Update on the miner:

We successfully created build system overlays for the four AMD repositories HIP, hipamd, ROCclr and ROCm-OpenCL-Runtime. CUDA also seems to work fine.

Instead of building all these repositories separately we now only need a single command to build all of them. We do not depend on the AMDs HIP-clang compiler, and we do not need NVIDIAs nvcc compiler. Everything compiles with a regular LLVM/Clang toolchain. Building this entire AMD stack now takes us less than 15 seconds 😂

We are currently working on adding a build system overlay for ROCm-Device-Libs as the last piece of the puzzle, which requires us to create a toolchain for compilation from LLVM IR to LLVM Bitcode. Then we will be able to compile GPU kernels for both AMD and NVIDIA with a single command 🥳.