Improve how to handle yield() in ARM.

The yield instruction was introduced in ARM processors more recent than armel.
So building in armel ends up in "Error: selected processor does not support
`yield' in ARM mode". Also, the __yield() intrinsic instruction is not
understood for armel by g++. So let's do nothing for armel.

Author: santiagorr

api/include/opentelemetry/common/spin_lock_mutex.h

@@ -68,8 +68,10 @@ class SpinLockMutex
 #  else
     __builtin_ia32_pause();
 #  endif
-#elif defined(__arm__)
-    __asm__ volatile("yield" ::: "memory");
+#elif defined(__armel__) || defined(__ARMEL__)
+    asm volatile ("nop" ::: "memory");
+#elif defined(__arm__) || defined(__aarch64__) // arm big endian / arm64
+    __asm__ __volatile__ ("yield" ::: "memory");
 #else
     // TODO: Issue PAGE/YIELD on other architectures.
 #endif

api/test/common/spinlock_benchmark.cc

@@ -102,8 +102,10 @@ static void BM_ProcYieldSpinLockThrashing(benchmark::State &s)
 #  else
           __builtin_ia32_pause();
 #  endif
-#elif defined(__arm__)
-          __asm__ volatile("yield" ::: "memory");
+#elif defined(__armel__) || defined(__ARMEL__)
+    asm volatile ("nop" ::: "memory");
+#elif defined(__arm__) || defined(__aarch64__) // arm big endian / arm64
+    __asm__ __volatile__ ("yield" ::: "memory");
 #endif
         }
       },