mirror of
https://github.com/bytecodealliance/wasm-micro-runtime.git
synced 2025-09-05 17:32:26 +00:00
WebAssembly Micro Runtime (WAMR)
aotassembly-scriptembeddedinterpreteriotjitpthreadruntimesgxwasi-nnwasi-threadswasmwasm-socketwebassembly
29d465b44e
for wasi_ephemeral_nn,
* implement u8 input
* stop dealing with quantization.
* wasi-nn doesn't have a concept of quantization or pre/post-processing.
i can't think of any ways to make the backend perform zero-point/scale
processing without risking to break other applications.
* there seems to be applications which just use u8 inputs/outputs for
a quantized model. (see [1] for an example.)
for certain kinds of inputs/outputs, it usually just works.
this commit keeps the legacy wasi_nn logic intact for now.
tested with [1] with [2] applied.
WAMR with this patch:
```
Read graph weights, size in bytes: 3561598
[wasi_nn.c:297 WARNING] load_by_name_with_config() not found
[wasi_nn_tensorflowlite.cpp:272 WARNING] Default encoding is CPU.
Loaded graph into wasi-nn with ID: Graph#0
Read input tensor, size in bytes: 150528
1.) [166](198)Aix galericulata
2.) [34](1)Gallus gallus domesticus
3.) [158](1)Coccothraustes coccothraustes
4.) [778](1)Sitta europaea
5.) [819](1)Anas platyrhynchos
```
wasmedge:
```
Read graph weights, size in bytes: 3561598
Loaded graph into wasi-nn with ID: Graph#0
Read input tensor, size in bytes: 150528
1.) [166](198)Aix galericulata
2.) [34](1)Gallus gallus domesticus
3.) [158](1)Coccothraustes coccothraustes
4.) [778](1)Sitta europaea
5.) [819](1)Anas platyrhynchos
```
and "Aix galericulata" seems like a reasonable classification
of the image to my eyes.
[1]
|
||
|---|---|---|
| .devcontainer | ||
| .github | ||
| build-scripts | ||
| ci | ||
| core | ||
| doc | ||
| language-bindings | ||
| product-mini | ||
| samples | ||
| test-tools | ||
| tests | ||
| wamr-compiler | ||
| wamr-sdk | ||
| wamr-wasi-extensions | ||
| zephyr | ||
| .clang-format | ||
| .clang-tidy | ||
| .dockerignore | ||
| .gitignore | ||
| ADOPTERS.md | ||
| ATTRIBUTIONS.md | ||
| CMakeLists.txt | ||
| CODE_OF_CONDUCT.md | ||
| CODEOWNERS | ||
| CONTRIBUTING.md | ||
| idf_component.yml | ||
| LICENSE | ||
| ORG_CODE_OF_CONDUCT.md | ||
| README.md | ||
| RELEASE_NOTES.md | ||
| SConscript | ||
| SECURITY.md | ||
| TSC_Charter.md | ||
WebAssembly Micro Runtime
A Bytecode Alliance project
Build WAMR | Build AOT Compiler | Embed WAMR | Export Native API | Build Wasm Apps | Samples
WebAssembly Micro Runtime (WAMR) is a lightweight standalone WebAssembly (Wasm) runtime with small footprint, high performance and highly configurable features for applications cross from embedded, IoT, edge to Trusted Execution Environment (TEE), smart contract, cloud native and so on. It includes a few parts as below:
- VMcore: A set of runtime libraries for loading and running Wasm modules. It supports rich running modes including interpreter, Ahead-of-Time compilation(AoT) and Just-in-Time compilation (JIT). WAMR supports two JIT tiers - Fast JIT, LLVM JIT, and dynamic tier-up from Fast JIT to LLVM JIT.
- iwasm: The executable binary built with WAMR VMcore which supports WASI and command line interface.
