English | 中文
CMake-Embedded
A VS Code extension that generates ready-to-build CMake projects for two
independent targets:
- MCU firmware - bare-metal STM32 / GD32 projects with a linker script, a GNU
startup file, a toolchain file, and CMake.
- Linux / SoC applications - cross compilation projects for Loongson
(loongarch64), ARM (aarch64), 32-bit ARM (armv7 hard-float), or RISC-V
(riscv64), with a toolchain file and CMake.
Both directions scan the opened workspace for C/C++ sources, header directories,
and preprocessor definitions, then generate the files needed to build.
Commands
| Command |
Direction |
Purpose |
CMake-Embedded: Generate MCU Project |
MCU |
Pick vendor, series, and part, then generate a bare-metal project |
CMake-Embedded: Generate Linux Project |
Linux |
Pick the cross toolchain, then generate a cross compilation project |
CMake-Embedded: Generate OpenOCD Flash Script |
MCU |
Generate flash.py only |
CMake-Embedded: Generate Cortex-Debug Configuration |
MCU |
Generate a .vscode/launch.json debug configuration only |
MCU direction
Features
- Generate
CMakeLists.txt from the current project directory
- Generate an MCU linker script and GNU assembler startup file
- Generate the C runtime support files under
system/
- Generate Debug and Release configure and build presets for Ninja
- Configure CMake Tools to use the generated presets
- Configure C/C++ IntelliSense from the CMake compile database
- Generate an optional OpenOCD
flash.py script independently from CMake
- Generate a Cortex-Debug OpenOCD launch configuration
- Generate
.elf, .hex, and .bin files after a successful build
- Prompt before overwriting generated files
The flash script supports J-Link OB, J-Link, ST-Link, and DapLink. The CMake
device selection and OpenOCD flash-target selection are independent, so a
project can use a custom CMake build and still generate a flash script. The
probe selection provides the OpenOCD interface configuration. GD32 F1/F4
targets use the compatible STM32 F1/F4 OpenOCD targets.
Demo

The device database currently includes common STM32/GD32 F1 and F4 parts plus
the STM32L4 family. The device picker groups them by vendor first (STM or
GD), then by series (STMF1x, STMF4x, STML4x, GDF1x, or GDF4x).
Usage
Open the MCU project folder in VS Code.
Open the Command Palette and run CMake-Embedded: Generate MCU Project.
Pick the vendor, series, and part, then confirm the generated files when asked.
Configure and build the project with the generated presets:
cmake --preset debug # or cmake --preset release
cmake --build --preset debug
The project generator asks whether flash.py should also be generated. If
selected, it asks for an OpenOCD flash target and probe separately from the
CMake build device. You can also run CMake-Embedded: Generate OpenOCD Flash Script by itself. This command asks only for the flash target and debug probe,
then writes flash.py in the project root without changing an existing CMake
project. The flash target database includes STM32 F0/F1/F2/F3/F4/F7/G0/G4/H7/
L0/L1/L4/L5/U5/WB/WL families and GD32 E23x/F1/F4/VF103 targets.
Set CMake-Embedded: Openocd Path (cmakeEmbedded.openocdPath) in VS Code
settings when OpenOCD is not available in PATH. The generated script stores
that value as its default and accepts an override when needed:
python flash.py
python flash.py --firmware build/my-project.elf
python flash.py --probe stlink
python flash.py --openocd D:\Tools\OpenOCD\bin\openocd.exe
python flash.py --dry-run
If Cortex-Debug cannot find the ARM GNU toolchain, set CMake-Embedded: Arm Toolchain Path (cmakeEmbedded.armToolchainPath) to the bin directory
containing the arm-none-eabi tools. The generated debug configuration uses
this path and sets the toolchain prefix to arm-none-eabi.
Run CMake-Embedded: Generate Cortex-Debug Configuration to add a standard
OpenOCD launch configuration to .vscode/launch.json. The generated entry
uses ${command:cmake.launchTargetPath} for the firmware, so it works with
the generated CMake project and with a user-provided CMake project. Existing
launch configurations are preserved. Press F5 in VS Code to start the
Cortex-Debug session.
