[REVIEW] Using 64-bit array lengths to increase scale of pca & tsvd #3983
Conversation
cjnolet
changed the title
|
I had started going down the path of updating the cublas/cusolver wrappers and then realized some of them are actually failing before the calls to cusolver (though I think the failure manifests during the cusolver calls because of the overflow). |
cpp/src/pca/pca.cuh
Outdated
| auto allocator = handle.get_device_allocator(); | ||
| device_buffer<math_t> components_all(allocator, stream, len); | ||
| device_buffer<math_t> explained_var_all(allocator, stream, prms.n_cols); | ||
| device_buffer<math_t> explained_var_ratio_all(allocator, stream, prms.n_cols); | ||
|
|
||
| <<<<<<< HEAD |
There was a problem hiding this comment.
Straggling line from conflict resolution
cpp/src/pca/pca.cuh
Outdated
| auto allocator = handle.get_device_allocator(); | ||
| device_buffer<math_t> components_all(allocator, stream, len); | ||
| device_buffer<math_t> explained_var_all(allocator, stream, prms.n_cols); | ||
| device_buffer<math_t> explained_var_ratio_all(allocator, stream, prms.n_cols); | ||
|
|
||
| <<<<<<< HEAD | ||
| printf("About to call calEig\n"); |
cpp/src/pca/pca.cuh
Outdated
| calEig<math_t, enum_solver>( | ||
| handle, in, components_all.data(), explained_var_all.data(), prms, stream); | ||
|
|
||
| printf("Called calEig\n"); |
cpp/src/pca/pca.cuh
Outdated
| explained_var_all.data(), prms.n_cols, explained_var, prms.n_components, 1, stream); | ||
| raft::matrix::truncZeroOrigin( | ||
| explained_var_ratio_all.data(), prms.n_cols, explained_var_ratio, prms.n_components, 1, stream); | ||
| ======= |
There was a problem hiding this comment.
Straggling line from conflict resolution
cpp/src/pca/pca.cuh
Outdated
| @@ -60,6 +75,7 @@ void truncCompExpVars(const raft::handle_t& handle, | |||
| explained_var_all.data(), prms.n_cols, explained_var, prms.n_components, 1, stream); | |||
| raft::matrix::truncZeroOrigin( | |||
| explained_var_ratio_all.data(), prms.n_cols, explained_var_ratio, prms.n_components, 1, stream); | |||
| >>>>>>> branch-21.08 | |||
There was a problem hiding this comment.
Straggling line from conflict resolution
|
@gpucibot merge |
Codecov Report
@@ Coverage Diff @@
## branch-21.08 #3983 +/- ##
===============================================
Coverage ? 85.77%
===============================================
Files ? 231
Lines ? 18261
Branches ? 0
===============================================
Hits ? 15664
Misses ? 2597
Partials ? 0
Flags with carried forward coverage won't be shown. Click here to find out more. Continue to review full report at Codecov.
|
…3983) Addresses rapidsai#2459 (likely not all of it) Authors: - Corey J. Nolet (https://github.com/cjnolet) Approvers: - Dante Gama Dessavre (https://github.com/dantegd) URL: rapidsai#3983
Addresses #2459 (likely not all of it)