CS infrastructure components for HPC applications
Please download the Axom-v0.9.0.tar.gz tarball below, which includes all of the Axom submodules as well.
Quadrilateral primitivecompute_bounding_box() operator for computing the bounding
box of a Quadrilateral
clip() operator for clipping a tetrahedron against the
half-space defined by a planecheckAndFixOrientation() function to primal::Tetrahedron
that swaps the order of vertices if the signed volume of the Tetrahedron is
negative, resulting in the signed volume becoming positive.FlatMap, a generic key-value store which aims for drop-in compatibility
with std::unordered_map, but utilizes an open-addressing design.Array::push_back() and Array::emplace_back().setTetPredFromBoundingBox() and setTetPred() functions to
quest::ProEReader and PProEReader that set a tet predicate, allowing
user code to read in a subset of a Pro/E ASCII tetrahedron mesh file.MarchingCubes has optimizations to improve GPU performance, particularly for
repeated computations. The constructor has changed and a new setMesh method
is added to set (or change) the mesh. New accessors present output data
without moving them from device to host. These accessors are an interim
solution and likely to be updated in the future.DistributedClosestPoint outputs are now controlled by the setOutput method.MarchingCubes allows user to select the underlying data-parallel implementation
fullParallel works best on GPUs.hybridParallel reduces the amount of data processed and works best with
MarchingCubesRuntimePolicy::seq.byPolicy (the default) selects the implementation based on the runtime policy.MarchingCubes and DistributedClosestPoint classes identify domains by their
state/domain_id parameters if provided, or the local iteration index if not.MarchingCubes and DistributedClosestPoint classes changed from requiring the Blueprint
coordset name to requiring the Blueprint topology name. The changed interface methods are:
DistributedClosestPoint::setObjectMesh
DistributedClosestPoint::computeClosestPoints
MarchingCubes::MarchingCubes
MarchingCubesSingleDomain::MarchingCubesSingleDomain
Polyhedron::volume() function changed from returning a signed
volume to an unsigned volume. The added Polyhedron::signedVolume() function
returns the signed volume.intersection_volume() operators changed from returning a signed
volume to an unsigned volume.BoundingBox::contains(BoundingBox) now returns true when the input is emptyC++20 standard: popCount() -> popcount(),
trailingZeros() -> countr_zero() and leadingZeros() -> countl_zero()
axom::utilities::swapEndian() -> byteswap() to conform to C++23 standardSamplingShaper now properly handles material names containing underscoresSamplingShaper can now be used with an mfem that is configured for (GPU) devicesPolygon area computation in 3D previously only worked when the polygon was aligned with the XY-plane. It now works for arbitrary polygons.vcpkg usage for automated Windows builds of our TPLs to its 2023.12.12 release
primal::clip(Triangle, BoundingBox)
std::locale error when when compiling src/axom/core/utilities/System.cpp using nvcccstdint for higher gcc version support (e.g. gcc-13)axom::Array, quest::Shaping and sidre::MFEMSidreDataCollection
Please download the Axom-v0.8.1.tar.gz tarball below, which includes all of the Axom submodules as well.
Please download the Axom-v0.8.0.tar.gz tarball below, which includes all of the Axom submodules as well.
