ1((double*)(s_DQ[1]+tidz), MD1, MQ1);
+
+      if (tidz == 0)
+      {
+         MFEM_FOREACH_THREAD(d,y,D1D)
+         {
+            MFEM_FOREACH_THREAD(q,x,Q1D)
+            {
+               B(q,d) = b(q,d);
+               G(q,d) = g(q,d);
+            }
+         }
+      }
+      MFEM_SYNC_THREAD;
+
+      for (int c = 0; c < VDIM; ++c)
+      {
+         MFEM_FOREACH_THREAD(dx,x,D1D)
+         {
+            MFEM_FOREACH_THREAD(dy,y,D1D)
+            {
+               X(dx,dy) = x(dx,dy,c,e);
+            }
+         }
+         MFEM_SYNC_THREAD;
+         MFEM_FOREACH_THREAD(dy,y,D1D)
+         {
+            MFEM_FOREACH_THREAD(qx,x,Q1D)
+            {
+               double u = 0.0;
+               double v = 0.0;
+               for (int dx = 0; dx < D1D; ++dx)
+               {
+                  const double input = X(dx,dy);
+                  u += input * B(qx,dx);
+                  v += input * G(qx,dx);
+               }
+               DQ0(dy,qx) = u;
+               DQ1(dy,qx) = v;
+            }
+         }
+         MFEM_SYNC_THREAD;
+         MFEM_FOREACH_THREAD(qy,y,Q1D)
+         {
+            MFEM_FOREACH_THREAD(qx,x,Q1D)
+            {
+               double u = 0.0;
+               double v = 0.0;
+               for (int dy = 0; dy < D1D; ++dy)
+               {
+                  u += DQ1(dy,qx) * B(qy,dy);
+                  v += DQ0(dy,qx) * G(qy,dy);
+               }
+               if (Q_LAYOUT == QVectorLayout::byNODES)
+               {
+                  y(qx,qy,c,0,e) = u;
+                  y(qx,qy,c,1,e) = v;
+               }
+               if (Q_LAYOUT == QVectorLayout::byVDIM)
+               {
+                  y(c,0,qx,qy,e) = u;
+                  y(c,1,qx,qy,e) = v;
+               }
+            }
+         }
+         MFEM_SYNC_THREAD;
+      }
+   });
+}
+
+template<QVectorLayout Q_LAYOUT,
+         int T_VDIM = 0, int T_D1D = 0, int T_Q1D = 0,
+         int MAX_D1D = 0, int MAX_Q1D = 0>
+static void Grad3D(const int NE,
+                   const double *b_,
+                   const double *g_,
+                   const double *x_,
+                   double *y_,
+                   const int vdim = 0,
+                   const int d1d = 0,
+                   const int q1d = 0)
+{
+   const int D1D = T_D1D ? T_D1D : d1d;
+   const int Q1D = T_Q1D ? T_Q1D : q1d;
+   const int VDIM = T_VDIM ? T_VDIM : vdim;
+
+   const auto b = Reshape(b_, Q1D, D1D);
+   const auto g = Reshape(g_, Q1D, D1D);
+   const auto x = Reshape(x_, D1D, D1D, D1D, VDIM, NE);
+   auto y = Q_LAYOUT ==QVectorLayout:: byNODES ?
