[WIP] GPU support in PEPSKit - #407

Draft

kshyatt wants to merge 102 commits into

mainfrom

ksh/cupeps

Commits on Sep 24, 2026

Initial steps to CUDA support
kshyatt
committed
f9048a6
Update Project.toml
kshyatt
committed
3f8e472
A few more small fixes
kshyatt
committed
f972bb2
Use proper storagetype for delta_t
kshyatt
committed
bff7b70
Restore singular value distance
kshyatt
committed
b2ac9d4
Update Project.toml
kshyatt
committed
6e44a59
Update Project.toml
kshyatt
committed
9d2f145
Cleanup and some actual tests

Katharine Hyatt
authored and
kshyatt
committed
9781943
Get rid of duped examples and fix formatting

Katharine Hyatt
authored and
kshyatt
committed
7c831ca
Try densitymatrices tests

Katharine Hyatt
authored and
kshyatt
committed
ac5de8f
Add a simple timestep example

Katharine Hyatt
authored and
kshyatt
committed
6ab850a
More progress

Katharine Hyatt
authored and
kshyatt
committed
b868ae4
refactor `_fliptwist_s`

lkdvos
authored and
kshyatt
committed
834b090
Fixes

Katharine Hyatt
authored and
kshyatt
committed
21c927b
Working CTMRG contractions tests

Katharine Hyatt
authored and
kshyatt
committed
1f1067a
Further updates

Katharine Hyatt
authored and
kshyatt
committed
3db0063
VUMPS test fix

Katharine Hyatt
authored and
kshyatt
committed
2b6628b
More init tests

Katharine Hyatt
authored and
kshyatt
committed
f0a03e7
Partially support cluster projectors tests

Katharine Hyatt
authored and
kshyatt
committed
5b02e38
Fix dumb typo

Katharine Hyatt
authored and
kshyatt
committed
bf31e3a
Working tf_ising_finiteT

Katharine Hyatt
authored and
kshyatt
committed
28fd127
Increase timeouts and add j1j2 test

Katharine Hyatt
authored and
kshyatt
committed
6857346
Fix CTMRG env init
kshyatt
committed
d3b3785
More init fixes
kshyatt
committed
2b16ace
Sitedep truncation also working

Katharine Hyatt
authored and
kshyatt
committed
10a90cb
Working ctmrg/partition_function (except C4v)

Katharine Hyatt
authored and
kshyatt
committed
d982416
Fix SUWeight

Katharine Hyatt
authored and
kshyatt
committed
a56ce75
Fix select algo

Katharine Hyatt
authored and
kshyatt
committed
6ac0f2c
Gaugefix working?

Katharine Hyatt
authored and
kshyatt
committed
acd5b65
Working BP tests

Katharine Hyatt
authored and
kshyatt
committed
ea491af
Fixed iteration appears to be working

Katharine Hyatt
authored and
kshyatt
committed
72b0d91
Actually add the test files whoops

Katharine Hyatt
authored and
kshyatt
committed
e1d5e25
Flavors working too

Katharine Hyatt
authored and
kshyatt
committed
29be9a4
Working ctmrg/unitcell

Katharine Hyatt
authored and
kshyatt
committed
b50560d
Working CTMRG/pepo

Katharine Hyatt
authored and
kshyatt
committed
67a3084
Formatter
kshyatt
committed
71664e3
Try simplifying the test script
kshyatt
committed
b2a8571
Try splitting groups
kshyatt
committed
b665ca5
Include in title
kshyatt
committed
021b5f9
Adjust paths
kshyatt
committed
9fbd7f2
Fix runtests.jl

Katharine Hyatt
authored and
kshyatt
committed
d24483e
Soothe VUMPS on 1.12
kshyatt
committed
97db82c
Bondenv tests

Katharine Hyatt
authored and
kshyatt
committed
b66e44b
Fix product_peps
kshyatt
committed
2cabe91
BP-Gaugefix test for CUDA

Katharine Hyatt
authored and
kshyatt
committed
61cb698
More working tests with downstream fixes

Katharine Hyatt
authored and
kshyatt
committed
b6dbd39
Bad branch

Katharine Hyatt
authored and
kshyatt
committed
33c92a9
Don't use QRIteration for C4V + CUDA
kshyatt
committed
5e9139e
Get cuda flavors working
kshyatt
committed
ee60dbf
Make the gaugefix loop GPU-friendly