axom::Array to conform more closely with the std::vector interface:
Array::front(): returns a reference to the first elementArray::back(): returns a reference to the last elementArray::resize(size, T value): resizes the array, and sets any new elements to value.ArrayView::empty() method to return whether the view is empty or not.area() function to primal::Polygon
ArrayViewIndirection indirection policy with slam::Map and slam::BivariateMap
const_iterator support to slam::BivariateMap and slam::SubMap
Hexahedron primitiveclip() operator for computing the intersection of a Hexahedron and another Tetrahedron as a Polyhedron
intersection_volume() operator for computing the volume of intersection between a primitive and a Tetrahedron
primal::Polyhedron::from_primitive() operator that returns a Polyhedron object from a given primitive.DataStore::getBufferInfo() and Group::getDataInfo() methods that insert information into a Conduit Node about buffers in a DataStore object or data in a Group subtree. The information can be accessed from the Node by the caller from specifically named fields in the Node.quest::ProEReader for reading in Pro/E tetrahedral meshesquest::IntersectionShaper class can now use a percent error to determine whether the revolved volume for a shape is sufficiently accurate or whether the shape must be further refined. This new dynamic method of shaping complements the existing segment-based curve refinement method and it is activated using Shaper::setRefinementType() and by calling Shaper::setPercentError() to set a refinement error percentage.MultiMat::setAllocatorID() or MultiMat::setFieldAllocatorID() can be called to change the memory space in which a field is allocated.MultiMat::addExternalField() function to support fields where the memory is managed externally. Fields which are externally-managed cannot be transposed between sparse and dense layouts, or moved between allocator spaces.MultiMat::removeField() function to remove fields from the Multimat instance.MultiMat::setCellMatRel() that supports setting a multi-material relation in a compressed sparse-row (CSR) representation.SamplingShaper before processing Klee inputslam::MappedVariableCardinality policy to accelerate mapping flat indices back to first-set indices when used in a StaticRelation
ArrayView(data, shape, strides) constructor to support column-major and custom striding layouts.ArrayView::subspan() overload for multi-dimensional subspansaxom::utilities::insertionSort() method.IntersectionShaper
axom::utilities::locale utility function to guard against platforms that do not have the requested locales via the std::locale function. If the system does not have the requested locale (e.g. en_US.UTF8), it returns the user's default locale.sidre_layout_json and conduit_layout_json to provide output of DataStore layout in a user-readable format that excludes the numerical arrays held by Views and Buffers.mint::mesh::UnstructuredMesh constructors, affecting capacity. A missing factor was added. If you worked around this by adding the factor yourself, you may want to undo that work-around.c2c version 1.8.0axom_). This change requires no change to downstream CMake users who import our targets. The exported CMake target axom includes all components, but users who do not import our targets will need to create the link line themselves. The following replacement can be used: -laxom -> -laxom_quest -laxom_multimat -laxom_slam -laxom_mint -laxom_klee -laxom_inlet -laxom_sidre -laxom_slic -laxom_lumberjack -laxom_core. If you only need a subset of the components, you can now use those targets directly, ie. axom::inlet.IntersectionShaper now implements material replacement rules.axom::Array move constructors are now noexcept.cli11, fmt, sol, and sparsehash, have been prefixed with axom:: to guard against conflicts.DistributedClosestPoint query now supports any blueprint-valid mesh format, including multidomain. Domain underloading and overloading can be expressed using multidomain format. Closest points are identified by cp_rank, cp_domain_index, and cp_index. The new cp_domain_index specifies the domain containing the closest point.DistributedClosestPoint interfacing variable names closest_point and min_distance have been changed to cp_coords and cp_distance, respectively, to match the naming convention of other interfacing variables.vcpkg ports for RAJA, Umpire with optional OpenMP feature for automated Windows build.ArrayView::subspan to prevent accessing invalid memory.vcpkg port for lua as optional dependency on Windows.PointInCell query to control the Newton solve from physical to reference space for a given element.IndirectionPolicy::data() now returns a reference to the underlying buffer. Rebinding an indirection to a new buffer is now achieved through IndirectionPolicy::ptr(), which returns a mutable pointer to the buffer.Shaper::applyTransforms() is no longer a public method.MultiMat::addField() and MultiMat::setVolfracField() API now use axom::ArrayView to accept data.MultiMat::makeOtherRelation() now runs on the GPU with an appropriately-set allocator ID.MultiMat::setCellMatRel(counts, indices) now runs on the GPU, and accepts GPU-side data.ArrayView::spacing() to ArrayView::minStride().path field when its format is "none"primal intersection operators against planes and triangles before using the geometry.lua dependencymfem dependency in device builds when mfem is configured with caliper
Array::resize(size, value) for Array<bool>
hasChildView() and hasChildGroup() changed to return false when Group holds items in list format. This fixes an IOManager issue causing failures for multi-domain meshes when writing Blueprint index file.importConduitTreeExternal() to handle lists with unnamed members the same as importConduitTree().flush() for parallel output log streams.src/axom/core/Types.hpp are deprecated because c++11 supports their equivalents.Please download the Axom-v0.7.0.tar.gz tarball below, which includes all of the Axom submodules as well.
view() method to axom::Array class to simplify creation of a corresponding axom::ArrayView
spin::ImplicitGrid.
The following functions run with the user-specified execution space (specified as a template argument
on ImplicitGrid):
ImplicitGrid::insert(nelems, bboxes): insert a batch of bounding boxes into the implicit gridImplicitGrid::getCandidatesAsArray(nquery, queryObjs, ...): query the implicit grid for a batch of
query objects, and generate a CSR-format array for the candidates.
