This work proposes a compilation flow using open-source compiler passes to
build a framework to achieve ninja performance from a generic linear algebra
high-level abstraction. We demonstrate this flow with a proof-of-concept MLIR
project that uses input IR in Linalg-on-Tensor from TensorFlow and PyTorch,
performs cache-level optimizations and lowering to micro-kernels for efficient
vectorization, achieving over 90% of the performance of ninja-written
equivalent programs. The contributions of this work include: (1) Packing
primitives on the tensor dialect and passes for cache-aware distribution of
tensors (single and multi-core) and type-aware instructions (VNNI, BFDOT,
BFMMLA), including propagation of shapes across the entire function; (2) A
linear algebra pipeline, including tile, fuse and bufferization strategies to
get model-level IR into hardware friendly tile calls; (3) A mechanism for
micro-kernel lowering to an open source library that supports various CPUs.

该研究提出了一种使用开源编译器的编译流程，构建一个框架，实现从通用的线性代数高级抽象中获得忍者级性能。通过用 TensorFlow 和 PyTorch 中的 Linalg-on-Tensor 中的输入 IR，执行缓存级优化和降低到微内核以实现高效矢量化，实现了超过 90% 与忍者编写等效程序性能的概念证明 MLIR 项目。该研究的贡献包括：（1）在张量方言上对基元进行打包以及用于缓存感知张量（单核和多核）的传播以及类型感知指令（VNNI、BFDOT、BFMMLA）的传播，包括在整个函数中进行形状传播；（2）线性代数流水线，包括切片、融合和缓冲策略，将模型级 IR 转化为硬件友好的切片调用；（3）将微内核降级到支持各种 CPU 的开源库的机制。