WebAssembly bindings

.github/workflows/main.yml

@@ -17,6 +17,7 @@ env:
   ccache_maxsize: 200M
   ccache_cmake: -DCMAKE_CXX_COMPILER_LAUNCHER=ccache -DCMAKE_C_COMPILER_LAUNCHER=ccache
   ccache_version: '4.5'
+  emsdk_version: 3.1.8 # For use in emscripten build
 
 jobs:
   formatting:
@@ -45,7 +46,7 @@ jobs:
           run:
             bash scripts/ci/format-check.sh
 
-        - uses: actions/upload-artifact@v2
+        - uses: actions/upload-artifact@v4
           if: always()
           with:
             name: format
@@ -88,7 +89,7 @@ jobs:
           run:
             bash scripts/ci/${{ matrix.name }}/04-package.sh
 
-        - uses: actions/upload-artifact@v2
+        - uses: actions/upload-artifact@v4
           if: ${{ failure() }}
           with:
             name: slimt
@@ -122,7 +123,7 @@ jobs:
             submodules: recursive
         - name: "Fetch pybind11"
           run: |
-            git clone https://github.com/pybind/pybind11 3rd-party/pybind11
+            git clone --branch v2.10.3 https://github.com/pybind/pybind11 3rd-party/pybind11
 
         - name: Generate ccache_vars for ccache based on machine
           shell: bash
@@ -194,7 +195,7 @@ jobs:
 
 
         - name: Build wheels
-          uses: pypa/cibuildwheel@v2.11.1
+          uses: pypa/cibuildwheel@v2.21.3
           # to supply options, put them in 'env', like:
           env:
             CIBW_BUILD_VERBOSITY: 3
@@ -285,7 +286,7 @@ jobs:
               # gdb -return-child-result -batch -ex 'set follow-fork-mode child' -ex 'run' -ex 'bt' -ex 'quit' --args python -m pytest --pyargs slimt -s
 
