Files
wamr/samples/workload/XNNPACK/README.md
liang.he ecd4fccc96 Upgrade XNNPACK workload (#2394)
- Sync source code to b9d4073a6913891ce9cbd8965c8d506075d2a45a, which is
  referred by tensorflow
- Upgrade emscripten to 3.1.44
- CMake outputs are .wasm files and .aot files
2023-08-01 19:13:00 +08:00

50 lines
1.0 KiB
Markdown

"XNNPACK" sample introduction
==============
This sample demonstrates how to build [XNNPACK](https://github.com/google/XNNPACK) benchmarks into WebAssembly with emsdk toolchain and run them with iwasm.
## Installation toolchains
please refer to [installation instructions](../README.md).
## Build XNNPACK
please build wamrc:
``` bash
cd <wamr-dir>/wamr-compiler
./build_llvm.sh
mkdir build && cd build
cmake ..
make
```
And then build xnnpack standalone wasm files
```bash
$ cd <wamr-dir>/samples/workload/XNNPACK
$ cmake -S . -B build
$ cmake --build build
```
Generated .wasm(and .aot) files are under *samples/workload/XNNPACK/build*.
## Run benchmarks
Firstly please build iwasm with simd, libc-emcc and lib-pthread supporting:
``` bash
$ cd <wamr-dir>/product-mini/platforms/linux/
$ mkdir build && cd build
$ cmake .. -DWAMR_BUILD_LIBC_EMCC=1 -DWAMR_BUILD_LIB_PTHREAD=1
$ make
```
Then run:
``` shell
$ cd <wamr-dir>/samples/workload/XNNPACK/build
$ iwasm averag_pooling_bench.aot # (or other aot files)
```