Support for fp16 in CAGRA and IVF-PQ #2085
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
achirkin
added
feature request
…n the raft_objs component
|
NB: this does not add the fp16 capabilities to the |
achirkin
added
3 - Ready for Review
and removed
2 - In Progress
tfeher
approved these changes
Jan 17, 2024
There was a problem hiding this comment.
Thanks @achirkin for this PR!
We have here a large amount of boilerplate code. Fortunately the nontrivial changes are relatively small, and are confined to
- mdspan_numpy_serializer.hpp
- device_load_stores.cuh
- test/neighbors/ann_cagra.cuh
The PR looks good to me!
| @@ -148,6 +149,26 @@ DI void sts(int32_t* addr, const int32_t (&x)[4]) | |||
| : "l"(s4), "r"(x[0]), "r"(x[1]), "r"(x[2]), "r"(x[3])); | |||
| } | |||
|
|
|||
| DI void sts(half* addr, const half& x) | |||
There was a problem hiding this comment.
Tagging @mdoijade to have a look at the changes in this file, since the load and store ops here are mostly used by IVF-Flat and contractions.cuh.
There was a problem hiding this comment.
The additions to device_loads_stores.cuh looks good to me, I agree it is good to have matching sts function call for lds for larger fp16 vector sizes.
enp1s0
reviewed
Jan 18, 2024
Co-authored-by: tsuki <12711693+enp1s0@users.noreply.github.com>
Co-authored-by: tsuki <12711693+enp1s0@users.noreply.github.com>
Co-authored-by: tsuki <12711693+enp1s0@users.noreply.github.com>
enp1s0
approved these changes
Jan 19, 2024
|
/merge |
cjnolet
added a commit
to cjnolet/raft
that referenced
this pull request
Feb 8, 2024
This reverts commit 72f48ae.
raydouglass
pushed a commit
that referenced
this pull request
Feb 9, 2024
RAFT C++ tests were not running for a portion of the 24.02 development cycle, until the merger of rapidsai/rapids-cmake#533. This PR fixes some failing tests and reverts PRs that caused test failures that were silent until now, specifically #2097 and #2085. These features will be revisited in a subsequent release. Authors: - Malte Förster (https://github.com/mfoerste4) - Corey J. Nolet (https://github.com/cjnolet) Approvers: - Ben Frederickson (https://github.com/benfred) - Bradley Dice (https://github.com/bdice)
achirkin
added a commit
to achirkin/raft
that referenced
this pull request
Feb 11, 2024
Add fp16 (CUDA half) support to CAGRA and its dependencies. Authors: - Artem M. Chirkin (https://github.com/achirkin) Approvers: - Tamas Bela Feher (https://github.com/tfeher) - tsuki (https://github.com/enp1s0) URL: rapidsai#2085
rapids-bot bot
pushed a commit
that referenced
this pull request
Feb 13, 2024
1. Add fp16 (CUDA half) support to CAGRA and its dependencies (#2085). 2. Fix the shared memory size error in the ivf-flat that got exposed by new tests in #2085. Regarding the point (2): Warp-sort top-k queue uses shared memory; the module provides the required shmem size calculation function decoupled from the queue object itself. As a result, it's easy to plug-in wrong types and get the calculation incorrectly. IVF-Flat scan kernel always kept the distances in the queue as floats, but we calculated the shmem size as if it used `AccT` (IVF-Flat's internal accumulation type). Hence, with adding the tests with fp16 inputs (and `AccT`), the allocated shmem became too small, which resulted in memory access violation errors. Authors: - Artem M. Chirkin (https://github.com/achirkin) Approvers: - Ben Frederickson (https://github.com/benfred) URL: #2172
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Add fp16 (CUDA half) support to CAGRA and its dependencies.