MaskRay wrote: > Hi, this broke our downstream. What is the motivation for this, and how is > this intended to be used with compiler-rt locations being set by > multilib.yaml? We have something similar to what ARM does for multilibs.yaml > but with different paths, and this prevents the driver from knowing anything > about a singular path at which it should find the correct compiler-rt. What > makes compiler-rt special in a way that other target libraries aren't? Here's > how a part of our deliverable looks like: > > ``` > riscv/lib/rv32imafc/ilp32f/ > ├── except > │ ├── libc.a > │ └── libm.a > ├── libclang_rt.builtins-riscv32.a > └── noexcept > ├── libc.a > └── libm.a > ``` > > and there are several such `<arch>/<abi>` directories
* The preferred filename has changed from `clang_rt.builtins{,-$arch}.a` to `$triple/clang_rt.builtins.a`. BareMetal hadn't been fixed before this PR. * All other ToolChains use an absolute path, while BareMetal inappropriately used `-lxxx`. * This was a necessary change to do #81037 I think you'll need to switch to LLVM_ENABLE_PER_TARGET_RUNTIME_DIR=on hierarchy, which is quite different from the older off hierarchy. https://github.com/llvm/llvm-project/pull/82424 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits