[mlir] Make Python bindings installable.

[stellaraccident]

Summary

• Links against libMLIR.so if the project is built for DYLIBs.
• Puts things in the right place in build and install time python/ trees so that RPaths line up.
• Adds install actions to install both the extension and sources.
• Copies py source files to the build directory to match (consistent layout between build/install time and one place to point a PYTHONPATH for tests and interactive use).
• Finally, "import mlir" from an installed LLVM just works.

[stellaraccident]

Fix typo.

[silvas]

Just a drive-by nit.

Please defer to others that have been tracking this work more closely for more substantial review.

edit: you seem to have caught the nit.

[zhanghb97]

LGTM, I use the following commands in the build directory:

cmake --build . --target check-mlir
cmake --build . --target install

I have a question that if I only want to install MLIRBindingsPythonExtension and MLIRBindingsPythonSources (without installing llvm, llvm-c, mlir, mlir-c, etc.), what should I do?

[stellaraccident]

In D89167#2323226, @zhanghb97 wrote:

LGTM, I use the following commands in the build directory:

cmake --build . --target check-mlir
cmake --build . --target install

I have a question that if I only want to install MLIRBindingsPythonExtension and MLIRBindingsPythonSources (without installing llvm, llvm-c, mlir, mlir-c, etc.), what should I do?

This should work:
cmake --build . --target install-MLIRBindingsPythonSources install-MLIRBindingsPythonExtension

Also, for interactive use, you don't need to install. Just build those two targets and set your PYTHONPATH to the python dir in your build directory.

[zhanghb97]

This should work:
cmake --build . --target install-MLIRBindingsPythonSources install-MLIRBindingsPythonExtension

It works. Thanks!

[ftynse]

mlir/lib/Bindings/Python/CMakeLists.txt

15

Did you mean ${PY_SRC_FILE} (singular) here?

[stephenneuendorffer]

mlir/test/lit.cfg.py

102–103

I don't think this should get littered in the llvm install. There should be a separate mlir-npcomp_obj_root, I think.

[stellaraccident]

Address comments/rebase.

[stellaraccident]

mlir/lib/Bindings/Python/CMakeLists.txt

15

Yes. (and this was only working because there is presently one source file, I expect)

mlir/test/lit.cfg.py

102–103

As discussed on discord, added a todo and will start a thread in a couple of days to work out overall issues that implicate needing to do this.

mlir/lib/Bindings/Python/CMakeLists.txt

+################################################################################
+# Copy python source tree.
+################################################################################
+
+set(PY_SRC_FILES
+  mlir/__init__.py
+)
+
+add_custom_target(MLIRBindingsPythonSources ALL
+  DEPENDS ${PY_SRC_FILES}
+)
+
+foreach(PY_SRC_FILE ${PY_SRC_FILES})
+  set(PY_DEST_FILE "${PROJECT_BINARY_DIR}/python/${PY_SRC_FILES}")
+  add_custom_command(
+    TARGET MLIRBindingsPythonSources PRE_BUILD
+    COMMENT "Copying python source ${PY_SRC_FILE} -> ${PY_DEST_FILE}"
+    DEPENDS "${PY_SRC_FILE}"
+    COMMAND "${CMAKE_COMMAND}" -E copy_if_different
+        "${CMAKE_CURRENT_SOURCE_DIR}/${PY_SRC_FILE}" "${PY_DEST_FILE}"
+  )
+endforeach()
+
+################################################################################
+# Build python extension
+################################################################################
+
 # Normally on unix-like platforms, extensions are built as "MODULE" libraries
 # and do not explicitly link to the python shared object. This allows for
 # some greater deployment flexibility since the extension will bind to
 # symbols in the python interpreter on load. However, it also keeps the
 # linker from erroring on undefined symbols, leaving this to (usually obtuse)
 # runtime errors. Building in "SHARED" mode with an explicit link to the
 # python libraries allows us to build with the expectation of no undefined
-# symbols, which is better for development.
-if(MLIR_PYTHON_BINDINGS_VERSION_LOCKED)
-  set(PYEXT_LINK_MODE SHARED)
-  set(PYEXT_LIBADD ${PYTHON_LIBRARIES})
-else()
+# symbols, which is better for development. Note that not all python
+# configurations provide build-time libraries to link against, in which
+# case, we fall back to MODULE linking.
+if(PYTHON_LIBRARIES STREQUAL "" OR NOT MLIR_PYTHON_BINDINGS_VERSION_LOCKED)
   set(PYEXT_LINK_MODE MODULE)
   set(PYEXT_LIBADD)
+else()
+  set(PYEXT_LINK_MODE SHARED)
+  set(PYEXT_LIBADD ${PYTHON_LIBRARIES})
 endif()
 
 # The actual extension library produces a shared-object or DLL and has
 # sources that must be compiled in accordance with pybind11 needs (RTTI and
 # exceptions).
 # TODO: Link the libraries separately once a helper function is available
 # to more generically add a pybind11 compliant library.
 add_library(MLIRBindingsPythonExtension ${PYEXT_LINK_MODE}
target_compile_options(MLIRBindingsPythonExtension PRIVATE
   $<$<CXX_COMPILER_ID:MSVC>:
     # Enable RTTI and exceptions.
     /EHsc /GR>
 )
 
