-
Notifications
You must be signed in to change notification settings - Fork 900
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
Replace defaulted stream value for libcudf APIs that use NVCOMP #10877
Replace defaulted stream value for libcudf APIs that use NVCOMP #10877
Conversation
I have verified that this is working in spark by building along with #10851 and then building a spark-rapids-jni and spark-rapids. I then ran a query and verified that ptds is being used. I think this is ready for review. |
This checkstyle failure does not appear to be due to these changes:
Should I fix the meta.yaml file? |
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.
One small comment.
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.
cpp changes look good.
Codecov Report
@@ Coverage Diff @@
## branch-22.06 #10877 +/- ##
================================================
+ Coverage 86.30% 86.32% +0.02%
================================================
Files 144 144
Lines 22665 22668 +3
================================================
+ Hits 19560 19569 +9
+ Misses 3105 3099 -6
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.
Ignore me.
…writer calls that use nvcomp
I have updated based on review comments from @jrhemstad and @vuule. I really appreciate all the reviews/comments. |
@jrhemstad, @vuule, this is ready for another review. |
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.
cuIO changes look good, thank you for addressing the feedback.
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.
CMake changes LGTM
Since we are going forward with including nvcomp-2.3 in 22.06, I think this can be merged now. |
@gpucibot merge |
…fault_stream_value (#11082) Closes #10864. Also closes #9614. This PR is a follow-up to #10877. It replaces all of the remaining instances of `rmm::cuda_stream_default` with `cudf::default_stream_value`. There are a lot of replacements and addition of includes, along with some reformatting due to clang-format, but like #10877, there should be no noticeable functional change here. Authors: - Jim Brennan (https://github.com/jbrennan333) Approvers: - Robert (Bobby) Evans (https://github.com/revans2) - https://github.com/nvdbaranec - Nghia Truong (https://github.com/ttnghia) - Karthikeyan (https://github.com/karthikeyann) URL: #11082
Closes #10862. The `PER_THREAD_DEFAULT_STREAM` build option was deprecated in branch-22.06 via #10877, and replaced with the new build option `CUDF_USE_PER_THREAD_DEFAULT_STREAM`. This PR removes `PER_THREAD_DEFAULT_STREAM`. I am putting this up as a draft because I am not certain if we want to do this in 22.08 or wait for a later release? Authors: - Jim Brennan (https://github.com/jbrennan333) Approvers: - Jason Lowe (https://github.com/jlowe) - Vyas Ramasubramani (https://github.com/vyasr) URL: #11134
This PR addresses #10862, but does not completely remove
PER_THREAD_DEFAULT_STREAM
. I just adds the new defineCUDF_USE_PER_THREAD_DEFAULT_STREAM
and adds a deprecation warning forPER_THREAD_DEFAULT_STREAM
.This PR also addresses #10864, but only changes the default parameter for
read_avro
andread_parquet
tocudf::default_stream_value
, which is defined asrmm::cuda_stream_per_thread
whenCUDF_USE_PER_THREAD_DEFAULT_STREAM
is defined. These cover the current interfaces to nvcomp.The goal for this PR is to ensure that we pass
rmm::cuda_stream_per_thread
to the nvcomp apis whenCUDF_USE_PER_THREAD_DEFAULT_STREAM
is defined. This is needed for nvcomp-2.3, which is provided as dynamic libraries where we cannot recompile with PTDS enabled. nvcomp-2.3 is being enabled in #10851.I have not marked this PR as closing the above issues, because we will need to follow up in a future PR to remove
PER_THREAD_DEFAULT_STREAM
and in another to replace all of the rest of the cuDF API call sites to use the newcudf::default_stream_value
.i am putting this up as a draft PR because I have not tested with nvcomp-2.3 yet.