Skip to content

Tracking memory resources#3

Open
achirkin wants to merge 13 commits intofea-unify-memory-resourcesfrom
fea-tracking-memory-resources
Open

Tracking memory resources#3
achirkin wants to merge 13 commits intofea-unify-memory-resourcesfrom
fea-tracking-memory-resources

Conversation

@achirkin
Copy link
Owner

@achirkin achirkin commented Mar 4, 2026

This PR is to show an isolated set of changes for rapidsai#2973

achirkin and others added 12 commits March 5, 2026 05:18
This option allows generating dependencies without `libucx` in the dependencies list, which is something we have to do for NVAIE/DLFW builds.

Authors:
  - Paul Taylor (https://github.com/trxcllnt)

Approvers:
  - James Lamb (https://github.com/jameslamb)

URL: rapidsai#2975
…apidsai#2974)

The per-row offset `l_offset = (offset + batch_id) * len` was stored as `IdxT`, which silently overflows when the total matrix size exceeds the range of a 32-bit index type. Both `offset` and `batch_id` are already `size_t`, so the multiplication naturally produces a `size_t`, but truncating it back to `IdxT` caused incorrect pointer arithmetic in the kernels.

Introduced a layout-policy abstraction (`dense_layout` / `csr_layout`) in a new header `select_k_layout.cuh`. This replaced the `len_or_indptr` boolean template parameter, to improve the API and push the related computations to compile-time for all select-k kernels.

Authors:
  - Yan Zaretskiy (https://github.com/yan-zaretskiy)

Approvers:
  - Artem M. Chirkin (https://github.com/achirkin)

URL: rapidsai#2974
This PR updates the repository to version 26.06.

This is part of the 26.04 release burndown process.
Fixes these `pre-commit` errors

```text
In file RAPIDS_BRANCH:1:9:
 release/26.04
warning: do not hard-code version, read from VERSION file instead

In file RAPIDS_BRANCH:1:9:
 release/26.04

verify-hardcoded-version-ucxx............................................Failed
- hook id: verify-hardcoded-version
- exit code: 1

In file UCXX_BRANCH:1:9:
 release/0.49
warning: do not hard-code version, read from UCXX_VERSION file instead

In file UCXX_BRANCH:1:9:
 release/0.49
```

See rapidsai/pre-commit-hooks#121 for details

Authors:
  - James Lamb (https://github.com/jameslamb)

Approvers:
  - Kyle Edwards (https://github.com/KyleFromNVIDIA)

URL: rapidsai#2980
Forward-merge release/26.04 into main
Use `cuda::mr::any_synchronous_resource` for host, pinned, and managed resource types and give the user explicit control for host, pinned, and managed resources.

#### New
  - `raft::resource::managed_memory_resource` and `raft::resource::pinned_memory_resource` are passed to managed and pinned mdarrays during construction via corresponding container policies. This allows the user to replace/modify these resources, for example, to add logging or memory pooling.
  - `raft::mr::get_default_host_resource` and `raft::mr::set_default_host_resource` can be used by the user to alter the default host resource the same way. It is not stored in `raft::resources` handle like the other two for two reasons:
    1. To mirror rmm default device resource getter/setter
    2. To avoid breaking the `raft::make_host_mdarray` overloads that do not take `raft::resources` as an argument (many instances across raft and cuvs).

#### Changed

 - Use `raft::mr::host_resource_ref` and `raft::mr::host_device_resource_ref` for the non-owning semantics (defined as `cuda::mr::synchronous_resource_ref` with appropriate access attributes)
 - Use `raft::host_resource` and `raft::host_device_resource` for owning semantics (defined as `cuda::mr::any_synchronous_resource` with appropriate access attributes)

With these changes, raft fully switches to `cuda::mr` types for host and host-device resources, while still using `rmm` types for device async resources. Changing the latter would break a lot of cuVS and is not needed - `rmm` will eventually fully converge to `cuda::mr` anyway.

#### Breaking changes
  - Rename container policies
  - Reuse of a single `host_container` for the three types of resources.
  - Switch to using `cuda::mr::any_synchronous_resource` from `std::pmr::memory_resource`

The effect of this changes should be limited, because the policies are hidden behind the mdarray templates and synonyms and the  `std::pmr::memory_resource` was introduced recently and haven't been used much.

Authors:
  - Artem M. Chirkin (https://github.com/achirkin)

Approvers:
  - Bradley Dice (https://github.com/bdice)
  - Tamas Bela Feher (https://github.com/tfeher)

URL: rapidsai#2968
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

6 participants