 # Configure the output to match python expectations.
 set_target_properties(
   MLIRBindingsPythonExtension PROPERTIES
-    LIBRARY_OUTPUT_DIRECTORY "${CMAKE_CURRENT_BINARY_DIR}"
+    # Build-time RPath layouts require to be a directory one up from the
+    # binary root.
+    LIBRARY_OUTPUT_DIRECTORY ${LLVM_BINARY_DIR}/python
     OUTPUT_NAME "_mlir"
     PREFIX "${PYTHON_MODULE_PREFIX}"
     SUFFIX "${PYTHON_MODULE_SUFFIX}${PYTHON_MODULE_EXTENSION}"
 )
 
 # pybind11 requires binding code to be compiled with -fvisibility=hidden
 # For static linkage, better code can be generated if the entire project
 # compiles that way, but that is not enforced here. Instead, include a linker
 # script that explicitly hides anything but the PyInit_* symbols, allowing gc
 # to take place.
 # TODO: Add a Windows .def file and figure out the right thing to do on MacOS.
 set_target_properties(
   MLIRBindingsPythonExtension PROPERTIES CXX_VISIBILITY_PRESET "hidden")
-if(NOT MSVC AND NOT APPLE)
-  set_target_properties(MLIRBindingsPythonExtension
-    PROPERTIES LINK_FLAGS
-    "-Wl,--version-script=${CMAKE_CURRENT_SOURCE_DIR}/unix_version.lds")
+
+set(PYEXT_DEPS)
+if(LLVM_BUILD_LLVM_DYLIB)
+  list(APPEND PYEXT_DEPS
+    # Depend on libMLIR.so first so that deps primarily come from the shared
+    # library.
+    MLIR
+  )
 endif()
 
-target_link_libraries(MLIRBindingsPythonExtension
-  PRIVATE
+# Full static dependencies are also added and will augment what is in the
+# shared lib if needed (or in fully static builds, will result in mondo-built
+# extension).
+list(APPEND PYEXT_DEPS
+  # Depend only on the MLIR C-API.
   MLIRCAPIIR
   MLIRCAPIRegistration
+)
+
+target_link_libraries(MLIRBindingsPythonExtension
+  PRIVATE
+  ${PYEXT_DEPS}
   ${PYEXT_LIBADD}
 )
+
+add_dependencies(MLIRBindingsPythonExtension MLIRBindingsPythonSources)
+llvm_setup_rpath(MLIRBindingsPythonExtension)
+
+################################################################################
+# Install
+################################################################################
+
+install(TARGETS MLIRBindingsPythonExtension
+  COMPONENT MLIRBindingsPythonExtension
+  LIBRARY DESTINATION python
+  ARCHIVE DESTINATION python
+  # NOTE: Even on DLL-platforms, extensions go in the lib directory tree.
+  RUNTIME DESTINATION python)
+
+# Note that we copy from the source tree just like for headers because
+# it will not be polluted with py_cache runtime artifacts (from testing and
+# such).
+install(
+  DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}/mlir
+  DESTINATION python
+  COMPONENT MLIRBindingsPythonSources
+  FILES_MATCHING PATTERN "*.py"
+)
+
+if (NOT LLVM_ENABLE_IDE)
+  add_llvm_install_targets(
+    install-MLIRBindingsPythonExtension
+    DEPENDS MLIRBindingsPythonExtension
+    COMPONENT MLIRBindingsPythonExtension)
+  add_llvm_install_targets(
+    install-MLIRBindingsPythonSources
+    DEPENDS MLIRBindingsPythonSources
+    COMPONENT MLIRBindingsPythonSources)
+endif()

mlir/lib/Bindings/Python/unix_version.lds

-{
-  global: PyInit__mlir;
-  local: *;
-};

mlir/test/lit.cfg.py

tools = [
     'mlir-tblgen',
     'mlir-translate',
     'mlir-capi-ir-test',
     'mlir-edsc-builder-api-test',
 ]
 
 # The following tools are optional
 tools.extend([
-    ToolSubst('%PYTHON', config.python_executable),
+    ToolSubst('%PYTHON', config.python_executable, unresolved='ignore'),
     ToolSubst('toy-ch1', unresolved='ignore'),
     ToolSubst('toy-ch2', unresolved='ignore'),
     ToolSubst('toy-ch3', unresolved='ignore'),
     ToolSubst('toy-ch4', unresolved='ignore'),
     ToolSubst('toy-ch5', unresolved='ignore'),
     ToolSubst('%cuda_wrapper_library_dir', config.cuda_wrapper_library_dir, unresolved='ignore'),
     ToolSubst('%linalg_test_lib_dir', config.linalg_test_lib_dir, unresolved='ignore'),
     ToolSubst('%mlir_runner_utils_dir', config.mlir_runner_utils_dir, unresolved='ignore'),
if config.target_triple:
     config.available_features.add('default_triple')
 
 # Add the python path for both the source and binary tree.
 # Note that presently, the python sources come from the source tree and the
 # binaries come from the build tree. This should be unified to the build tree
 # by copying/linking sources to build.
 if config.enable_bindings_python:
     llvm_config.with_environment('PYTHONPATH', [
-        os.path.join(config.mlir_src_root, "lib", "Bindings", "Python"),
-        os.path.join(config.mlir_obj_root, "lib", "Bindings", "Python"),
+        # There is
+        os.path.join(config.llvm_obj_root, 'python'),
     ], append_path=True)