Skip to content

Conversation

@edwinsolisf
Copy link
Contributor

@edwinsolisf edwinsolisf commented Jun 20, 2024

Fixes cmake configuring issue where specified cmake policy is not available prior to a certain cmake version: https://cmake.org/cmake/help/latest/policy/CMP0146.html


  • Is this a new feature or a bug fix?: Bug fix
  • More detail if necessary to describe all commits in pull request: checks the cmake version before setting policy
  • Why these changes are necessary: Needed for correct cmake configuration
  • Potential impact on specific hardware, software or backends: None
  • New functions and their functionality: None
  • Can this PR be backported to older versions?: No
  • Future changes not implemented in this PR: None

Changes to Users

  • No additional options added to the build
  • No action required by the user

Checklist

  • Rebased on latest master
  • Code compiles
  • Tests pass
  • Functions added to unified API
  • Functions documented

@christophe-murphy christophe-murphy merged commit d374808 into arrayfire:3539-build-oneapi-version-2024-incompatible-for-mkl Sep 30, 2024
christophe-murphy added a commit that referenced this pull request Feb 13, 2025
* Updated cmake files to support oneAPI version 2024.1. Currently doesn't support compiling on Windows, this will be added later.

* Some tests were showing oneAPI errors due to nested calls of submit to the SYCL queue which is not supported. This has been fixed by moving calls to get() out of the submit calls. If a get call is made to a node that has not yet been evaluated, it will need to submit work to the SYCL queue.

* Modify test cases to check if library functions are supported on the current backend. If a function is not supported the test is skipped. This works with both the C API which returns an error flag and the C++ API which throws an exception.

* Modified half support check to check the native vector width for half precision as well as the fp16 aspect for the oneAPI backend. Some devices advertise the fp16 aspect but their native vector width for half precision is zero which results in errors when calling OpenCL routines with half precision arguments.

* Fix for bug in index function introduced when implementing a fix for nested oneAPI queue submissions.

* Fixed bug in irreduce_dim_launcher where incorrect templated calls were made to ireduceDimKernelSMEM for 1, 2 and 4 y threads

* Fix bug in wrap function where one dimension of the output array was missing from the global problem size.

* Check for failure of asserts in apiWrapper and return from test to prevent segfaults in subsequent asserts.

* Modified ASSERT_SUCCESS macro to skip unsupported tests rather than failing when AF_SKIP_UNSUPPORTED_TESTS CMake option is enabled

* Fixed cmake policy issue for oneapi fix (#3569)

* Cmake function CMakeDetermineCompileFeatures has been changed to CMakeDetermineCompilerSupport in version 3.30. Added support for this.

* Removed unsupported compute capabilites from all architectures list for CUDA 12

* Rename variables for input and output arrays of join method

* Fix issue where NOT_SUPPORTED errors were storing the error message as the back end name

* Added macro for asserting success of C++ API functions that throw exceptions. If the not supported exception is thrown, the test can be skipped if the AF_SKIP_UNSUPPORTED_TESTS variable is on.

* Remove macros that check for unsupported exceptions and skip tests. A SKIP_BACKEND macro will be made instead that will need to be explicitly added to each test that calls a function unsupported for a given backend.

* The UNSUPPORTED_BACKEND macro has now been added to all tests that are not supported by the oneAPI back end. If AF_SKIP_UNSUPPORTED_TESTS is set to ON then all tests with this macro will be skipped. These will need to be removed as oneAPI support is added for each feature.

* Update getBackendName test helper function to support all back ends. It is now used for the UNSUPPORTED_BACKEND macro.

---------

Co-authored-by: Edwin Lester Solís Fuentes <68087165+edwinsolisf@users.noreply.github.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants