Refactor `rmm::device_scalar` in terms of `rmm::device_uvector` by harrism · Pull Request #789 · rapidsai/rmm

harrism · 2021-06-01T00:25:10Z

This PR refactors device_scalar to use a single-element device_uvector for its storage. This simplifies the implementation of device_scalar. Also changes the API of device_scalar so that its asynchronous / stream-ordered methods use the same API style (with explicit stream parameter) as device_uvector and device_buffer.

Closes #570

This is a breaking change. When it is merged, PRs are likely to need to be merged immediately in other libraries to account for the API changes.

cuDF: Adapt cudf::scalar classes to changes in rmm::device_scalar NVIDIA/cudf#8411
cuGraph: Update to changed rmm::device_scalar API cugraph#1637
RAFT: Pass stream to device_scalar::value() calls. NVIDIA/raft#259
~~cuML~~ (unused)
~~cuSpatial~~ (unused)

…value_to_zero_async

rongou · 2021-06-01T15:23:12Z

-    T const &initial_value,
-    cuda_stream_view stream             = cuda_stream_view{},
+    value_type const &initial_value,
+    cuda_stream_view stream,


Why remove the default? It seems that it would be a common use case to use the default stream, especially with PTDS enabled.

Because we want to make all stream-ordered APIs explicit. See e.g. #418. Note that device_uvector and now device_buffer have no default stream parameters.

rongou · 2021-06-01T15:30:55Z

-    RMM_CUDA_TRY(
-      cudaMemcpyAsync(element_ptr(element_index), &v, sizeof(v), cudaMemcpyDefault, s.value()));
-    s.synchronize_no_throw();
+    if constexpr (std::is_fundamental<value_type>::value) {


Maybe this can be done using SFINAE?

It certainly can, but why would I want to go back to SFINAE? I replaced SFINAE with if constexpr, which allows having a single implementation of this function rather than three!

jrhemstad · 2021-06-02T13:09:31Z

+  using value_type      = T;
+  using reference       = value_type &;
+  using const_reference = value_type const &;
+  using pointer         = value_type *;
+  using const_pointer   = value_type const *;
+  using iterator        = pointer;
+  using const_iterator  = const_pointer;


I would define these in terms of device_uvector. Also, I don't think the iterator typedefs are needed for device_scalar?

Suggested change

using value_type = T;

using reference = value_type &;

using const_reference = value_type const &;

using pointer = value_type *;

using const_pointer = value_type const *;

using iterator = pointer;

using const_iterator = const_pointer;

using value_type = typename device_uvector<T>::value_type;

using reference = typename device_uvector<T>::reference;

using const_reference = typename device_uvector<T>::const_reference;

using pointer = typename device_uvector<T>::pointer;

using const_pointer = typename device_uvector<T>::const_pointer;

I had been thinking about whether it would be useful for device_scalar to provide begin() and end()...

Done, thanks.

…of uvector

harrism · 2021-06-08T03:22:21Z

@gpucibot merge

rapidsai/rmm#789 refactors `rmm::device_scalar`, which all of `cudf::scalar` depends on. Notably, it renames some methods, makes stream parameters explicit, and deletes streamless constructors. As a result, the present PR deletes the default and non-stream copy constructors of all the `cudf::*_scalar` classes. This should be merged immediately after rapidsai/rmm#789 because that PR will break the build. Authors: - Mark Harris (https://github.com/harrism) Approvers: - https://github.com/brandon-b-miller - Vukasin Milovanovic (https://github.com/vuule) - Robert (Bobby) Evans (https://github.com/revans2) - Robert Maynard (https://github.com/robertmaynard) URL: #8411

In comms/test.hpp `device_scalar::value` was not being passed an explicit stream, which means that the default stream was being synced. rapidsai/rmm#789 will remove the default from this parameter, and would have therefore broken the RAFT build. So this PR fixes the oversynchronization and ensures RAFT will build after the RMM PR is merged. Note this PR includes the cmake changes from #258 (just so I could build locally). Once #258 is merged this PR's changes will be simplified. Authors: - Mark Harris (https://github.com/harrism) Approvers: - Dante Gama Dessavre (https://github.com/dantegd) - Chuck Hastings (https://github.com/ChuckHastings) URL: #259

In comms/test.hpp `device_scalar::value` was not being passed an explicit stream, which means that the default stream was being synced. rapidsai/rmm#789 will remove the default from this parameter, and would have therefore broken the RAFT build. So this PR fixes the oversynchronization and ensures RAFT will build after the RMM PR is merged. Note this PR includes the cmake changes from NVIDIA#258 (just so I could build locally). Once NVIDIA#258 is merged this PR's changes will be simplified. Authors: - Mark Harris (https://github.com/harrism) Approvers: - Dante Gama Dessavre (https://github.com/dantegd) - Chuck Hastings (https://github.com/ChuckHastings) URL: NVIDIA#259

harrism added 2 commits May 25, 2021 14:37

Refactor device_scalar in terms of device_uvector

209bc6b

Add device_uvector::set_element_to_zero_async and device_scalar::set_…

6506390

…value_to_zero_async

harrism added breaking Breaking change improvement Improvement / enhancement to an existing function labels Jun 1, 2021

harrism requested a review from jrhemstad June 1, 2021 00:25

harrism self-assigned this Jun 1, 2021

harrism requested a review from a team as a code owner June 1, 2021 00:25

harrism requested a review from rongou June 1, 2021 00:25

github-actions Bot added the cpp Pertains to C++ code label Jun 1, 2021

harrism changed the title ~~Fea device scalar refactor~~ Refactor rmm::device_scalar in terms of rmm::device_uvector Jun 1, 2021

harrism mentioned this pull request Jun 1, 2021

Adapt cudf::scalar classes to changes in rmm::device_scalar NVIDIA/cudf#8411

Merged

rongou reviewed Jun 1, 2021

View reviewed changes

jrhemstad reviewed Jun 2, 2021

View reviewed changes

Comment thread include/rmm/device_scalar.hpp Outdated

jrhemstad reviewed Jun 2, 2021

View reviewed changes

Comment thread include/rmm/device_uvector.hpp Outdated

harrism added 3 commits June 3, 2021 09:15

Merge branch 'branch-21.08' into fea-device_scalar-refactor

898279d

Remove unused param names and define value_type and friends in terms …

bfc4e8a

…of uvector

Remove unused param names in deleted func

4625dff

harrism requested review from jrhemstad and rongou June 2, 2021 23:29

rongou approved these changes Jun 3, 2021

View reviewed changes

jrhemstad approved these changes Jun 3, 2021

View reviewed changes

harrism mentioned this pull request Jun 8, 2021

Pass stream to device_scalar::value() calls. NVIDIA/raft#259

Merged

Merge branch 'branch-21.08' into fea-device_scalar-refactor

683b559

rapids-bot Bot merged commit aa2a2f3 into rapidsai:branch-21.08 Jun 8, 2021

Navigation Menu

Uh oh!

Refactor `rmm::device_scalar` in terms of `rmm::device_uvector` - #789

Refactor `rmm::device_scalar` in terms of `rmm::device_uvector`#789
rapids-bot[bot] merged 6 commits into
rapidsai:branch-21.08from
harrism:fea-device_scalar-refactor

harrism commented Jun 1, 2021 •

edited

Loading

Uh oh!

rongou Jun 1, 2021

Uh oh!

harrism Jun 2, 2021

Uh oh!

rongou Jun 1, 2021

Uh oh!

harrism Jun 2, 2021

Uh oh!

jrhemstad Jun 2, 2021

Uh oh!

harrism Jun 2, 2021

Uh oh!

harrism Jun 2, 2021

Uh oh!

Uh oh!

Uh oh!

harrism commented Jun 8, 2021

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Uh oh!

Conversation

harrism commented Jun 1, 2021 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

rongou Jun 1, 2021

Choose a reason for hiding this comment

Uh oh!

harrism Jun 2, 2021

Choose a reason for hiding this comment

Uh oh!

rongou Jun 1, 2021

Choose a reason for hiding this comment

Uh oh!

harrism Jun 2, 2021

Choose a reason for hiding this comment

Uh oh!

jrhemstad Jun 2, 2021

Choose a reason for hiding this comment

Uh oh!

harrism Jun 2, 2021

Choose a reason for hiding this comment

Uh oh!

harrism Jun 2, 2021

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

harrism commented Jun 8, 2021

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

harrism commented Jun 1, 2021 •

edited

Loading