多面体编译

https://polly.llvm.org/docs/Architecture.html
The Architecture
Polly is a loop optimizer for LLVM. Starting from LLVM-IR it detects and extracts interesting loop kernels. For each kernel a mathematical model is derived which precisely describes the individual computations and memory accesses in the kernels. Within Polly a variety of analysis and code transformations are performed on this mathematical model. After all optimizations have been derived and applied, optimized LLVM-IR is regenerated and inserted into the LLVM-IR module.

Polly in the LLVM pass pipeline
The standard LLVM pass pipeline as it is used in -O1/-O2/-O3 mode of clang/opt consists of a sequence of passes that can be grouped in different conceptual phases. The first phase, we call it here Canonicalization, is a scalar canonicalization phase that contains passes like -mem2reg,