In addition, ImplicitGrid::getQueryObject() returns an object that may be used within a GPU kernel
to query the implicit grid.ImplicitGrid
const versions of begin() and end() for Array and ArrayView
axom::Array with explicitly specified
memory spaceaxom::Array that uses both uninitialized data (ArrayOptions::Uninitialized)
and custom allocatorsoperator<()) to StackArray. This allows it to be used as a key for std::map
popCount(), trailingZeros() and leadingZeros()
slam::DynamicSet
ImpicitGrid::getCandidates() from a grid cell of the lattice.
This makes it easier to iterate over the bins of the spatial index.axom::Array<T>/ArrayView<T> iterators, to allow passing iterator
pairs to standard library functionsaxom::Array<T> allocated in device-only memory.axom::Array<T> using
operator[] and operator()
hip compiler. Support for running device
kernels with hip will be added in the future.spin::UniformGrid.
The following functions run with a user-specified execution space (specified as a template argument
on UniformGrid):
UniformGrid::initialize(): creates/re-creates a uniform grid with an array of objects and their
corresponding bounding boxesUniformGrid::getCandidatesAsArray(): query the uniform grid for objects that share a grid
cell with the query bounding box
In addition, UniformGrid::getQueryObject() returns an object that may be used within a GPU kernel
to query the uniform grid.UniformGrid. Two policies are provided:
DynamicGridStorage stores the bins as an array of arrays (default)FlatGridStorage stores the bins as a flat array of elements, where each bin is a slice of the
arrayaxom::Array
axom::Array::resize(ArrayOptions::Uninitialized, dims) to support resizes
without constructing or initializing new elementsScatteredInterpolation query to quest, which enables interpolating scalar fields at arbitrary
points from a given 2D or 3D point mesh in the Mesh Blueprint format. The current implementation
generates a Deluanay complex over the points and performs linear interpolation over the
triangle/tetrahedron at the query points.SLIC_ERROR_ROOT(msg) and SLIC_ERROR_ROOT_IF(EXP, msg). This can be set via
slic::initialize(bool is_root = true) or slic::setIsRoot().Views and Groups of a sidre::Group.
These can be accessed via the range-for syntax as for(auto& view: grp.views()){...},
Or using the iterator syntax as
for(auto& it = grp.views().begin(), itEnd = grp.views().end(); it ! itEnd ; ++it) {...}, and similarly for the groups of a group.Attributes and Bufferss of a sidre::DataStore,
with a similar syntax, e.g. for(auto& buf : datastore.buffers()){...}.ImplicitGrid::getCandidatesAsArray() to accept query points/bounding boxes
as an axom::ArrayView.primal::closest_point(point,sphere) overload to find the closest point on a sphere to a given pointSignedDistance query to return the closest point on the surface
to the query point and the surface normal at that point. Also exposes this functionality
in quest's signed_distance C API.lerp) of two numbersCurvedPolygon class to primal representing a polygon with BezierCurves as edgesCurvedPolygon
BezierCurve and CurvedPolygon objectsArrayViewIndirection storage policy to Slamslam::DynamicVariableRelation
multimat, to simplify the handing of multi-material meshes and
fields.Polygon and CurvedPolygon objects.constants.hpp to primal to track geometric constants. Initially includes
a value for PRIMAL_TINY, a small constant that can be added to
denominators to avoid division by zero.DistributedClosestPoint query now supports "domain underloading" -- ranks that are passed in can
have empty object meshes and/or empty query meshesclip() operator for computing the intersection of a Tetrahedron and another Tetrahedron as a Polyhedron
slic::outputLocalMessages() to output messages from the current rank to the console for MPI-enabled LogStreams.BLT_CXX_STD.axom::Array now default-initializes its data by default. To create an Array with uninitialized
elements, pass an axom::ArrayOptions::Uninitialized as the first constructor argument.axom::ArrayView<const T> can now be created from a const Array<T>
ExecSpace template parameter to spin::ImplicitGrid.
