CMAKE_CUDA_ARCHITECTURES doesn't change when build-system invokes cmake #7579
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
Consider the following: ``` cmake -DCMAKE_CUDA_ARCHITECTURES="" . #build for detected touch <cudf_dir>/cpp/CMakeLists.txt ninja #should be build for detected cmake -DCMAKE_CUDA_ARCHITECTURES= . #build for all touch <cudf_dir>/cpp/CMakeLists.txt ninja #should be build for all cmake -DCMAKE_CUDA_ARCHITECTURES="" . #build for detected touch <cudf_dir>/cpp/CMakeLists.txt ninja #should be build for detected ``` Before these changes the invocations of `ninja` would always go back to building for all when ever `ninja` was invoked. The issue is that once a CMake cache variable exists it can't be removed via `-DCMAKE_CUDA_ARCHITECTURES=` and therefore becomes sticky. To resolve the issue you can now pass `-DCMAKE_CUDA_ARCHITECTURES=ALL` to consistently get all archs, and now the build-system should not change what CUDA archs you are building for.
github-actions
bot
added
CMake
kkraus14
approved these changes
Mar 12, 2021
|
@gpucibot merge |
|
rerun tests |
Anything I can do to resolve that issue from CI? |
Will be fixed in #7580 |
|
rerun tests |
Codecov Report
@@ Coverage Diff @@
## branch-0.19 #7579 +/- ##
===============================================
+ Coverage 81.86% 81.99% +0.12%
===============================================
Files 101 101
Lines 16884 16989 +105
===============================================
+ Hits 13822 13930 +108
+ Misses 3062 3059 -3
Continue to review full report at Codecov.
|
trxcllnt
added a commit
to trxcllnt/node-rapids
that referenced
this pull request
Mar 13, 2021
robertmaynard
added a commit
to trxcllnt/cudf
that referenced
this pull request
Mar 15, 2021
rapids-bot bot
pushed a commit
that referenced
this pull request
Mar 15, 2021
Fixes regression from #7579 in auto-detecting GPU architectures when `-DCMAKE_CUDA_ARCHITECTURES=` is passed on the CLI. Now that the cached `CMAKE_CUDA_ARCHITECTURES` isn't unset before calling `enable_language(CUDA)`, this call throws an error and configuration fails. This change ensures we call `enable_language(CUDA)` after any potential rewrites of `CMAKE_CUDA_ARCHITECTURES`. This PR also aligns with RMM's `EvalGPUArchs.cmake` logic and prints `SUPPORTED_CUDA_ARCHITECTURES` instead of `"ALL"` in the case the current machine is a CPU-only node. Related: rapidsai/rmm#727 Authors: - Paul Taylor (@trxcllnt) - Robert Maynard (@robertmaynard) Approvers: - Keith Kraus (@kkraus14) URL: #7593
hyperbolic2346
pushed a commit
to hyperbolic2346/cudf
that referenced
this pull request
Mar 25, 2021
…ke (rapidsai#7579) Consider the following: ``` cmake -DCMAKE_CUDA_ARCHITECTURES="" . #build for detected touch <cudf_dir>/cpp/CMakeLists.txt ninja #should be build for detected cmake -DCMAKE_CUDA_ARCHITECTURES= . #build for all touch <cudf_dir>/cpp/CMakeLists.txt ninja #should be build for all cmake -DCMAKE_CUDA_ARCHITECTURES="" . #build for detected touch <cudf_dir>/cpp/CMakeLists.txt ninja #should be build for detected ``` Before these changes the invocations of `ninja` would always go back to building for all when ever `ninja` was invoked. The issue is that once a CMake cache variable exists it can't be removed via `-DCMAKE_CUDA_ARCHITECTURES=` and therefore becomes sticky. To resolve the issue you can now pass `-DCMAKE_CUDA_ARCHITECTURES=ALL` to consistently get all archs, and now the build-system should not change what CUDA archs you are building for. Authors: - Robert Maynard (@robertmaynard) Approvers: - Keith Kraus (@kkraus14) URL: rapidsai#7579
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.
Consider the following:
Before these changes the invocations of
ninjawould alwaysgo back to building for all when ever
ninjawas invoked. The issueis that once a CMake cache variable exists it can't be removed
via
-DCMAKE_CUDA_ARCHITECTURES=and therefore becomes sticky.To resolve the issue you can now pass
-DCMAKE_CUDA_ARCHITECTURES=ALLto consistently get all archs, and now the build-system should not
change what CUDA archs you are building for.