feat: initial WASM support (VowpalWabbit#4562)

Co-authored-by: olgavrou <[email protected]>

.github/workflows/asan.yml

@@ -94,7 +94,7 @@ jobs:
           cmakeListsTxtPath: "${{ github.workspace }}/CMakeLists.txt"
           configurePreset: "${{ matrix.preset }}"
       - name: Build
-        run: cmake --build build -t vw_cli_bin spanning_tree vw_core_test
+        run: cmake --build build -t vw_cli_bin vw_spanning_tree vw_spanning_tree_bin vw_core_test
       - name: Run unit tests
         run: ./build/vowpalwabbit/core/vw_core_test --gtest_filter=-\*WIiterations
       - name: Run python test script

.github/workflows/wasm.yml

@@ -0,0 +1,39 @@
+name: Wasm
+
+on:
+  push:
+    branches:
+      - master
+      - 'releases/**'
+  pull_request:
+    branches:
+      - '*'
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.head_ref || github.sha }}
+  cancel-in-progress: true
+
+jobs:
+  job:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v2
+        with:
+          submodules: true
+      - uses: lukka/get-cmake@latest
+      - uses: mymindstorm/setup-emsdk@v11
+      - uses: actions/setup-node@v3
+        with:
+            node-version: 16
+      - name: Configure
+        run: emcmake cmake --preset wasm -G Ninja -DCMAKE_BUILD_TYPE=MinSizeRel -DCMAKE_TOOLCHAIN_FILE=$(pwd)/ext_libs/vcpkg/scripts/buildsystems/vcpkg.cmake
+      - name: Build
+        run: cmake --build build --target vw-wasm
+      - uses: actions/upload-artifact@v3
+        with:
+          path: wasm/out
+      - name: Test
+        working-directory: wasm
+        run: |
+          npm install
+          npm test

CMakeLists.txt

@@ -154,6 +154,7 @@ option(vw_BUILD_NET_CORE "Build .NET Core targets" OFF)
 option(vw_BUILD_NET_FRAMEWORK "Build .NET Framework targets" OFF)
 option(VW_USE_ASAN "Compile with AddressSanitizer" OFF)
 option(VW_USE_UBSAN "Compile with UndefinedBehaviorSanitizer" OFF)
+option(VW_BUILD_WASM "Add WASM target" OFF)
 
 if(VW_USE_ASAN)
   add_compile_definitions(VW_USE_ASAN)
@@ -260,6 +261,10 @@ if(BUILD_BENCHMARKS)
   add_subdirectory(test/benchmarks)
 endif()
 
+if(VW_BUILD_WASM)
+  add_subdirectory(wasm)
+endif()
+
 if(CMAKE_PROJECT_NAME STREQUAL PROJECT_NAME AND BUILD_TESTING)
   add_subdirectory(test)
 

CMakePresets.json

@@ -158,6 +158,58 @@
           "value": "On"
         }
       }
+    },
+    {
+      "name": "wasm",
+      "description": "CMAKE_TOOLCHAIN_FILE has to be passed explicitly since emcmake overwrites it if we define it here.",
+      "inherits": "default",
+      "cacheVariables": {
+        "VCPKG_CHAINLOAD_TOOLCHAIN_FILE":
+        {
+          "type": "FILEPATH",
+          "value": "$env{EMSDK}/upstream/emscripten/cmake/Modules/Platform/Emscripten.cmake"
+        },
+        "VW_BUILD_WASM": {
+          "type": "BOOL",
+          "value": "On"
+        },
+        "USE_LATEST_STD": {
+          "type": "BOOL",
+          "value": "On"
+        },
+        "VW_FEAT_FLATBUFFERS": {
+          "type": "BOOL",
+          "value": "Off"
+        },
+        "VW_FEAT_CSV": {
+          "type": "BOOL",
+          "value": "Off"
+        },
+        "VW_FEAT_LDA": {
+          "type": "BOOL",
+          "value": "Off"
+        },
+        "VW_FEAT_SEARCH": {
+          "type": "BOOL",
+          "value": "Off"
+        },
+        "VW_FEAT_NETWORKING": {
+          "type": "BOOL",
+          "value": "Off"
+        },
+        "BUILD_TESTING": {
+          "type": "BOOL",
+          "value": "Off"
+        },
+        "VW_BUILD_VW_C_WRAPPER": {
+          "type": "BOOL",
+          "value": "Off"
+        },
+        "VCPKG_TARGET_TRIPLET": {
+          "type": "STRING",
+          "value": "wasm32-emscripten"
+        }
+      }
     }
   ],
   "buildPresets": [

vowpalwabbit/core/src/confidence_sequence.cc

@@ -204,7 +204,7 @@ double confidence_sequence::approxpolygammaone(double b) const
 
