[BACKEND][CPU] Make the CPU backend buildable and runnable in Mac M1. (…

…#18)

Add header for unique_ptr in CPU launcher.

python/triton/runtime/build.py

@@ -1,5 +1,6 @@
 import contextlib
 import sys
+import platform
 import io
 import sysconfig
 import os
@@ -20,6 +21,7 @@ def quiet():
 
 def _build(name, src, srcdir, library_dirs, include_dirs, libraries):
     suffix = sysconfig.get_config_var('EXT_SUFFIX')
+    system = platform.system()
     so = os.path.join(srcdir, '{name}{suffix}'.format(name=name, suffix=suffix))
     # try to avoid setuptools if possible
     cc = os.environ.get("CC")
@@ -44,6 +46,9 @@ def _build(name, src, srcdir, library_dirs, include_dirs, libraries):
     include_dirs = include_dirs + [srcdir, py_include_dir, *custom_backend_dirs]
     # for -Wno-psabi, see https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111047
     cc_cmd = [cc, src, "-O3", "-shared", "-fPIC", "-Wno-psabi", "-o", so]
+    # Use dynamic lookup to load Python library on Mac
+    if system == "Darwin":
+        cc_cmd += ["-undefined", "dynamic_lookup"]
     cc_cmd += [f'-l{lib}' for lib in libraries]
     cc_cmd += [f"-L{dir}" for dir in library_dirs]
     cc_cmd += [f"-I{dir}" for dir in include_dirs if dir is not None]

third_party/cpu/backend/driver.py

@@ -143,6 +143,7 @@ def format_of(ty):
 #include <optional>
 #include <stdio.h>
 #include <string>
+#include <memory>
 
 #define NPY_NO_DEPRECATED_API NPY_1_7_API_VERSION
 #include <Python.h>