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.

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

[gpuCI] Forward-merge branch-22.08 to branch-22.10 [skip gpuci] #4833

Merged
merged 3 commits into from
Jul 26, 2022

Conversation

GPUtester
Copy link
Contributor

Forward-merge triggered by push to branch-22.08 that creates a PR to keep branch-22.10 up-to-date. If this PR is unable to be immediately merged due to conflicts, it will remain open for the team to manually merge.

Allard Hendriksen added 3 commits July 22, 2022 23:21
Fixes issue #2387.

For large data sizes, the batch size of the DBSCAN algorithm is small in order to fit the distance matrix in memory.

This results in a matrix that has dimensions num_points x batch_size, both for the distance and adjacency matrix.

The conversion of the boolean adjacency matrix to CSR format is performed in the 'adjgraph' step. This step was slow when the batch size was small, as described in issue #2387.

In this commit, the adjgraph step is sped up. This is done in two ways:

1. The adjacency matrix is now stored in row-major batch_size x num_points format --- it was transposed before. This required changes    in the vertexdeg step.

2. The csr_row_op kernel has been replaced by the adj_to_csr kernel.    This kernel can divide the work over multiple blocks even when the    number of rows (batch size) is small. It makes optimal use of memory    bandwidth because rows of the matrix are laid out contiguously in memory.

Authors:
  - Allard Hendriksen (https://github.com/ahendriksen)
  - Corey J. Nolet (https://github.com/cjnolet)

Approvers:
  - Corey J. Nolet (https://github.com/cjnolet)
  - Tamas Bela Feher (https://github.com/tfeher)

URL: #4803
This functionality has been moved to RAFT.

Authors:
  - Allard Hendriksen (https://github.com/ahendriksen)

Approvers:
  - Tamas Bela Feher (https://github.com/tfeher)
  - Corey J. Nolet (https://github.com/cjnolet)

URL: #4829
This PR removes the naive versions of the DBSCAN algorithms. They were not used anymore and were largely incorrect, as described in #3414. 

This fixes issue #3414.

Authors:
  - Allard Hendriksen (https://github.com/ahendriksen)

Approvers:
  - Corey J. Nolet (https://github.com/cjnolet)

URL: #4804
@GPUtester GPUtester requested a review from a team as a code owner July 26, 2022 18:44
@GPUtester GPUtester merged commit 628f4c7 into branch-22.10 Jul 26, 2022
@GPUtester
Copy link
Contributor Author

SUCCESS - forward-merge complete.

jakirkham pushed a commit to jakirkham/cuml that referenced this pull request Feb 27, 2023
[gpuCI] Forward-merge branch-22.08 to branch-22.10 [skip gpuci]
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
Projects
None yet
Development

Successfully merging this pull request may close these issues.

1 participant