[openmp] Fix strict aliasing issue in cmpxchg routine

[serge-sans-paille]

Summary

Avoid warning under -fstrict-aliasing by using a call to memcpy to perform type punning.

Compiler output seems unaffected, see https://godbolt.org/z/E4hhvox1x

[jdoerfert]

LG

openmp/runtime/src/kmp_os.h

 
 #ifndef KMP_OS_H
 #define KMP_OS_H
 
 #include "kmp_config.h"
 #include <atomic>
 #include <stdarg.h>
 #include <stdlib.h>
+#include <string.h>
 
 #define KMP_FTN_PLAIN 1
 #define KMP_FTN_APPEND 2
 #define KMP_FTN_UPPER 3
 /*
 #define KMP_FTN_PREPEND 4
 #define KMP_FTN_UAPPEND 5
 */

 #define KMP_XCHG_FIXED16(p, v)                                                 \
   __sync_lock_test_and_set((volatile kmp_uint16 *)(p), (kmp_uint16)(v))
 #define KMP_XCHG_FIXED32(p, v)                                                 \
   __sync_lock_test_and_set((volatile kmp_uint32 *)(p), (kmp_uint32)(v))
 #define KMP_XCHG_FIXED64(p, v)                                                 \
   __sync_lock_test_and_set((volatile kmp_uint64 *)(p), (kmp_uint64)(v))
 
 inline kmp_real32 KMP_XCHG_REAL32(volatile kmp_real32 *p, kmp_real32 v) {
-  kmp_int32 tmp =
-      __sync_lock_test_and_set((volatile kmp_uint32 *)(p), *(kmp_uint32 *)&v);
-  return *(kmp_real32 *)&tmp;
+  volatile kmp_uint32 *up;
+  kmp_uint32 uv;
+  memcpy(&up, &p, sizeof(up));
+  memcpy(&uv, &v, sizeof(uv));
+  kmp_int32 tmp = __sync_lock_test_and_set(up, uv);
+  kmp_real32 ftmp;
+  memcpy(&ftmp, &tmp, sizeof(tmp));
+  return ftmp;
 }
 
 inline kmp_real64 KMP_XCHG_REAL64(volatile kmp_real64 *p, kmp_real64 v) {
-  kmp_int64 tmp =
-      __sync_lock_test_and_set((volatile kmp_uint64 *)(p), *(kmp_uint64 *)&v);
-  return *(kmp_real64 *)&tmp;
+  volatile kmp_uint64 *up;
+  kmp_uint64 uv;
+  memcpy(&up, &p, sizeof(up));
+  memcpy(&uv, &v, sizeof(uv));
+  kmp_int64 tmp = __sync_lock_test_and_set(up, uv);
+  kmp_real64 dtmp;
+  memcpy(&dtmp, &tmp, sizeof(tmp));
+  return dtmp;
 }
 
 #else
 
 extern kmp_int8 __kmp_test_then_add8(volatile kmp_int8 *p, kmp_int8 v);
 extern kmp_int8 __kmp_test_then_or8(volatile kmp_int8 *p, kmp_int8 v);
 extern kmp_int8 __kmp_test_then_and8(volatile kmp_int8 *p, kmp_int8 v);
 extern kmp_int32 __kmp_test_then_add32(volatile kmp_int32 *p, kmp_int32 v);