trixirt

[Issue]: failure to handle missing python module joblib

2

### Problem Description On RHEL and SUSE, joblib is not available in the distro. This import will fail https://github.com/ROCm/hipBLASLt/blob/develop/tensilelite/Tensile/Common/Parallel.py#L31 For rocBLAS, this is handled here https://github.com/ROCm/Tensile/blob/develop/Tensile/Utilities/ConditionalImports.py#L30 ### Operating System RHEL,...

[Issue]: hardcoded toolchain paths break on Fedora and OpenSUSE

2

### Problem Description There are several hard coded paths to the rocm-llvm toolchain like https://github.com/ROCm/hipBLASLt/blob/develop/tensilelite/Tensile/Ops/gen_assembly.sh#L36 and https://github.com/ROCm/hipBLASLt/blob/develop/library/src/amd_detail/rocblaslt/src/kernels/compile_code_object.sh#L37 on Fedora and OpenSUSE the path to the compiler is > hipconfig -l...

[Feature]: Replace fmt:: with std::

4

### Suggestion Description The fmt package is not available on all linux distros. So the requirement here https://github.com/ROCm/rocSOLVER/blob/develop/CMakeLists.txt#L67 This will fail for RHEL and OracleLinux the fmt::format and fmt::print calls...

fp16 python interface is used here https://github.com/Maratyszcza/NNPACK/blob/master/src/x86_64-fma/blas/shdotxf.py#L4 but has been removed here https://github.com/Maratyszcza/FP16/commit/07ff4fd681127d86a06cb8d369cb96988de402e6 This breaks the building of nnpack.

Run this fedora container on a fedora rawhide host. https://github.com/trixirt/rocm-distro-containers/blob/main/fedora/rawhide/rocblas/check/Dockerfile with the args docker run --device /dev/kfd --device /dev/dri -it --rm --cpus=1 This produces a backtrace [----------] 498 tests from...

Run this fedora container on a fedora rawhide host. https://github.com/trixirt/rocm-distro-containers/blob/main/fedora/rawhide/rocblas/check/Dockerfile with the args docker run --device /dev/kfd --device /dev/dri -it --rm --cpus=1 This produces a backtrace ... [Detaching after vfork...

[Bug]: use roctracer/roctx.h without cmake find_package

2

### Describe the bug A clear and concise description of what the bug or problem is. On 6.4.0, roctx.h is included here https://github.com/ROCm/rocBLAS/blob/rocm-6.4.0/library/src/include/logging.hpp#L44 without having a cmake find_package(roctracer, REQUIRED) Will...

[Issue]: Segfault in vxCreateTensorFromHandle

1

### Problem Description Running rocAL basic_test produces this backtrace _pthread_kill_implementation (threadid=, signo=signo@entry=6, no_tid=no_tid@entry=0) at pthread_kill.c:44 44 return INTERNAL_SYSCALL_ERROR_P (ret) ? INTERNAL_SYSCALL_ERRNO (ret) : 0; Missing rpms, try: dnf --enablerepo='*debug*' install...

Download is done here https://github.com/ROCm/ROCmValidationSuite/blob/master/CMakeLists.txt#L391 before check of user setting inc dir here https://github.com/ROCm/ROCmValidationSuite/blob/master/CMakeLists.txt#L411 If the user passes in the variable like this -DMXDATAGENERATOR_INC_DIR=${PWD}/mxDataGenerator/lib/include download is not needed. Also building...

lib64 vs lib

1

This is representative of a general problem https://github.com/ROCm/ROCmValidationSuite/blob/master/CMakeLists.txt#L63 On at least RHEL, Fedora, SUSE and OpenSUSE the library install location is lib64, not lib.