I have refactored the standard-to-llvm conversion according to the standard dialect splitting: operations that belong to the memref
and math
dialects are now converted by -convert-memref-to-llvm
and -convert-math-to-llvm
passes, respectively. These new passes live in separate libraries. These passes must run before -convert-std-to-llvm
that also performs the cast cleanup.
Also, the common utilities for converting to the LLVM dialect are factored out into lib/Conversions/LLVMCommon, it is no longer necessary to depend on StandardToLLVM to use those.