[LLVM] add wasm32 wheels (#267)

This PR factors out the LLVM distro portion of the build of
`mlir-python-bindings-wasm` - i.e., it creates a `wasm32-empscripten`
wheel/distro in addition to our current set of platforms. This is in
preparation for #268 which will build
out various `mlir-wgpu` things on top of this distro.

One tedious/weeds thing to call out: just building the distro via
cross-compiling (using [Emscripten's
toolchain](https://github.com/emscripten-core/emscripten/blob/main/cmake/Modules/Platform/Emscripten.cmake))
didn't work because `find_package(Python3 Development)` fails to
successfully find Python dev libs. I believe it has something to do with
[CMP0190](https://cmake.org/cmake/help/latest/policy/CMP0190.html) but I
couldn't debug it all the way. But `pyodide build` succeeds (because it
somehow isn't cross-compiling technically???) so that's good enough.

In addition some drive-by cleanup (remove `macos-13` from the GHA
workflows).

Author: makslevental

.github/actions/setup_base/action.yml

@@ -10,7 +10,10 @@ inputs:
   os:
     required: true
     description: ''
-  arch:
+  host-arch:
+    required: true
+    description: ''
+  target-arch:
     required: true
     description: ''
   container:
@@ -119,18 +122,18 @@ runs:
           dnf install -y sudo ncurses-compat-libs tmate python3-pip git
         
           # install ccache
-          if [[ "${{ inputs.arch }}" == "x86_64" ]]; then
+          if [[ "${{ inputs.host-arch }}" == "x86_64" ]]; then
             curl -sLO https://github.com/ccache/ccache/releases/download/v4.10.2/ccache-4.10.2-linux-x86_64.tar.xz
             tar -xf ccache-4.10.2-linux-x86_64.tar.xz
             pushd ccache-4.10.2-linux-x86_64 && make install && popd
-          elif [[ "${{ inputs.arch }}" == "aarch64" ]]; then
+          elif [[ "${{ inputs.host-arch }}" == "aarch64" ]]; then
             dnf install -y ccache
           fi
         
           # install gh-cli
-          if [[ "${{ inputs.arch }}" == "x86_64" ]]; then
+          if [[ "${{ inputs.host-arch }}" == "x86_64" ]]; then
             rpm_arch=amd64
-          elif [[ "${{ inputs.arch }}" == "aarch64" ]]; then
+          elif [[ "${{ inputs.host-arch }}" == "aarch64" ]]; then
             rpm_arch=arm64
           fi
           curl -sLO https://github.com/cli/cli/releases/download/v2.61.0/gh_2.61.0_linux_${rpm_arch}.rpm
@@ -235,3 +238,38 @@ runs:
           echo MACOSX_DEPLOYMENT_TARGET=11.0 >> $GITHUB_ENV
           echo CMAKE_OSX_DEPLOYMENT_TARGET=11.0 >> $GITHUB_ENV
         fi
+
+    - name: Install pyodide
+      if: inputs.target-arch == 'wasm32'
+      shell: bash
+      run: |
+        
+        pip install pyodide-build>=0.28.0
+        echo "EMSCRIPTEN_VERSION=$(pyodide config get emscripten_version)" >> $GITHUB_ENV
+
+    - name: Setup Emscripten
+      if: inputs.target-arch == 'wasm32'
+      shell: bash
+      run: |
+        
+        git clone https://github.com/emscripten-core/emsdk.git
+        pushd emsdk
+        ./emsdk install ${{ env.EMSCRIPTEN_VERSION }}
+        ./emsdk activate ${{ env.EMSCRIPTEN_VERSION }} 
+        echo "EMSDK=$PWD" >> $GITHUB_ENV
+        popd
+
+    - name: Setup emscripten ccache
+      if: inputs.target-arch == 'wasm32'
+      shell: bash
+      run: |
+        
+        source "$EMSDK/emsdk_env.sh"
+        
+        $CMAKE_C_COMPILER_LAUNCHER -z
+        # https://github.com/juj/ccache/blob/c4284c78e9ff286ee5208f449af4c2aaba062d37/cmake/StandardWarnings.cmake#L5
+        # https://github.com/juj/ccache/blob/c4284c78e9ff286ee5208f449af4c2aaba062d37/CMakeLists.txt#L61
+        # error: builtin __has_trivial_copy is deprecated; use __is_trivially_copyable instead [-Werror,-Wdeprecated-builtins]
+        VERBOSE=1 CCACHE_COMPILERCHECK="string:$(g++ --version | head -n 1)" CC=gcc CXX=g++ emsdk install ccache-git-emscripten-64bit
+        $CMAKE_C_COMPILER_LAUNCHER -s
+        emsdk activate ccache-git-emscripten-64bit