+            Reshape(y_, Q1D, Q1D, Q1D, VDIM, 3, NE):
+            Reshape(y_, VDIM, 3, Q1D, Q1D, Q1D, NE);
+
+   MFEM_FORALL_3D(e, NE, Q1D, Q1D, Q1D,
+   {
+      const int D1D = T_D1D ? T_D1D : d1d;
+      const int Q1D = T_Q1D ? T_Q1D : q1d;
+      const int VDIM = T_VDIM ? T_VDIM : vdim;
+      constexpr int MQ1 = T_Q1D ? T_Q1D : MAX_Q1D;
+      constexpr int MD1 = T_D1D ? T_D1D : MAX_D1D;
+
+      const int tidz = MFEM_THREAD_ID(z);
+      MFEM_SHARED double s_B[MQ1*MD1];
+      MFEM_SHARED double s_G[MQ1*MD1];
+      DeviceTensor<2,double> B(s_B, Q1D, D1D);
+      DeviceTensor<2,double> G(s_G, Q1D, D1D);
+
+      MFEM_SHARED double sm0[3][MQ1*MQ1*MQ1];
+      MFEM_SHARED double sm1[3][MQ1*MQ1*MQ1];
+      DeviceTensor<3,double> X((double*)(sm0+2), MD1, MD1, MD1);
+      DeviceTensor<3,double> DDQ0((double*)(sm0+0), MD1, MD1, MQ1);
+      DeviceTensor<3,double> DDQ1((double*)(sm0+1), MD1, MD1, MQ1);
+      DeviceTensor<3,double> DQQ0((double*)(sm1+0), MD1, MQ1, MQ1);
+      DeviceTensor<3,double> DQQ1((double*)(sm1+1), MD1, MQ1, MQ1);
+      DeviceTensor<3,double> DQQ2((double*)(sm1+2), MD1, MQ1, MQ1);
+
+      if (tidz == 0)
+      {
+         MFEM_FOREACH_THREAD(d,y,D1D)
+         {
+            MFEM_FOREACH_THREAD(q,x,Q1D)
+            {
+               B(q,d) = b(q,d);
+               G(q,d) = g(q,d);
+            }
+         }
+      }
+      MFEM_SYNC_THREAD;
+
+      for (int c = 0; c < VDIM; ++c)
+      {
+         MFEM_FOREACH_THREAD(dx,x,D1D)
+         {
+            MFEM_FOREACH_THREAD(dy,y,D1D)
+            {
+               MFEM_FOREACH_THREAD(dz,z,D1D)
+               {
+                  X(dx,dy,dz) = x(dx,dy,dz,c,e);
+               }
+            }
+         }
+         MFEM_SYNC_THREAD;
+         MFEM_FOREACH_THREAD(dz,z,D1D)
+         {
+            MFEM_FOREACH_THREAD(dy,y,D1D)
+            {
+               MFEM_FOREACH_THREAD(qx,x,Q1D)
+               {
+                  double u = 0.0;
+                  double v = 0.0;
+                  for (int dx = 0; dx < D1D; ++dx)
+                  {
+                     const double input = X(dx,dy,dz);
+                     u += input * B(qx,dx);
+                     v += input * G(qx,dx);
+                  }
+                  DDQ0(dz,dy,qx) = u;
+                  DDQ1(dz,dy,qx) = v;
+               }
+            }
+         }
+         MFEM_SYNC_THREAD;
+         MFEM_FOREACH_THREAD(dz,z,D1D)
+         {
+            MFEM_FOREACH_THREAD(qy,y,Q1D)
+            {
+               MFEM_FOREACH_THREAD(qx,x,Q1D)
+               {
+                  double u = 0.0;
+                  double v = 0.0;
+                  double w = 0.0;
+                  for (int dy = 0; dy < D1D; ++dy)
+                  {
+                     u += DDQ1(dz,dy,qx) * B(qy,dy);
+                     v += DDQ0(dz,dy,qx) * G(qy,dy);
+                     w += DDQ0(dz,dy,qx) * B(qy,dy);
+                  }
+                  DQQ0(dz,qy,qx) = u;
+                  DQQ1(dz,qy,qx) = v;
+                  DQQ2(dz,qy,qx) = w;
+               }
+            }
+         }
+         MFEM_SYNC_THREAD;
+         MFEM_FOREACH_THREAD(qz,z,Q1D)
+         {
+            MFEM_FOREACH_THREAD(qy,y,Q1D)
+            {
+               MFEM_FOREACH_THREAD(qx,x,Q1D)
+               {
+                  double u = 0.0;
+                  double v = 0.0;
+                  double w = 0.0;
+                  for (int dz = 0; dz < D1D; ++dz)
+                  {
+                     u += DQQ0(dz,qy,qx) * B(qz,dz);
