Fixed cling argument --cuda-path #6707

SimeonEhrig · 2020-10-26T14:16:32Z

The cling argument --cuda-path is necessary if the CUDA SDK is not installed under /usr/local/cuda, e.g. on HPC systems. The integration tests are also updated to handle a CUDA SDK under a non-standard location.

Additional diagnostic

To find the bug, I also add some diagnostic functions for the PTX compiler.

Now error messages are prefixed in CUDA mode to better decide which of the two compiler pipelines is causing the error.

example

- example before:
error: cannot find CUDA installation.  Provide its path via --cuda-path, or pass -nocudainc to build without CUDA includes.
error: cannot find libdevice for sm_20. Provide path to different CUDA installation via --cuda-path, or pass -nocudalib to build without linking with libdevice.
error: cannot find CUDA installation.  Provide its path via --cuda-path, or pass -nocudainc to build without CUDA includes.

- example after:
cling: error: cannot find CUDA installation.  Provide its path via --cuda-path, or pass -nocudainc to build without CUDA includes.
cling-ptx: error: cannot find libdevice for sm_20. Provide path to different CUDA installation via --cuda-path, or pass -nocudalib to build without linking with libdevice.
cling-ptx: error: cannot find CUDA installation.  Provide its path via --cuda-path, or pass -nocudainc to build without CUDA includes.

Now, the class cudaIncrementalDeviceCompiler available through reflection via the gCling object.

phsft-bot · 2020-10-26T14:16:44Z

Starting build on ROOT-debian10-i386/cxx14, ROOT-performance-centos8-multicore/default, ROOT-fedora30/cxx14, ROOT-fedora31/noimt, ROOT-ubuntu16/nortcxxmod, mac1014/python3, mac1015/cxx17, windows10/cxx14
How to customize builds

phsft-bot · 2020-10-26T14:31:39Z

Build failed on ROOT-fedora31/noimt.
Running on root-fedora-31-2.cern.ch:/home/sftnight/build/workspace/root-pullrequests-build
See console output.

Failing tests:

phsft-bot · 2020-10-26T14:36:57Z

Build failed on ROOT-performance-centos8-multicore/default.
Running on null:/data/sftnight/workspace/root-pullrequests-build
See console output.

Failing tests:

phsft-bot · 2020-10-26T14:39:22Z

Build failed on ROOT-fedora30/cxx14.
Running on root-fedora30-1.cern.ch:/build/workspace/root-pullrequests-build
See console output.

Failing tests:

phsft-bot · 2020-10-26T14:40:04Z

Build failed on mac1015/cxx17.
Running on macphsft18.dyndns.cern.ch:/Users/sftnight/build/jenkins/workspace/root-pullrequests-build
See console output.

Failing tests:

phsft-bot · 2020-10-26T14:47:59Z

Build failed on ROOT-ubuntu16/nortcxxmod.
Running on sft-ubuntu-1604-1.cern.ch:/build/workspace/root-pullrequests-build
See console output.

Failing tests:

phsft-bot · 2020-10-26T14:53:25Z

Build failed on mac1014/python3.
Running on macphsft17.dyndns.cern.ch:/build/jenkins/workspace/root-pullrequests-build
See console output.

Failing tests:

phsft-bot · 2020-10-26T15:03:06Z

Build failed on ROOT-debian10-i386/cxx14.
Running on pcepsft11.dyndns.cern.ch:/home/sftnight/build/workspace/root-pullrequests-build
See console output.

Failing tests:

Axel-Naumann · 2020-10-30T14:11:23Z

@SimeonEhrig great, thanks! Could you check out https://github.com/root-project/roottest.git and modify the expected diagnostics (*.ref files in the corresponding directories)? You can create a PR for roottest, and if you use the same branch name as for this PR, this PR will pick it up when testing! Happy to explain offline if this is too complex / convoluted :-)

SimeonEhrig · 2020-11-02T14:02:40Z

@SimeonEhrig great, thanks! Could you check out https://github.com/root-project/roottest.git and modify the expected diagnostics (*.ref files in the corresponding directories)? You can create a PR for roottest, and if you use the same branch name as for this PR, this PR will pick it up when testing! Happy to explain offline if this is too complex / convoluted :-)