Generated layout
<project>/
├── CMakeLists.txt
├── CMakePresets.json
├── .mcu-cmake.json
├── .vscode/
│ ├── settings.json
│ ├── c_cpp_properties.json
│ └── launch.json (optional Cortex-Debug configuration)
├── <mcu>.ld
├── startup_<device>.S
├── flash.py (optional OpenOCD flash script)
├── cmake/
│ └── <device>-toolchain.cmake
└── system/
├── syscalls.c
└── sysmem.c
The build directories are build/debug and build/release. Firmware output files are generated
there together with the map file and memory usage report.
Linux direction
Features
- Generate
CMakeLists.txt from the current project directory
- Generate the cross toolchain file
cmake/linux-toolchain.cmake
- Generate Debug and Release Ninja configure and build presets
- Generate
.linux-cmake.json, recording the selected toolchain and project structure
- Configure CMake Tools to use the generated presets and C/C++ IntelliSense from
the CMake compile database
- Prompt before overwriting generated files
The toolchain file only contains compiler command names (for example
loongarch64-linux-gnu-gcc), never absolute paths, so the same project builds on
any machine.
| Name |
Prefix |
CMAKE_SYSTEM_PROCESSOR |
| Loongson (loongarch64) |
loongarch64-linux-gnu- |
loongarch64 |
| ARM (aarch64) |
aarch64-linux-gnu- |
aarch64 |
| ARM 32-bit (armv7 hard-float) |
arm-linux-gnueabihf- |
arm |
| RISC-V (riscv64) |
riscv64-linux-gnu- |
riscv64 |
There are no extension settings for toolchain paths. The toolchain is located
through the environment only:
CROSS_COMPILE, for example export CROSS_COMPILE=loongarch64-linux-gnu-
CC / CXX
PATH, matching a supported gcc + g++ prefix pair
All four families are always offered in the picker. An entry that was found in
the environment is annotated with its source and directory; an entry that was not
found can still be generated, and only requires the toolchain bin directory on
PATH before building.
Usage
Open one project root folder in VS Code (one project per root).
Open the Command Palette and run CMake-Embedded: Generate Linux Project.
Pick the cross toolchain (Loongson, ARM aarch64, ARM 32-bit, or RISC-V) and confirm the generated files.
Configure and build the project with the generated presets:
cmake --preset debug # or cmake --preset release
cmake --build --preset debug
Generated layout
<project>/
├── CMakeLists.txt
├── CMakePresets.json
├── .linux-cmake.json
├── .vscode/
│ ├── settings.json
│ └── c_cpp_properties.json
└── cmake/
└── linux-toolchain.cmake
The build directories are build/debug and build/release. The user code
section of an existing CMakeLists.txt (between # CMAKE-EMBEDDED USER CODE BEGIN and END) is preserved when regenerating, so hand-written
target_sources, target_link_libraries, and similar blocks survive.
Requirements
- VS Code 1.85 or newer
- CMake Tools extension
- C/C++ extension
- CMake 3.22 or newer
- Ninja
Depending on the direction:
- MCU: Arm GNU Toolchain with
arm-none-eabi-gcc, arm-none-eabi-g++,
arm-none-eabi-objcopy, and arm-none-eabi-size available in PATH. The
Cortex-Debug extension is only needed to generate a debug configuration.
- Linux: a cross toolchain with
loongarch64-linux-gnu-gcc,
aarch64-linux-gnu-gcc, arm-linux-gnueabihf-gcc, or riscv64-linux-gnu-gcc
(and the matching g++) available in PATH.
Development
npm install
npm test
Press F5 in VS Code to launch an Extension Development Host, open a project,
and run CMake-Embedded: Generate MCU Project or CMake-Embedded: Generate Linux Project from the Command Palette.
To debug the Linux direction inside WSL, open this extension project in a WSL
window first and press F5 there, so the extension host runs on the WSL side
and sees the WSL PATH.
Project Status
This repository is an early extension prototype. Device profiles and generated
templates are intentionally kept small so more MCU families can be added
incrementally.