This patch removes patterns for a few operations which allows mathToLLVM conversion to convert the operations into LLVM intrinsics instead since they are supported directly by the AMDGPU backend.
parseMinus()
host_shared
hasSummary
hasDescription
cppType
cppClassName