I think I understand what you want. Unfortunately I have no idea what is causing the errors. The tests have nothing to do with my changes and I cannot reproduce the error on my system. I think the test scripts have problems catching the expected errors.

phsft-bot · 2020-12-04T08:58:55Z

Starting build on ROOT-debian10-i386/cxx14, ROOT-performance-centos8-multicore/default, ROOT-fedora30/cxx14, ROOT-fedora31/noimt, ROOT-ubuntu16/nortcxxmod, mac1015/cxx17, windows10/cxx14
How to customize builds

phsft-bot · 2020-12-04T09:00:42Z

Build failed on mac1015/cxx17.
Running on macitois17.cern.ch:/Users/sftnight/build/workspace/root-pullrequests-build
See console output.

Errors:

[2020-12-04T09:00:41.423Z] CMake Error at cmake/modules/SearchInstalledSoftware.cmake:1616 (message):
[2020-12-04T09:00:41.424Z] CMake Error at /Volumes/HD2/build/workspace/root-pullrequests-build/rootspi/jenkins/root-build.cmake:1077 (message):

phsft-bot · 2020-12-04T09:13:12Z

Build failed on ROOT-fedora31/noimt.
Running on root-fedora-31-2.cern.ch:/home/sftnight/build/workspace/root-pullrequests-build
See console output.

Failing tests:

phsft-bot · 2020-12-04T09:22:02Z

Build failed on ROOT-fedora30/cxx14.
Running on root-fedora30-1.cern.ch:/build/workspace/root-pullrequests-build
See console output.

Failing tests:

phsft-bot · 2020-12-04T09:31:17Z

Build failed on ROOT-ubuntu16/nortcxxmod.
Running on sft-ubuntu-1604-1.cern.ch:/build/workspace/root-pullrequests-build
See console output.

Failing tests:

phsft-bot · 2020-12-04T09:58:40Z

Build failed on ROOT-performance-centos8-multicore/default.
Running on null:/data/sftnight/workspace/root-pullrequests-build
See console output.

Failing tests:

phsft-bot · 2020-12-04T19:25:45Z

Build failed on ROOT-debian10-i386/cxx14.
Running on pcepsft11.dyndns.cern.ch:/home/sftnight/build/workspace/root-pullrequests-build
See console output.

Failing tests:

Axel-Naumann · 2020-12-14T21:37:01Z

interpreter/cling/lib/Interpreter/CIFactory.cpp

    DiagnosticOptions& DiagOpts = InvocationPtr->getDiagnosticOpts();
    llvm::IntrusiveRefCntPtr<DiagnosticsEngine> Diags =
-        SetupDiagnostics(DiagOpts);
+        SetupDiagnostics(DiagOpts, COpts.CUDADevice ? "cling-ptx" : "cling");


I bet that this change is causing the test failures due to the output being different than expected. I am okay with showing cling as part of the diagnostic even for ROOT, but we will need to update the .ref files for the test suite. Can you take care of that, @SimeonEhrig or shall I?

You are right. I thought I implemented it in such a way that it doesn't affect the output when CUDA mode is not enabled to avoid this kind of problems with the CI, but that's wrong.

@Axel-Naumann
Can you please change the .ref files. I don`t have time to get involved at the moment. And please link the changes in this issue. Maybe then I understand how it works and can do it myself next time.

phsft-bot · 2021-01-26T18:36:47Z

Starting build on ROOT-debian10-i386/cxx14, ROOT-performance-centos8-multicore/default, ROOT-fedora30/cxx14, ROOT-fedora31/noimt, ROOT-ubuntu16/nortcxxmod, mac1014/python3, mac11.0/cxx17, windows10/cxx14
How to customize builds

phsft-bot · 2021-01-26T18:49:39Z

Build failed on ROOT-fedora31/noimt.
Running on root-fedora-31-2.cern.ch:/home/sftnight/build/workspace/root-pullrequests-build
See console output.

Failing tests:

phsft-bot · 2021-01-26T19:09:00Z

Build failed on ROOT-ubuntu16/nortcxxmod.
Running on sft-ubuntu-1604-2.cern.ch:/build/workspace/root-pullrequests-build
See console output.

Failing tests:

phsft-bot · 2021-01-26T19:20:36Z

Build failed on windows10/cxx14.
Running on null:C:\build\workspace\root-pullrequests-build
See console output.

Errors:

[2021-01-26T18:37:05.247Z] ghprbPullLongDescription=The cling argument --cuda-path is necessary if the CUDA SDK is not installed under /usr/local/cuda, e.g. on HPC systems. The integration tests are also updated to handle a CUDA SDK under a non-standard location.\r\n\r\n# Additional diagnostic\r\n\r\nTo find the bug, I also add some diagnostic functions for the PTX compiler.\r\n\r\n1. Now error messages are prefixed in CUDA mode to better decide which of the two compiler pipelines is causing the error.\r\n\r\nexample\r\n\r\n- example before:\r\nerror: cannot find CUDA installation. Provide its path via --cuda-path, or pass -nocudainc to build without CUDA includes.\r\nerror: cannot find libdevice for sm_20. Provide path to different CUDA installation via --cuda-path, or pass -nocudalib to build without linking with libdevice.\r\nerror: cannot find CUDA installation. Provide its path via --cuda-path, or pass -nocudainc to build without CUDA includes.\r\n\r\n- example after:\r\ncling: error: cannot find CUDA installation. Provide its path via --cuda-path, or pass -nocudainc to build without CUDA includes.\r\ncling-ptx: error: cannot find libdevice for sm_20. Provide path to different CUDA installation via --cuda-path, or pass -nocudalib to build without linking with libdevice.\r\ncling-ptx: error: cannot find CUDA installation. Provide its path via --cuda-path, or pass -nocudainc to build without CUDA includes.\r\n\r\n\r\n2. Now, the class cudaIncrementalDeviceCompiler available through reflection via the gCling object.
[2021-01-26T19:15:54.928Z] ghprbPullLongDescription=The cling argument --cuda-path is necessary if the CUDA SDK is not installed under /usr/local/cuda, e.g. on HPC systems. The integration tests are also updated to handle a CUDA SDK under a non-standard location.\r\n\r\n# Additional diagnostic\r\n\r\nTo find the bug, I also add some diagnostic functions for the PTX compiler.\r\n\r\n1. Now error messages are prefixed in CUDA mode to better decide which of the two compiler pipelines is causing the error.\r\n\r\nexample\r\n\r\n- example before:\r\nerror: cannot find CUDA installation. Provide its path via --cuda-path, or pass -nocudainc to build without CUDA includes.\r\nerror: cannot find libdevice for sm_20. Provide path to different CUDA installation via --cuda-path, or pass -nocudalib to build without linking with libdevice.\r\nerror: cannot find CUDA installation. Provide its path via --cuda-path, or pass -nocudainc to build without CUDA includes.\r\n\r\n- example after:\r\ncling: error: cannot find CUDA installation. Provide its path via --cuda-path, or pass -nocudainc to build without CUDA includes.\r\ncling-ptx: error: cannot find libdevice for sm_20. Provide path to different CUDA installation via --cuda-path, or pass -nocudalib to build without linking with libdevice.\r\ncling-ptx: error: cannot find CUDA installation. Provide its path via --cuda-path, or pass -nocudainc to build without CUDA includes.\r\n\r\n\r\n2. Now, the class cudaIncrementalDeviceCompiler available through reflection via the gCling object.

phsft-bot · 2021-01-26T19:28:43Z

Build failed on ROOT-performance-centos8-multicore/default.
Running on null:/data/sftnight/workspace/root-pullrequests-build
See console output.

Failing tests:

phsft-bot · 2021-01-26T20:10:36Z

Build failed on mac11.0/cxx17.
Running on macphsft20.dyndns.cern.ch:/Users/sftnight/build/workspace/root-pullrequests-build
See console output.

