Add Cython/Python copy_to_host and to_device
#268
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
Pad the size of the memoryview to workaround Cython issue creating size 0 memoryviews. Then trim the added length afterwards. Should allow us to pass size 0 memoryviews to `copy_to_host`.
This should avoid raising unnecessarily when the host and/or device pointers are `nullptr`.
Go ahead and pull these into Cython for simplicity. After all we are basically just calling `cudaMemcpyAsync` at this point.
Creates an analogous method to `copy_to_host` on `DeviceNDArray`s.
pentschev
reviewed
Jan 30, 2020
Is that still not added or did I miss something? |
kkraus14
reviewed
Jan 30, 2020
Co-Authored-By: Peter Andreas Entschev <[email protected]>
jakirkham
commented
Jan 30, 2020
jakirkham
force-pushed
the
add_cy_copy_to_host
branch
from
049bba4 to
38c04b1
Compare
To better align with how other libraries handle streams, skip the synchronization step. This should happen anyways for the default stream (if the user doesn't specify one). Also should give users who want to manage their concurrency more control.
Add a `to_device` function that the static method uses.
jakirkham
changed the title
copy_to_host functioncopy_to_host and to_device functions
jakirkham
changed the title
copy_to_host and to_device functionscopy_to_host and to_device
kkraus14
reviewed
Jan 31, 2020
kkraus14
reviewed
Jan 31, 2020
kkraus14
reviewed
Jan 31, 2020
|
Pushed some fixes. Please take a look and let me know your thoughts 🙂 |
jakirkham
commented
Jan 31, 2020
kkraus14
reviewed
Jan 31, 2020
kkraus14
suggested changes
Jan 31, 2020
kkraus14
added
2 - In Progress
kkraus14
approved these changes
Feb 3, 2020
Also make sure we return the exact same input provided to us (as opposed to a Cython memoryview).
kkraus14
approved these changes
Feb 3, 2020
|
Added docstrings, merged with upstream, and did a little cleanup. Tests passed for me locally. Please let me know if there is anything else needed. Thanks all for the reviews! 😄 |
|
rerun tests |
kkraus14
added
5 - Ready to Merge
This was referenced Feb 4, 2020
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.
Exposes Cython/Python functions for
copy_to_hostandto_devicefor copying from device to host taking only a pointer and abytes-like` buffer. This should help users that have at least a device pointer and the number of bytes of their object to copy it over to host. Hopefully this will allow code constructing temporary device arrays to be removed. As a result this should cutdown on the performance penalty incurred by this construction.Also adds a
copy_to_hostmethod toDeviceBufferto match what Numba does withDeviceNDArrays. Further renamesfrombytestoto_deviceto match Numba. Should make it easier for Python users to leverage this functionality through duck-typing in Python. Thus avoiding handlingDeviceBufferandDeviceNDArraydifferently.Due to changes in the Cython-level, the
copy_to_hostfunction at the C++ level became redundant and has been dropped.