+                     v += DQQ1(dz,qy,qx) * B(qz,dz);
+                     w += DQQ2(dz,qy,qx) * G(qz,dz);
+                  }
+                  if (Q_LAYOUT == QVectorLayout::byNODES)
+                  {
+                     y(qx,qy,qz,c,0,e) = u;
+                     y(qx,qy,qz,c,1,e) = v;
+                     y(qx,qy,qz,c,2,e) = w;
+                  }
+                  if (Q_LAYOUT == QVectorLayout::byVDIM)
+                  {
+                     y(c,0,qx,qy,qz,e) = u;
+                     y(c,1,qx,qy,qz,e) = v;
+                     y(c,2,qx,qy,qz,e) = w;
+                  }
+               }
+            }
+         }
+         MFEM_SYNC_THREAD;
+      }
+   });
+}
+
+} // namespace mfem

fem/quadinterpolator_grad_by_nodes.cpp

@@ -0,0 +1,109 @@
+// Copyright (c) 2010-2020, 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 "quadinterpolator.hpp"
+#include "quadinterpolator_grad.hpp"
+
+namespace mfem
+{
+
+template<>
+void QuadratureInterpolator::Derivatives<QVectorLayout::byNODES>(
+   const Vector &e_vec, Vector &q_der) const
+{
+   const int NE = fespace->GetNE();
+   if (NE == 0) { return; }
+
+   const int vdim = fespace->GetVDim();
+   const int dim = fespace->GetMesh()->Dimension();
+   const FiniteElement *fe = fespace->GetFE(0);
+   const IntegrationRule *ir =
+      IntRule ? IntRule : &qspace->GetElementIntRule(0);
+   const DofToQuad &maps = fe->GetDofToQuad(*ir, DofToQuad::TENSOR);
+   const int D1D = maps.ndof;
+   const int Q1D = maps.nqpt;
+   const double *B = maps.B.Read();
+   const double *G = maps.G.Read();
+   const double *X = e_vec.Read();
+   double *Y = q_der.Write();
+
+   constexpr QVectorLayout L = QVectorLayout::byNODES;
+
+   const int id = (dim<<12) | (vdim<<8) | (D1D<<4) | Q1D;
+
+   switch (id)
+   {
+      case 0x2133: return Grad2D<L,1,3,3,16>(NE,B,G,X,Y);
+      case 0x2134: return Grad2D<L,1,3,4,16>(NE,B,G,X,Y);
+      case 0x2143: return Grad2D<L,1,4,3,16>(NE,B,G,X,Y);
+      case 0x2144: return Grad2D<L,1,4,4,16>(NE,B,G,X,Y);
+
+      case 0x2222: return Grad2D<L,2,2,2,16>(NE,B,G,X,Y);
+      case 0x2223: return Grad2D<L,2,2,3,8>(NE,B,G,X,Y);
+      case 0x2224: return Grad2D<L,2,2,4,4>(NE,B,G,X,Y);
+      case 0x2225: return Grad2D<L,2,2,5,4>(NE,B,G,X,Y);
+      case 0x2226: return Grad2D<L,2,2,6,2>(NE,B,G,X,Y);
+
+      case 0x2233: return Grad2D<L,2,3,3,2>(NE,B,G,X,Y);
+      case 0x2234: return Grad2D<L,2,3,4,4>(NE,B,G,X,Y);
+      case 0x2243: return Grad2D<L,2,4,3,4>(NE,B,G,X,Y);
+      case 0x2236: return Grad2D<L,2,3,6,2>(NE,B,G,X,Y);
+
+      case 0x2244: return Grad2D<L,2,4,4,2>(NE,B,G,X,Y);
+      case 0x2245: return Grad2D<L,2,4,5,2>(NE,B,G,X,Y);
+      case 0x2246: return Grad2D<L,2,4,6,2>(NE,B,G,X,Y);
+      case 0x2247: return Grad2D<L,2,4,7,2>(NE,B,G,X,Y);
+
+      case 0x2256: return Grad2D<L,2,5,6,2>(NE,B,G,X,Y);
+
+      case 0x3124: return Grad3D<L,1,2,4>(NE,B,G,X,Y);
+      case 0x3133: return Grad3D<L,1,3,3>(NE,B,G,X,Y);
+      case 0x3134: return Grad3D<L,1,3,4>(NE,B,G,X,Y);
+      case 0x3136: return Grad3D<L,1,3,6>(NE,B,G,X,Y);
+      case 0x3144: return Grad3D<L,1,4,4>(NE,B,G,X,Y);
+      case 0x3148: return Grad3D<L,1,4,8>(NE,B,G,X,Y);
+
+      case 0x3323: return Grad3D<L,3,2,3>(NE,B,G,X,Y);
+      case 0x3324: return Grad3D<L,3,2,4>(NE,B,G,X,Y);
+      case 0x3325: return Grad3D<L,3,2,5>(NE,B,G,X,Y);
+      case 0x3326: return Grad3D<L,3,2,6>(NE,B,G,X,Y);
+
+      case 0x3333: return Grad3D<L,3,3,3>(NE,B,G,X,Y);
+      case 0x3334: return Grad3D<L,3,3,4>(NE,B,G,X,Y);
