Memory operations form an important instruction for most of the target architecture. Some of the architectures have sophisticated instructions to move data in and out of the memory. Some even perform binary operations directly on the memory operands, while some of them load data from memory into registers and then perform operations on them (CISC vs RISC). Many load-store operations are also done by LLVM instrinsics. For examples, please refer to http://llvm.org/docs/LangRef.html#masked-vector-load-and-store-intrinsics.

LLVM IR provides a common playfield for all the architectures. It provides elementary instructions for data operations on memory or on aggregate data types. The architectures, while lowering LLVM IR, may combine IR instructions to emit their specific instructions. In this chapter, we went through some advanced IR instructions and also looked into examples of them. For a detailed study, refer to http://llvm.org/docs/LangRef.html, which provides the authoritative resource for LLVM IR instructions.

In the next chapter, you will study how LLVM IR can be optimized to reduce instructions and emit a clean code.