.github/workflows/build_llvm.yml

@@ -20,23 +20,23 @@ on:
         default: false
   pull_request:
     paths:
-      - ".github/actions/setup_base"
+      - ".github/actions/setup_base/action.yml"
       - ".github/workflows/build_llvm.yml"
-      - "scripts/cmake/llvm_cache.cmake"
-      - "scripts/cmake/llvm_wasm_cache.cmake"
+      - "scripts/cmake/*"
+      - "scripts/llvm_wasm/*"
       - "scripts/build_llvm.sh"
-      - "scripts/build_llvm_wasi.sh"
+      - "scripts/build_mlir_native_tools.sh"
       - "third_party/llvm-project"
   push:
     branches:
       - main
     paths:
-      - ".github/actions/setup_base"
+      - ".github/actions/setup_base/action.yml"
       - ".github/workflows/build_llvm.yml"
-      - "scripts/cmake/llvm_cache.cmake"
-      - "scripts/cmake/llvm_wasm_cache.cmake"
+      - "scripts/cmake/*"
+      - "scripts/llvm_wasm/*"
       - "scripts/build_llvm.sh"
-      - "scripts/build_llvm_wasi.sh"
+      - "scripts/build_mlir_native_tools.sh"
       - "third_party/llvm-project"
 
 concurrency:
@@ -48,6 +48,7 @@ concurrency:
   cancel-in-progress: true
 
 jobs:
+
   build:
 
     strategy:
@@ -58,33 +59,27 @@ jobs:
             runs-on: "ubuntu-22.04"
             container: "quay.io/pypa/manylinux_2_28_x86_64"
             os: "almalinux"
-            arch: x86_64
+            host-arch: x86_64
+            target-arch: x86_64
+
           - name: "manylinux_aarch64"
             runs-on: "ubuntu-22.04-arm"
             container: "quay.io/pypa/manylinux_2_34_aarch64"
             os: "almalinux"
-            arch: aarch64
-#          - name: "ubuntu_aarch64"
-#            runs-on: "ubuntu-22.04-arm"
-#            os: "ubuntu"
-#            arch: aarch64
-#          - name: "wasm_wasi"
-#            runs-on: "ubuntu-22.04"
-#            container: "quay.io/pypa/manylinux_2_28_x86_64"
-#            os: "almalinux"
-#            arch: "wasm32-wasi"
+            host-arch: "aarch64"
+            target-arch: "aarch64"
+
           - name: "windows_amd64"
             runs-on: "windows-2022"
             os: "windows"
-            arch: amd64
+            host-arch: amd64
+            target-arch: amd64
+
           - name: "macos_arm64"
             runs-on: "macos-14"
             os: "macos"
-            arch: arm64
-#          - name: "macos_x86_64"
-#            runs-on: "macos-13"
-#            os: "macos"
-#            arch: x86_64
+            host-arch: arm64
+            target-arch: arm64
 
     runs-on: ${{ matrix.runs-on }}
 
@@ -127,7 +122,8 @@ jobs:
           cache-key: ${{ env.cache-key }}
           restore-key: "mlir_${{ matrix.name }}_clang"
           os: ${{ matrix.os }}
-          arch: ${{ matrix.arch }}
+          host-arch: ${{ matrix.host-arch }}
+          target-arch: ${{ matrix.target-arch }}
           container: ${{ matrix.container }}
           runs-on: ${{ matrix.runs-on }}
 
@@ -150,14 +146,11 @@ jobs:
           ###############################
           # Build MLIR/LLVM distro
           ###############################
-          if [[ "${{ matrix.arch }}" == "wasm32-wasi" ]]; then
-            scripts/build_llvm_wasi.sh
-          else
-            scripts/build_llvm.sh
-            echo "*********************** SMOKE TEST *********************************"
-            "$LLVM_INSTALL_DIR/bin/mlir-tblgen" --version
-            echo "*********************** SMOKE TEST *********************************"
-          fi
+          
+          scripts/build_llvm.sh
+          echo "*********************** SMOKE TEST *********************************"
+          "$LLVM_INSTALL_DIR/bin/mlir-tblgen" --version
+          echo "*********************** SMOKE TEST *********************************"
           ccache -s
           
           ###############################
@@ -178,7 +171,7 @@ jobs:
           elif [ x"${{ matrix.os }}" == x"windows" ]; then
             PLAT="win"
           fi
-          PLAT="${PLAT}_${{ matrix.arch }}"
+          PLAT="${PLAT}_${{ matrix.host-arch }}"
           
           pushd projects/mlir-native-tools
           $python3_command setup.py bdist_wheel --dist-dir ../../wheelhouse --plat $PLAT
@@ -256,9 +249,117 @@ jobs:
           limit-access-to-actor: true
           install-dependencies: ${{ startsWith(matrix.runs-on, 'macos') || startsWith(matrix.runs-on, 'windows') }}
 
