AscendNPU IR is an MLIR-based intermediate representation for Ascend NPU operator compilation, providing multi-level abstraction and compiler optimizations, with flexible integration for ecosystem frameworks and fine-grained performance tuning.
My Final Year Project (FYP) focuses on the extension and optimization of AscendNPU IR. The project is conducted in collaboration with Huawei Compiler Technologies Laboratory, under the supervision of Prof. Lionel Parreaux.
Check some of my notes here.