Failing tests:

phsft-bot · 2021-01-26T20:14:03Z

Build failed on ROOT-debian10-i386/cxx14.
Running on pcepsft10.dyndns.cern.ch:/build/workspace/root-pullrequests-build
See console output.

Failing tests:

phsft-bot · 2021-01-26T20:23:10Z

Build failed on ROOT-fedora30/cxx14.
Running on root-fedora30-2.cern.ch:/home/sftnight/build/workspace/root-pullrequests-build
See console output.

Failing tests:

phsft-bot · 2021-02-04T19:15:49Z

Build failed on windows10/cxx14.
Running on null:C:\build\workspace\root-pullrequests-build
See console output.

Errors:

[2021-02-04T17:31:23.761Z] ghprbPullLongDescription=The cling argument --cuda-path is necessary if the CUDA SDK is not installed under /usr/local/cuda, e.g. on HPC systems. The integration tests are also updated to handle a CUDA SDK under a non-standard location.\r\n\r\n# Additional diagnostic\r\n\r\nTo find the bug, I also add some diagnostic functions for the PTX compiler.\r\n\r\n1. Now error messages are prefixed in CUDA mode to better decide which of the two compiler pipelines is causing the error.\r\n\r\nexample\r\n\r\n- example before:\r\nerror: cannot find CUDA installation. Provide its path via --cuda-path, or pass -nocudainc to build without CUDA includes.\r\nerror: cannot find libdevice for sm_20. Provide path to different CUDA installation via --cuda-path, or pass -nocudalib to build without linking with libdevice.\r\nerror: cannot find CUDA installation. Provide its path via --cuda-path, or pass -nocudainc to build without CUDA includes.\r\n\r\n- example after:\r\ncling: error: cannot find CUDA installation. Provide its path via --cuda-path, or pass -nocudainc to build without CUDA includes.\r\ncling-ptx: error: cannot find libdevice for sm_20. Provide path to different CUDA installation via --cuda-path, or pass -nocudalib to build without linking with libdevice.\r\ncling-ptx: error: cannot find CUDA installation. Provide its path via --cuda-path, or pass -nocudainc to build without CUDA includes.\r\n\r\n\r\n2. Now, the class cudaIncrementalDeviceCompiler available through reflection via the gCling object.
[2021-02-04T18:11:39.761Z] ghprbPullLongDescription=The cling argument --cuda-path is necessary if the CUDA SDK is not installed under /usr/local/cuda, e.g. on HPC systems. The integration tests are also updated to handle a CUDA SDK under a non-standard location.\r\n\r\n# Additional diagnostic\r\n\r\nTo find the bug, I also add some diagnostic functions for the PTX compiler.\r\n\r\n1. Now error messages are prefixed in CUDA mode to better decide which of the two compiler pipelines is causing the error.\r\n\r\nexample\r\n\r\n- example before:\r\nerror: cannot find CUDA installation. Provide its path via --cuda-path, or pass -nocudainc to build without CUDA includes.\r\nerror: cannot find libdevice for sm_20. Provide path to different CUDA installation via --cuda-path, or pass -nocudalib to build without linking with libdevice.\r\nerror: cannot find CUDA installation. Provide its path via --cuda-path, or pass -nocudainc to build without CUDA includes.\r\n\r\n- example after:\r\ncling: error: cannot find CUDA installation. Provide its path via --cuda-path, or pass -nocudainc to build without CUDA includes.\r\ncling-ptx: error: cannot find libdevice for sm_20. Provide path to different CUDA installation via --cuda-path, or pass -nocudalib to build without linking with libdevice.\r\ncling-ptx: error: cannot find CUDA installation. Provide its path via --cuda-path, or pass -nocudainc to build without CUDA includes.\r\n\r\n\r\n2. Now, the class cudaIncrementalDeviceCompiler available through reflection via the gCling object.

Axel-Naumann · 2021-02-05T08:00:59Z

The windows "error" is spurious, caused by the error log parser matching the commit log.

interpreter/cling/lib/Interpreter/CIFactory.cpp

interpreter/cling/lib/Interpreter/IncrementalCUDADeviceCompiler.cpp

interpreter/cling/lib/Interpreter/Interpreter.cpp

interpreter/cling/test/lit.cfg

- add the prefix "cling" (normal interpreter error) or "cling-ptx" (ptx interpreter -> just in CUDA mode) to every interpreter error message - example before: error: cannot find CUDA installation. Provide its path via --cuda-path, or pass -nocudainc to build without CUDA includes. error: cannot find libdevice for sm_20. Provide path to different CUDA installation via --cuda-path, or pass -nocudalib to build without linking with libdevice. error: cannot find CUDA installation. Provide its path via --cuda-path, or pass -nocudainc to build without CUDA includes. - example after: cling: error: cannot find CUDA installation. Provide its path via --cuda-path, or pass -nocudainc to build without CUDA includes. cling-ptx: error: cannot find libdevice for sm_20. Provide path to different CUDA installation via --cuda-path, or pass -nocudalib to build without linking with libdevice. cling-ptx: error: cannot find CUDA installation. Provide its path via --cuda-path, or pass -nocudainc to build without CUDA includes.

- make public that it is accessible via gCling object during the Cling runtime.

- now the custom path of `--cuda-path` is correctly set in the ptx compiler, allowing the use of CUDA SDK's which are not installed on in the default location

…ult location - To enable the CUDA test, lit detects the `libcudart.so` in `LD_LIBRARY_PATH`. Now lit also set the CUDA SDK root of `libcudart.so` as cling parameter (`--cuda-path`) in the tests. - Pass through the environment variable `CUDA_VISIBLE_DEVICES`.

- see: root-project#6707

phsft-bot · 2021-02-05T10:28:05Z

Starting build on ROOT-debian10-i386/cxx14, ROOT-performance-centos8-multicore/default, ROOT-fedora30/cxx14, ROOT-fedora31/noimt, ROOT-ubuntu16/nortcxxmod, mac1014/python3, mac11.0/cxx17, windows10/cxx14
How to customize builds

phsft-bot · 2021-02-05T11:34:11Z

Build failed on mac11.0/cxx17.
Running on macphsft20.dyndns.cern.ch:/Users/sftnight/build/workspace/root-pullrequests-build
See console output.

Failing tests:

phsft-bot · 2021-02-05T12:14:51Z

Build failed on windows10/cxx14.
Running on null:C:\build\workspace\root-pullrequests-build
See console output.

Errors:

[2021-02-05T10:28:28.693Z] ghprbPullLongDescription=The cling argument --cuda-path is necessary if the CUDA SDK is not installed under /usr/local/cuda, e.g. on HPC systems. The integration tests are also updated to handle a CUDA SDK under a non-standard location.\r\n\r\n# Additional diagnostic\r\n\r\nTo find the bug, I also add some diagnostic functions for the PTX compiler.\r\n\r\n1. Now error messages are prefixed in CUDA mode to better decide which of the two compiler pipelines is causing the error.\r\n\r\nexample\r\n\r\n- example before:\r\nerror: cannot find CUDA installation. Provide its path via --cuda-path, or pass -nocudainc to build without CUDA includes.\r\nerror: cannot find libdevice for sm_20. Provide path to different CUDA installation via --cuda-path, or pass -nocudalib to build without linking with libdevice.\r\nerror: cannot find CUDA installation. Provide its path via --cuda-path, or pass -nocudainc to build without CUDA includes.\r\n\r\n- example after:\r\ncling: error: cannot find CUDA installation. Provide its path via --cuda-path, or pass -nocudainc to build without CUDA includes.\r\ncling-ptx: error: cannot find libdevice for sm_20. Provide path to different CUDA installation via --cuda-path, or pass -nocudalib to build without linking with libdevice.\r\ncling-ptx: error: cannot find CUDA installation. Provide its path via --cuda-path, or pass -nocudainc to build without CUDA includes.\r\n\r\n\r\n2. Now, the class cudaIncrementalDeviceCompiler available through reflection via the gCling object.
[2021-02-05T11:10:55.739Z] ghprbPullLongDescription=The cling argument --cuda-path is necessary if the CUDA SDK is not installed under /usr/local/cuda, e.g. on HPC systems. The integration tests are also updated to handle a CUDA SDK under a non-standard location.\r\n\r\n# Additional diagnostic\r\n\r\nTo find the bug, I also add some diagnostic functions for the PTX compiler.\r\n\r\n1. Now error messages are prefixed in CUDA mode to better decide which of the two compiler pipelines is causing the error.\r\n\r\nexample\r\n\r\n- example before:\r\nerror: cannot find CUDA installation. Provide its path via --cuda-path, or pass -nocudainc to build without CUDA includes.\r\nerror: cannot find libdevice for sm_20. Provide path to different CUDA installation via --cuda-path, or pass -nocudalib to build without linking with libdevice.\r\nerror: cannot find CUDA installation. Provide its path via --cuda-path, or pass -nocudainc to build without CUDA includes.\r\n\r\n- example after:\r\ncling: error: cannot find CUDA installation. Provide its path via --cuda-path, or pass -nocudainc to build without CUDA includes.\r\ncling-ptx: error: cannot find libdevice for sm_20. Provide path to different CUDA installation via --cuda-path, or pass -nocudalib to build without linking with libdevice.\r\ncling-ptx: error: cannot find CUDA installation. Provide its path via --cuda-path, or pass -nocudainc to build without CUDA includes.\r\n\r\n\r\n2. Now, the class cudaIncrementalDeviceCompiler available through reflection via the gCling object.