Katharine Hyatt
authored and
kshyatt
committed
affe1a7
Branch
kshyatt
committed
43d749d
CUDA expvals test works
kshyatt
committed
8121dd6
bp/expvals tests working

Katharine Hyatt
authored and
kshyatt
committed
9f20acb
GPUified SVD and passing tests

Katharine Hyatt
authored and
kshyatt
committed
1261547
Working eigh

Katharine Hyatt
authored and
kshyatt
committed
7260963
Switch to CUDACore and more tests working

Katharine Hyatt
authored and
kshyatt
committed
678ef00
CUSOLVER doesn't have a QRIteration heev
kshyatt
committed
35b2996
Point at branch for now
kshyatt
committed
c9cb1ad
Switch to CUDACore and more tests working

Katharine Hyatt
authored and
kshyatt
committed
7e31399
Gradient tests

Katharine Hyatt
authored and
kshyatt
committed
6283e3c
Fix for c4v

Katharine Hyatt
authored and
kshyatt
committed
af651ee
Use the new braiding branch over at TK

Katharine Hyatt
authored and
kshyatt
committed
f4e1540
Use alloc_caches from GPUArrays

Katharine Hyatt
authored and
kshyatt
committed
7aa97a1
Leave a note about which GPU algo to use

Katharine Hyatt
authored and
kshyatt
committed
987539c
Use DivideAndConquer for ROCm tests

Katharine Hyatt
authored and
kshyatt
committed
650af37
Let the cache depth vary by algo

Katharine Hyatt
authored and
kshyatt
committed
1790515
Fix AMDGPU rev

Katharine Hyatt
authored and
kshyatt
committed
f7d4b82
CUDA/CUDACore badness STRIKES AGAIN
kshyatt
committed
05457d7
Make sure buffers aren't overwritten in rrule_via_ad
kshyatt
committed
b147048
use Jacobi for the fixed_iterscheme tests for speeeeeeeed

Katharine Hyatt
authored and
kshyatt
committed
7043ba6
Fixes for holding cache memory for too long
kshyatt
committed
6550429
Fix sources AGAIN
kshyatt
committed
70a9501
Try turning off the most memory intensive part
kshyatt
committed
b167403
Add a quick check to make sure no empty spaces were created

Katharine Hyatt
authored and
kshyatt
committed
e347f16
Batched SVD support and more caching improvements

Katharine Hyatt
authored and
kshyatt
committed
43b6c58
More batching
kshyatt
committed
237c268
Force real-valued accumulator
kshyatt
committed
40f5f3c
Another small fix
kshyatt
committed
71ea078
Move batching logic into MAK as much as possible
kshyatt
committed
6ea240c
Try to make formatter happy
kshyatt
committed
5e90e25
Fix missing docstring refs
kshyatt
committed
6e83400
Fix stale checks in the GPU gradient tests
kshyatt
committed
26cbe25
Incremental trimming of duplicated tests
kshyatt
committed
68685b0
Use new TO version
kshyatt
committed
9cf720f
Fixups
kshyatt
committed
ef09f93
Use backend and allocator for the VUMPS transfer matrices
kshyatt
committed
22bdd26
Test fixes
kshyatt
committed
2cca8bc
Fix bp/unitcell test
kshyatt
committed
2a55870
More incremental fixes
kshyatt
committed
bb2e924
Another missed AT in tf_ising_finiteT
kshyatt
committed
381f3b2
Various ctmrg fixes
kshyatt
committed
f5c414f
Formatter
kshyatt
committed
b5b7c24
Fix the c4v gradients
kshyatt
committed
148a698
Update caching strategy
kshyatt
committed
b16a40f
Trim down the GPU tests so they run faster on CI
kshyatt
committed
0f304ca
Don't run a test on AMDGPU that needs geev
kshyatt
committed
b78287f
Formatter
kshyatt
committed
af9eaa2
Fix pullback for GMRES
kshyatt
committed
65eeaf8
Soothe CI a bit more
kshyatt
committed
a2d7259
Extend timeout and use 1.13
kshyatt
committed
6e8b5de
Fix codecov token
kshyatt
committed
37dca7e

Commits on Sep 29, 2026

Try the guarding branch
kshyatt
committed
f305470