+  build-wasm:
+
+    runs-on: ubuntu-22.04
+
+    name: "Build llvm-mlir wasm"
+
+    defaults:
+      run:
+        shell: bash
+
+    permissions:
+      id-token: write
+      contents: write
+
+    env:
+      cache-key: mlir_wasm_emscripten_${{ format('{0}-{1}', github.ref_name, github.run_number) }}
+
+    steps:
+      - name: "Check out repository"
+        uses: actions/[email protected]
+        with:
+          submodules: true
+
+      - name: "Setup base"
+        uses: ./.github/actions/setup_base
+        id: setup_base
+        with:
+          cache-key: ${{ env.cache-key }}
+          restore-key: "mlir_wasm_emscripten"
+          os: "ubuntu"
+          host-arch: "x86_64"
+          target-arch: "wasm32"
+
+      - name: Build MLIR and LLVM
+        run: |
+          
+          ###############################
+          # Build host mlir-native-tools wheel
+          ###############################
+          
+          scripts/build_mlir_native_tools.sh
+          export LLVM_NATIVE_TOOL_DIR="$PWD/mlir-native-tools-install/bin"
+          export LLVM_NATIVE_TOOL_DIR="$LLVM_NATIVE_TOOL_DIR"
+          export LLVM_TABLEGEN="$LLVM_NATIVE_TOOL_DIR/llvm-tblgen"
+          export MLIR_TABLEGEN="$LLVM_NATIVE_TOOL_DIR/mlir-tblgen"
+          export MLIR_LINALG_ODS_YAML_GEN="$LLVM_NATIVE_TOOL_DIR/mlir-linalg-ods-yaml-gen"
+          
+          source "$EMSDK/emsdk_env.sh"
+          
+          export PATH="$EMSDK/ccache/git-emscripten_64bit/bin:$PATH"
+          export CCACHE_COMPILERCHECK="string:$($EMSDK/upstream/bin/clang --version | head -n 1)"
+          export CCACHE="$EMSDK/ccache/git-emscripten_64bit/bin/ccache"
+          export "CMAKE_C_COMPILER_LAUNCHER=$CCACHE"
+          export "CMAKE_CXX_COMPILER_LAUNCHER=$CCACHE"
+          
+          export LLVM_SOURCE_DIR="$PWD/third_party/llvm-project"
+          pushd $LLVM_SOURCE_DIR && LLVM_SHA_SHORT="$(git rev-parse --short HEAD)" && popd
+          export WHEEL_VERSION="$(date +'%Y%m%d')+$LLVM_SHA_SHORT"
+          echo "WHEEL_VERSION=$WHEEL_VERSION" >> $GITHUB_OUTPUT
+          
+          ###############################
+          # Build MLIR/LLVM distro
+          ###############################
+          
+          $CCACHE -z
+          pyodide build scripts/llvm_wasm -o wheelhouse --compression-level 10
+          $CCACHE -s
+
+      - name: Upload ccache log
+        uses: actions/upload-artifact@v4
+        with:
+          name: llvm_ccache_log_mlir_wasm32
+          path: '/tmp/ccache.log'
+
+      - name: "Save cache"
+        uses: actions/cache/save@v3
+        if: (!cancelled() && github.event_name == 'push' && github.ref_name == 'main')
+        with:
+          path: ${{ steps.setup_base.outputs.cache-dir }}
+          key: ${{ env.cache-key }}
+
+      - name: Upload LLVM distro artifacts
+        uses: actions/upload-artifact@v4
+        with:
+          name: mlir_ubuntu_wasm32_artifact
+          path: llvm-install
+          if-no-files-found: warn
+
+      - name: Upload mlir-wheel artifact
+        uses: actions/upload-artifact@v4
+        with:
+          name: mlir_wheel_ubuntu_wasm32_artifact
+          path: wheelhouse/mlir_wheel*
+          if-no-files-found: warn
+
+      - name: Release current commit
+        if: (github.event_name == 'push' && github.ref_name == 'main') || github.event_name == 'workflow_dispatch'
+        uses: ncipollo/[email protected]
+        with:
+          artifacts: "*.tar.gz,wheelhouse/*.whl"
+          token: "${{ secrets.GITHUB_TOKEN }}"
+          tag: "llvm"
+          name: "llvm"
+          removeArtifacts: false
+          allowUpdates: true
+          replacesArtifacts: true
+          makeLatest: true
+
   call-build-python-bindings:
 
-    needs: [build]
+    needs: [build, build-wasm]
 
     permissions:
       contents: write