[CMake] Find HomeBrew OpenMP on MacOS (pytorch#145870)

Either via `OMP_PREFIX` envvar or by searching in `/opt/homebrew/opt/libomp` folder

Modify libomp bundling logic in setup.py to change absolute path to libomp.dylib to a relative one if necessary
Pull Request resolved: pytorch#145870
Approved by: https://github.com/Skylion007, https://github.com/atalman
ghstack dependencies: pytorch#145871

Author: malfet

cmake/Modules/FindOpenMP.cmake

@@ -79,6 +79,13 @@ cmake_policy(SET CMP0012 NEW) # if() recognizes numbers and booleans
 cmake_policy(SET CMP0054 NEW) # if() quoted variables not dereferenced
 cmake_policy(SET CMP0057 NEW) # if IN_LIST
 
+
+if(NOT "$ENV{OMP_PREFIX}" STREQUAL "")
+  set(OpenMP_PREFIX "$ENV{OMP_PREFIX}")
+elseif(${CMAKE_SYSTEM_NAME} STREQUAL "Darwin" AND EXISTS /opt/homebrew/opt/libomp)
+  set(OpenMP_PREFIX "/opt/homebrew/opt/libomp")
+endif()
+
 function(_OPENMP_FLAG_CANDIDATES LANG)
   if(NOT OpenMP_${LANG}_FLAG)
     unset(OpenMP_FLAG_CANDIDATES)
@@ -93,7 +100,7 @@ function(_OPENMP_FLAG_CANDIDATES LANG)
     else()
       # AppleClang may need a header file, search for omp.h with hints to brew
       # default include dir
-      find_path(__header_dir "omp.h" HINTS "/usr/local/include")
+      find_path(__header_dir "omp.h" HINTS "/usr/local/include" "${OpenMP_PREFIX}/include")
     endif()
     set(OMP_FLAG_AppleClang "-Xpreprocessor -fopenmp" "-Xpreprocessor -fopenmp -I${__header_dir}")
 
@@ -273,6 +280,16 @@ function(_OPENMP_GET_FLAGS LANG FLAG_MODE OPENMP_FLAG_VAR OPENMP_LIB_NAMES_VAR)
       mark_as_advanced(OpenMP_libomp_LIBRARY)
     endif()
 
+    # Use OpenMP_PREFIX if defined
+    if (NOT OpenMP_libomp_LIBRARY AND NOT "${OpenMP_PREFIX}" STREQUAL "")
+      find_library(OpenMP_libomp_LIBRARY
+        NAMES omp gomp iomp5
+        HINTS "${OpenMP_PREFIX}/lib"
+        DOC "libomp location for OpenMP"
+      )
+      mark_as_advanced(OpenMP_libomp_LIBRARY)
+    endif()
+
     if(OpenMP_libomp_LIBRARY MATCHES "iomp5")
       set(OpenMP_libiomp5_LIBRARY "${MKL_OPENMP_LIBRARY}" CACHE STRING "libiomp5 location for OpenMP")
       if("-fopenmp=libiomp5" IN_LIST OpenMP_${LANG}_FLAG_CANDIDATES)
@@ -666,5 +683,6 @@ unset(OpenMP_Fortran_TEST_SOURCE)
 unset(OpenMP_C_CXX_CHECK_VERSION_SOURCE)
 unset(OpenMP_Fortran_CHECK_VERSION_SOURCE)
 unset(OpenMP_Fortran_INCLUDE_LINE)
+unset(OpenMP_PREFIX)
 
 cmake_policy(POP)

setup.py

@@ -572,35 +572,52 @@ def _embed_libomp(self):
                 assert rpath.startswith("path ")
                 rpaths.append(rpath.split(" ", 1)[1].rsplit("(", 1)[0][:-1])
 
-        omp_lib_name = (
-            "libomp.dylib" if os.uname().machine == "arm64" else "libiomp5.dylib"
-        )
-        omp_rpath_lib_path = os.path.join("@rpath", omp_lib_name)
-        if omp_rpath_lib_path not in libs:
+        omplib_path = get_cmake_cache_vars()["OpenMP_libomp_LIBRARY"]
+        omplib_name = get_cmake_cache_vars()["OpenMP_C_LIB_NAMES"] + ".dylib"
+        omplib_rpath_path = os.path.join("@rpath", omplib_name)
+
+        # This logic is fragile and checks only two cases:
+        # - libtorch_cpu depends on `@rpath/libomp.dylib`e (happens when built inside miniconda environment)
+        # - libtorch_cpu depends on `/abs/path/to/libomp.dylib` (happens when built with libomp from homebrew)
+        if not any(c in libs for c in [omplib_path, omplib_rpath_path]):
             return
 
         # Copy libomp/libiomp5 from rpath locations
+        target_lib = os.path.join(self.build_lib, "torch", "lib", omplib_name)
+        libomp_relocated = False
         for rpath in rpaths:
-            source_lib = os.path.join(rpath, omp_lib_name)
+            source_lib = os.path.join(rpath, omplib_name)
             if not os.path.exists(source_lib):
                 continue
-            target_lib = os.path.join(self.build_lib, "torch", "lib", omp_lib_name)
             self.copy_file(source_lib, target_lib)
             # Delete old rpath and add @loader_lib to the rpath
             # This should prevent delocate from attempting to package another instance
             # of OpenMP library in torch wheel as well as loading two libomp.dylib into
             # the address space, as libraries are cached by their unresolved names
-            subprocess.check_call(
-                [
-                    "install_name_tool",
-                    "-rpath",
-                    rpath,
+            install_name_tool_args = [
+                "-rpath",
+                rpath,
+                "@loader_path",
+            ]
+            libomp_relocated = True
+            break
+        if not libomp_relocated and os.path.exists(omplib_path):
+            self.copy_file(omplib_path, target_lib)
+            install_name_tool_args = [
+                "-change",
+                omplib_path,
+                omplib_rpath_path,
+            ]
+            if "@loader_path" not in rpaths:
+                install_name_tool_args += [
+                    "-add_rpath",
                     "@loader_path",
-                    libtorch_cpu_path,
                 ]
-            )
-            break
-
+            libomp_relocated = True
+        if libomp_relocated:
+            install_name_tool_args.insert(0, "install_name_tool")
+            install_name_tool_args.append(libtorch_cpu_path)
+            subprocess.check_call(install_name_tool_args)
         # Copy omp.h from OpenMP_C_FLAGS and copy it into include folder
         omp_cflags = get_cmake_cache_vars()["OpenMP_C_FLAGS"]
         if not omp_cflags: