The HPC group has been working on a number of HPC focused optimisations in LLVM and collaborating with TCWG to agree on benchmarks and results analysis. LLVM has reached a sufficient level as a compiler for system programming. However, there are several problems as compilers for HPC applications. In this report, we show the deficiency of LLVM as an HPC compiler and propose an improvement plan for them. We also report on some of our work related to it, which include activating software pipelining for AArch64, optimizing the control flow graph for HPC applications, improving register allocation for kernel parts of HPC applications, and so on.