Axel-Naumann

That inline won't work, I meant to suggest to actually inline :-) See proposed changes.

Axel-Naumann · 2021-02-05T14:46:59Z

interpreter/cling/lib/Interpreter/Interpreter.cpp

+  IncrementalCUDADeviceCompiler* Interpreter::getCUDACompiler() const {
+    return m_CUDACompiler.get();
+  }
+


Suggested change

IncrementalCUDADeviceCompiler* Interpreter::getCUDACompiler() const {

return m_CUDACompiler.get();

}

interpreter/cling/include/cling/Interpreter/Interpreter.h

phsft-bot · 2021-02-05T15:33:56Z

Build failed on ROOT-debian10-i386/cxx14.
Running on pcepsft10.dyndns.cern.ch:/build/workspace/root-pullrequests-build
See console output.

Failing tests:

- see: root-project#6707

phsft-bot · 2021-02-05T15:49:29Z

Starting build on ROOT-debian10-i386/cxx14, ROOT-performance-centos8-multicore/default, ROOT-fedora30/cxx14, ROOT-fedora31/noimt, ROOT-ubuntu16/nortcxxmod, mac1014/python3, mac11.0/cxx17, windows10/cxx14
How to customize builds

phsft-bot · 2021-02-05T16:10:16Z

Build failed on ROOT-performance-centos8-multicore/default.
Running on null:/data/sftnight/workspace/root-pullrequests-build
See console output.

Failing tests:

phsft-bot · 2021-02-05T16:54:41Z

Build failed on mac11.0/cxx17.
Running on macphsft23.dyndns.cern.ch:/Users/sftnight/build/workspace/root-pullrequests-build
See console output.

Failing tests:

projectroot.tree.ntuple.v7.test.gtest_tree_ntuple_v7_test_ntuple_extended

phsft-bot · 2021-02-05T17:43:28Z

Build failed on windows10/cxx14.
Running on null:C:\build\workspace\root-pullrequests-build
See console output.

Errors:

[2021-02-05T15:51:10.498Z] ghprbPullLongDescription=The cling argument --cuda-path is necessary if the CUDA SDK is not installed under /usr/local/cuda, e.g. on HPC systems. The integration tests are also updated to handle a CUDA SDK under a non-standard location.\r\n\r\n# Additional diagnostic\r\n\r\nTo find the bug, I also add some diagnostic functions for the PTX compiler.\r\n\r\n1. Now error messages are prefixed in CUDA mode to better decide which of the two compiler pipelines is causing the error.\r\n\r\nexample\r\n\r\n- example before:\r\nerror: cannot find CUDA installation. Provide its path via --cuda-path, or pass -nocudainc to build without CUDA includes.\r\nerror: cannot find libdevice for sm_20. Provide path to different CUDA installation via --cuda-path, or pass -nocudalib to build without linking with libdevice.\r\nerror: cannot find CUDA installation. Provide its path via --cuda-path, or pass -nocudainc to build without CUDA includes.\r\n\r\n- example after:\r\ncling: error: cannot find CUDA installation. Provide its path via --cuda-path, or pass -nocudainc to build without CUDA includes.\r\ncling-ptx: error: cannot find libdevice for sm_20. Provide path to different CUDA installation via --cuda-path, or pass -nocudalib to build without linking with libdevice.\r\ncling-ptx: error: cannot find CUDA installation. Provide its path via --cuda-path, or pass -nocudainc to build without CUDA includes.\r\n\r\n\r\n2. Now, the class cudaIncrementalDeviceCompiler available through reflection via the gCling object.
[2021-02-05T16:35:12.117Z] ghprbPullLongDescription=The cling argument --cuda-path is necessary if the CUDA SDK is not installed under /usr/local/cuda, e.g. on HPC systems. The integration tests are also updated to handle a CUDA SDK under a non-standard location.\r\n\r\n# Additional diagnostic\r\n\r\nTo find the bug, I also add some diagnostic functions for the PTX compiler.\r\n\r\n1. Now error messages are prefixed in CUDA mode to better decide which of the two compiler pipelines is causing the error.\r\n\r\nexample\r\n\r\n- example before:\r\nerror: cannot find CUDA installation. Provide its path via --cuda-path, or pass -nocudainc to build without CUDA includes.\r\nerror: cannot find libdevice for sm_20. Provide path to different CUDA installation via --cuda-path, or pass -nocudalib to build without linking with libdevice.\r\nerror: cannot find CUDA installation. Provide its path via --cuda-path, or pass -nocudainc to build without CUDA includes.\r\n\r\n- example after:\r\ncling: error: cannot find CUDA installation. Provide its path via --cuda-path, or pass -nocudainc to build without CUDA includes.\r\ncling-ptx: error: cannot find libdevice for sm_20. Provide path to different CUDA installation via --cuda-path, or pass -nocudalib to build without linking with libdevice.\r\ncling-ptx: error: cannot find CUDA installation. Provide its path via --cuda-path, or pass -nocudainc to build without CUDA includes.\r\n\r\n\r\n2. Now, the class cudaIncrementalDeviceCompiler available through reflection via the gCling object.

Failing tests:

SimeonEhrig · 2021-02-10T09:33:01Z

@Axel-Naumann I applied your suggestions. I think, the CI is also fine. The MacOS 11 and CentOS 8 looks not related to my changes, and the Windows 10 job has the parser problems.

Axel-Naumann · 2021-02-10T09:47:24Z

Great, please go ahead and hit "rebase and merge"!

SimeonEhrig · 2021-02-10T10:04:52Z

Great, please go ahead and hit "rebase and merge"!

done 😁

- see: root-project/root#6707

- see: root-project#6707

SimeonEhrig requested a review from Axel-Naumann as a code owner October 26, 2020 14:16

bellenot assigned Axel-Naumann Oct 26, 2020

SimeonEhrig force-pushed the clingPtxArgs branch from 05b0c45 to d0d4e40 Compare December 4, 2020 08:58

Axel-Naumann requested changes Dec 14, 2020

View reviewed changes

SimeonEhrig force-pushed the clingPtxArgs branch from d0d4e40 to d673399 Compare January 26, 2021 18:36

Axel-Naumann requested changes Feb 5, 2021

View reviewed changes

SimeonEhrig added 5 commits February 5, 2021 10:09

Make Incremental CUDA Device Compiler Interface public

37db04d

- make public that it is accessible via gCling object during the Cling runtime.

