Replace remaining instances of rmm::cuda_stream_default with cudf::default_stream_value

[jbrennan333]

Closes #10864.
Also closes #9614.

This PR is a follow-up to #10877. It replaces all of the remaining instances of rmm::cuda_stream_default with cudf::default_stream_value.

There are a lot of replacements and addition of includes, along with some reformatting due to clang-format, but like #10877, there should be no noticeable functional change here.

[jbrennan333]

I should clarify that when cuDF is compiled with LIBCUDF_USE_PER_THREAD_DEFAULT_STREAM, this does ensure that any calls to outside libraries that might not have been compiled for PTDS will be called with the per thread default stream instead of the default stream, so in that respect it is a functional change. All known instances where we needed that change were addressed in #10877, specifically code that called into nvcomp-2.3.

[jbrennan333]

It looks like the gpuCI/cudf/build/cuda-arm64/11.5 build has been broken for a while?

02:04:20 Version of installed CUDA didn't match package
02:04:20 Build step 'Execute shell' marked build as failure
02:04:20 [Set GitHub commit status (universal)] ERROR on repos [GHRepository@25f3499f[nodeId=MDEwOlJlcG9zaXRvcnk5MDUwNjkxOA==,description=cuDF - GPU DataFrame Library ,homepage=[http://rapids.ai,name=cudf,fork=false,archived=false,visibility=public,size=100218,milestones={},language=C++,commits={},source](http://rapids.ai%2Cname%3Dcudf%2Cfork%3Dfalse%2Carchived%3Dfalse%2Cvisibility%3Dpublic%2Csize%3D100218%2Cmilestones%3D%7B%7D%2Clanguage%3Dc++%2Ccommits%3D%7B%7D%2Csource/)=<null>,parent=<null>,isTemplate=false,compareUsePaginatedCommits=false,url=https://api.github.com/repos/rapidsai/cudf,id=90506918,nodeId=<null>,createdAt=2017-05-07T03:43:37Z,updatedAt=2022-06-07T20:05:49Z]] (sha:af5df59) with context:gpuCI/cudf/build/cuda-arm64/11.5

[jbrennan333]

I believe the CI failures are unrelated to my changes, so I am going to open this up for review.

[karthikeyann]

Here are few instances of potential replacement.

cpp/src/column/column.cu:185:  rmm::cuda_stream_view stream{};
cpp/src/io/utilities/hostdevice_vector.hpp:164:  rmm::cuda_stream_view stream{};
cpp/src/io/csv/reader_impl.cu:91:  selected_rows_offsets(rmm::cuda_stream_view stream) : all{0, stream}, selected{all} {}
cpp/tests/column/column_device_view_test.cu:39:  rmm::cuda_stream_view stream{};
cpp/tests/column/column_device_view_test.cu:57:  rmm::cuda_stream_view stream{};
cpp/tests/table/experimental_row_operator_tests.cu:53:  rmm::cuda_stream_view stream{};
cpp/tests/table/experimental_row_operator_tests.cu:76:  rmm::cuda_stream_view stream{};
cpp/tests/table/experimental_row_operator_tests.cu:101:  rmm::cuda_stream_view stream{};
cpp/tests/table/experimental_row_operator_tests.cu:125:  rmm::cuda_stream_view stream{};
cpp/tests/table/table_view_tests.cu:45:  rmm::cuda_stream_view stream{};
cpp/include/cudf/column/column.hpp:182:                     rmm::cuda_stream_view stream = rmm::cuda_stream_view{});
cpp/benchmarks/synchronization/synchronization.hpp:36:        rmm::cuda_stream_view stream{}; // default stream, could be another stream

Also, unused stream member in cpp/src/transform/row_bit_count.cu:164

[vuule]

Could we patch rmm locally to throw if a stream view of a default stream is created and run tests? Looks like it's easy to miss a few when checking manually.

[codecov]

Codecov Report

❗ No coverage uploaded for pull request base (branch-22.08@ef6a390). Click here to learn what that means.
The diff coverage is n/a.

@@               Coverage Diff               @@
##             branch-22.08   #11082   +/-   ##
===============================================
  Coverage                ?   86.34%           
===============================================
  Files                   ?      144           
  Lines                   ?    22729           
  Branches                ?        0           
===============================================
  Hits                    ?    19625           
  Misses                  ?     3104           
  Partials                ?        0           

---

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 ef6a390...186a4ef. Read the comment docs.

[ttnghia]

This change is huge. Assuming that there are a lot of other PRs that are under review, I wonder if after merging this PR then after the pending PRs are merged some will reverse back this result.

It is ideal to merge this lastly, after all other PRs, right before code freeze.

[jbrennan333]

This change is huge. Assuming that there are a lot of other PRs that are under review, I wonder if after merging this PR then after the pending PRs are merged some will reverse back this result.
It is ideal to merge this lastly, after all other PRs, right before code freeze.

Thanks for the comment @ttnghia. We actually put off doing this fully at the end of the 22.06 sprint, and instead only put in the changes needed to fix nvcomp (#10877). I didn't want to make such a huge change at the end of the sprint. Even though the change is low risk, there is still a chance to introduce bugs.
I would prefer to go ahead with this one now. I will take the action to check again closer to the end of the 22.08 and put up another PR if needed.

[jbrennan333]

Could we patch rmm locally to throw if a stream view of a default stream is created and run tests? Looks like it's easy to miss a few when checking manually.

Thanks for the suggestion @vuule. I will try this.

[jbrennan333]

rerun tests

[ttnghia]

Rerun tests.

[jbrennan333]

@karthikeyann I made the changes you requested. Thanks again for the review!

[jbrennan333]

Thanks @karthikeyann, @nvdbaranec, @ttnghia and @vuule for the reviews! Can someone please merge this for me?

[ttnghia]

@gpucibot merge