ExecSpace is now the second template parameter (out of three) and defaults to axom::SEQ_EXEC.DataStore, MFEMSidreDataCollection will now save only
its domain and global groupsinlet::Field fails a range or valid value constraint, the provided value and
corresponding range/set of valid values are now included in the error messageprimal::Plane, primal::Sphere, primal::orientation()
and primal::in_sphere()
primal::in_sphere() -- the computations are now based on (D+1)x(D+1) determinants
instead of (D+2)x(D+2) determinants for D-dimensional spheresprimal::Segment, primal::Triangle and primal::Tetrahedron
primal::Sphere to use more of primal, e.g. uses primal::Point instead of T* to represent pointscircumsphere() functions to primal::Triangle and primal::Tetrahedron to return the Sphere
that circumscribes the triangle/tetrahedron's verticesprimal::Vector from a primal::Point to yield a new primal::Point
quest::findTriMeshIntersections*() implementations for BVH and ImplicitGrid
BVH::find*() batch functions now return the total number of candidate intersections foundaxom::Array<T> to be iterated over with begin()/end()
AXOM_VERSION_EXTRA in favor of axom::gitSHA() and adding the SHA to axom::getVersion() and
axom::about()
ArrayOps, to avoid generating unnecessary copies in
fill/destroy operations on otherwise trivially-copyable/destructible types.axom::Array now consistently propagates the allocator ID on copy, move, and swap operations when possible.
This is a breaking change; copy-construction of a dynamic array from a device array will no longer automatically
move the array to host memory, and will instead maintain the same allocator ID as the source array.BVH::TraverserType::traverse_tree() now supports passing in arbitrary query objects
for BVH traversal.inlet::LuaReader::solState() to be a protected function that now returns a std::shared_ptr<axom::sol::state>.
This is an advanced feature that could cause users to break an input file state after verification. This also allows us
to not expose axom/sol.hpp to all users of Inlet. This greatly reduces compile times. Using this feature requires
both a derived class and including axom/sol.hpp in the user code.createView of
axom::sidre::Group which accept int ndims, IndexType *shape
arguments to be createViewWithShape or createViewWithShapeAndAllocate.std-compliant
implementationsidre/core/sidre.hpp.
We recommend using the automatically generated header file axom/sidre.hpp to include
sidre functionality.sidre::ItemCollection base class that were not common to all derived classes
and added a new derived class sidre::IndexedCollection
BVH::findPoints/Rays/BoundingBoxes() candidate search methods now accept an axom::ArrayView<IndexType>
for the offsets and counts output arrays, and return candidates as an axom::Array<IndexType>.primal::Polygon::centroid() to primal::Polygon::vertexMean() because it was not actually computing the centroid.axom:sidre:IndexType is now the same type as
axom:IndexType. Before, Sidre always used int64_t. Now it
respects the define AXOM_USE_64BIT_INDEXTYPE.ArrayIndirection to CArrayIndirection
and the axom::Array-based indirection policy was renamed from CoreArrayIndirection to ArrayIndirection.primal::detail::intersect_ray now correctly identifies intersections between collinear Segment and Ray objects.axom::Arraysmfem::FiniteElementSpaces in sidre::MFEMSidreDataCollection
quest::PointInCell when not using RAJAaxom::Array<T> for non-trivial types T which allocate memoryaxom::ArrayList for axom configurations without Umpire to fix a compiler error (XL compiler)quest::Delaunay to avoid "multiply-defined" linker errorsaxom::Array<T> fill operations on uninitialized memoryaxom::Array<T>::resize(new_size) with new_size < curr_size
quest::SignedDistance when closest point is along an edge
with a sharp dihedral angle and the adjacent triangles have significantly different areasaxom::Array<T> behavior on copy-construction when T is a non-trivial typeusing statement in SidreDataTypesIds.h with typedef
since C syntax is required in this file.sidre::Array<T> construction where the size is set to the underlying buffer
capacity, instead of the actual number of elementsaxom::Array<T>::insert behavior with non-trivial types.Please download the Axom-v0.6.1.tar.gz tarball below, which includes all of the Axom submodules as well.
AXOM_DEBUG_DEFINE to control whether the AXOM_DEBUG compiler define is enabled.
By DEFAULT, it is enabled for Debug and RelWithDebInfo configurations, but this can be overriden
by setting AXOM_DEBUG_DEFINE to ON or OFF.axom::Array is now GPU-compatible, in particular via a memory space template parameter and via
extensions to axom::ArrayView that allow for copying into kernels and transfers between memory spaces.primal::Points and primal::VectorsAXOM_NOT_USED macro to AXOM_UNUSED_PARAM for better consistency with other Axom macrosexplicit to axom::Inlet::InletVector constructors and added a constructor that accepts a double*
AXOM_ENABLE_MFEM_SIDRE_DATACOLLECTION configuration option is now ON by default (rather than OFF).
