Enable pmap for ERI value computation and sparse einsum and DFT integration#134

Draft

mihaipgc wants to merge 14 commits intomainfrom pmapIntorEri

+12,357-167

Commits on Oct 26, 2023

Got nanodft to use eri computed (and screened) on IPU. Poplar automatically realizes to only compute integrals once. Preparing to shard eri computing on different ipus.
alexandermath
committed
added shell padding; padding with -1 indices which are handled in poplar vertex by doing a no-op
mihaipgc
committed
fixed padding; simplified pmap slice ids and passed ndevices from args; pmap now working
mihaipgc
committed

Commits on Oct 27, 2023

Commits on Oct 30, 2023

reduced memory consumption in integral computation
mihaipgc
committed

Commits on Oct 31, 2023

fixed pmap memory issues (removed need for dynamic_slice) by slicing and passing shell indices through pmap
mihaipgc
committed

Commits on Nov 1, 2023

100x+ speedup in shell computation with c++ and openmp; can generate shell indices and sizes for 100 atoms
mihaipgc
committed

Commits on Nov 2, 2023

further reduced memory and preparing for benchmarks
mihaipgc
committed