WALRUS: WebAssembly Lightweight RUntime

This project aims to provide a lightweight WebAssembly runtime engine. It now fully supports WebAssembly specs with an simple interpreter, but we plan to optimize interpreting as well as adopting JIT compiler for better performance.

Cloning

Clone as normal, but don't forget to get the submodules as well:

$ git clone --recursive https://github.com/Samsung/walrus
$ cd walrus
$ git submodule update --init

This will fetch the testsuite and gtest repos, which are needed for some tests.

Building using CMake

You'll need CMake. You can then run CMake, the normal way:

Note: Due to the need for supporting old targets only supporting cmake 2.8, the defined cmake version is 2.8, however due to many environments shipping newer cmake versions which do not support cmake versions below 3.5, you may need to override the minimum version using -DCMAKE_POLICY_VERSION_MINIMUM=3.5

$ cmake -DCMAKE_POLICY_VERSION_MINIMUM=3.5 -H. -Bout/release/x64 -DWALRUS_ARCH=x64 -DWALRUS_HOST=linux -DWALRUS_MODE=release -DWALRUS_OUTPUT=shell -GNinja
$ ninja -Cout/release/x64
$ ./out/release/x64/walrus test.wasm // run walrus executable

This will produce build files using CMake's default build generator. Read the CMake documentation for more information.

Perf

You'll need Perf.

0. To compile with perf support, use -DWALRUS_JITPERF=1

1. Set the path where temporary files are created with WALRUS_PERF_DIR environment variable. Run Walrus with: sudo perf record -k 1 walrus --jit WALRUS_PARAMETERS The -k 1 option sets the monotonic clock, -k mono is also correct.

   Three files are generated:

   • perf.data - it is generated by perf
   • jit-XXXXXX.dump - The dump of jit generated functions, XXXXXX is the PID of process
   • jit-XXXXXX-codedump.txt - Walrus ByteCode dump, XXXXXX is the PID of process

2. You should chown perf.data, because you don't need sudo after that.

3. Inject JITDump file into perf.data with perf inject --jit -i perf.data -o perf.data.jitted

   It'll generate many shared object files, and perf.data.jitted

4. View the report with perf report -i perf.data.jitted