This option should be disabled if mfem was configured with MFEM_USE_SIDRE.AXOM_DEBUG compiler define is now properly exported via the axom CMake target when it is enabledEPS to primal::closest_point() operator. This effectively snaps
closest points to the triangle boundaries vertices and edges when they are within EPS,
improving consistency when, e.g., querying multiple triangles from the same mesh.SignedDistance queries for query points closest to edges or vertices
of the input triangle meshPlease download the Axom-v0.6.0.tar.gz tarball below, which includes all of the Axom submodules.
AXOM_ENABLE_TOOLS
c2c ("contours to codes") library for parsing 2D spline data.
c2c is currently only available for Axom configurations on LLNL platforms.intersect(Ray, Segment) can now return the parametric coordinates of the intersection
along both the ray and the segment (when the intersection exists)intersect(Segment, BoundingBox) can now return the parametric coordinates bounding the
portion of the segment contained within the BoundingBox (when the intersection exists)InOutOctree class to work with 2D line segment meshes. Previously,
it only worked with 3D triangle meshesc2c ".contour" files in Quest. Contours that enclose a 2D region can be linearized
into line segment meshes and loaded into Quest's InOutOctree for in/out queries.inout C API to support 2D queries using the c2c library, when Axom is configured with c2c
axom::Array modeled after std::vector. Previous axom::Array renamed to axom::MCArray. Future changes to both arrays are expected.data_collection_util tool to generate Mesh Blueprint compliant high order distributed meshes from
an mfem mesh or over a Cartesian domainaxom::utilities::getHostName() and axom::utilities::getUserName().axom::primal::ZipIterable<T> type to convert structure-of-arrays data to a given
Primal geometric primitive.computeDistances() function to SignedDistance class for batched
signed-distance queries.getTraverser() function to BVH, enabling the customized traversal of a
BVH from within a device kernel.Octahedron primitivePolyhedron primitive for representing convex polyhedra bounded by planar polygons in 3Dclip() operator for computing the intersection of a Tetrahedron and an Octahedron as a Polyhedron
klee, for specifying non-conformal shape overlays for materials onto simulation meshes.
This component defines a schema for defining, transforming and overlaying 2D and 3D shapes
and validates klee input files. See the klee documentation for more information.View::clear().axom::ArrayView that provides view/indexing semantics over a raw pointer.
This replaces the external buffer logic previously provided by axom::Array.MFEMSidreDataCollection now reuses FESpace/QSpace objects with the same basis
Harden configuration options for BLT tools (style, code quality, etc.) against accidentally being enabled for users. Developers will
always give a full path (e.g. CLANGFORMAT_EXECUTABLE)
Inlet: Writers are passed directly to Inlet::write instead of being registered
Inlet objects can now be constructed without a user-provided sidre::DataStore
Conduit version changed to v. 0.7.2
Renames AXOM_DEBUG_VAR macro to AXOM_UNUSED_VAR since there were many cases where the latter
was not the appropriate name. This macro elides warnings about unused variables
Inlet's isUserProvided can now be used to query the status of subobjects of a Container via a name parameter
Upgrades our vcpkg usage for automated Windows builds of our TPLs to its 2021.05.12 release
Upgrades built-in cli11 library to its v1.9.1 release
Quest's inout C API has two new functions: inout_set_dimension() and inout_set_segments_per_knot_span().
The latter is only applicable for 2D queries on c2c contours
Spin: Refactored BVH public API based on user suggestions
BVH constructor only handles setting up default values, while the actual building of the BVH is
now done in a BVH::initialize(primal::BoundingBox*, int) method.
Alternate Umpire allocator IDs are supplied via BVH::setAllocatorID(int).
Other BVH methods have been modified to accept or return Primal primitives.
Spin: Removed hard dependency on RAJA and Umpire from BVH.
Moved slam::IteratorBase to axom::IteratorBase.
sidre::Array now derives from axom::MCArray.
axom::Array is now multidimensional; it intends to behave like std::vector in the 1D case
and numpy.ndarray in the multidimensional case
Quest: SignedDistance has been modified to use spin::BVH instead of BVHTree. This
enables signed-distance queries to run on the GPU, as specified via a new template
parameter.
