gpudev update #92
Merged
gpudev update #92
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Contributor
cjknight
commented
May 14, 2024
cjknight
commented
- Merged with latest dev
- Some "light" tuning of get_jk() CUDA kernels (light because need straightforward port to portable programming model)
- Added mini-app for evaluating tall-skinny matrix transposes
- Updated Makefile and arch files to support mini-app builds
- Enabled "not with_k" branch in get_jk()
- Added exploratory code to bypass dfobj.loop() copy ; not needed at moment
- Added support for per-device streams and handles
- Added multi-gpu support for get_jk()
- Temporarily disabled gpu-offload of hessop_get_veff because of issue with multi-gpu support
- Added OpenMP thread affinity info to output
- Updated arch and builds after Polaris@ALCF software upgrade
Hack H1EZip to not "unroll" the Hilbert-space layer and the single state layer together, so that averaging over the latter can be abstracted into the fcibox.fcisolvers individual RDM generators.
LASSIS unittest fails because lroots now needs to be passed from one level of LASSIS prepare_states to another, along with charges, spins, smults, and wfnsyms.
This now breaks lasscf_rdm instead, because the guess items are not CI vectors but RDM arrays.
lasscf_rdm transparently works for now
Include unambiguous statement that they average over local states, resolving issue MatthewRHermes#34.
The casdm1rs and casdm2r go into the slot for "rootspace 0" and "rootspace 1", respectively, that the usual algorithm would make. Also use setUpModule in test_lasscf_rdm.py
Use n(elec,hole)(u,d) instead of n(elec,hole)(a,b) to compute number of charge-separated single excitations. Changes total energies in test_c2h4n4, but all delta-E tests in test_lassis_targets_slow still pass.
A much simpler, but less flexible/more costly choice.
Ignore both the s-change of each fragment and the putative m value of the hopping electron, and assume that electrons can always come out of both doubly- and singly-occupied orbitals and can always go into both singly-occupied and unoccupied orbitals.
Test on H2 for lasscf geometry optimization
must implement smart lindep handling for the branch to be viable
since spin-shuffle spoils bitwise agreement
storage space
Still need to fix signs, but in a way which doesn't break the op_o1 op_o1 agreement check
hopefully make tests work
LASSI op_o1 now aggressively identifies linearly-equivalent rootspaces and caches the relevant unitary transformation matrices, in order to reduce the number of interactions computed explicitly, except for in the ContractHamCI class, where I still don't know how exactly to implement this.
Profile the timing of lassi op_o1 more thoroughly. Move two fns from sitools to citools (the latter being safer to import). Tighten exact-equivalence tolerance in lassi op_o1 screening.
For op_o1 profiling, use more descriptive names in the output and add it around the HamS2Ovlp full overlap matrix steps.
"unique" messages demoted to DEBUG
Adding keyword to plot only active space orbitals
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.