Skip to content

Conversation

@LeiWang1999
Copy link
Member

This pull request introduces several new features and improvements to the tilelang library, with a focus on enhancing the carver module's functionality and adding support for new templates. The most important changes include adding new test cases, incorporating the FlashAttentionTemplate, and updating the node and policy classes to support more complex operations.

New Features and Enhancements:

  • Added new test cases for matrix multiplication and Flash Attention configurations in testing/python/carver/test_tilelang_carver_generate_hints.py and testing/python/carver/test_tilelang_carver_recommend_hints.py. [1] [2]
  • Introduced FlashAttentionTemplate for attention-like operations with flash memory in the tilelang/carver module. [1] [2]
  • Updated the README.md to document the new FlashAttentionTemplate and removed the TODO item for Flash Attention support.

Codebase Simplification and Refactoring:

  • Enhanced the PrimFuncNode class and added OutputNode and Edge classes to support more complex graph structures in tilelang/carver/roller/node.py. [1] [2] [3] [4] [5] [6]
  • Modified the DefaultPolicy class to use the new node classes and support multiple output nodes in tilelang/carver/roller/policy/default.py. [1] [2] [3]

Minor Fixes:

  • Fixed alignment issues in tilelang/carver/roller/bestfit.py to improve memory allocation logic.
  • Corrected the get_rstep method in tilelang/carver/roller/hint.py to properly map reduce steps to nodes.

…consistency and clarity; add apply_simplify function to simplify PrimFunc or IRModule.
…remove unnecessary whitespace in multiple files.
…r result validation in test_tilelang_kernel_gemm_mma_intrinsic.py
…riting high-performance kernels with thread primitives
… for improved code documentation and clarity
@LeiWang1999 LeiWang1999 merged commit b427ec4 into tile-ai:main Feb 11, 2025
2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant