Distributed PDHCG - #5

Merged

Lhongpei merged 20 commits into

mainfrom

distributed_and_spmvop

Apr 30, 2026

Commits on Mar 29, 2026

init
Lhongpei
committed
378aef5

Commits on Apr 2, 2026

Preparation for distributed version: 1. move all kernel to `pdhcg_kernels.cu`; 2. move all func using SpMV to `pdhg_core_op.cu`; 3. remove feasibility polish func
Lhongpei
committed
7ec890e
update lic for permute.cu
Lhongpei
committed
bbb04bd
support permutation in QP with both sparse and lowrank second order obj term
Lhongpei
committed
bd32312
fix error in `permute_problem_return_new`
Lhongpei
committed
cacd9e7
add distributed component, required for developing distributed version
Lhongpei
committed
3ed2018

Commits on Apr 3, 2026

update
Lhongpei
committed
1568544

Commits on Apr 7, 2026

refactor initialization to decouple diagonal Q processing
Lhongpei
committed
ad356b1
move preprocess logic to precondition part
Lhongpei
committed
5787ed7

Commits on Apr 10, 2026

integrate single- and multi-GPU version; there are still some issues
Lhongpei
committed
3069def
fix a bug caused by lack of initalization of `n_start`
Lhongpei
committed
a45bba2
move cudaFree(0) into single GPU version
Lhongpei
committed
1ef7749

Commits on Apr 13, 2026

Now, multi-GPU version can successfully solve some instances; fix a error in create result when problem is LP
Lhongpei
committed
f6e6ca5

Commits on Apr 14, 2026

fix a bug of losing global variable number
Lhongpei
committed
7df2538
refractor distributed solver main function; now presolve is done before permutation.
Lhongpei
committed
b6322d1
add SpMVOp Support
Lhongpei
committed
750fc89

Commits on Apr 18, 2026

SpMVOp is now available with CUDA >= 13.0
Lhongpei
committed
1e0546a

Commits on Apr 20, 2026

update `CMAKE_CUDA_ARCHITECTURES`
Lhongpei
committed
874d898

Commits on Apr 29, 2026

fix distributed version error caused by mpi detection
Lhongpei
committed
1e90890

Commits on Apr 30, 2026

Update Documentation
Lhongpei
committed
cce1961