Code Generation for Statically Pipelined Architecture

I’m working on a project with a machine language for a statically pipelined architecture. This article provides an overview of the idea: http://www.cs.fsu.edu/~whalley/papers/cal11.pdf. Right now, we’re using the VPO compiler to produce this assembly code. We’ve been exploring using LLVM as a front-end to provide high level optimization.

We’d like to know if there’s a good way to go about producing this assembly language directly. The problem I’m looking at is that one instruction can control multiple individual effects. For example, the instruction " M[ALUR2] = sw RS2; RS2 = r[20]; ALUR2 = ALUR2 add SE" does a store, a register read, and an ALU operation (keeping intermediate values in internal registers). We have encoding restrictions that allow us to fit such a group of effects into a 32 bit instruction.

What would be the best way to accomplish this code generation using LLVM?