diff --git a/libc/include/stdatomic.h b/libc/include/stdatomic.h
index 6854d8b..3db25a7 100644
--- a/libc/include/stdatomic.h
+++ b/libc/include/stdatomic.h
@@ -32,13 +32,33 @@
 
 #include <sys/cdefs.h>
 
-#if defined(__cplusplus) && defined(_USING_LIBCXX) && \
-    (__has_feature(cxx_atomic) || _GNUC_VER >= 407)
+
+#if defined(__cplusplus) && defined(_USING_LIBCXX)
+# ifdef __clang__
+#  if __has_feature(cxx_atomic)
+#   define _STDATOMIC_HAVE_ATOMIC
+#  endif
+# else /* gcc */
+#  if __GNUC_PREREQ(4, 7)
+#   define _STDATOMIC_HAVE_ATOMIC
+#  endif
+# endif
+#endif
+
+#ifdef _STDATOMIC_HAVE_ATOMIC
 
 /* We have a usable C++ <atomic>; use it instead.  */
 
 #include <atomic>
 
+#undef _Atomic
+        /* Also defined by <atomic> for gcc.  But not used in macros. */
+        /* Also a clang intrinsic.                                    */
+        /* Should not be used by client code before this file is      */
+        /* included.  The definitions in <atomic> themselves see      */
+        /* the old definition, as they should.                        */
+        /* Client code sees the following definition.                 */
+
 #define _Atomic(t) std::atomic<t>
 
 using std::atomic_is_lock_free;
@@ -123,14 +143,30 @@
  * bits as a T.
  */
 
-#if __has_extension(c_atomic) || __has_extension(cxx_atomic)
-#define	__CLANG_ATOMICS
-#elif __GNUC_PREREQ(4, 7)
-#define	__GNUC_ATOMICS
-#elif defined(__GNUC__)
-#define	__SYNC_ATOMICS
+#include <stddef.h>  /* For ptrdiff_t.                          */
+#include <stdint.h>  /* TODO: Should pollute namespace less.    */
+#if __STDC_VERSION__ >= 201112L
+# include <uchar.h>  /* For char16_t and char32_t.              */
+#endif
+
+#ifdef __clang__
+# if __has_extension(c_atomic) || __has_extension(cxx_atomic)
+#  define       __CLANG_ATOMICS
+# else
+#  error "stdatomic.h does not support your compiler"
+# endif
+# if __has_builtin(__sync_swap)
+#  define __HAS_BUILTIN_SYNC_SWAP
+# endif
 #else
-#error "stdatomic.h does not support your compiler"
+# if __GNUC_PREREQ(4, 7)
+#  define	__GNUC_ATOMICS
+# else
+#  define	__SYNC_ATOMICS
+#  ifdef __cplusplus
+#   define       __ATOMICS_AVOID_DOT_INIT
+#  endif
+# endif
 #endif
 