 
-        - uses: actions/upload-artifact@v2
+        - uses: actions/upload-artifact@v4
           with:
             name: wheels
             path: ./wheelhouse/*.whl
@@ -367,3 +368,128 @@ jobs:
       run: |
         python3 -m pip install twine
         twine upload *.whl
+
+  build-wasm:
+    name: "emscripten"
+    runs-on: ubuntu-latest
+    steps:
+
+      - name: Checkout
+        uses: actions/checkout@v2
+        with:
+          submodules: recursive
+
+      - name: Set ccache environment for emcc
+        run: |
+          # We are hardcoding this to mtime instead of env pickup. Rest use content.
+          echo "CCACHE_COMPILER_CHECK=mtime" >> $GITHUB_ENV
+
+          echo "CCACHE_BASEDIR=${{ env.ccache_basedir }}" >> $GITHUB_ENV
+          echo "CCACHE_COMPRESS=${{ env.ccache_compress }}" >> $GITHUB_ENV
+          echo "CCACHE_COMPRESSLEVEL=${{ env.ccache_compresslevel }}" >> $GITHUB_ENV
+          echo "CCACHE_DIR=${{ env.ccache_dir }}" >> $GITHUB_ENV
+          echo "CCACHE_MAXSIZE=${{ env.ccache_maxsize }}" >> $GITHUB_ENV
+          # https://emscripten.org/docs/compiling/Building-Projects.html#using-a-compiler-wrapper
+          echo "EM_COMPILER_WRAPPER=ccache" >> $GITHUB_ENV
+
+      # This need to be run before setup, so ccache build caching doesn't complain.
+      - name: Obtain emsdk sources
+        run: |
+            git clone --depth 1 https://github.com/emscripten-core/emsdk.git
+
+      - name: Cache-op for build-cache through ccache
+        uses: actions/cache@v2
+        with:
+          path: |
+              ${{ env.ccache_dir }}
+              ${{ github.workspace }}/emsdk/ccache/git-emscripten_64bit/
+          key: ccache-${{ github.job }}-${{ env.emsdk_version }}-${{ steps.ccache_vars.outputs.hash }}-${{ github.ref }}-${{ steps.ccache_vars.outputs.timestamp }}
+          restore-keys: |-
+            ccache-${{ github.job }}-${{ env.emsdk_version }}-${{ steps.ccache_vars.outputs.hash }}-${{ github.ref }}
+            ccache-${{ github.job }}-${{ env.emsdk_version }}-${{ steps.ccache_vars.outputs.hash }}
+            ccache-${{ github.job }}-${{ env.emsdk_version }}
+
+      - name: Setup Emscripten toolchain
+        run: |
+            (cd emsdk && ./emsdk install ${{ env.emsdk_version }} ccache-git-emscripten-64bit)
+            (cd emsdk && ./emsdk activate ${{ env.emsdk_version }} ccache-git-emscripten-64bit)
+            # mtime of this file is checked by ccache, we set it to avoid cache misses.
+            touch -m -d '1 Jan 2021 12:00' emsdk/.emscripten
+
+            # These needs to be done in the activated shell.
+            eval $(./emsdk/emsdk construct_env \
+                | sed 's/export PATH=\(.*\);/echo \1 >> $GITHUB_PATH;/' \
+                | sed 's/export \(.*\);/echo \1 >> $GITHUB_ENV;/' );
+
+            # This looks more permanent than version pinned, so keeping temporarily to avoid failures.
+            echo "${{ github.workspace }}/emsdk/ccache/git-emscripten_64bit/bin" >> $GITHUB_PATH
+
+      - name: Generate ccache_vars for ccache based on machine
+        shell: bash
+        id: ccache_vars
+        run: |-
+          echo "::set-output name=hash::$(echo ${{ env.ccache_compilercheck }})"
+          echo "::set-output name=timestamp::$(date '+%Y-%m-%dT%H.%M.%S')"
+
+      - name: Verify Emscripten setup
+        run: |
+            emcc --version
+            emcmake cmake --version
+            emmake make --version
+
+      - name: ccache prolog
+        run: |-
+          ccache -s # Print current cache stats
+          ccache -z # Zero cache entry
+
+      - name: "Configure builds"
+        run: |
+          mkdir -p build-wasm
+          cd build-wasm
+          emcmake cmake -DWITH_BLAS=OFF -DWITH_RUY=ON -DWITH_GEMMOLOGY=OFF -DSLIMT_USE_INTERNAL_PCRE2=ON -DBUILD_WASM=ON ..
+
+
+      - name: "Compile"
+        working-directory: build-wasm
+        run: |
+          emmake make -j2
+
+      - name: ccache epilog
+        run: |
+          ccache -s # Print current cache stats
+
+      # The following is useless as it works within Firefox, that too only in a
+      # privileged context. Might choose to enable should Mozilla be
+      # interested.
+      #
+      # - name: Import GEMM library from a separate wasm module
+      #   working-directory: build-wasm
+      #   run: bash ../wasm/patch-artifacts-import-gemm-module.sh
+
+      # Setup nodejs-18, as nodejs-14 provided by emsdk fails when running
+      # and newer version of node allows us to use fetch().
+      - name: Setup nodejs
+        uses: actions/setup-node@v3
+        with:
+          node-version: 18
+
+      # - name: Test run
+      #   working-directory: wasm
+      #   run: |
+      #     cp ../build-wasm/bergamot-translator-worker.{js,wasm} ./
+      #     npm install jsdom
+
+      #     # --unhandled-rejections make the script exit with a non-zero code (at least on node-14).
+      #     # So leaving this here.
+      #     node --unhandled-rejections=strict node-test.js
+
+      # # Upload both together.
+      # - name: Upload wasm artifact
+      #   uses: actions/upload-artifact@v4
+      #   with:
+      #     name: wasm-artefacts
+      #     if-no-files-found: error
+      #     path: |
+      #         ${{github.workspace}}/build-wasm/bergamot-translator-worker.js
+      #         ${{github.workspace}}/build-wasm/bergamot-translator-worker.wasm
+      #         ${{github.workspace}}/build-wasm/bergamot-translator-worker.js.bak

