fix: objective constant bug with presolve (MIT-Lu-Lab#90)

* fix: objective constant bug with presolve

* update to version 0.2.9

Add cusparseSpMVOp backend with compile-time auto-detection (MIT-Lu-L…

…ab#88)

* working SpMVop

* update

* improve SpMVOp via cupdlpx_spmv_ctx_t

* update feasibility polishing spmv

* Apply formatter

* fix spmvop version issue

* add more CUDA versions to CI

* CI: update cuda-toolkit setting

* update build.yml

* update build.yml

* update SpMV backend in readme

* update version to v0.2.8

Bug fixed: add cmake to broken sdist (MIT-Lu-Lab#87)

CUDA Graph (MIT-Lu-Lab#71)

* refactor iteration loop

* fix bug

* refactor: takeout the first iteration from the inner loop

* working CUDA graph

* fuse halpern update kernel

* add delta solution scaling

* change fixed point error scaling to 2

* update

* update compute_next_dual_solution_kernel

* move cumulative_time_sec into check_termination_criteria and fix feasibility polishing timing bug

* add __restrict__ for all kernels

* remove old kernels and support cuda graph in feasibility polishing

* fix feasibility polishing time limit check

* update

* update preconditioning log

* update

* update

* update CMAKE_CUDA_ARCHITECTURES

* bump to v0.2.5

Add GPU implementation of preconditioner - v2 (MIT-Lu-Lab#31)

* New feat: GPU precondition

* New feat: synchronized A/At scaling

* Todo: infeasible is stucked

* New feat: preconditioner prints

* Bug fixed: objective & bound rescale

* feat: add logging for precondition time

* Bug fixed: numerical issue

* Clean code: clang-format and align naming style

* Clean code: rename variables

* Refactor: move finite-bound computation

* Numerical issue: back to old version

* Code refactor: CUB DeviceReduce for bound norm

* Bug fixed: typo

* Clean code: Limit CUB headers to device_reduce to reduce compile time

* Clean code

* Bug fixed: warm start solution

* Bug fixed: illegal memory access in bound

* Bug fixed: norm computation on scaled bound

* Revert "Bug fixed: norm computation on scaled bound"

This reverts commit d372a20.

* Bug fixed: unused sv params

* New feat: add verbose logging

* New ver: 0.2.4

Update PSLP to 0.0.4 (MIT-Lu-Lab#59)

* update PSLP to 0.0.4 and fix the primal objective value bug

* update to v0.2.3

* fix instance solved during presolve issue

* add cudaGetLastError at end

release v0.2.2 (MIT-Lu-Lab#58)

Update PSLP to v0.0.3 (MIT-Lu-Lab#56)

* add support for PSLP v0.0.3

* add presolve_stats in results

* comment out presolve stats

* release v0.2.1

Integrate PSLP Presolver (MIT-Lu-Lab#41)

* add PSLP presolver

* fix: correct the reduced cost in PSLP postsolve

* add presolve stats

* test PSLP relax_bounds false

* remove PSLP_INF

* add presolve_setup_time and presolve_time record, improve reduce prob memory

* set relax_bounds to True (default)

* fix Segmentation fault when presolve detected infeasible

* refactor: PSLP presolve

* add verbose check for log

* set use_presolve to true by default

* rename use_presolve to presolve

* refactor: add presolve in python interface

* docs: add presolve in readme

* feat: release v0.2.0

* fix: off-by-one iteration error and ensure consistent final metrics

release: v0.1.5 (MIT-Lu-Lab#53)