- wamrc: The AOT compiler to compile Wasm file into AOT file
- Useful components and tools for building real solutions with WAMR vmcore:
- App-framework: A framework for supporting APIs for the Wasm applications
- App-manager: A framework for dynamical loading the Wasm module remotely
- WAMR-IDE: An experimental VSCode extension for developping WebAssembly applications with C/C++
Key features
- Full compliant to the W3C Wasm MVP
- Small runtime binary size (core vmlib on cortex-m4f with tail-call/bulk memory/shared memory support, text size from bloaty)
- ~58.9K for fast interpreter
- ~56.3K for classic interpreter
- ~29.4K for aot runtime
- ~21.4K for libc-wasi library
- ~3.7K for libc-builtin library
- Near to native speed by AOT and JIT
- Self-implemented AOT module loader to enable AOT working on Linux, Windows, MacOS, Android, SGX and MCU systems
- Choices of Wasm application libc support: the built-in libc subset for the embedded environment or WASI for the standard libc
- The simple C APIs to embed WAMR into host environment, see how to integrate WAMR and the API list
- The mechanism to export native APIs to Wasm applications, see how to register native APIs
- Multiple modules as dependencies, ref to document and sample
- Multi-thread, pthread APIs and thread management, ref to document and sample
- wasi-threads, ref to document and sample
- Linux SGX (Intel Software Guard Extension) support, ref to document
- Source debugging support, ref to document
- XIP (Execution In Place) support, ref to document
- Berkeley/Posix Socket support, ref to document and sample
- Multi-tier JIT and Running mode control
- Language bindings: Go, Python, Rust
Wasm post-MVP features
- wasm-c-api, ref to document and sample
- 128-bit SIMD, ref to samples/workload
- Reference Types, ref to document and sample
- Bulk memory operations, Shared memory, Memory64
- Tail-call, Garbage Collection, Exception Handling
- Extended Constant Expressions
Supported architectures and platforms
The WAMR VMcore supports the following architectures:
- X86-64, X86-32
- ARM, THUMB (ARMV7 Cortex-M7 and Cortex-A15 are tested)
- AArch64 (Cortex-A57 and Cortex-A53 are tested)
- RISCV64, RISCV32 (RISC-V LP64 and RISC-V LP64D are tested)
- XTENSA, MIPS, ARC
The following platforms are supported, click each link below for how to build iwasm on that platform. Refer to WAMR porting guide for how to port WAMR to a new platform.
- Linux, Linux SGX (Intel Software Guard Extension), MacOS, Android, Windows, Windows (MinGW, MSVC)
- Zephyr, AliOS-Things, VxWorks, NuttX, RT-Thread, ESP-IDF(FreeRTOS)
Getting started
- Build VM core and Build wamrc AOT compiler
- Build iwasm (mini product): Linux, SGX, MacOS and Windows
- Embed into C/C++, Embed into Python, Embed into Go, Embed in Rust
- Register native APIs for Wasm applications
- Build wamrc AOT compiler
- Build Wasm applications
- Port WAMR to a new platform
- VS Code development container
- Samples and Benchmarks
- End-user APIs documentation
Performance and memory
- Blog: The WAMR memory model
- Blog: Understand WAMR heaps and stacks
- Blog: Introduction to WAMR running modes
- Memory usage tuning: the memory model and how to tune the memory usage
- Memory usage profiling: how to profile the memory usage
- Performance tuning: how to tune the performance
- Benchmarks: checkout these links for how to run the benchmarks: PolyBench, CoreMark, Sightglass, JetStream2
- Performance and footprint data: the performance and footprint data
Project Technical Steering Committee
The WAMR PTSC Charter governs the operations of the project TSC. The current TSC members:
- dongsheng28849455 - Dongsheng Yan, dongsheng.yan@sony.com
- loganek - Marcin Kolny, mkolny@amazon.co.uk
- lum1n0us - Liang He， liang.he@intel.com
- no1wudi Qi Huang, huangqi3@xiaomi.com
- qinxk-inter - Xiaokang Qin， xiaokang.qxk@antgroup.com
- ttrenner - Trenner, Thomas， trenner.thomas@siemens.com
- wei-tang - Wei Tang， tangwei.tang@antgroup.com
- wenyongh - Wenyong Huang， wenyong.huang@intel.com
- woodsmc - Woods, Chris， chris.woods@siemens.com
- xujuntwt95329 - Jun Xu， Jun1.Xu@intel.com
- xwang98 - Xin Wang， xin.wang@intel.com (chair)
- yamt - Takashi Yamamoto, yamamoto@midokura.com
License
WAMR uses the same license as LLVM: the Apache 2.0 license with the LLVM
exception. See the LICENSE file for details. This license allows you to freely
use, modify, distribute and sell your own products based on WAMR.
Any contributions you make will be under the same license.