 double confidence_sequence::lblogwealth(double sumXt, double v, double eta, double s, double lb_alpha) const
 {
-#if !defined(__APPLE__) && !defined(_WIN32)
+#if !defined(__APPLE__) && !defined(_WIN32) && !defined(__EMSCRIPTEN__)
   double zeta_s = std::riemann_zeta(s);
 #else
   double zeta_s = 10.584448464950803;  // std::riemann_zeta(s) -- Assuming s=1.1 is constant

wasm/CMakeLists.txt

@@ -0,0 +1,9 @@
+if(VW_INSTALL)
+    message(FATAL_ERROR "Install not supported for WASM build" )
+endif()
+
+set(CMAKE_RUNTIME_OUTPUT_DIRECTORY "${CMAKE_CURRENT_LIST_DIR}/out/")
+
+add_executable(vw-wasm wasm_wrapper.cc)
+set_target_properties(vw-wasm PROPERTIES LINK_FLAGS "-fexceptions -s WASM=1 -s NO_DYNAMIC_EXECUTION=1 --bind -s ALLOW_MEMORY_GROWTH=1 -s EXPORTED_FUNCTIONS=\"['_malloc', '_free']\"")
+target_link_libraries(vw-wasm PUBLIC vw_explore vw_core "-fexceptions")

wasm/example.js

@@ -0,0 +1,44 @@
+
+const VWModule = require('./out/vw-wasm.js');
+
+// You must wait for the WASM runtime to be ready before using the module.
+VWModule['onRuntimeInitialized'] = () => {
+    try {
+        // Create a model with default options
+        let model = new VWModule.VWModel("");
+
+        let example_line = "0 | price:.23 sqft:.25 age:.05 2006";
+        // For multi_ex learners, the input to parse should have newlines in it.
+        // One call to parse should only ever get input for either a single
+        // example or a single multi_ex grouping.
+        let parsedExample = model.parse(example_line);
+
+        // Prediction returns a normal javascript value which corresponds to the
+        // expected prediction type. Here is just a number.
+        console.log(`prediction: ${model.predict(parsedExample)}`);
+
+        model.learn(parsedExample);
+
+        // Any examples which were given to either learn and/or predict must be
+        // given to finishExample.
+        model.finishExample(parsedExample);
+
+        // Every object returned by parse must be deleted manually.
+        // Additionally, shallow or deep example copies must be deleted.
+        //      let shallowCopyExample = parsedExample;
+        //      shallowCopyExample.delete();
+        //
+        //      let deepCopyExample = parsedExample.clone(model);
+        //      deepCopyExample.delete();
+        parsedExample.delete();
+
+        // VWModel must also always be manually deleted.
+        model.delete();
+    }
+    catch (e) {
+        // Exceptions that are produced by the module must be passed through
+        // this transformation function to get the error info.
+        console.error(VWModule.getExceptionMessage(e));
+    }
+}
+

wasm/package.json

@@ -0,0 +1,9 @@
+{
+  "devDependencies": {
+    "mocha": "^9.1.2"
+  },
+  "scripts": {
+    "test1": "mocha --delay",
+    "test": "node --experimental-wasm-threads ./node_modules/mocha/bin/mocha --delay"
+  }
+}

wasm/readme.md

@@ -0,0 +1,46 @@
+# VW - WASM
+
+This module is currently very experimental. It is not currently built standalone and still requires the glue JS to function. Ideally it will be standalone WASM module eventually.
+
+## Use docker container
+### Build
+```sh
+# Run in VW root directory
+docker run --rm -v $(pwd):/src -it emscripten/emsdk emcmake cmake -G "Unix Makefiles" --preset wasm -DCMAKE_BUILD_TYPE=MinSizeRel -DCMAKE_TOOLCHAIN_FILE=/src/ext_libs/vcpkg/scripts/buildsystems/vcpkg.cmake
+docker run --rm -v $(pwd):/src -it emscripten/emsdk emcmake cmake --build /src/build --target vw-wasm -j $(nproc)
+```
+
+Artifacts are placed in `wasm/out`
+
+### Test
+Assumes required build artifacts are in `wasm/out`
+
+```sh
+# Run in VW root directory
+docker run --workdir="/src/wasm" --rm -v $(pwd):/src -t node:16-alpine npm install
+docker run --workdir="/src/wasm" --rm -v $(pwd):/src -t node:16-alpine npm test
+```
+
+## Or, setup local environment to build
+
+### Install Emscripten and activate environment
+Instructions here: https://emscripten.org/docs/getting_started/downloads.html
+```sh
+git clone https://github.com/emscripten-core/emsdk.git
+cd emsdk
+./emsdk install latest
+./emsdk activate latest
+source ./emsdk_env.sh
+```
+
+### Build
+Make sure Emscripten is activated.
+```sh
+emcmake cmake --preset wasm -DCMAKE_BUILD_TYPE=MinSizeRel -DCMAKE_TOOLCHAIN_FILE=$(pwd)/ext_libs/vcpkg/scripts/buildsystems/vcpkg.cmake
+cmake --build build --target vw-wasm
+```
+
+### Test
+```sh
+npm test
+```