This BOF is not happening with current packages, maybe keep this as a reference... The build failure is for a non official package. Even so, it is asking /usr/share/cmake/Modules/CMakeDetermineHIPCompiler.cmake that comes with cmake: # equery b /usr/share/cmake/Modules/CMakeDetermineHIPCompiler.cmake * Searching for /usr/share/cmake/Modules/CMakeDetermineHIPCompiler.cmake ... dev-build/cmake-3.31.0 (/usr/share/cmake/Modules/CMakeDetermineHIPCompiler.cmake) The issue here is that for some reason, when trying to compile composable-kernel it asks for CMAKE_HIP_COMPILER_ROCM_ROOT path and gets '/usr/lib' and then tries to build the root path /usr/lib/lib and /usr/lib/lib64. That cmake file has a rule where if it fails to get the CMAKE_HIP_COMPILER_ROCM_ROOT, it will run `hipconfig --rocmpath` and get the value from it so I used that as a WA as it returns '/usr' as expected. The actual error is: " CMake Error at /usr/share/cmake/Modules/CMakeDetermineHIPCompiler.cmake:217 (message): The ROCm root directory: /usr/lib does not contain the HIP runtime CMake package, expected at one of: /usr/lib/lib/cmake/hip-lang/hip-lang-config.cmake /usr/lib/lib64/cmake/hip-lang/hip-lang-config.cmake Call Stack (most recent call first): CMakeLists.txt:26 (project) " The change I did in that file is to move this rule up from line 158 to the head of the rules that look for CMAKE_HIP_COMPILER_ROCM_ROOT. if(NOT CMAKE_HIP_COMPILER_ROCM_ROOT) execute_process( COMMAND hipconfig --rocmpath OUTPUT_VARIABLE _CMAKE_HIPCONFIG_ROCMPATH RESULT_VARIABLE _CMAKE_HIPCONFIG_RESULT ) if(_CMAKE_HIPCONFIG_RESULT EQUAL 0 AND EXISTS "${_CMAKE_HIPCONFIG_ROCMPATH}") set(CMAKE_HIP_COMPILER_ROCM_ROOT "${_CMAKE_HIPCONFIG_ROCMPATH}") endif() endif() if(NOT CMAKE_HIP_COMPILER_ROCM_ROOT) message(FATAL_ERROR "Failed to find ROCm root directory.") endif()
Not really sure what to do with this if you cannot provide the name of a failing Gentoo package.
Hi, The package is composable-kernel, however the provided 6.1.1 version does not have this issue, and I am trying to bring this version to a newer 6.2.4 version where it has a bunch of issues where this is one of them. Again, I am also not sure that we have anything to do here, and I wanted to raise a flag at this point because the 6.1.1 package has a lot of bugs that are fixed in 6.2.x and we will need to move to it sooner or later.