[OpenMP][AIX]Define struct kmp_base_tas_lock with the order of two members swapped for big-endian (#79188)

The direct lock data structure has bit `0` (the least significant bit)
of the first 32-bit word set to `1` to indicate it is a direct lock. On
the other hand, the first word (in 32-bit mode) or first two words (in
64-bit mode) of an indirect lock are the address of the entry allocated
from the indirect lock table. The runtime checks bit `0` of the first
32-bit word to tell if this is a direct or an indirect lock. This works
fine for 32-bit and 64-bit little-endian because its memory layout of a
64-bit address is (`low word`, `high word`). However, this causes
problems for big-endian where the memory layout of a 64-bit address is
(`high word`, `low word`). If an address of the indirect lock table
entry is something like `0x110035300`, i.e., (`0x1`, `0x10035300`), it
is treated as a direct lock. This patch defines `struct
kmp_base_tas_lock` with the ordering of the two 32-bit members flipped
for big-endian PPC64 so that when checking/setting tags in member
`poll`, the second word (the low word) is used. This patch also changes
places where `poll` is not already explicitly specified for
checking/setting tags.

4 files changed, 20 insertions, 10 deletions

diff --git a/openmp/runtime/src/kmp_csupport.cpp b/openmp/runtime/src/kmp_csupport.cpp
index 9eeaeb8..878e78b 100644
--- a/openmp/runtime/src/kmp_csupport.cpp
+++ b/openmp/runtime/src/kmp_csupport.cpp
@@ -1533,8 +1533,9 @@ void __kmpc_critical_with_hint(ident_t *loc, kmp_int32 global_tid,
   kmp_dyna_lockseq_t lockseq = __kmp_map_hint_to_lock(hint);
   if (*lk == 0) {
     if (KMP_IS_D_LOCK(lockseq)) {
-      KMP_COMPARE_AND_STORE_ACQ32((volatile kmp_int32 *)crit, 0,
-                                  KMP_GET_D_TAG(lockseq));
+      KMP_COMPARE_AND_STORE_ACQ32(
+          (volatile kmp_int32 *)&((kmp_base_tas_lock_t *)crit)->poll, 0,
+          KMP_GET_D_TAG(lockseq));
     } else {
       __kmp_init_indirect_csptr(crit, loc, global_tid, KMP_GET_I_TAG(lockseq));
     }
diff --git a/openmp/runtime/src/kmp_gsupport.cpp b/openmp/runtime/src/kmp_gsupport.cpp
index 8818965..4dc8a90 100644
--- a/openmp/runtime/src/kmp_gsupport.cpp
+++ b/openmp/runtime/src/kmp_gsupport.cpp
@@ -144,7 +144,7 @@ void KMP_EXPAND_NAME(KMP_API_NAME_GOMP_BARRIER)(void) {
 
 // Mutual exclusion
 
-// The symbol that icc/ifort generates for unnamed for unnamed critical sections
+// The symbol that icc/ifort generates for unnamed critical sections
 // - .gomp_critical_user_ - is defined using .comm in any objects reference it.
 // We can't reference it directly here in C code, as the symbol contains a ".".
 //
diff --git a/openmp/runtime/src/kmp_lock.cpp b/openmp/runtime/src/kmp_lock.cpp
index 85c54f4..0ad14f8 100644
--- a/openmp/runtime/src/kmp_lock.cpp
+++ b/openmp/runtime/src/kmp_lock.cpp
@@ -2689,7 +2689,7 @@ void __kmp_spin_backoff(kmp_backoff_t *boff) {
 // lock word.
 static void __kmp_init_direct_lock(kmp_dyna_lock_t *lck,
                                    kmp_dyna_lockseq_t seq) {
-  TCW_4(*lck, KMP_GET_D_TAG(seq));
+  TCW_4(((kmp_base_tas_lock_t *)lck)->poll, KMP_GET_D_TAG(seq));
   KA_TRACE(
       20,
       ("__kmp_init_direct_lock: initialized direct lock with type#%d\n", seq));
@@ -3180,8 +3180,8 @@ kmp_indirect_lock_t *__kmp_allocate_indirect_lock(void **user_lock,
   lck->type = tag;
 
   if (OMP_LOCK_T_SIZE < sizeof(void *)) {
-    *((kmp_lock_index_t *)user_lock) = idx
-                                       << 1; // indirect lock word must be even
+    *(kmp_lock_index_t *)&(((kmp_base_tas_lock_t *)user_lock)->poll) =
+        idx << 1; // indirect lock word must be even
   } else {
     *((kmp_indirect_lock_t **)user_lock) = lck;
   }
diff --git a/openmp/runtime/src/kmp_lock.h b/openmp/runtime/src/kmp_lock.h
index f21179b..e2a0cda 100644
--- a/openmp/runtime/src/kmp_lock.h
+++ b/openmp/runtime/src/kmp_lock.h
@@ -50,7 +50,7 @@ typedef struct ident ident_t;
 // recent versions), but we are bounded by the pointer-sized chunks that
 // the Intel compiler allocates.
 
-#if KMP_OS_LINUX && defined(KMP_GOMP_COMPAT)
+#if (KMP_OS_LINUX || KMP_OS_AIX) && defined(KMP_GOMP_COMPAT)
 #define OMP_LOCK_T_SIZE sizeof(int)
 #define OMP_NEST_LOCK_T_SIZE sizeof(void *)
 #else
@@ -120,8 +120,15 @@ extern void __kmp_validate_locks(void);
 
 struct kmp_base_tas_lock {
   // KMP_LOCK_FREE(tas) => unlocked; locked: (gtid+1) of owning thread
+#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ && __LP64__
+  // Flip the ordering of the high and low 32-bit member to be consistent
+  // with the memory layout of the address in 64-bit big-endian.
+  kmp_int32 depth_locked; // depth locked, for nested locks only
+  std::atomic<kmp_int32> poll;
+#else
   std::atomic<kmp_int32> poll;
   kmp_int32 depth_locked; // depth locked, for nested locks only
+#endif
 };
 
 typedef struct kmp_base_tas_lock kmp_base_tas_lock_t;
@@ -1138,11 +1145,13 @@ extern int (**__kmp_indirect_test)(kmp_user_lock_p, kmp_int32);
 
 // Extracts direct lock tag from a user lock pointer
 #define KMP_EXTRACT_D_TAG(l)                                                   \
-  (*((kmp_dyna_lock_t *)(l)) & ((1 << KMP_LOCK_SHIFT) - 1) &                   \
-   -(*((kmp_dyna_lock_t *)(l)) & 1))
+  ((kmp_dyna_lock_t)((kmp_base_tas_lock_t *)(l))->poll &                       \
+   ((1 << KMP_LOCK_SHIFT) - 1) &                                               \
+   -((kmp_dyna_lock_t)((kmp_tas_lock_t *)(l))->lk.poll & 1))
 
 // Extracts indirect lock index from a user lock pointer
-#define KMP_EXTRACT_I_INDEX(l) (*(kmp_lock_index_t *)(l) >> 1)
+#define KMP_EXTRACT_I_INDEX(l)                                                 \
+  ((kmp_lock_index_t)((kmp_base_tas_lock_t *)(l))->poll >> 1)
 
 // Returns function pointer to the direct lock function with l (kmp_dyna_lock_t
 // *) and op (operation type).