v3: Updates for macOS Flang from Homebrew

CHANGELOG.md

@@ -9,12 +9,16 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Fixed
 
+- Fixes for using Homebrew flang on macOS
+
 ### Removed
 
 ### Added
 
 ### Changed
 
+- Enable flang architecture flags
+
 ### Deprecated
 
 ## [3.74.0] - 2026-03-09

compiler/flags/LLVMFlang_Fortran.cmake

@@ -35,30 +35,36 @@ set (ARCH_CONSISTENCY "")
 set (FTZ "")
 set (ALIGN_ALL "")
 set (NO_ALIAS "")
+set (STACK_ARRAYS "-fstack-arrays")
 
 set (NO_RANGE_CHECK "")
 
+set(NO_FMA "-ffp-contract=off")
+
+# NOTE: There is currently a bug(?) in flang that says you have to pass
+# -fno-integrated-as when doing save temps (see https://github.com/llvm/llvm-project/pull/119624)
+# It does seem to work, but a PR says this might have issues with offloading:
+# NOTE 2: I found a bug with this and flang: https://github.com/llvm/llvm-project/issues/184802
+# Turning off for now.
+#set(SAVE_TEMPS "-fno-integrated-as -save-temps=obj")
+
 cmake_host_system_information(RESULT proc_description QUERY PROCESSOR_DESCRIPTION)
 
-# NOT SURE ABOUT ANY OF THIS...
 if ( ${CMAKE_HOST_SYSTEM_PROCESSOR} STREQUAL aarch64 )
-  set (FLANG_TARGET_ARCH "-mcpu=armv8.2-a+crypto+crc+fp16+rcpc+dotprod")
+  set (FLANG_TARGET_ARCH "armv8.2-a+crypto+crc+fp16+rcpc+dotprod")
 elseif (${proc_description} MATCHES "Apple M")
-  set (FLANG_TARGET_ARCH "-mcpu=apple-m1")
+  set (FLANG_TARGET_ARCH "apple-m1")
 elseif (${proc_description} MATCHES "EPYC")
-  set (FLANG_TARGET_ARCH "-mcpu=znver2")
+  set (FLANG_TARGET_ARCH "znver2")
 elseif (${proc_description} MATCHES "Intel|INTEL")
-  set (FLANG_TARGET_ARCH "-mcpu=haswell")
+  set (FLANG_TARGET_ARCH "haswell")
 elseif ( ${CMAKE_HOST_SYSTEM_PROCESSOR} STREQUAL "x86_64" )
   message(WARNING "Unknown processor type. Defaulting to a generic x86_64 processor. Performance may be suboptimal.")
   set (FLANG_TARGET_ARCH "x86-64")
 else ()
   message(FATAL_ERROR "Unknown processor. Please file an issue at https://github.com/GEOS-ESM/ESMA_cmake")
 endif ()
 
-# ...SO WE JUST TURN OFF FLANG_TARGET_ARCH FOR NOW
-set (FLANG_TARGET_ARCH "")
-
 ####################################################
 
 # Common Fortran Flags
@@ -68,12 +74,12 @@ set (common_Fortran_fpe_flags "${TRACEBACK}")
 
 # GEOS Debug
 # ----------
-set (GEOS_Fortran_Debug_Flags "${FOPT0} ${DEBINFO}")
+set (GEOS_Fortran_Debug_Flags "${FOPT0} ${DEBINFO} ${SAVE_TEMPS} ${NO_FMA}")
 set (GEOS_Fortran_Debug_FPE_Flags "${common_Fortran_fpe_flags}")
 
 # GEOS Release
 # ------------
-set (GEOS_Fortran_Release_Flags "${FOPT3} ${FLANG_TARGET_ARCH} ${DEBINFO}")
+set (GEOS_Fortran_Release_Flags "${FOPT3} -march=${FLANG_TARGET_ARCH} -funroll-loops ${STACK_ARRAYS} ${DEBINFO}")
 set (GEOS_Fortran_Release_FPE_Flags "${common_Fortran_fpe_flags}")
 
 # Create a NoVectorize version for consistency. No difference from Release for Flang

operating_system/osx_extras.cmake

@@ -56,3 +56,13 @@ set (ENABLE_RELATIVE_RPATHS FALSE)
 
 set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -Wl,-headerpad_max_install_names")
 set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} -Wl,-headerpad_max_install_names")
+
+# 5) Flang compiler workarounds for macOS shared libraries
+# CMake natively uses Apple's '-dynamiclib' and '-install_name' for shared libraries,
+# but the LLVM flang frontend strictly rejects them.
+# We force flang to use '-shared' (which it correctly translates down to the linker)
+# and use -Xlinker to safely pass the soname without space-parsing errors.
+if(CMAKE_Fortran_COMPILER_ID MATCHES "Flang" OR CMAKE_Fortran_COMPILER_ID MATCHES "LLVMFlang")
+  set(CMAKE_SHARED_LIBRARY_CREATE_Fortran_FLAGS "-shared -Wl,-headerpad_max_install_names")
+  set(CMAKE_SHARED_LIBRARY_SONAME_Fortran_FLAG "-Xlinker -install_name -Xlinker ")
+endif()