Bugfix for argument --cuda-path in ptx compiler

dd1d1b5

- now the custom path of `--cuda-path` is correctly set in the ptx compiler, allowing the use of CUDA SDK's which are not installed on in the default location

cling interpreter: add prefix to error messages, only in CUDA mode

822106a

SimeonEhrig added a commit to SimeonEhrig/root that referenced this pull request Feb 5, 2021

Apply Axel's suggestion for PR "Fixed cling argumenent --cuda-path"

5d042dc

- see: root-project#6707

SimeonEhrig force-pushed the clingPtxArgs branch from 2c9371d to 5d042dc Compare February 5, 2021 10:27

Axel-Naumann requested changes Feb 5, 2021

View reviewed changes

Apply Axel's suggestion for PR "Fixed cling argumenent --cuda-path"

02cfe25

- see: root-project#6707

SimeonEhrig force-pushed the clingPtxArgs branch from 5d042dc to 02cfe25 Compare February 5, 2021 15:49

Axel-Naumann approved these changes Feb 10, 2021

View reviewed changes

SimeonEhrig merged commit 962e763 into root-project:master Feb 10, 2021

FonsRademakers pushed a commit to root-project/cling that referenced this pull request Feb 10, 2021

Apply Axel's suggestion for PR "Fixed cling argumenent --cuda-path"

da1bb78

- see: root-project/root#6707

SimeonEhrig mentioned this pull request Feb 10, 2021

Fixed cling argument --cuda-path root-project/cling#363

Closed

SimeonEhrig mentioned this pull request Feb 18, 2021

Allows to configure CUDA sm level for Cling CUDA tests #7190

Merged

nicknagi pushed a commit to nicknagi/root that referenced this pull request Mar 30, 2021

Apply Axel's suggestion for PR "Fixed cling argumenent --cuda-path"

72b6fbd

- see: root-project#6707

	IncrementalCUDADeviceCompiler* Interpreter::getCUDACompiler() const {
	return m_CUDACompiler.get();
	}

Fixed cling argument --cuda-path #6707

Fixed cling argument --cuda-path #6707

Uh oh!

Conversation

SimeonEhrig commented Oct 26, 2020

Additional diagnostic

Uh oh!

phsft-bot commented Oct 26, 2020

Uh oh!

phsft-bot commented Oct 26, 2020

Failing tests:

Uh oh!

phsft-bot commented Oct 26, 2020

Failing tests:

Uh oh!

phsft-bot commented Oct 26, 2020

Failing tests:

Uh oh!

phsft-bot commented Oct 26, 2020

Failing tests:

Uh oh!

phsft-bot commented Oct 26, 2020

Failing tests:

Uh oh!

phsft-bot commented Oct 26, 2020

Failing tests:

Uh oh!

phsft-bot commented Oct 26, 2020

Failing tests:

Uh oh!

Axel-Naumann commented Oct 30, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

SimeonEhrig commented Nov 2, 2020

Uh oh!

phsft-bot commented Dec 4, 2020

Uh oh!

phsft-bot commented Dec 4, 2020

Errors:

Uh oh!

phsft-bot commented Dec 4, 2020

Failing tests:

Uh oh!

phsft-bot commented Dec 4, 2020

Failing tests:

Uh oh!

phsft-bot commented Dec 4, 2020

Failing tests:

Uh oh!

phsft-bot commented Dec 4, 2020

Failing tests:

Uh oh!

phsft-bot commented Dec 4, 2020

Failing tests:

Uh oh!

Axel-Naumann Dec 14, 2020

Choose a reason for hiding this comment

Uh oh!

SimeonEhrig Dec 15, 2020

Choose a reason for hiding this comment

Uh oh!

phsft-bot commented Jan 26, 2021

Uh oh!

phsft-bot commented Jan 26, 2021

Failing tests:

Uh oh!

phsft-bot commented Jan 26, 2021

Failing tests:

Uh oh!

phsft-bot commented Jan 26, 2021

Errors:

Uh oh!

phsft-bot commented Jan 26, 2021

Failing tests:

Uh oh!

phsft-bot commented Jan 26, 2021

Failing tests:

Uh oh!

phsft-bot commented Jan 26, 2021

Failing tests:

Axel-Naumann commented Oct 30, 2020 •

edited

Loading