[OpenMP] arm64_32 port for Apple WatchOS (#87246)

detect `aarch64_32` with compiler defined macro `__ARM64_ARCH_8_32__`
reuse ARM `__kmp_unnamed_critical_addr` and add `KMP_PREFIX_UNDERSCORE`
macro like AARCH64
reuse AARCH64 `__kmp_invoke_microtask`


build log for watchos armv7k + arm64_32 and watchos simulator x86_64 +
arm64

https://github.com/nihui/action-protobuf/actions/runs/8520684611/job/23337305030

13 files changed, 46 insertions, 20 deletions

diff --git a/openmp/runtime/CMakeLists.txt b/openmp/runtime/CMakeLists.txt
index 041b60ef..57ed54b 100644
--- a/openmp/runtime/CMakeLists.txt
+++ b/openmp/runtime/CMakeLists.txt
@@ -30,7 +30,7 @@ if(${OPENMP_STANDALONE_BUILD})
   # If adding a new architecture, take a look at cmake/LibompGetArchitecture.cmake
   libomp_get_architecture(LIBOMP_DETECTED_ARCH)
   set(LIBOMP_ARCH ${LIBOMP_DETECTED_ARCH} CACHE STRING
-    "The architecture to build for (x86_64/i386/arm/ppc/ppc64/ppc64le/aarch64/mic/mips/mips64/riscv64/loongarch64/ve/s390x/wasm32).")
+    "The architecture to build for (x86_64/i386/arm/ppc/ppc64/ppc64le/aarch64/aarch64_32/mic/mips/mips64/riscv64/loongarch64/ve/s390x/wasm32).")
   # Should assertions be enabled?  They are on by default.
   set(LIBOMP_ENABLE_ASSERTIONS TRUE CACHE BOOL
     "enable assertions?")
@@ -55,6 +55,8 @@ else() # Part of LLVM build
     set(LIBOMP_ARCH ppc64)
   elseif(LIBOMP_NATIVE_ARCH MATCHES "powerpc")
     set(LIBOMP_ARCH ppc)
+  elseif(LIBOMP_NATIVE_ARCH MATCHES "aarch64_32")
+    set(LIBOMP_ARCH aarch64_32)
   elseif(LIBOMP_NATIVE_ARCH MATCHES "aarch64")
     set(LIBOMP_ARCH aarch64)
   elseif(LIBOMP_NATIVE_ARCH MATCHES "arm64")
@@ -91,7 +93,7 @@ if(LIBOMP_ARCH STREQUAL "aarch64")
   endif()
 endif()
 
-libomp_check_variable(LIBOMP_ARCH 32e x86_64 32 i386 arm ppc ppc64 ppc64le aarch64 aarch64_a64fx mic mips mips64 riscv64 loongarch64 ve s390x wasm32)
+libomp_check_variable(LIBOMP_ARCH 32e x86_64 32 i386 arm ppc ppc64 ppc64le aarch64 aarch64_32 aarch64_a64fx mic mips mips64 riscv64 loongarch64 ve s390x wasm32)
 