 /*
@@ -139,33 +175,53 @@
 
 #ifdef __GCC_ATOMIC_BOOL_LOCK_FREE
 #define	ATOMIC_BOOL_LOCK_FREE		__GCC_ATOMIC_BOOL_LOCK_FREE
+#elif defined(__SYNC_ATOMICS)
+#define	ATOMIC_BOOL_LOCK_FREE           2 /* For all modern platforms */
 #endif
 #ifdef __GCC_ATOMIC_CHAR_LOCK_FREE
 #define	ATOMIC_CHAR_LOCK_FREE		__GCC_ATOMIC_CHAR_LOCK_FREE
+#elif defined(__SYNC_ATOMICS)
+#define	ATOMIC_CHAR_LOCK_FREE           2
 #endif
 #ifdef __GCC_ATOMIC_CHAR16_T_LOCK_FREE
 #define	ATOMIC_CHAR16_T_LOCK_FREE	__GCC_ATOMIC_CHAR16_T_LOCK_FREE
+#elif defined(__SYNC_ATOMICS)
+#define	ATOMIC_CHAR16_T_LOCK_FREE       2
 #endif
 #ifdef __GCC_ATOMIC_CHAR32_T_LOCK_FREE
 #define	ATOMIC_CHAR32_T_LOCK_FREE	__GCC_ATOMIC_CHAR32_T_LOCK_FREE
+#elif defined(__SYNC_ATOMICS)
+#define	ATOMIC_CHAR32_T_LOCK_FREE       2
 #endif
 #ifdef __GCC_ATOMIC_WCHAR_T_LOCK_FREE
 #define	ATOMIC_WCHAR_T_LOCK_FREE	__GCC_ATOMIC_WCHAR_T_LOCK_FREE
+#elif defined(__SYNC_ATOMICS)
+#define	ATOMIC_WCHAR_T_LOCK_FREE        2
 #endif
 #ifdef __GCC_ATOMIC_SHORT_LOCK_FREE
 #define	ATOMIC_SHORT_LOCK_FREE		__GCC_ATOMIC_SHORT_LOCK_FREE
+#elif defined(__SYNC_ATOMICS)
+#define	ATOMIC_SHORT_LOCK_FREE          2
 #endif
 #ifdef __GCC_ATOMIC_INT_LOCK_FREE
 #define	ATOMIC_INT_LOCK_FREE		__GCC_ATOMIC_INT_LOCK_FREE
+#elif defined(__SYNC_ATOMICS)
+#define	ATOMIC_INT_LOCK_FREE            2
 #endif
 #ifdef __GCC_ATOMIC_LONG_LOCK_FREE
 #define	ATOMIC_LONG_LOCK_FREE		__GCC_ATOMIC_LONG_LOCK_FREE
+#elif defined(__SYNC_ATOMICS)
+#define	ATOMIC_LONG_LOCK_FREE           2
 #endif
 #ifdef __GCC_ATOMIC_LLONG_LOCK_FREE
 #define	ATOMIC_LLONG_LOCK_FREE		__GCC_ATOMIC_LLONG_LOCK_FREE
+#elif defined(__SYNC_ATOMICS)
+#define	ATOMIC_LLONG_LOCK_FREE          1 /* maybe */
 #endif
 #ifdef __GCC_ATOMIC_POINTER_LOCK_FREE
 #define	ATOMIC_POINTER_LOCK_FREE	__GCC_ATOMIC_POINTER_LOCK_FREE
+#elif defined(__SYNC_ATOMICS)
+#define	ATOMIC_POINTER_LOCK_FREE        2
 #endif
 
 /*
@@ -176,7 +232,11 @@
 #define	ATOMIC_VAR_INIT(value)		(value)
 #define	atomic_init(obj, value)		__c11_atomic_init(obj, value)
 #else
+#ifdef __ATOMICS_AVOID_DOT_INIT
+#define	ATOMIC_VAR_INIT(value)		{ value }
+#else
 #define	ATOMIC_VAR_INIT(value)		{ .__val = (value) }
+#endif
 #define	atomic_init(obj, value)		((void)((obj)->__val = (value)))
 #endif
 
@@ -228,7 +288,7 @@
  */
 
 static __inline void
-atomic_thread_fence(memory_order __order __unused)
+atomic_thread_fence(memory_order __order __attribute__((unused)))
 {
 
 #ifdef __CLANG_ATOMICS
@@ -241,7 +301,7 @@
 }
 
 static __inline void
-atomic_signal_fence(memory_order __order __unused)
+atomic_signal_fence(memory_order __order __attribute__((unused)))
 {
 
 #ifdef __CLANG_ATOMICS
@@ -263,7 +323,7 @@
 	((void)(obj), (_Bool)1)
 #elif defined(__CLANG_ATOMICS)
 #define	atomic_is_lock_free(obj) \
-	__atomic_is_lock_free(sizeof(*(obj)), obj)
+	__c11_atomic_is_lock_free(sizeof(*(obj)))
 #elif defined(__GNUC_ATOMICS)
 #define	atomic_is_lock_free(obj) \
 	__atomic_is_lock_free(sizeof((obj)->__val), &(obj)->__val)
@@ -276,7 +336,7 @@
  * 7.17.6 Atomic integer types.
  */
 
