mac m1 cross compile (#1204)

* mac m1 cross compile

Add support for M1 cross compile

* Remove redundant ExecutionEngine

It is registered as part of RegisterEverything

* nuke non-universal zstd

disable LTC

.github/workflows/buildAndTest.yml

@@ -13,7 +13,7 @@ jobs:
       fail-fast: true
       matrix:
         os: [macos-12, ubuntu-22.04]
-        targetarch: [x86_64, AArch64]
+        targetarch: [x86_64, arm64]
         python-version: ["3.10"]
         torch-binary: [ON, OFF]
         llvmtype: [source, binary]
@@ -27,7 +27,8 @@ jobs:
           - llvmbuildtype: out-of-tree
             torch-binary: OFF
           # Disable M1 builds until https://github.com/llvm/torch-mlir/issues/1094 is fixed
-          - targetarch: AArch64
+          - targetarch: arm64
+            os: ubuntu-22.04
           # macOS we only do source builds to reduce options
           - os: macos-12
             torch-binary: OFF
@@ -35,6 +36,8 @@ jobs:
             llvmtype: source
           - os: macos-12
             llvmtype: out-of-tree
+          - os: macos-12
+            targetarch: x86_64
     runs-on: ${{ matrix.os }}
 
     steps:
@@ -46,9 +49,37 @@ jobs:
       with:
         cache-suffix: ${{ matrix.os }}-${{ matrix.targetarch }}-${{ matrix.llvmtype }}-${{ matrix.llvmbuildtype }}
 
+    - name: Configure llvm-cross-compile
+      # libzstd on GH Runners are only x86_64 to remove them.
+      if: matrix.targetarch == 'arm64'
+      run: |
+        sudo rm -rf /usr/local/lib/libzstd.*.dylib
+        sudo rm -rf /usr/local/lib/cmake/zstd/*
+        cd $GITHUB_WORKSPACE
+        cmake -GNinja -Bbuild_${{ matrix.targetarch }} \
+          -DCMAKE_BUILD_TYPE=Release \
+          -DCMAKE_LINKER=lld \
+          -DCMAKE_C_COMPILER_LAUNCHER=ccache -DCMAKE_CXX_COMPILER_LAUNCHER=ccache \
+          -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ \
+          -DPython3_EXECUTABLE=$(which python) \
+          -DLLVM_ENABLE_ASSERTIONS=ON \
+          -DLLVM_ENABLE_PROJECTS=mlir \
+          -DLLVM_EXTERNAL_PROJECTS="torch-mlir;torch-mlir-dialects" \
+          -DLLVM_EXTERNAL_TORCH_MLIR_SOURCE_DIR="$GITHUB_WORKSPACE" \
+          -DLLVM_EXTERNAL_TORCH_MLIR_DIALECTS_SOURCE_DIR="${GITHUB_WORKSPACE}/external/llvm-external-projects/torch-mlir-dialects" \
+          -DMLIR_ENABLE_BINDINGS_PYTHON=ON \
+          -DTORCH_MLIR_ENABLE_LTC=OFF \
+          -DTORCH_MLIR_ENABLE_MHLO=OFF \
+          -DTORCH_MLIR_USE_INSTALLED_PYTORCH=${{ matrix.torch-binary }} \
+          -DCMAKE_OSX_ARCHITECTURES=${{ matrix.targetarch }} \
+          -DMACOSX_DEPLOYMENT_TARGET=12.0 \
+          -DLLVM_TARGETS_TO_BUILD="AArch64" \
+          -DLLVM_USE_HOST_TOOLS=ON \
+          $GITHUB_WORKSPACE/externals/llvm-project/llvm
+
     - name: Configure llvm-binary-torch-src-or-binary
       # Should be the fastest builds for CI and fails fast
-      # OSX CMake flags are ignore on Linux
+      # OSX CMake flags are ignored on Linux
       if: matrix.llvmtype == 'binary'
       run: |
         cd $GITHUB_WORKSPACE
@@ -102,12 +133,18 @@ jobs:
           -DTORCH_MLIR_USE_INSTALLED_PYTORCH=${{ matrix.torch-binary }} \
           -DTORCH_MLIR_ENABLE_LTC=OFF \
           -DPython3_EXECUTABLE=$(which python) \
+          -DLLVM_TARGETS_TO_BUILD=host \
           .
 
     - name: Build torch-mlir
-      # Build step for both in-tree and out-of-tree workflows
+      if: matrix.targetarch == 'x86_64'
       run: |
         cmake --build build
+    
+    - name: Build torch-mlir (cross-compile)
+      if: matrix.targetarch == 'arm64'
+      run: |
+        cmake --build build_${{ matrix.targetarch }}
 
     - name: Run torch-mlir unit tests
       if: matrix.llvmtype == 'binary'

python/CMakeLists.txt

@@ -130,7 +130,6 @@ set(_source_components
   MLIRPythonSources
   MLIRPythonExtension.Core
   MLIRPythonExtension.RegisterEverything
-  MLIRPythonExtension.ExecutionEngine
   TorchMLIRPythonSources
   TorchMLIRPythonExtensions
 )