Configuration Options for different hardware and compiler toolchains

These options should be documented.
Options should be configurable in a consistent way and potentially exported from the IDE project

Compiler toolchain should be configurable to ATfE (CLANG) and potentially exported from the IDE project

Addressed by the three-step flow on main:

The NPU and Vela configuration lives in the mlops: node of cmsis-executorch.csolution.yml. cbuild setup exports it to *.cbuild-mlops.yml, and create_ai_layer.py passes it to ExecuTorch's EthosUCompileSpec; vela: misc: forwards any further Vela option (Pass every Vela option of the mlops: node to Vela #23). The model parameters (input shape, calibration samples) are under mlops: model: as well (Use the implicit simulator target and model parameters in mlops: #28). It is documented in the README and in documentation/mlops-flow.md.
There is no Docker/GCC library build any more: the runtime is part of the CMSIS build, and the compiler is chosen with cbuild --toolchain. CI builds and runs with AC6, GCC and CLANG (ATfE).

Navigation Menu