Skip to content

Add fallback memory resource for TCC devices#257

Merged
leofang merged 14 commits intomainfrom
ksimpson/tcc_memory_resource
Dec 6, 2024
Merged

Add fallback memory resource for TCC devices#257
leofang merged 14 commits intomainfrom
ksimpson/tcc_memory_resource

Conversation

@keenan-simpson
Copy link
Contributor

@keenan-simpson keenan-simpson commented Nov 28, 2024 •
edited
Loading

For devices which don't support memory pools, we need to provide an alternate default memory resource.

This basic WAR implementation works. I used a colossus lease for a Tesla T4 on Friday Nov 29 and these were the results:

using the DefaultAsyncMempool --> python - m pytest tests/test_memory.py
=============================================== short test summary info ===============================================
FAILED tests/test_memory.py::test_buffer_initialization - cuda.core.experimental._utils.CUDAError: CUDA_ERROR_NOT_SUPPORTED: operation not supported

using the implementation in this branch --> python - m pytest tests/test_memory.py
collected 4 items

tests\test_memory.py .... (SUCCESS)

close #208

@keenan-simpson keenan-simpson added P0 High priority - Must do! feature New feature or request cuda.core Everything related to the cuda.core module labels Nov 28, 2024
@keenan-simpson keenan-simpson added this to the cuda.core beta 2 milestone Nov 28, 2024
@keenan-simpson keenan-simpson self-assigned this Nov 28, 2024
@copy-pr-bot
Copy link
Contributor

copy-pr-bot bot commented Nov 28, 2024

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

@keenan-simpson keenan-simpson added bug Something isn't working and removed feature New feature or request labels Nov 28, 2024
@keenan-simpson keenan-simpson force-pushed the ksimpson/tcc_memory_resource branch from 42346c5 to 319a372 Compare November 29, 2024 19:17
leofang
leofang requested changes Nov 30, 2024
View reviewed changes
@keenan-simpson keenan-simpson marked this pull request as ready for review December 3, 2024 21:26
leofang
leofang reviewed Dec 4, 2024
View reviewed changes
leofang
leofang requested changes Dec 4, 2024
View reviewed changes
@keenan-simpson
Copy link
Contributor Author

keenan-simpson commented Dec 4, 2024

re-tested using synchronous malloc and free on Tesla T4 colossus instance.

on main branch:

(test_env) C:\cuda-python\cuda_core>python -m pytest tests\test_memory.py

platform win32 -- Python 3.12.7, pytest-8.3.4, pluggy-1.5.0
rootdir: C:\cuda-python\cuda_core
configfile: pyproject.toml
collected 4 items

tests\test_memory.py FFFF [100%]

with change:

(test_env) C:\cuda-python\cuda_core>python -m pytest tests\test_memory.py

platform win32 -- Python 3.12.7, pytest-8.3.4, pluggy-1.5.0
rootdir: C:\cuda-python\cuda_core
configfile: pyproject.toml
collected 4 items

tests\test_memory.py .... [100%]

leofang
leofang reviewed Dec 6, 2024
View reviewed changes
@leofang
Copy link
Member

leofang commented Dec 6, 2024

/ok to test

@leofang
Copy link
Member

leofang commented Dec 6, 2024

Windows failures are known (#271) and irrelevant. Let's merge. Thanks, Keenan!

@leofang leofang merged commit c6a1a94 into main Dec 6, 2024
@leofang leofang deleted the ksimpson/tcc_memory_resource branch December 6, 2024 22:04
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@leofang leofang leofang approved these changes

@vzhurba01 vzhurba01 Awaiting requested review from vzhurba01

Assignees

@keenan-simpson keenan-simpson

Labels

bug Something isn't working cuda.core Everything related to the cuda.core module P0 High priority - Must do!

Projects

None yet

Milestone

cuda.core beta 2

Development

Successfully merging this pull request may close these issues.

Check TCC/WDDM modes during Device initialization

2 participants

@keenan-simpson @leofang