Skip to content

mattnappo/comco

Repository files navigation

Comco: A MLIR-Based Intermediate Representation for CUDA Kernel Fusion

Comco is an MLIR dialect providing higher level machine learning primitives and a set of incremental lowerings to MLIR's gpu/nvgpu dialects. Comco also performs basic kernel fusion on tensor and broadcasting operations (i.e. GEMM + AllReduce).

About

MLIR dialect for higher level machine learning primitives

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published