This is a small module that uses LLVM intrinsics to set the rounding mode for Float64.
This is a workaround for the deprecation of setrounding for Float64, see Deprecate setrounding.
This is specially needed for the implementation of the :fast multiplication of matrices in IntervalLinearAlgebra, which uses directed rounding together with high performance Blas routine to compute enclosures of matrix multiplication.
One feature of the LLVM intrinsic is that they can be used to set the rounding mode on AMDGPU; it may be worth investigating this direction. Discussion at LLVM: Intrinsic for setting rounding mode