 set(LIBOMP_LIB_TYPE normal CACHE STRING
   "Performance,Profiling,Stubs library (normal/profile/stubs)")
@@ -167,6 +169,7 @@ set(IA32 FALSE)
 set(INTEL64 FALSE)
 set(ARM FALSE)
 set(AARCH64 FALSE)
+set(AARCH64_32 FALSE)
 set(AARCH64_A64FX FALSE)
 set(PPC64BE FALSE)
 set(PPC64LE FALSE)
@@ -196,6 +199,8 @@ elseif("${LIBOMP_ARCH}" STREQUAL "ppc64le") # PPC64LE architecture
   set(PPC64 TRUE)
 elseif("${LIBOMP_ARCH}" STREQUAL "aarch64") # AARCH64 architecture
   set(AARCH64 TRUE)
+elseif("${LIBOMP_ARCH}" STREQUAL "aarch64_32") # AARCH64_32 architecture
+  set(AARCH64_32 TRUE)
 elseif("${LIBOMP_ARCH}" STREQUAL "aarch64_a64fx") # AARCH64_A64FX architecture
   set(AARCH64_A64FX TRUE)
 elseif("${LIBOMP_ARCH}" STREQUAL "mic") # Intel(R) Many Integrated Core Architecture
diff --git a/openmp/runtime/cmake/LibompGetArchitecture.cmake b/openmp/runtime/cmake/LibompGetArchitecture.cmake
index d7f8187..2d5c662 100644
--- a/openmp/runtime/cmake/LibompGetArchitecture.cmake
+++ b/openmp/runtime/cmake/LibompGetArchitecture.cmake
@@ -35,6 +35,8 @@ function(libomp_get_architecture return_arch)
       #error ARCHITECTURE=arm
     #elif defined(__arm__) || defined(_M_ARM) || defined(_ARM)
       #error ARCHITECTURE=arm
+    #elif defined(__ARM64_ARCH_8_32__)
+      #error ARCHITECTURE=aarch64_32
     #elif defined(__aarch64__) || defined(_M_ARM64)
       #error ARCHITECTURE=aarch64
     #elif defined(__powerpc64__) && defined(__LITTLE_ENDIAN__)
diff --git a/openmp/runtime/cmake/LibompUtils.cmake b/openmp/runtime/cmake/LibompUtils.cmake
index 139eabb..58b93b3 100644
--- a/openmp/runtime/cmake/LibompUtils.cmake
+++ b/openmp/runtime/cmake/LibompUtils.cmake
@@ -101,6 +101,8 @@ function(libomp_get_legal_arch return_arch_string)
     set(${return_arch_string} "PPC64LE" PARENT_SCOPE)
   elseif(${AARCH64})
     set(${return_arch_string} "AARCH64" PARENT_SCOPE)
+  elseif(${AARCH64_32})
+    set(${return_arch_string} "AARCH64_32" PARENT_SCOPE)
   elseif(${AARCH64_A64FX})
     set(${return_arch_string} "AARCH64_A64FX" PARENT_SCOPE)
   elseif(${MIPS})
diff --git a/openmp/runtime/cmake/config-ix.cmake b/openmp/runtime/cmake/config-ix.cmake
index 76f471a..337fe2e 100644
--- a/openmp/runtime/cmake/config-ix.cmake
+++ b/openmp/runtime/cmake/config-ix.cmake
@@ -326,6 +326,7 @@ else()
       (LIBOMP_ARCH STREQUAL i386) OR
 #      (LIBOMP_ARCH STREQUAL arm) OR
       (LIBOMP_ARCH STREQUAL aarch64) OR
+      (LIBOMP_ARCH STREQUAL aarch64_32) OR
       (LIBOMP_ARCH STREQUAL aarch64_a64fx) OR
       (LIBOMP_ARCH STREQUAL ppc64le) OR
       (LIBOMP_ARCH STREQUAL ppc64) OR
diff --git a/openmp/runtime/src/kmp_gsupport.cpp b/openmp/runtime/src/kmp_gsupport.cpp
index 4dc8a90..86cf164 100644
--- a/openmp/runtime/src/kmp_gsupport.cpp
+++ b/openmp/runtime/src/kmp_gsupport.cpp
@@ -358,7 +358,7 @@ void KMP_EXPAND_NAME(KMP_API_NAME_GOMP_ORDERED_END)(void) {
 // (IA-32 architecture) or 64-bit signed (Intel(R) 64).
 
 #if KMP_ARCH_X86 || KMP_ARCH_ARM || KMP_ARCH_MIPS || KMP_ARCH_WASM ||          \
-    KMP_ARCH_PPC
+    KMP_ARCH_PPC || KMP_ARCH_AARCH64_32
 #define KMP_DISPATCH_INIT __kmp_aux_dispatch_init_4
 #define KMP_DISPATCH_FINI_CHUNK __kmp_aux_dispatch_fini_chunk_4
 #define KMP_DISPATCH_NEXT __kmpc_dispatch_next_4
diff --git a/openmp/runtime/src/kmp_os.h b/openmp/runtime/src/kmp_os.h
index a628070..24b40ed 100644
--- a/openmp/runtime/src/kmp_os.h
+++ b/openmp/runtime/src/kmp_os.h
@@ -179,7 +179,7 @@ typedef unsigned long long kmp_uint64;
 #endif /* KMP_OS_UNIX */
 
 #if KMP_ARCH_X86 || KMP_ARCH_ARM || KMP_ARCH_MIPS || KMP_ARCH_WASM ||          \
-    KMP_ARCH_PPC
+    KMP_ARCH_PPC || KMP_ARCH_AARCH64_32
 #define KMP_SIZE_T_SPEC KMP_UINT32_SPEC
 #elif KMP_ARCH_X86_64 || KMP_ARCH_PPC64 || KMP_ARCH_AARCH64 ||                 \
     KMP_ARCH_MIPS64 || KMP_ARCH_RISCV64 || KMP_ARCH_LOONGARCH64 ||             \
@@ -1051,7 +1051,7 @@ extern kmp_real64 __kmp_xchg_real64(volatile kmp_real64 *p, kmp_real64 v);
 
 #if KMP_ARCH_PPC64 || KMP_ARCH_ARM || KMP_ARCH_AARCH64 || KMP_ARCH_MIPS ||     \
     KMP_ARCH_MIPS64 || KMP_ARCH_RISCV64 || KMP_ARCH_LOONGARCH64 ||             \
-    KMP_ARCH_VE || KMP_ARCH_S390X || KMP_ARCH_PPC
+    KMP_ARCH_VE || KMP_ARCH_S390X || KMP_ARCH_PPC || KMP_ARCH_AARCH64_32
 #if KMP_OS_WINDOWS
 #undef KMP_MB
 #define KMP_MB() std::atomic_thread_fence(std::memory_order_seq_cst)
diff --git a/openmp/runtime/src/kmp_platform.h b/openmp/runtime/src/kmp_platform.h
index c06f46d..200fdf6 100644
--- a/openmp/runtime/src/kmp_platform.h
+++ b/openmp/runtime/src/kmp_platform.h
@@ -105,6 +105,7 @@
 #define KMP_ARCH_X86 0
 #define KMP_ARCH_X86_64 0
 #define KMP_ARCH_AARCH64 0
+#define KMP_ARCH_AARCH64_32 0
 #define KMP_ARCH_PPC64_ELFv1 0
 #define KMP_ARCH_PPC64_ELFv2 0
 #define KMP_ARCH_PPC64_XCOFF 0
@@ -157,6 +158,9 @@
 #define KMP_ARCH_PPC_XCOFF 1
 #undef KMP_ARCH_PPC
 #define KMP_ARCH_PPC 1
+#elif defined __ARM64_ARCH_8_32__
+#undef KMP_ARCH_AARCH64_32
+#define KMP_ARCH_AARCH64_32 1
 #elif defined __aarch64__
 #undef KMP_ARCH_AARCH64
 #define KMP_ARCH_AARCH64 1
@@ -244,7 +248,7 @@
 /* Specify 32 bit architectures here */
 #define KMP_32_BIT_ARCH                                                        \
   (KMP_ARCH_X86 || KMP_ARCH_ARM || KMP_ARCH_MIPS || KMP_ARCH_WASM ||           \
-   KMP_ARCH_PPC)
+   KMP_ARCH_PPC || KMP_ARCH_AARCH64_32)
 
 // Platforms which support Intel(R) Many Integrated Core Architecture
 #define KMP_MIC_SUPPORTED                                                      \
@@ -254,7 +258,8 @@
 #if (1 != KMP_ARCH_X86 + KMP_ARCH_X86_64 + KMP_ARCH_ARM + KMP_ARCH_PPC64 +     \
               KMP_ARCH_AARCH64 + KMP_ARCH_MIPS + KMP_ARCH_MIPS64 +             \
               KMP_ARCH_RISCV64 + KMP_ARCH_LOONGARCH64 + KMP_ARCH_VE +          \
-              KMP_ARCH_S390X + KMP_ARCH_WASM + KMP_ARCH_PPC)
+              KMP_ARCH_S390X + KMP_ARCH_WASM + KMP_ARCH_PPC +                  \
+              KMP_ARCH_AARCH64_32)
 #error Unknown or unsupported architecture
 #endif
 
diff --git a/openmp/runtime/src/kmp_runtime.cpp b/openmp/runtime/src/kmp_runtime.cpp
index c715a57..95acf4d 100644
--- a/openmp/runtime/src/kmp_runtime.cpp
+++ b/openmp/runtime/src/kmp_runtime.cpp
@@ -8926,7 +8926,7 @@ __kmp_determine_reduction_method(
        // KMP_OS_SOLARIS || KMP_OS_WASI || KMP_OS_AIX
 
 #elif KMP_ARCH_X86 || KMP_ARCH_ARM || KMP_ARCH_AARCH || KMP_ARCH_MIPS ||       \
-    KMP_ARCH_WASM || KMP_ARCH_PPC
+    KMP_ARCH_WASM || KMP_ARCH_PPC || KMP_ARCH_AARCH64_32
 
 #if KMP_OS_LINUX || KMP_OS_DRAGONFLY || KMP_OS_FREEBSD || KMP_OS_NETBSD ||     \
     KMP_OS_OPENBSD || KMP_OS_WINDOWS || KMP_OS_HURD || KMP_OS_SOLARIS ||       \
diff --git a/openmp/runtime/src/z_Linux_asm.S b/openmp/runtime/src/z_Linux_asm.S
index 14987c2..2019490 100644
--- a/openmp/runtime/src/z_Linux_asm.S
+++ b/openmp/runtime/src/z_Linux_asm.S
@@ -108,7 +108,7 @@ KMP_PREFIX_UNDERSCORE(\proc):
 # endif // KMP_OS_DARWIN
 #endif // KMP_ARCH_X86 || KMP_ARCH_x86_64
 
-#if (KMP_OS_LINUX || KMP_OS_DARWIN || KMP_OS_WINDOWS) && (KMP_ARCH_AARCH64 || KMP_ARCH_ARM)
+#if (KMP_OS_LINUX || KMP_OS_DARWIN || KMP_OS_WINDOWS) && (KMP_ARCH_AARCH64 || KMP_ARCH_AARCH64_32 || KMP_ARCH_ARM)
 
 # if KMP_OS_DARWIN
 #  define KMP_PREFIX_UNDERSCORE(x) _##x  // extra underscore for OS X* symbols
@@ -176,7 +176,7 @@ KMP_PREFIX_UNDERSCORE(\proc):
 .endm
 # endif // KMP_OS_DARWIN
 
-#endif // (KMP_OS_LINUX || KMP_OS_DARWIN || KMP_OS_WINDOWS) && (KMP_ARCH_AARCH64 || KMP_ARCH_ARM)
+#endif // (KMP_OS_LINUX || KMP_OS_DARWIN || KMP_OS_WINDOWS) && (KMP_ARCH_AARCH64 || KMP_ARCH_AARCH64_32 || KMP_ARCH_ARM)
 
 .macro COMMON name, size, align_power
 #if KMP_OS_DARWIN
@@ -1236,7 +1236,7 @@ KMP_LABEL(kmp_1_exit):
 #endif /* KMP_ARCH_X86_64 */
 
 // '
-#if (KMP_OS_LINUX || KMP_OS_DARWIN || KMP_OS_WINDOWS) && KMP_ARCH_AARCH64
+#if (KMP_OS_LINUX || KMP_OS_DARWIN || KMP_OS_WINDOWS) && (KMP_ARCH_AARCH64 || KMP_ARCH_AARCH64_32)
 
 //------------------------------------------------------------------------
 // int
@@ -1360,7 +1360,7 @@ KMP_LABEL(kmp_1):
 	DEBUG_INFO __kmp_invoke_microtask
 // -- End  __kmp_invoke_microtask
 
-#endif /* (KMP_OS_LINUX || KMP_OS_DARWIN || KMP_OS_WINDOWS) && KMP_ARCH_AARCH64 */
+#endif /* (KMP_OS_LINUX || KMP_OS_DARWIN || KMP_OS_WINDOWS) && (KMP_ARCH_AARCH64 || KMP_ARCH_AARCH64_32) */
 
 #if (KMP_OS_LINUX || KMP_OS_DARWIN || KMP_OS_WINDOWS) && KMP_ARCH_ARM
 
@@ -1505,7 +1505,7 @@ KMP_LABEL(kmp_1):
 	DEBUG_INFO __kmp_invoke_microtask
 // -- End  __kmp_invoke_microtask
 
-#endif /* (KMP_OS_LINUX || KMP_OS_DARWIN || KMP_OS_WINDOWS) && KMP_ARCH_AARCH64 */
+#endif /* (KMP_OS_LINUX || KMP_OS_DARWIN || KMP_OS_WINDOWS) && KMP_ARCH_ARM */
 
 #if KMP_ARCH_PPC64
 
@@ -2405,18 +2405,21 @@ __kmp_invoke_microtask:
 
 #endif /* KMP_ARCH_S390X */
 
-#if KMP_ARCH_ARM || KMP_ARCH_MIPS
+#if KMP_ARCH_ARM || KMP_ARCH_MIPS || KMP_ARCH_AARCH64_32
+#ifndef KMP_PREFIX_UNDERSCORE
+# define KMP_PREFIX_UNDERSCORE(x) x
+#endif
     .data
     COMMON .gomp_critical_user_, 32, 3
     .data
     .align 4
-    .global __kmp_unnamed_critical_addr
-__kmp_unnamed_critical_addr:
+    .global KMP_PREFIX_UNDERSCORE(__kmp_unnamed_critical_addr)
+KMP_PREFIX_UNDERSCORE(__kmp_unnamed_critical_addr):
     .4byte .gomp_critical_user_
 #ifdef __ELF__
-    .size __kmp_unnamed_critical_addr,4
+    .size KMP_PREFIX_UNDERSCORE(__kmp_unnamed_critical_addr),4
 #endif
-#endif /* KMP_ARCH_ARM */
+#endif /* KMP_ARCH_ARM || KMP_ARCH_MIPS || KMP_ARCH_AARCH64_32 */
 
 #if KMP_ARCH_PPC64 || KMP_ARCH_AARCH64 || KMP_ARCH_MIPS64 ||                   \
     KMP_ARCH_RISCV64 || KMP_ARCH_LOONGARCH64 || KMP_ARCH_VE ||                 \
diff --git a/openmp/runtime/src/z_Linux_util.cpp b/openmp/runtime/src/z_Linux_util.cpp
index 29db9d0..ee5e6e1 100644
--- a/openmp/runtime/src/z_Linux_util.cpp
+++ b/openmp/runtime/src/z_Linux_util.cpp
@@ -2635,7 +2635,8 @@ finish: // Clean up and exit.
 #if !(KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_MIC ||                            \
       ((KMP_OS_LINUX || KMP_OS_DARWIN) && KMP_ARCH_AARCH64) ||                 \
       KMP_ARCH_PPC64 || KMP_ARCH_RISCV64 || KMP_ARCH_LOONGARCH64 ||            \
-      KMP_ARCH_ARM || KMP_ARCH_VE || KMP_ARCH_S390X || KMP_ARCH_PPC_XCOFF)
+      KMP_ARCH_ARM || KMP_ARCH_VE || KMP_ARCH_S390X || KMP_ARCH_PPC_XCOFF ||   \
+      KMP_ARCH_AARCH64_32)
 
 // Because WebAssembly will use `call_indirect` to invoke the microtask and
 // WebAssembly indirect calls check that the called signature is a precise
diff --git a/openmp/runtime/test/ompt/callback.h b/openmp/runtime/test/ompt/callback.h
index 05b2006..f07176c 100644
--- a/openmp/runtime/test/ompt/callback.h
+++ b/openmp/runtime/test/ompt/callback.h
@@ -186,7 +186,7 @@ ompt_label_##id:
 #define print_possible_return_addresses(addr) \
   printf("%" PRIu64 ": current_address=%p or %p\n", ompt_get_thread_data()->value, \
          ((char *)addr) - 8, ((char *)addr) - 12)
-#elif KMP_ARCH_AARCH64
+#elif KMP_ARCH_AARCH64 || KMP_ARCH_AARCH64_32
 // On AArch64 the NOP instruction is 4 bytes long, can be followed by inserted
 // store instruction (another 4 bytes long).
 // FIXME: PR #65696 addded a third possibility (12 byte offset) to make the
diff --git a/openmp/runtime/tools/lib/Platform.pm b/openmp/runtime/tools/lib/Platform.pm
index 6efd932..36847d5 100644
--- a/openmp/runtime/tools/lib/Platform.pm
+++ b/openmp/runtime/tools/lib/Platform.pm
@@ -53,6 +53,8 @@ sub canon_arch($) {
 			$arch = "ppc64le";
         } elsif ( $arch =~ m{\Appc64} ) {
         	$arch = "ppc64";
+        } elsif ( $arch =~ m{\Aaarch64_32} ) {
+                $arch = "aarch64_32";
         } elsif ( $arch =~ m{\Aaarch64} ) {
                 $arch = "aarch64";
         } elsif ( $arch =~ m{\Amic} ) {
@@ -97,6 +99,7 @@ sub canon_mic_arch($) {
         "32e" => "Intel(R) 64",
         "arm" => "ARM",
         "aarch64" => "AArch64",
+        "aarch64_32" => "AArch64_32",
         "loongarch64" => "LoongArch64",
         "mic" => "Intel(R) Many Integrated Core Architecture",
         "mips" => "MIPS",
@@ -222,6 +225,8 @@ sub target_options() {
         $_host_arch = "ppc64le";
     } elsif ( $hardware_platform eq "ppc64" ) {
         $_host_arch = "ppc64";
+    } elsif ( $hardware_platform eq "aarch64_32" ) {
+        $_host_arch = "aarch64_32";
     } elsif ( $hardware_platform eq "aarch64" ) {
         $_host_arch = "aarch64";
     } elsif ( $hardware_platform eq "mips64" ) {
diff --git a/openmp/runtime/tools/lib/Uname.pm b/openmp/runtime/tools/lib/Uname.pm
index 9dde444..447680c 100644
--- a/openmp/runtime/tools/lib/Uname.pm
+++ b/openmp/runtime/tools/lib/Uname.pm
@@ -150,6 +150,8 @@ if ( 0 ) {
         $values{ hardware_platform } = "ppc64le";
     } elsif ( $values{ machine } =~ m{\Appc64\z} ) {
         $values{ hardware_platform } = "ppc64";
+    } elsif ( $values{ machine } =~ m{\Aaarch64_32\z} ) {
+        $values{ hardware_platform } = "aarch64_32";
     } elsif ( $values{ machine } =~ m{\Aaarch64\z} ) {
         $values{ hardware_platform } = "aarch64";
     } elsif ( $values{ machine } =~ m{\Amips64\z} ) {