add setup.py.

.gitignore

@@ -4,3 +4,28 @@
 *.ptx
 *.cubin
 *.fatbin
+
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[cod]
+*$py.class
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+share/python-wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST

cmake/generic.cmake

@@ -7,7 +7,7 @@ set(CMAKE_BUILD_TYPE Release)
 
 set(CMAKE_CXX_STANDARD
     20
-    CACHE STRING "The C++ standard whoese features are requested." FORCE)
+    CACHE STRING "The C++ standard whose features are requested." FORCE)
 set(CMAKE_CXX_STANDARD_REQUIRED ON)
 
 set(CMAKE_CUDA_STANDARD
@@ -48,6 +48,12 @@ set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -std=c++20)
 set(CUDA_NVCC_FLAGS_DEBUG ${CUDA_NVCC_FLAGS_DEBUG} -std=c++20 -O0)
 set(CUDA_NVCC_FLAGS_RELEASE ${CUDA_NVCC_FLAGS_RELEASE} -std=c++20 -O3)
 
+if(${CUDA_VERSION_MAJOR} VERSION_GREATER_EQUAL "11")
+  add_definitions("-DENABLE_BF16")
+  message(STATUS "CUDA_VERSION ${CUDA_VERSION_MAJOR}.${CUDA_VERSION_MINOR} "
+                 "is greater or equal than 11.0, enable -DENABLE_BF16 flag.")
+endif()
+
 message(STATUS "tilefusion: CUDA detected: " ${CUDA_VERSION})
 message(STATUS "tilefusion: CUDA nvcc is: " ${CUDA_NVCC_EXECUTABLE})
 message(STATUS "tilefusion: CUDA toolkit directory: " ${CUDA_TOOLKIT_ROOT_DIR})

pyproject.toml

@@ -15,11 +15,26 @@ classifiers = [
     "Operating System :: OS Independent",
     "Topic :: Software Development :: Libraries",
 ]
+# NOTE: setuptools's `install_requires` can overwritten in
+ # `pyproject.toml`'s `dependencies` field.
+ # Make sure to keep this field in sync with what is in `requirements.txt`.
+dependencies = [
+    "torch",
+]
 
 [project.urls]
 Homepage = "https://github.com/microsoft/TileFusion"
 Issues = "https://github.com/microsoft/TileFusion/issues"
 
+[build-system]
+requires = [
+    "cmake",
+    "packaging",
+    "setuptools>=49.4.0",
+    "wheel",
+]
+build-backend = "setuptools.build_meta"
+
 [tool.ruff]
 line-length = 80
 exclude = [

pytilefusion/__version__.py

@@ -0,0 +1 @@
+__version__ = '0.0.0'

