NVIDIA / NVIDIA/cccl

Update `cub::DeviceTransform` vectorized path tuning

Open
#7,265 0 comments 0 reactions 1 assignee Claimed by @bernhardmgruber View on GitHub
Dominant language
C++
Stars
2.5k
Forks
487
Avg merge
2d 7h
Merged PRs (30d)
296

Description

To test whether tuning works after #6914 I ran a small job for babelstream and here are some results regarding only the vectorized code path on B200:
```
cub.bench.transform.babelstream[T{ct}=I8, OffsetT{ct}=I32]:
variant score mins means maxs
279 bif_-8.alg_1.tpb_128.vsp2_5.vpt_1 (_7.db) 1.185134 1.000000 1.167870 1.390244
271 bif_-8.alg_1.tpb_128.vsp2_5.vpt_1 (.db) 1.185134 1.000000 1.167870 1.390244
103 bif_-16.alg_1.tpb_256.vsp2_5.vpt_1 (_0.db) 1.184023 1.000000 1.166830 1.389566
92 bif_-16.alg_1.tpb_256.vsp2_3.vpt_2 (_3.db) 1.183893 1.000000 1.167303 1.338542
25 bif_-12.alg_1.tpb_128.vsp2_5.vpt_1 (.db) 1.182305 0.994819 1.164960 1.389185
cub.bench.transform.babelstream[T{ct}=I8, OffsetT{ct}=I64]:
variant score mins means maxs
29 bif_-4.alg_1.tpb_128.vsp2_3.vpt_2 (.db) 1.211975 1.000000 1.194684 1.415144
37 bif_-4.alg_1.tpb_128.vsp2_3.vpt_2 (_7.db) 1.211975 1.000000 1.194684 1.415144
122 bif_4.alg_1.tpb_128.vsp2_4.vpt_2 (_0.db) 1.209728 0.888889 1.190481 1.412790
13 bif_-16.alg_1.tpb_128.vsp2_3.vpt_2 (_2.db) 1.204793 0.996094 1.187273 1.415163
58 bif_0.alg_1.tpb_128.vsp2_3.vpt_2 (_2.db) 1.203520 0.994819 1.185968 1.415144
```
Let's add those to the current tunings.

Contributor guide

Open the contributing guide

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.