Add a new CUDA Next library and a first entry in it with hierarchy_dimensions type template #1485
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
miscco
added
feature request
miscco
reviewed
Mar 5, 2024
Fix missing _CCCL_HOST_DEVICE and add missing ones Change index_type to dimensions_index_type License year fixes
Add cmake files for package management, for now hardcode version 0.1 Make CCCL package include CUDA_Next Add C++17 target properties to CUDA_Next
Co-authored-by: Michael Schellenberger Costa <[email protected]>
1 task
|
Some assumptions I'm making as I update the CMake. Let me know if anything here is incorrect (@jrhemstad for vis):
1b and 1c are not consistent for our other projects for legacy reasons, so I want to make sure we stay consistent with this one. |
|
@alliepiper Regarding 1. I think it's a good name, I started with CUDA_Next, but CudaNext seems better. |
ahendriksen
requested changes
Mar 12, 2024
There was a problem hiding this comment.
Very nice! I found one typo in the README.
Over-arching comment: I think to accurately estimate the usefulness of an experimental feature, it has to be documented well. Do we have an idea how we want to document the CUDA next library? Can we add that as part of this PR or as part of a next PR?
|
Another question: I'm guessing that we'll want to install these headers under a prefix, e.g. I think |
I agree that |
Co-authored-by: Allard Hendriksen <[email protected]>
|
@pciolkosz I've noticed that the current tests have a couple of issues:
I can WAR this for now by unsetting the build type for CudaNext, but this should be fixed before we merged, since it usually means there's some UB somewhere. The runtime failure is: To save some time, CUDA error 98 is "invalid device function". Maybe it's related to the lambda being used? I have verified that |
|
/ok to test |
|
/ok to test |
[skip-tests] since this is only changing infra.
|
/ok to test |
25 tasks
🟩 CI Results [ Failed: 0 | Passed: 365 | Total: 365 ]
|
| # | Runner |
|---|---|
| 275 | linux-amd64-cpu16 |
| 40 | linux-amd64-gpu-v100-latest-1 |
| 28 | linux-arm64-cpu16 |
| 22 | windows-amd64-cpu16 |
👃 Inspect Changes
Modifications in project?
| Project | |
|---|---|
| +/- | CCCL Infrastructure |
| libcu++ | |
| +/- | CUB |
| Thrust | |
| +/- | CUDA Experimental |
Modifications in project or dependencies?
| Project | |
|---|---|
| +/- | CCCL Infrastructure |
| +/- | libcu++ |
| +/- | CUB |
| +/- | Thrust |
| +/- | CUDA Experimental |
|
pre-commit.ci autofix |
|
pre-commit.ci autofix |
|
/ok to test |
🟩 CI Results [ Failed: 0 | Passed: 365 | Total: 365 ]
|
| # | Runner |
|---|---|
| 275 | linux-amd64-cpu16 |
| 40 | linux-amd64-gpu-v100-latest-1 |
| 28 | linux-arm64-cpu16 |
| 22 | windows-amd64-cpu16 |
👃 Inspect Changes
Modifications in project?
| Project | |
|---|---|
| +/- | CCCL Infrastructure |
| libcu++ | |
| CUB | |
| Thrust | |
| +/- | CUDA Experimental |
Modifications in project or dependencies?
| Project | |
|---|---|
| +/- | CCCL Infrastructure |
| +/- | libcu++ |
| +/- | CUB |
| +/- | Thrust |
| +/- | CUDA Experimental |
alliepiper
approved these changes
May 9, 2024
|
The infrastructure is complete and CI is ready but disabled -- once the code is ready, just add it back in and uncomment the |
|
/ok to test |
🟨 CI Results [ Failed: 1 | Passed: 305 | Total: 306 ]
|
| # | Runner |
|---|---|
| 232 | linux-amd64-cpu16 |
| 32 | linux-amd64-gpu-v100-latest-1 |
| 24 | linux-arm64-cpu16 |
| 18 | windows-amd64-cpu16 |
👃 Inspect Changes
Modifications in project?
| Project | |
|---|---|
| +/- | CCCL Infrastructure |
| libcu++ | |
| CUB | |
| Thrust | |
| +/- | CUDA Experimental |
Modifications in project or dependencies?
| Project | |
|---|---|
| +/- | CCCL Infrastructure |
| +/- | libcu++ |
| +/- | CUB |
| +/- | Thrust |
| +/- | CUDA Experimental |
🟩 CI Results [ Failed: 0 | Passed: 306 | Total: 306 ]
|
| # | Runner |
|---|---|
| 232 | linux-amd64-cpu16 |
| 32 | linux-amd64-gpu-v100-latest-1 |
| 24 | linux-arm64-cpu16 |
| 18 | windows-amd64-cpu16 |
👃 Inspect Changes
Modifications in project?
| Project | |
|---|---|
| +/- | CCCL Infrastructure |
| libcu++ | |
| CUB | |
| Thrust | |
| +/- | CUDA Experimental |
Modifications in project or dependencies?
| Project | |
|---|---|
| +/- | CCCL Infrastructure |
| +/- | libcu++ |
| +/- | CUB |
| +/- | Thrust |
| +/- | CUDA Experimental |
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.
This change introduces a new CUDA Next library that aims to be a collection of experimental features. It would allow CUDA users to test and provide feedback on those features before they end up in other libraries in a more stable manner.
This change adds license file and some minimal cmake bits that needs to be extended in future changes.
The first entry in the CUDA Next library added in this pull request is hierarchy_dimensions type template that allows to describe a hierarchy of CUDA threads with a mix of static and dynamic information. It can be used to pass into kernels and then calculate aggregates at compile time (like count threads in each CUDA block to create a statically sized array). It can also be used to make libraries aware of the shape of currently running grid and optimize some thread id calculation with compile time values.
hierarchy_dimensions type template is basically a tuple of level_dimensions entries that describe levels. Each level consist of two things, first one is a type to describe what that level it is, for example block_level, cluster_level, etc. Second one is cuda::std::extents object to describe dimensions of that level with both static and dynamic values, the same way it describes cuda::std::mdspan objects.
This is an initial implementation, there is a number of TODOs, names are not final and the interface can change depending on the feedback received.
This type is is also a building block for other libraries and features that are in the pipeline.
NVBUG 4541889