requirements.txt

@@ -0,0 +1,5 @@
+cmake
+packaging
+setuptools>=49.4.0
+torch
+wheel

setup.py

@@ -0,0 +1,152 @@
+# -------------------------------------------------------------------------
+# Copyright (c) Microsoft Corporation. All rights reserved.
+# Licensed under the MIT License.
+# --------------------------------------------------------------------------
+
+import os
+import subprocess
+from pathlib import Path
+
+from setuptools import Command, Extension, find_packages, setup
+from setuptools.command.build_ext import build_ext
+
+cur_path = Path(__file__).parent
+
+
+def get_requirements():
+    """Get Python package dependencies from requirements.txt."""
+    with open(cur_path / "requirements.txt") as f:
+        requirements = f.read().strip().split("\n")
+    requirements = [req for req in requirements if "https" not in req]
+    return requirements
+
+
+class CMakeExtension(Extension):
+    """ specify the root folder of the CMake projects"""
+
+    def __init__(self, name, cmake_lists_dir=".", **kwargs):
+        Extension.__init__(self, name, sources=[], **kwargs)
+        self.cmake_lists_dir = os.path.abspath(cmake_lists_dir)
+
+
+class CMakeBuildExt(build_ext):
+    """launches the CMake build."""
+
+    def build_extension(self, ext: CMakeExtension) -> None:
+        # Ensure that CMake is present and working
+        try:
+            subprocess.check_output(["cmake", "--version"])
+        except OSError:
+            raise RuntimeError("Cannot find CMake executable") from None
+
+        debug = int(
+            os.environ.get("DEBUG", 0)
+        ) if self.debug is None else self.debug
+        cfg = "Debug" if debug else "Release"
+
+        parallel_level = os.environ.get("CMAKE_BUILD_PARALLEL_LEVEL", None)
+        if parallel_level is not None:
+            self.parallel = int(parallel_level)
+        else:
+            self.parallel = os.cpu_count()
+
+        for ext in self.extensions:
+            extdir = os.path.abspath(
+                os.path.dirname(self.get_ext_fullpath(ext.name))
+            )
+
+            cmake_args = [
+                "-DCMAKE_BUILD_TYPE=%s" % cfg,
+                "-DCMAKE_LIBRARY_OUTPUT_DIRECTORY_{}={}".format(
+                    cfg.upper(), extdir
+                ), "-DCMAKE_ARCHIVE_OUTPUT_DIRECTORY_{}={}".format(
+                    cfg.upper(), self.build_temp
+                )
+            ]
+
+            # Adding CMake arguments set as environment variable
+            if "CMAKE_ARGS" in os.environ:
+                cmake_args += [
+                    item for item in os.environ["CMAKE_ARGS"].split(" ") if item
+                ]
+
+            if not os.path.exists(self.build_temp):
+                os.makedirs(self.build_temp)
+
+            build_args = []
+            build_args += ["--config", cfg]
+            # Set CMAKE_BUILD_PARALLEL_LEVEL to control the parallel build level
+            # across all generators.
+            if (
+                "CMAKE_BUILD_PARALLEL_LEVEL" not in os.environ and
+                hasattr(self, "parallel") and self.parallel
+            ):
+                build_args += [f"-j{self.parallel}"]
+
+            build_temp = Path(self.build_temp) / ext.name
+            if not build_temp.exists():
+                build_temp.mkdir(parents=True)
+
+            # Config
+            subprocess.check_call(["cmake", ext.cmake_lists_dir] + cmake_args,
+                                  cwd=self.build_temp)
+
+            # Build
+            subprocess.check_call(["cmake", "--build", "."] + build_args,
+                                  cwd=self.build_temp)
+
+
+class clean(Command):
+    user_options = []
+
+    def initialize_options(self):
+        pass
+
+    def finalize_options(self):
+        pass
+
+    def run(self):
+        import glob
+        import re
+        import shutil
+
+        with open(".gitignore") as f:
+            ignores = f.read()
+            pat = re.compile(r"^#( BEGIN NOT-CLEAN-FILES )?")
+            for wildcard in filter(None, ignores.split("\n")):
+                match = pat.match(wildcard)
+                if match:
+                    if match.group(1):
+                        # Marker is found and stop reading .gitignore.
+                        break
+                    # Ignore lines which begin with '#'.
+                else:
+                    # Don't remove absolute paths from the system
+                    wildcard = wildcard.lstrip("./")
+
+                    for filename in glob.glob(wildcard):
+                        print(f"cleaning '{filename}'")
+                        try:
+                            os.remove(filename)
+                        except OSError:
+                            shutil.rmtree(filename, ignore_errors=True)
+
+
+description = ("TileFusion: A Python wrapper for tilefusion C++ library.")
+
+with open(os.path.join("pytilefusion", '__version__.py')) as f:
+    exec(f.read())
+
+setup(
+    name="tilefusion",
+    python_requires=">=3.10",
+    packages=find_packages(exclude=[""]),
+    install_requires=get_requirements(),
+    version=__version__,  # noqa F821
+    description=description,
+    ext_modules=[CMakeExtension("tilefusion")],
+    cmdclass={
+        "build_ext": CMakeBuildExt,
+        "clean": clean,
+    },
+)

src/CMakeLists.txt

@@ -21,7 +21,15 @@ set_target_properties(
              CUDA_SEPARABLE_COMPILATION ON)
 
 target_compile_options(
-  ${TARGET} PUBLIC $<$<COMPILE_LANGUAGE:CUDA>: -Werror,-Wall -rdc=true
-                   -std=c++20 -fconcepts -fpermissive>)
+  ${TARGET}
+  PUBLIC $<$<COMPILE_LANGUAGE:CUDA>:
+         -Werror,-Wall
+         -rdc=true
+         -std=c++20
+         -fconcepts
+         -fpermissive
+         --use_fast_math
+         --generate-line-info
+         >)
 target_compile_features(${TARGET} PUBLIC cxx_std_20 cuda_std_20)
 target_link_libraries(${TARGET} "${TORCH_LIBRARIES}")