coll: circulant graph allreduce algorithm - #7748

Open

hzhou wants to merge 8 commits into

pmodels:mainfrom

hzhou:2603_cga_allreduce

Commits on Mar 18, 2026

coll/cga: fix reduce chunk_count
hzhou
committed
87cb737
coll/cga: use MPI_Aint to prevent overflow
hzhou
committed
44dfdcd
coll/cga: use malloc for packbuf if CHUNK_SIZE is 0
hzhou
committed
84748c4
coll/cga: allow switching coll_type
hzhou
committed
4a32670
coll/cga: packbuf dependency in allreduce
hzhou
committed
09396ce
coll: add intra_circ_graph allreduce algorithm
hzhou
committed
e5be20c
test: modify coll/allred.c to avoid exceeding precision
hzhou
committed
5866780

Commits on Mar 19, 2026

coll/cga: add staging to avoid direct gpu reduction
hzhou
committed
1856eee