-#if !__has_extension(c_atomic) && !__has_extension(cxx_atomic)
+#ifndef __CLANG_ATOMICS
 /*
  * No native support for _Atomic(). Place object in structure to prevent
  * most forms of direct non-atomic access.
@@ -397,7 +457,7 @@
     desired, success, failure)						\
 	atomic_compare_exchange_strong_explicit(object, expected,	\
 		desired, success, failure)
-#if __has_builtin(__sync_swap)
+#ifdef __HAS_BUILTIN_SYNC_SWAP
 /* Clang provides a full-barrier atomic exchange - use it if available. */
 #define	atomic_exchange_explicit(object, desired, order)		\
 	((void)(order), __sync_swap(&(object)->__val, desired))
@@ -477,7 +537,7 @@
 	atomic_bool	__flag;
 } atomic_flag;
 
-#define	ATOMIC_FLAG_INIT		{ ATOMIC_VAR_INIT(0) }
+#define	ATOMIC_FLAG_INIT		{ ATOMIC_VAR_INIT(false) }
 
 static __inline bool
 atomic_flag_test_and_set_explicit(volatile atomic_flag *__object,
diff --git a/tests/Android.mk b/tests/Android.mk
index b370b92..8184bf7 100644
--- a/tests/Android.mk
+++ b/tests/Android.mk
@@ -237,7 +237,8 @@
 
 # -----------------------------------------------------------------------------
 # Tests for the device using bionic's .so. Run with:
-#   adb shell /data/nativetest/bionic-unit-tests/bionic-unit-tests
+#   adb shell /data/nativetest/bionic-unit-tests/bionic-unit-tests32
+#   adb shell /data/nativetest/bionic-unit-tests/bionic-unit-tests64
 # -----------------------------------------------------------------------------
 bionic-unit-tests_whole_static_libraries := \
     libBionicTests \
@@ -269,7 +270,8 @@
 
 # -----------------------------------------------------------------------------
 # Tests for the device linked against bionic's static library. Run with:
-#   adb shell /data/nativetest/bionic-unit-tests-static/bionic-unit-tests-static
+#   adb shell /data/nativetest/bionic-unit-tests-static/bionic-unit-tests-static32
+#   adb shell /data/nativetest/bionic-unit-tests-static/bionic-unit-tests-static64
 # -----------------------------------------------------------------------------
 bionic-unit-tests-static_whole_static_libraries := \
     libBionicTests \
diff --git a/tests/stdatomic_test.cpp b/tests/stdatomic_test.cpp
index 5e88c88..b7fb19b 100644
--- a/tests/stdatomic_test.cpp
+++ b/tests/stdatomic_test.cpp
@@ -14,11 +14,10 @@
  * limitations under the License.
  */
 
-#include <gtest/gtest.h>
-
-#if !defined(__GLIBC__) /* TODO: fix our prebuilt toolchains! */
-
 #include <stdatomic.h>
+#include <gtest/gtest.h>
+#include <pthread.h>
+#include <stdint.h>
 
 TEST(stdatomic, LOCK_FREE) {
   ASSERT_TRUE(ATOMIC_BOOL_LOCK_FREE);
@@ -64,14 +63,17 @@
 
 TEST(stdatomic, atomic_is_lock_free) {
   atomic_char small;
-  atomic_intmax_t big;
   ASSERT_TRUE(atomic_is_lock_free(&small));
+#if defined(__clang__) || __GNUC_PREREQ(4, 7)
+  // Otherwise stdatomic.h doesn't handle this.
+  atomic_intmax_t big;
   // atomic_intmax_t(size = 64) is not lock free on mips32.
 #if defined(__mips__) && !defined(__LP64__)
   ASSERT_FALSE(atomic_is_lock_free(&big));
 #else
   ASSERT_TRUE(atomic_is_lock_free(&big));
 #endif
+#endif
 }
 
