Habana TPC kernel library with supporting firmware, drivers and tools.
Quickly and easily deploy a variety of models and algorithms on Gaudi. With the software package, you have access to an extensive set of Tensor Processing Core (TPC™) kernel libraries (1400+ kernels) and open TPC for programming that provides a complete set of tools (compiler, assembler, and simulator). These tools enable you to develop customized TPC kernels that can augment the existing Habana kernels. In addition, the TPC Kernel Repository provides templates and examples to make it easy to develop your own kernels.
TPC Tools Installation Guide
Access the necessary tools required to build TPC kernel plug-ins for the Graph Compiler.
TPC User Guide
Create custom TPC kernels.
TPC Debugger User Guide
Install and use Habana® TPC Eclipse-based debugger for both source level TPC-C and TPC-Assembly.
TensorFlow Custom OP Framework
Expose described API to write TensorFlow Operators for the Habana Accelerator.
Habana Custom Kernels
A stand-alone set of template examples that show how to create and build the custom kernels, which can be used in Tensorflow custom ops. Contains a tested set of TPC kernels, Glue code and unit tests.
Gaudi: Intro to TPC Kernels
A how-to tutorial on building custom TPC Kernels.(Click here for Mandarin version)