GitLab

Changes to interface, added internal namespace, new convention is for internal classes to be lower case and for external classes to be capitalized

 - diagonal contractions with not all edge lengths equal
 - read_all had a short circuit
 - CTF_VERSION 100 now defined

Also fixed a bug with A*A, which I introduced ~5 commits ago when fixing multi-tensor symmetric contractions.

This commit contains a number of major changes and removals of unused source which aims at slimming down the source code in preparation for a release.

Switched to using CommData_t by value instead of by reference. MPI_Comms now freed via an std::set.

Got rid of the deprecated subdirectory.

Got rid of inner mapping functionalities which were not used due to being inefficient.

(somewhat questionable decision) got rid of old unit tests contained in unit_test subdirectory, including the code needed for the VERIFY functionality used in dist_tensor_op.cxx.
The old unit tests were rather limited and at this point somewhat redundant, but there may be reason to use VERIFY functionality in the future.

Hooked up add_to/from_subworld to fast redistribution kernel.  It should now be possible to use this function to add tensors
from an arbitrary sub MPI comm to/from a superset comm.

Also dist_tensor_internal.cxx is no longer included via cyclopstf.cxx though the organization here can be further improved.

Implemented 'slow' (via dummy tensor and slice) versions off add_to/from_subworld. They work for a basic use-case, see examples/subworld_gemm.cxx.

Got rid of was/is padding, there is always padding. Got rid of cloning amongst different worlds, because it is not a well defined interface.
Added dummy interface functions for new cross world functions, add_to_subworld() and add_from_subworld();

1. the CTF memory tracking thinks there are leaks when there are not for reasons I cannot currently fathom, so currently -DPRODUCTION is on by default in all cfgs
2. CTF_cond_free cannot be used in PRODUCTION mode (since there is no memory tracking), which is a pain for freeing communicators in dist_tensor_internal.cxx, current solution is to leak this little bit of memory

Implemented permuted slice. Working for multiworld reads/write. -1s not tested yet, and same world fast-redistribution not
hooked up, though implemented.

Reads and writes which happen to the upper part or diagonal of SH/SY/AS tensors should now
be handled properly. Also corrected handling of contractions and sums over SY symmetries by
forcing unpacking (cannot use overcount factor due to that damn diagonal, though more
efficient solutions than packing are certainly possible).

Also, simplified interface.

examples/slice_gemm.cxx for basic usage and
examples/strassen.cxx for more advanced usage

Standardized all allocs to CTF_free and CTF_alloc and implemented a memory management system that keeps track of memory usage.