 TEST(stdatomic, atomic_flag) {
@@ -167,4 +169,80 @@
   ASSERT_EQ(0x002, atomic_load(&i));
 }
 
-#endif
+// And a rudimentary test of acquire-release memory ordering:
+
+constexpr static uint_least32_t BIG = 10000000ul; // Assumed even below.
+
+struct three_atomics {
+  atomic_uint_least32_t x;
+  char a[123];  // Everything in different cache lines,
+                // increase chance of compiler getting alignment wrong.
+  atomic_uint_least32_t y;
+  char b[4013];
+  atomic_uint_least32_t z;
+};
+
+// Very simple acquire/release memory ordering sanity check.
+static void* writer(void* arg) {
+  three_atomics* a = reinterpret_cast<three_atomics*>(arg);
+  for (uint_least32_t i = 0; i <= BIG; i+=2) {
+    atomic_store_explicit(&a->x, i, memory_order_relaxed);
+    atomic_store_explicit(&a->z, i, memory_order_relaxed);
+    atomic_store_explicit(&a->y, i, memory_order_release);
+    atomic_store_explicit(&a->x, i+1, memory_order_relaxed);
+    atomic_store_explicit(&a->z, i+1, memory_order_relaxed);
+    atomic_store_explicit(&a->y, i+1, memory_order_release);
+  }
+  return 0;
+}
+
+static void* reader(void* arg) {
+  three_atomics* a = reinterpret_cast<three_atomics*>(arg);
+  uint_least32_t xval = 0, yval = 0, zval = 0;
+  size_t repeat = 0;
+  size_t repeat_limit = 1000;
+  while (yval != BIG + 1) {
+    yval = atomic_load_explicit(&a->y, memory_order_acquire);
+    zval = atomic_load_explicit(&a->z, memory_order_relaxed);
+    xval = atomic_load_explicit(&a->x, memory_order_relaxed);
+    // If we see a given value of y, the immediately preceding
+    // stores to z and x, or later ones, should also be visible.
+    if (zval < yval) {
+      // Cant just ASSERT, since we are in a non-void function.
+      ADD_FAILURE() << "acquire-release ordering violation: "
+                    << zval << " < " << yval << ", " << xval << "\n";
+      return 0; // Only report once.
+    }
+    if (xval < yval) {
+      // Cant just ASSERT, since we are in a non-void function.
+      ADD_FAILURE() << "acquire-release ordering violation: "
+                    << xval << " < " << yval << ", " << zval <<  "\n";
+      return 0; // Only report once.
+    }
+    if (repeat < repeat_limit) ++repeat;
+  }
+  // The following assertion is not technically guaranteed to hold.
+  // But if it fails to hold, this test was useless, and we have a
+  // serious scheduling issue that we should probably know about.
+  EXPECT_EQ(repeat, repeat_limit);
+  return 0;
+}
+
+TEST(stdatomic, ordering) {
+  // Run a memory ordering sanity test.
+  void* result;
+  three_atomics a;
+  atomic_init(&a.x, 0ul);
+  atomic_init(&a.y, 0ul);
+  atomic_init(&a.z, 0ul);
+  pthread_t t1,t2;
+  ASSERT_EQ(0, pthread_create(&t1, 0, reader, &a));
+  ASSERT_EQ(0, pthread_create(&t2, 0, writer, &a));
+  ASSERT_EQ(0, pthread_join(t1, &result));
+  EXPECT_EQ(0, result);
+  ASSERT_EQ(0, pthread_join(t2, &result));
+  EXPECT_EQ(0, result);
+  EXPECT_EQ(atomic_load_explicit(&a.x, memory_order_consume), BIG + 1);
+  EXPECT_EQ(atomic_load_explicit(&a.y, memory_order_seq_cst), BIG + 1);
+  EXPECT_EQ(atomic_load(&a.z), BIG + 1);
+}