Home

TPP-MLIR Wiki

Design and implementation details

• Default Pipeline
• Pass Analysis
• Self Attention block
• Intel AMX Tiling
• MXFP Dialect

Components that live fully upstream

• Data Layout & Target Information-dialect
• Tensor Pack and Unpack
• Linalg matmul with affine maps

Discussions and unfinished drafts

• Linalg Dialect Wish-list
• Generics-in-MLIR
• Linalg-Generic-Nesting
• Linalg-Named-Ops
• Tensor-of-Tensors
• Blocking/tiling-on-vector

Outdated discussions

• More-generic-matcher-for-BRGEMM
• Rising-Linalg-to-TPP-on-Tensor
• TPP-in-destination-passing-style

Development productivity

• ViM-tools-and-configuration
• Multiple-LLVM-Workspaces
• CMake-Options-for-LLVM