-
Notifications
You must be signed in to change notification settings - Fork 915
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Use cuFile direct device reads/writes by default in cuIO #9722
Conversation
Codecov Report
@@ Coverage Diff @@
## branch-22.02 #9722 +/- ##
===============================================
Coverage ? 10.49%
===============================================
Files ? 119
Lines ? 20305
Branches ? 0
===============================================
Hits ? 2130
Misses ? 18175
Partials ? 0 Continue to review full report at Codecov.
|
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Seems changing the default for LIBCUDF_CUFILE_POLICY
would qualify this as a breaking change?
while (std::getline(user_config_file, line)) { | ||
std::string const tag = "\"allow_compat_mode\""; | ||
if (line.find(tag) != std::string::npos) { | ||
// TODO: only replace the true/false value |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I feel like the fact the file is overwritten and not actually modified should be documented in the .rst
somehow.
FYI -- I recently upgraded my dev environment to CUDA 11.5 via the runfile and it seems like neither the runfile nor conda packages (either provided by NVIDIA or in conda-forge) currently include |
I don't know. The change does not break any code. I'm okay with changing the label (whose description, btw, is not helpful here :D). |
I don't see how this would break cudf for users. This PR does not make cufile a requirement. If the headers are not present during compilation, we will compile without any cufile capability. If the library is not present at runtime, we will detect an issue when loading it and fall back to the host path. Even if something goes wrong when opening a file with cufile, we fall back to the host path. Thank you for bringing this up, will look into whether cufile should be included in the packages. |
Co-authored-by: David Wendt <[email protected]>
Apologies, I misread the code. Awesome to see this moving forward 😄 |
I thought 'breaking' was a way to communicate that we are changing a default or a previously documented behavior. |
…nto fea-use-gds-default
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Happy to see this happening! Mostly looks good, just a few minor suggestions before we push this out.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Overall this looks good! I have some comments on docs. 📜
Co-authored-by: Bradley Dice <[email protected]> Co-authored-by: Vyas Ramasubramani <[email protected]>
Co-authored-by: Bradley Dice <[email protected]>
Co-authored-by: Bradley Dice <[email protected]>
…nto fea-use-gds-default
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Two minor typos, otherwise LGTM. 👍
Co-authored-by: Bradley Dice <[email protected]>
@gpucibot merge |
Signed-off-by: Peixin Li <[email protected]> related to #9722 skip cufile test in JNI build while we have a separate pipeline for GDS testing Authors: - Peixin (https://github.com/pxLi) Approvers: - Tim Liu (https://github.com/NvTimLiu) - Gary Shen (https://github.com/GaryShen2008) URL: #9744
Making this change early in 22.02 to test through internal use + nightly builds before the release.