CMakeLists.txt

@@ -35,6 +35,14 @@ option(SLIMT_PYTHON_LINK_STATIC
        "link-method to produce python package (static/shared)" ON)
 option(SLIMT_GENERATED_UNIT_TESTS "Generate unit tests to run using Python" OFF)
 
+option(BUILD_CLI "Build command-line applications" ON)
+option(BUILD_PYTHON "Build Python bindings library" OFF)
+option(BUILD_WASM "Build WebAssemby bindings library" OFF)
+
+if(BUILD_WASM)
+  set(BUILD_CLI OFF)
+endif(BUILD_WASM)
+
 include(MacroEnsureOutOfSourceBuild)
 macro_ensure_out_of_source_build(
   "${PROJECT_NAME} requires an out of source build. Please create a separate build directory and run 'cmake /path/to/${PROJECT_NAME} [options]' there."
@@ -119,6 +127,12 @@ if(WITH_BLAS)
     )
     add_library(AppleLibs::accelerate ALIAS apple_accelerate)
     list(APPEND SLIMT_PRIVATE_LIBS AppleLibs::accelerate)
+    # 'cblas_sgemm' is deprecated: first deprecated in macOS 13.3 - An updated
+    # CBLAS interface supporting ILP64 is available.  Please compile with
+    # -DACCELERATE_NEW_LAPACK to access the new headers and
+    # -DACCELERATE_LAPACK_ILP64 for ILP64 support.
+    list(APPEND SLIMT_COMPILE_DEFINITIONS ACCELERATE_NEW_LAPACK
+         ACCELERATE_LAPACK_ILP64)
   endif(APPLE)
 endif(WITH_BLAS)
 
@@ -219,10 +233,13 @@ if(BUILD_PYTHON)
   else(USE_PYBIND11_SOURCE)
     find_package(pybind11 REQUIRED)
   endif(USE_PYBIND11_SOURCE)
-
   add_subdirectory(bindings/python)
 endif(BUILD_PYTHON)
 
+if(BUILD_WASM)
+  add_subdirectory(bindings/wasm)
+endif(BUILD_WASM)
+
 if(BUILD_JNI)
   add_subdirectory(bindings/java)
 endif(BUILD_JNI)

app/CMakeLists.txt

@@ -1,9 +1,11 @@
-add_executable(slimt_cli main.cc)
-set_target_properties(slimt_cli PROPERTIES OUTPUT_NAME "slimt-cli")
-target_link_libraries(slimt_cli PUBLIC slimt)
+if(BUILD_CLI)
+  add_executable(slimt_cli main.cc)
+  set_target_properties(slimt_cli PROPERTIES OUTPUT_NAME "slimt-cli")
+  target_link_libraries(slimt_cli PUBLIC slimt-static)
 
-set(SLIMT_BINARIES slimt_cli)
+  set(SLIMT_BINARIES slimt_cli)
 
-if(UNIX)
-  install(TARGETS ${SLIMT_BINARIES} DESTINATION ${CMAKE_INSTALL_BINDIR})
-endif(UNIX)
+  if(UNIX)
+    install(TARGETS ${SLIMT_BINARIES} DESTINATION ${CMAKE_INSTALL_BINDIR})
+  endif(UNIX)
+endif(BUILD_CLI)

