GitLab

Renamed Accumulator->Transform, since Accumulator does not make so much sense for Endomorphism

implemented sequential dense=sparse*dense contraction kernel and allowed contraction algorithms to be constructed using it.  Added spmv benchmark which now works sequentially.

Both sequential key permutation kernels and logic, as well as sparse distributed communication still needs to be implemented to get dense=sparse*dense working in parallel.

Added tests for sparse univariate custom function that performs integration of a sparse matrix of forces based on a cutoff. It reveals limitations in the current sparse summation implementation.

Got univariate functions working, added some two tests for them, including force integration. Added a function to compute the additive inverse of a tensor to be used for groups, as the normal infrastructure cannot currently handle this case, since it uses the notion of an inverse of the multiplicative identity

Wrote sparse scale, added tests for endomorphism, including for custom defined types.  Had to fix a number of bugs in algebraic structures in the process.

Endomorphism now works with custom functions on dense and sparse tensors.

Also updated the dependencies in src/*/Makefile.

Fixes to custom elementwise functions. Partially completed work on an expression based interface for them. Added endomorphism test.

Corrections to sparse summations to make it work when data needs to be replicated and to make it faster when it is reduced.

Also a new test, speye.cxx for sparse replicate/reduce/diagonal that creates a sparse identity tensor

Implemented sparse summation, added sparse tensor constructor to the interface, and added a simple sparse sum test which seems to be working

Implemented --build-dir option making it possible to have multiple external builds configured differently sharing the same source

1. Added capability (via constructor) to repack tensor data into a different packed symmetric layout without doing the normal symmetrization permutations that happen during sum
2. Fixed bug associated with C["ij"]=A["ij"]*B["ijkl"] where C and A are SY and B is fully NS.
3. Getting (2) to work in parallel required resolving a bug associated with desymmetrization (in the two unfold_broken_Sym functions), which were changing symmetry without changing sym_table, and as a result yielding extra mapping restrictions.

The bug fix to (3) might improve performance generally.

Renived sym3, which only makes sense when without sym sum, updated from v1.0 interface to namespace interface in examples/tests/studies, got fast algorithms working

Started writing interface for semirings, which tries to expose type-templated semirings, but type-free tensors (on typed rings)
by autocasting templated semirings into (char *)-based semirings, pretty wild, but the point is to have an elegant typed interface
without instatiating everything for every type defined by the user.

Made MPI_Comm_split happen on-demand rather than at start-up for all the torus topologies.

Added MPI_Finalized(..) check to make sure that CTF exits gracefully (without freeing the main comm)
if a user calls MPI_Finalize() before destroying a CTF_World

Reorganized examples folder into tests/benchmarks/studies/examples and added configure mechanism adapted from Aquarius and NuLAB

makefile

available memory for now.

Cleaned up build system, now make should work immediately on linux, etc. Deprecated some unushed tests and code.

as Devin's interface.

Added a new interface which is not fully complete nor tested