Musing from C4ML 2021

Towards Automatic Scheduling for Tensorized Computation

Polyhedral Building Blocks for High-Performance Code Generation in MLIR.

A high-performance polyhedral math library as a foundation for AI compilers.

PolyDL: Polyhedral Compiler Optimizations for Deep Learning Workloads.

Understanding the Poplar Graph Compiler for IPUs.

Memory access planning for NPUs.

Polyhedral compilation techniques for code generation on spatial architectures.

Learning to optimize neural networks quickly.

An MLIR-Based end-to-end dynamic shape compiler.

Realize implicit GEMM-based convolutions on AMD GPU using MLIR.

oneDNN Graph API: unify deep learning framework integration and maximize compute efficiency for multiple AI hardware.