bindings/wasm/CMakeLists.txt

@@ -0,0 +1,84 @@
+# See https://github.com/emscripten-core/emscripten/blob/main/src/settings.js
+list(
+  APPEND
+  WASM_COMPILE_FLAGS
+  -O3
+  # Preserve whitespaces in JS even for release builds; this doesn't increase
+  # wasm binary size
+  $<$<CONFIG:Release>:-g1>
+  # Relevant Debug info only for release with debug builds as this increases
+  # wasm binary size
+  $<$<CONFIG:RelWithDebInfo>:-g2>
+  -fPIC
+  -mssse3
+  -msimd128
+  # -fno-exceptions # Can't do that because spdlog uses exceptions
+  -sDISABLE_EXCEPTION_CATCHING=1
+  -sSTRICT=1)
+list(
+  APPEND
+  WASM_LINK_FLAGS
+  -O3
+  # Preserve whitespaces in JS even for release builds; this doesn't increase
+  # wasm binary size
+  $<$<CONFIG:Release>:-g1>
+  # Relevant Debug info only for release with debug builds as this increases
+  # wasm binary size
+  $<$<CONFIG:RelWithDebInfo>:-g2>
+  -lembind
+  # Save some code, and some speed
+  -sASSERTIONS=0
+  -sDISABLE_EXCEPTION_CATCHING=1
+  # the intgemm functions we call will be undefined since these are linked at
+  # runtime by our own javascript.
+  #
+  # jerin: Disabling this particular flag, as it errors. em++: error: legacy
+  # setting used in strict mode: LLD_REPORT_UNDEFINED -sLLD_REPORT_UNDEFINED
+  -sERROR_ON_UNDEFINED_SYMBOLS=0
+  # Cause we can!
+  -sSTRICT=1
+  # You know we need it
+  -sALLOW_MEMORY_GROWTH=1
+  -sENVIRONMENT=web,worker
+  # No need to call main(), there's nothing there.
+  -sINVOKE_RUN=0
+  # No need for filesystem code in the generated Javascript
+  -sFILESYSTEM=0
+  # If you turn this on, it will mangle names which makes the dynamic linking
+  # hard.
+  -sDECLARE_ASM_MODULE_EXPORTS=0
+  # Export all of the intgemm functions in case we need to fall back to using
+  # the embedded intgemm
+  # -sEXPORTED_FUNCTIONS=[_int8PrepareAFallback,_int8PrepareBFallback,_int8PrepareBFromTransposedFallback,_int8PrepareBFromQuantizedTransposedFallback,_int8PrepareBiasFallback,_int8MultiplyAndAddBiasFallback,_int8SelectColumnsOfBFallback]
+  # Necessary for mozintgemm linking. This prepares the `wasmMemory` variable
+  # ahead of time as opposed to delegating that task to the wasm binary itself.
+  # This way we can link MozIntGEMM module to the same memory as the main
+  # bergamot-translator module.
+  -sIMPORTED_MEMORY=1
+  # Dynamic execution is either frowned upon or blocked inside browser
+  # extensions
+  -sDYNAMIC_EXECUTION=0
+  # no main, pass --no-entry to suppress, said linker.
+  --no-entry)
+
+add_executable(wasm-slimt slimt.cpp)
+
+# Generate version file that can be included in the wasm artifacts
+configure_file(${CMAKE_CURRENT_SOURCE_DIR}/project_version.js.in
+               ${CMAKE_CURRENT_BINARY_DIR}/project_version.js @ONLY)
+
+target_link_libraries(wasm-slimt PRIVATE slimt-static)
+
+# This compile definition is required for generating binding code properly
+target_compile_definitions(wasm-slimt PRIVATE WASM_BINDINGS)
+target_compile_options(wasm-slimt PRIVATE ${WASM_COMPILE_FLAGS})
+target_link_options(wasm-slimt PRIVATE ${WASM_LINK_FLAGS})
+target_link_options(
+  wasm-slimt PRIVATE
+  --extern-pre-js=${CMAKE_CURRENT_BINARY_DIR}/project_version.js)
+
+set_target_properties(
+  wasm-slimt
+  PROPERTIES OUTPUT_NAME "slimt"
+             SUFFIX ".js"
+             RUNTIME_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR})

bindings/wasm/project_version.js.in

@@ -0,0 +1 @@
+var SLIMT_VERSION_FULL = "@PROJECT_VERSION_STRING_FULL@";