[Review] Fix memory bug for SVM with large n_rows

[tfeher]

closes #3233

This PR introduces 64bit int arithmetic while accessing SVM kernel tile to avoid overflow.

[tfeher]

This PR depends on rapidsai/raft#131, I will update the raft path once that is merged.

[JohnZed]

Looks good! Just a few tiny questions/suggestions. I marked this as blocked so we remember to merge into RAFT first and re-enable the test.

[tfeher]

Thanks @JohnZed for the review, I have answered the comments. I have also updated the RAFT path.

[codecov-io]

Codecov Report

Merging #3420 (e113428) into branch-0.18 (550121b) will increase coverage by 0.24%.
The diff coverage is 85.45%.

@@               Coverage Diff               @@
##           branch-0.18    #3420      +/-   ##
===============================================
+ Coverage        71.48%   71.73%   +0.24%     
===============================================
  Files              207      212       +5     
  Lines            16748    17059     +311     
===============================================
+ Hits             11973    12237     +264     
- Misses            4775     4822      +47     

Impacted Files	Coverage Δ
python/cuml/datasets/classification.py	80.80% <ø> (+0.19%)	⬆️
python/cuml/decomposition/incremental_pca.py	94.70% <ø> (ø)
python/cuml/svm/svc.pyx	95.16% <ø> (-0.62%)	⬇️
python/cuml/svm/svm_base.pyx	94.27% <ø> (-0.63%)	⬇️
python/cuml/dask/ensemble/base.py	20.27% <12.50%> (+0.95%)	⬆️
python/cuml/dask/cluster/kmeans.py	54.00% <33.33%> (ø)
python/cuml/dask/decomposition/base.py	39.53% <50.00%> (ø)
...ython/cuml/dask/ensemble/randomforestclassifier.py	30.00% <50.00%> (+0.51%)	⬆️
python/cuml/dask/ensemble/randomforestregressor.py	35.08% <50.00%> (+0.54%)	⬆️
python/cuml/dask/linear_model/linear_regression.py	59.09% <50.00%> (ø)
... and 48 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update a22681c...e113428. Read the comment docs.

[tfeher]

rerun tests

[JohnZed]

Looks great! I just updated RAFT to the latest commit (there is a tiny addition after what you had previously)

[JohnZed]

@gpucibot merge

[cjnolet]

Looks like a 404 error on one of the faiss packages:

13:19:24 conda.CondaMultiError: HTTP 404 NOT FOUND for url <https://conda.anaconda.org/rapidsai-nightly/linux-64/faiss-proc-1.0.0-cuda.tar.bz2>
13:19:24 Elapsed: 00:00.009541

I have not seen this in other PRs yet so I'm going to rerun tests.

[cjnolet]

rerun tests

[tfeher]

rerun tests

[tfeher]

There is an unrelated error:

cuml/test/test_nearest_neighbors.py::test_ivfpq_pred[8-128-4000-False-4-16-8] Faiss assertion 'err == cudaSuccess' failed in void faiss::gpu::freeMemorySpace(faiss::gpu::MemorySpace, void*) at gpu/utils/MemorySpace.cpp:72; details: Failed to cudaFree pointer 0x7ff740def200 (error 700 an illegal memory access was encountered)
Fatal Python error: Aborted

I will merge branch-0.18 and have another try.