Reduce Size of Fused Reduction in Kmeans|| Iterations - #2438

Open

tarang-jain wants to merge 62 commits into

NVIDIA:mainfrom

tarang-jain:kmeanspp

Commits on Jun 17, 2026

add example
divyegala
committed
41b0713

Commits on Jun 24, 2026

initial integration
divyegala
committed
b10c02c
attempt to fix tile linkage
divyegala
committed
1b934dd
working test, remove example
divyegala
committed
c7f7cbd
style check
divyegala
committed
86c9311

Commits on Jun 25, 2026

start integrating other metrics
divyegala
committed
e352629
if constexpr exit
divyegala
committed
d6560fc
passing KMeans tests
divyegala
committed
6742853

Commits on Jun 30, 2026

undo kvp, add constrains and alignments to tile export
divyegala
committed
db6b385
add reproducible benchmark scripts
divyegala
committed
0a3da06
benchmark sweep
divyegala
committed
9da31ae

Commits on Jul 2, 2026

index dtype
divyegala
committed
35291ec

Commits on Jul 9, 2026

new reduce
divyegala
committed
f48ec5f
try int64 vector constant
divyegala
committed
842863e
new tile shape for sm90
divyegala
committed
9fb768d

Commits on Jul 11, 2026

using prebuilt cubins
divyegala
committed
88ddda8

Commits on Jul 15, 2026

use 13.4 tile
divyegala
committed
099324a

Commits on Jul 18, 2026

relaxed cubin, sm100 dispatch, new reduction strat
divyegala
committed
fe77c61

Commits on Jul 22, 2026

split tile for relaxed and strict sm120 abi
divyegala
committed
7e086b1
remove i64, batch i32 to save binary size
divyegala
committed
82d14a6

Commits on Jul 28, 2026

working through
divyegala
committed
b0efee6

Commits on Jul 29, 2026

new tiles and reduction shapes for sm120
divyegala
committed
9b2d2cb

Commits on Aug 4, 2026

new kernel from huy
divyegala
committed
54125c0

Commits on Aug 6, 2026

add changes
tarang-jain
committed
4173cd5
add changes
tarang-jain
committed
05301a3
cutile pip at build time
divyegala
committed
27f2069
merge upstream
divyegala
committed
9f38f18
Delete cpp/src/cluster/detail/.nfs000000001cb8b01800001a98
divyegala
authored
a69cb01
Delete benchmark_kmeans.py
divyegala
authored
2185f1c
Delete run_benchmark_kmeans.sh
divyegala
authored
2fdd7f7
correct dependencies
divyegala
committed
2fb6253
package and style check
divyegala
committed
afc3cb5
Delete cpp/src/cluster/detail/.nfs000000001cb8b01800001a98
divyegala
authored
e28ecd9

Commits on Aug 7, 2026

Commits on Aug 10, 2026

fix upstream merge
divyegala
committed
f1bf39d
fix test
divyegala
committed
25fe50b

Commits on Aug 11, 2026

inertia calculation with unexpanded l2 metric
divyegala
committed
f225d05
correct from host to build deps
divyegala
committed
52e6be6
batch
divyegala
committed
e6b36d1
use cutile from conda
divyegala
committed
d5c56ae
fix ==
divyegala
committed
ccec474
use sass within family
divyegala
committed
d63649e
address reviews
divyegala
committed
b7ad920
Merge remote-tracking branch 'upstream/main' into cutile-python-to-cpp
divyegala
committed
720af43
delete resource
divyegala
committed
8f8e721
python and cutile in build not host
divyegala
committed
c92ed15
correctly express python
divyegala
committed
53d04f5
fix c
divyegala
committed
7f4c205

Commits on Aug 12, 2026

Merge branch 'main' of https://github.com/rapidsai/cuvs into kmeanspp
tarang-jain
committed
1a08b55
use find package
divyegala
committed
9715ac3
Merge branch 'main' into cutile-python-to-cpp
divyegala
authored
a6ddd7b

Commits on Aug 13, 2026

use build prefix python
divyegala
committed
9cdc427
Merge remote-tracking branch 'origin/cutile-python-to-cpp' into cutile-python-to-cpp
divyegala
committed
114bef8
hint python early
divyegala
committed
766eefc
explicilty find python
divyegala
committed
810a582
Merge branch 'main' of https://github.com/rapidsai/cuvs into kmeanspp
tarang-jain
committed
1234cdd
Merge branch 'cutile-python-to-cpp' of https://github.com/divyegala/cuvs into kmeanspp
tarang-jain
committed
66c3b40

Commits on Aug 14, 2026