Cory Bloor

Reopening, as this still a matter of ongoing discussion. The status quo is not acceptable.

While I agree with the overall sentiment, the situation is not quite as dire as you're suggesting. > This is in large part why [no widely-used Linux distribution](https://repology.org/project/rocm-smi/versions) ships packaging...

> * [`rocblas`](https://github.com/ROCmSoftwarePlatform/rocBLAS)'s build has quite a few configuration knobs and yet has no installation documentation, instead pointing to the out-of-date ROCm 4.3 installation guide, which makes no mention of...

> My personal todo list includes standardizing the CMake build commands used for {roc,hip}{BLAS,SOLVER,FFT,SPARSE,RAND}, but that will be a long process. I discovered that there was already someone working on...

> Has there been any progress in packaging documentation? There has not been anything published. It remains a work-in-progress. > having a list of the relevant packages, the locations of...

The way that separate architectures and lazy loading were implemented was really not ideal. The complexity of building all the necessary data structures during initialization should really be pushed to...

> to make work my rx5700 with llamacpp? want to ask frist before do the test and break my ubuntu lol Use librocblas-dev and libhipblas-dev from Ubuntu 23.10 or later....

Just FYI, on the develop branch it is now possible to use a `-DBUILD_WITH_PIP=OFF` CMake option to entirely avoid the use of pip/virtualenv during the rocBLAS build. When passed, the...

[The LLVM Offload Bundler docs on target id notation](https://clang.llvm.org/docs/ClangOffloadBundler.html#target-id) explain a bit about the syntax. When a code object is built for `gfx90a:xnack-`, it will only run on GPUs in...

> Since only gfx90a are built with :xnack- and other gfx90? are :xnack- only, what happened if some one is running on xnack enabled GPUs? The rocr-runtime will not let...