Litcius/Paper detail

MLIR-based code generation for GPU tensor cores

Navdeep Katel, Vivek Khandelwal, Uday Bondhugula

202219 citationsDOI

Abstract

The state-of-the-art in high-performance deep learning today is primarily driven by manually developed libraries optimized and highly tuned by expert programmers using low-level abstractions with significant effort. This effort is often repeated for similar hardware and future ones. In this work, we pursue and evaluate the more modular and reusable approach of using compiler IR infrastructure to generate libraries by encoding all the required optimizations as a sequence of transformations and customized passes on an IR. We believe that until the recent introduction of MLIR (Multi-level intermediate representation), it had been hard to represent and transform computation at various levels of abstraction within a single IR.

Topics & Concepts

Computer scienceCompilerAbstractionModular designEncoding (memory)Parallel computingCode (set theory)ComputationComputer architectureDeep learningRepresentation (politics)State (computer science)Intermediate languageProgramming languageCode generationArtificial intelligenceKey (lock)Operating systemPoliticsPhilosophyLawPolitical scienceSet (abstract data type)EpistemologyParallel Computing and Optimization TechniquesAdvanced Neural Network ApplicationsDistributed and Parallel Computing Systems