Note: TiledKernel is no longer an active project, and the successor is ThrillerFlow.

TileGraph is an experimental DNN compiler that utilizes static code generation and kernel fusion techniques.

设计文档 | Design Docs

git clone [email protected]:KuangjuX/TileGraph.git
git submodule update --init --recursive

cd 3rd-party/fmt && mkdir build && cd build && cmake -DCMAKE_POSITION_INDEPENDENT_CODE=TRUE .. && make

make test

make examples

• InfiniTensor/RefactorGraph: 分层解耦的深度学习推理引擎
• microsoft/nnfusion: A flexible and efficient deep neural network (DNN) compiler that generates high-performance executable from a DNN model description.

• BOLT: BRIDGING THE GAP BETWEEN AUTO-TUNERS AND HARDWARE-NATIVE PERFORMANCE(MLSys'22)
• Welder: Scheduling Deep Learning Memory Access via Tile-graph(OSDI'23)
• Graphene: An IR for Optimized Tensor Computations on GPUs(ASPLOS'23)