[llvm-branch-commits] [clang] clang/AMDGPU: Fix double linking opencl libs with --libclc-lib (PR #204865)

[Matt Arsenault via llvm-branch-commits]

arsenm wrote:

> > Not sure I follow, these can't coexist. They both have the exact same OpenCL functions with itanium mangling
> 
> The headers that HIP/CUDA use eagerly rewrite `sin` and `std::sin` to `__ocml_sin_f64` via force-included headers. So if those headers are still included there's no way to resolve math without using the same symbols.

The long term plan is to stop depending on those altogether. Any shim is wasted effort that just makes the end goal more difficult 

https://github.com/llvm/llvm-project/pull/204865