Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

IVF-PQ: tweak launch configuration #1069

Merged
merged 3 commits into from
Dec 8, 2022
+7 −1
Merged

IVF-PQ: tweak launch configuration #1069

Changes from 2 commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
0a2f023
Tweak the launch cofiguration in some non-optimal cases.
achirkin Dec 5, 2022
ced1e7f
Simplify the selection condition
achirkin Dec 8, 2022
4706f81
Fix a typo (different signedness comparison warning)
achirkin Dec 8, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 7 additions & 1 deletion cpp/include/raft/spatial/knn/detail/ivf_pq_search.cuh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1173,7 +1173,13 @@ struct ivfpq_compute_similarity {
// If we don't have enough repeating probes (locality_hint < tmp.blocks_per_sm),
// the locality is not going to improve with increasing the number of blocks per SM.
// Hence, the only metric here is the occupancy.
select_it = tmp.occupancy > cur.occupancy;
bool improves_occupancy = tmp.occupancy > cur.occupancy;
// Otherwise, the performance still improves with a smaller block size,
// given there is enough work to do
bool improves_parallelism =
tmp.occupancy == cur.occupancy &&
7 * tmp.blocks_per_sm * dev_props.multiProcessorCount <= n_blocks;
select_it = improves_occupancy || improves_parallelism;
} else {
// If we don't use shared memory for the lookup table, increasing the number of blocks
// is very taxing on the global memory usage.
Expand Down