Spin: Removed BVHTree class in favor of BVH.
Quest's signed_distance C API: Removed functions related to old BVHTree class
and added functions related to BVH class
void signed_distance_set_max_levels( int maxLevels )
void signed_distance_set_max_occupancy( int maxOccupancy )
void signed_distance_set_allocator( int allocatorID )
void signed_distance_set_execution_space( SignedDistExec execSpace )
All built-in third-party libraries (fmt, cli11, sol, and sparsehash) have been guarded to allow downstream users to
have their own versions. This includes moving their headers under include/axom instead of include/ and
moving their C++ namespace under axom (eg. fmt to axom::fmt). If you don't use our built-n TPLs this has no
affect on you, but if you do these are some the changes you will need to make:
| Library | Namespace changes | Header include changes |
|---|---|---|
| fmt | fmt:: → axom::fmt:: |
#include "fmt/fmt.hpp" → #include "axom/fmt.hpp" |
| sol | sol:: → axom::sol:: |
#include "sol/sol.hpp" → #include "axom/sol.hpp" |
| sparsehash | google:: → axom::google:: |
#include "sparsehash → #include "axom/sparsehash |
| cli11 | CLI:: → axom::CLI:: |
#include "CLI11/CLI11.hpp" → #include "axom/CLI11.hpp" |
Moved axom::MCArray and the sidre::Array it was based on into mint
as axom::deprecated::MCArray and sidre::deprecated::MCArray, respectively.
sidre::Array is now based on axom::Array.
utilities::string::split now returns a vector instead of using an out-parameter,
Inlet's string utilities were moved to Core, and splitLastNTokens was renamed to rsplitN
axom::Array-related classes have been moved into individual files.
RAJA dependency updated to 0.14.0
Umpire dependency updated to 0.6.0. Support for versions prior to v2.1.0 was removed.
Conduit dependency updated to 0.7.2+ (develop as of Sept 13, 2021). This was required because Spack
is now using HDF5's CMake build system.
Internal BLT dependency updated to 0.4.1
intersect(Ray, Segment) calculation for Segments that do not have unit lengthaxom/config.hpp.
The latter are now only added in MSVC configurations.Mint's VTK output for fields of type int64 and float
MFEMSidreDataCollection
MFEMSidreDataCollection for C++14 standard library feature that was not available in [email protected]
MFEMSidreDataCollection until after setting
the nodal GridFunction (when applicable)R and Z coordinates when linearizing NURBS curves in c2c readerInOutOctree cases with grazing trianglesPlease download the Axom-v0.5.0.tar.gz tarball below, which includes all of the Axom submodules as well.
AXOM_ENABLE_MFEM_SIDRE_DATACOLLECTION CMake option is enabled.slic::setAbortFunction to configure a custom callback when SLIC aborts.batched option to quest's InOutOctree containment query example application.
This uses a kernel to test for containment on an array of points.
The query uses OpenMP threading, when available.NOMINMAX to axom/config.hpp.in to avoid problems with
the Windows min and max macros.cpp14 variant to Spack package to allow Inlet::LuaReader to be used more easily.void and strings in Lua-defined functionsget<std::vector<T>> for retrieving arrays without index informationWriter for generating JSON schemas which can be used by text editors
for autocompletioninlet::Inlet that retrieves the set of unexpected names
in the input fileContainers as strict, which fail verification when unexpected
entries are presentMFEMSidreDataCollection for registering QFunctions
(data associated with quadrature points on a mesh)Inlet object can be reconstructed from a Sidre group,
excluding callback functionsMFEMSidreDataCollection. These correspond to
matsets
and
specsets
in the Mesh BlueprintDataStore::setConduitSLICMessageHandlers() method.SchemaCreator to an abstract class and added missing functionsReader class from Inlet and Sol Lua state
from the LuaReader classoperator[] and
T get<T>()
std::shared_ptr<T> has been replaced with T& in non-owning contexts
and std::unique_ptr<T> in owning contexts - specifically, within Inlet's internal
tree structureaxom::slic::UnitTestLogger to axom::slic:SimpleLogger because it's used in
more than just unit tests.conduit to version 0.6.0 released Nov 2, 2020.sparsehash to version 2.0.4 released Aug 11, 2020.primal::Vector in Lua for use in input-file-defined functionsMFEMSidreDataCollection will now reconstruct fields and the mesh when a
datastore is Loaded inTable interface to eliminate ambiguity and duplicated functionalityDocWriter to Writer and refactored its interfaceTable to Container
operator bool for Field and Container has been replaced with more precise isUserProvided
and exists, which also returns true if a default value was specified.fmt to master branch snapshot, March 26, 2021.SphinxWriter will now print only one element schema per container instead of
printing the same schema for each element in the containerAXOM_EXPORT instead of AXOM_API
style build
target if version is unknown or wrong.Please download the Axom-v0.4.0.tar.gz tarball below, which includes all of the Axom submodules as well.
