feat: device registration #103
Merged
Conversation
This file contains hidden or 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
kilinchange
force-pushed
the
device_registry
branch
6 times, most recently
from
ee4e27b to
2e2b4c3
Compare
kilinchange
changed the title
kilinchange
force-pushed
the
device_registry
branch
8 times, most recently
from
c60d854 to
ab3274a
Compare
kilinchange
force-pushed
the
device_registry
branch
4 times, most recently
from
f0b0494 to
2b5ad33
Compare
…tics - Split internal implementation headers into a separate include group - Drop redundant explicit default initialization for Device - Add `impl` suffix to CUDA guard implementation files - Unify Arange initialization via DeviceGuardImpl
kilinchange
force-pushed
the
device_registry
branch
from
eaacac4 to
7a43321
Compare
chen2021673
reviewed
Feb 9, 2026
chen2021673
reviewed
Feb 10, 2026
kilinchange
force-pushed
the
device_registry
branch
from
1f1247f to
e2c91ef
Compare
Chamberlain0w0
requested changes
Feb 11, 2026
| break; | ||
| } | ||
| } | ||
| impl->MemcpyAsync(tensor->DataPtr(), buffer.data(), num_elements * sizeof(float), |
Contributor
There was a problem hiding this comment.
意识到一个问题，这里的 buffer.data() 是 cpu 端的，如果用 cudaMemcpyAsync 的话，确实有可能在实际 memcpy 之前，buffer.data() 就已被释放？H2D/D2H 的 memcpy 是不是都显式同步比较合适？
Collaborator
Author
There was a problem hiding this comment.
新提 pr 统一修复框架内 d2h/h2d 的 host buffer 生命周期问题
kilinchange
force-pushed
the
device_registry
branch
2 times, most recently
from
976a497 to
6613c02
Compare
kilinchange
requested review from
Chamberlain0w0,
JYMiracle305 and
chen2021673
Chamberlain0w0
approved these changes
Feb 11, 2026
kilinchange
force-pushed
the
device_registry
branch
from
6613c02 to
a6e41fb
Compare
- Drop legacy hardware-specific branching - Convert DeviceGuardImpl base methods to fatal-only fallbacks - Explicitly implement supported CPU runtime behaviors - Validate CUDA device type and index bounds in CudaGuardImpl - Widen DeviceCount return type to prevent truncation
kilinchange
force-pushed
the
device_registry
branch
from
a6e41fb to
4f0fa84
Compare
Collaborator
Author
|
JYMiracle305
approved these changes
Feb 11, 2026
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
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.
device 注册设计文档：https://gxtctab8no8.feishu.cn/docx/F0CzdkVXCoaxRgxYc3AcFewtnOc?from=from_copylink
Device 现在是一个只维护 type/index 的简单类型，所有运行时方法都由 DeviceGuard/DeviceGuardImpl 提供。
本次 pr 涉及主要改动：