Reduce GPU/host synchronization overhead in barrier termination check - #1808

Open

yuwenchen95 wants to merge 9 commits into

NVIDIA:mainfrom

yuwenchen95:polish_termination_check

Commits on Aug 26, 2026

Reorder computation for residuals, mu and objectives at the begin of barrier method and the end of each barrier iteration, which reduce the number of synchronization required
yuwenchen95
committed
17aa1cb
Rename new vectors
yuwenchen95
committed
a6f49a6

Commits on Aug 27, 2026

Code clean up: rename functions and create a new struct barrier_reduce_helper_t addressing operations related to termination check
yuwenchen95
committed
ea4f38a
Merge branch 'main' into polish_termination_check
yuwenchen95
committed
c2dd449
restore 'CHECK_OBJECTIVE_GAP' back
yuwenchen95
committed
520e995

Commits on Aug 28, 2026