EPS that is used to determine intersections between
triangles in primal:intersect() as an optional final parameter.mesh_tester utility for calculating
triangle-triangle intersection.axom::execution_space< ExecSpace >::onDevice() to check if execution
space is on device.AXOM_SUPPRESS_HD_WARN to silence host device compiler
warnings.SignedDistance class and C API to toggle whether
the distance query computes the sign.batched option to quest's signed distance query example application.
This computes all distance queries on an array of points using a single call to computeDistance.
The query uses OpenMP threading, when available.AXOM_ENABLE_ANNOTATIONS, to enable/disable code
annotations in Axom. Default is OFF.AXOM_PERF_MARK_FUNCTION macro, or at a more fine grain level,
different sections of code can be annotated by wrapping them within an
AXOM_PERF_MARK_SECTION block. As a first cut, this works with NVTX tools.
However, the hooks are in place to add support for Caliper in the future.axom::nvtx:set_color() and
axom::nvtx::set_category() to set the corresponding parameters respectively.
This facilitates in the performance evaluation by allowing developers to easily
filter out calls by category or visually by setting a different color to use
in GUI tools, such as, NVVP and NSight.float or double type. The functionality is equivalent to that provided by
std::numeric_limits, but, the code is host/device decorated accordingly such that it
can also be called on the device.axom-config.cmake file to axom's installation to streamline incorporating axom
into user applications. See <axom-install>/examples/axom for example usages.C++ and Lua binding.
Sol is automatically enabled when LUA_DIR is found.
The version of Sol used in this release is v2.20.6, which requires C++14.AXOM_ENABLE_CUB option, since CUB is no lonher used directly in
Axom code. Instead, we use RAJA::stable_sort with RAJA-v0.12.1 and fallback
to std::stable_sort with older versions of RAJA and when the code is built
without RAJA.Uncrustify to [clang-format].
Axom's clang-format rules depend on clang 10.mesh_tester utility. Interface
now uses a -m, --method option to select the spatial index, and -p, policy
option now accepts a string or integer value.AXOM_USE_MPI3option to AXOM_ENABLE_MPI3 for consistency.BVH::findPoints() and BVH::findRays().umpire::Allocator object to specify an allocator, we now use the corresponding
integer ID associated with the allocator.SIDRE_datastore_new is now SIDRE_DataStore_new.axom::slic::message Level enums are changed
from enum-name_enumerator to namespace_enumerator.
ex. level_error is now message_error.datastore_new is now SidreDataStore
iomanager_new is now IOManager
primal::intersect(Segment, BoundingBox) and added regression tests.AXOM_USE_64BIT_INDEXTYPE configurations. Axom can once again
be configured with 64-bit index types.__threadfence_system() after the parent is computed and stored back
to global memory to ensure that the write is visible to all threads.mint_cell_types.cpp test was causing a segfault in the compiler. The main
issue triggering this compiler bug was the use of constexpr when defining the
static cell_info array of structs. The code has been modified to use const
instead.generate_aabbs_and_centroids method
in the spin_bvh.cpp unit test. It seems that the compiler does not handle
the lambda capture of the arrays correctly which leads to a segfault. A
workaround for the IBM XL compiler is provided.Please download the Axom-v0.3.3.tar.gz tarball below, which includes all of the Axom submodules as well.
axom::reallocate(_, 0) to always return a valid pointer.
This matches the semantics of Umpire's reallocate(_, 0).
Note: Umpire's PR-292 fixed a bug in its handling of this case and Axom
includes a workaround to get the new behavior until we upgrade to Umpire v2.0+.convert_sidre_protocol example. Data truncation functionality now
works properly when multiple Views point to the same data.Please download the Axom-v0.3.2.tar.gz tarball below, which includes all of the Axom submodules as well
intersect operator was also added to
compute the intersection points between a pair of Bezier curves of arbitrary order.