+      case 0x3335: return Grad3D<L,3,3,5>(NE,B,G,X,Y);
+      case 0x3336: return Grad3D<L,3,3,6>(NE,B,G,X,Y);
+      case 0x3344: return Grad3D<L,3,4,4>(NE,B,G,X,Y);
+      case 0x3346: return Grad3D<L,3,4,6>(NE,B,G,X,Y);
+      case 0x3347: return Grad3D<L,3,4,7>(NE,B,G,X,Y);
+      case 0x3348: return Grad3D<L,3,4,8>(NE,B,G,X,Y);
+      default:
+      {
+         constexpr int MD1 = 8;
+         constexpr int MQ1 = 8;
+         MFEM_VERIFY(D1D <= MD1, "Orders higher than " << MD1-1
+                     << " are not supported!");
+         MFEM_VERIFY(Q1D <= MQ1, "Quadrature rules with more than "
+                     << MQ1 << " 1D points are not supported!");
+         if (dim == 2)
+         {
+            return Grad2D<L,0,0,0,0,MD1,MQ1>(NE,B,G,X,Y,vdim,D1D,Q1D);
+         }
+         if (dim == 3)
+         {
+            return Grad3D<L,0,0,0,MD1,MQ1>(NE,B,G,X,Y,vdim,D1D,Q1D);
+         }
+      }
+   }
+   mfem::out << "Unknown kernel 0x" << std::hex << id << std::endl;
+   MFEM_ABORT("Kernel not supported yet");
+}
+
+} // namespace mfem

fem/quadinterpolator_grad_by_vdim.cpp

@@ -0,0 +1,76 @@
+// Copyright (c) 2010-2020, 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 "quadinterpolator.hpp"
+#include "quadinterpolator_grad.hpp"
+
+namespace mfem
+{
+
+template<>
+void QuadratureInterpolator::Derivatives<QVectorLayout::byVDIM>(
+   const Vector &e_vec, Vector &q_der) const
+{
+   const int NE = fespace->GetNE();
+   if (NE == 0) { return; }
+
+   const int vdim = fespace->GetVDim();
+   const int dim = fespace->GetMesh()->Dimension();
+   const FiniteElement *fe = fespace->GetFE(0);
+   const IntegrationRule *ir =
+      IntRule ? IntRule : &qspace->GetElementIntRule(0);
+   const DofToQuad &maps = fe->GetDofToQuad(*ir, DofToQuad::TENSOR);
+   const int D1D = maps.ndof;
+   const int Q1D = maps.nqpt;
+   const double *B = maps.B.Read();
+   const double *G = maps.G.Read();
+   const double *X = e_vec.Read();
+   double *Y = q_der.Write();
+
+   constexpr QVectorLayout L = QVectorLayout::byVDIM;
+
+   const int id = (dim<<12) | (vdim<<8) | (D1D<<4) | Q1D;
+
+   switch (id)
+   {
+      case 0x2134: return Grad2D<L,1,3,4,8>(NE,B,G,X,Y);
+      case 0x2146: return Grad2D<L,1,4,6,4>(NE,B,G,X,Y);
+      case 0x2158: return Grad2D<L,1,5,8,2>(NE,B,G,X,Y);
+
+      case 0x2234: return Grad2D<L,2,3,4,8>(NE,B,G,X,Y);
+      case 0x2246: return Grad2D<L,2,4,6,4>(NE,B,G,X,Y);
+      case 0x2258: return Grad2D<L,2,5,8,2>(NE,B,G,X,Y);
+
+      case 0x3134: return Grad3D<L,1,3,4>(NE,B,G,X,Y);
+      case 0x3146: return Grad3D<L,1,4,6>(NE,B,G,X,Y);
+      case 0x3158: return Grad3D<L,1,5,8>(NE,B,G,X,Y);
+
+      case 0x3334: return Grad3D<L,3,3,4>(NE,B,G,X,Y);
+      case 0x3346: return Grad3D<L,3,4,6>(NE,B,G,X,Y);
+      case 0x3358: return Grad3D<L,3,5,8>(NE,B,G,X,Y);
+      default:
+      {
+         constexpr int MD1 = 8;
+         constexpr int MQ1 = 8;
+         MFEM_VERIFY(D1D <= MD1, "Orders higher than " << MD1-1
+                     << " are not supported!");
+         MFEM_VERIFY(Q1D <= MQ1, "Quadrature rules with more than "
+                     << MQ1 << " 1D points are not supported!");
+         if (dim == 2) { Grad2D<L,0,0,0,0,MD1,MQ1>(NE,B,G,X,Y,vdim,D1D,Q1D); }
+         if (dim == 3) { Grad3D<L,0,0,0,MD1,MQ1>(NE,B,G,X,Y,vdim,D1D,Q1D); }
+         return;
+      }
+   }
+   mfem::out << "Unknown kernel 0x" << std::hex << id << std::endl;
+   MFEM_ABORT("Kernel not supported yet");
+}
+
+} // namespace mfem