diff --git a/libc/Android.bp b/libc/Android.bp
index 3c5c9f1..c34023c 100644
--- a/libc/Android.bp
+++ b/libc/Android.bp
@@ -1659,8 +1659,14 @@
     },
     native_bridge_supported: false,
     // It is never correct to depend on this directly. This is only
-    // needed for the runtime apex, and in base_system.mk.
-    visibility: ["//bionic/apex"],
+    // needed for the runtime apex, and in base_system.mk, and system_image_defaults
+    // which is default module for soong-defined system image.
+    visibility: [
+        "//bionic/apex",
+        "//build/make/target/product/generic",
+        //TODO(b/381985636) : Remove visibility to Soong-defined GSI once resolved
+        "//build/make/target/product/gsi",
+    ],
 }
 
 genrule {
diff --git a/libc/bionic/posix_timers.cpp b/libc/bionic/posix_timers.cpp
index 65749a4..9516059 100644
--- a/libc/bionic/posix_timers.cpp
+++ b/libc/bionic/posix_timers.cpp
@@ -141,7 +141,7 @@
   // Otherwise, this must be SIGEV_THREAD timer...
   timer->callback = evp->sigev_notify_function;
   timer->callback_argument = evp->sigev_value;
-  atomic_init(&timer->deleted, false);
+  atomic_store_explicit(&timer->deleted, false, memory_order_relaxed);
 
   // Check arguments that the kernel doesn't care about but we do.
   if (timer->callback == nullptr) {
diff --git a/libc/bionic/pthread_barrier.cpp b/libc/bionic/pthread_barrier.cpp
index 1618222..ff048a6 100644
--- a/libc/bionic/pthread_barrier.cpp
+++ b/libc/bionic/pthread_barrier.cpp
@@ -95,8 +95,8 @@
     return EINVAL;
   }
   barrier->init_count = count;
-  atomic_init(&barrier->state, WAIT);
-  atomic_init(&barrier->wait_count, 0);
+  atomic_store_explicit(&barrier->state, WAIT, memory_order_relaxed);
+  atomic_store_explicit(&barrier->wait_count, 0, memory_order_relaxed);
   barrier->pshared = false;
   if (attr != nullptr && (*attr & 1)) {
     barrier->pshared = true;
diff --git a/libc/bionic/pthread_cond.cpp b/libc/bionic/pthread_cond.cpp
index f444676..197fd19 100644
--- a/libc/bionic/pthread_cond.cpp
+++ b/libc/bionic/pthread_cond.cpp
@@ -140,10 +140,10 @@
   if (attr != nullptr) {
     init_state = (*attr & COND_FLAGS_MASK);
   }
-  atomic_init(&cond->state, init_state);
+  atomic_store_explicit(&cond->state, init_state, memory_order_relaxed);
 
 #if defined(__LP64__)
-  atomic_init(&cond->waiters, 0);
+  atomic_store_explicit(&cond->waiters, 0, memory_order_relaxed);
 #endif
 
   return 0;
diff --git a/libc/bionic/pthread_create.cpp b/libc/bionic/pthread_create.cpp
index 54bfa20..3fa8ee6 100644
--- a/libc/bionic/pthread_create.cpp
+++ b/libc/bionic/pthread_create.cpp
@@ -159,11 +159,11 @@
 int __init_thread(pthread_internal_t* thread) {
   thread->cleanup_stack = nullptr;
 
-  if (__predict_true((thread->attr.flags & PTHREAD_ATTR_FLAG_DETACHED) == 0)) {
-    atomic_init(&thread->join_state, THREAD_NOT_JOINED);
-  } else {
-    atomic_init(&thread->join_state, THREAD_DETACHED);
+  ThreadJoinState state = THREAD_NOT_JOINED;
+  if (__predict_false((thread->attr.flags & PTHREAD_ATTR_FLAG_DETACHED) != 0)) {
+    state = THREAD_DETACHED;
   }
+  atomic_store_explicit(&thread->join_state, state, memory_order_relaxed);
 
   // Set the scheduling policy/priority of the thread if necessary.
   bool need_set = true;
diff --git a/libc/bionic/pthread_internal.cpp b/libc/bionic/pthread_internal.cpp
index c6426ed..4f2ad0c 100644
--- a/libc/bionic/pthread_internal.cpp
+++ b/libc/bionic/pthread_internal.cpp
@@ -259,8 +259,7 @@
   g_func = func;
   g_arg = arg;
 
-  static _Atomic(bool) g_retval;
-  atomic_init(&g_retval, true);
+  static _Atomic(bool) g_retval(true);
 
   auto handler = [](int, siginfo_t*, void*) {
     ErrnoRestorer restorer;
diff --git a/libc/bionic/pthread_mutex.cpp b/libc/bionic/pthread_mutex.cpp
index 0a452e9..c99717a 100644
--- a/libc/bionic/pthread_mutex.cpp
+++ b/libc/bionic/pthread_mutex.cpp
@@ -509,8 +509,8 @@
     memset(mutex, 0, sizeof(pthread_mutex_internal_t));
 
     if (__predict_true(attr == nullptr)) {
-        atomic_init(&mutex->state, MUTEX_TYPE_BITS_NORMAL);
-        return 0;
+      atomic_store_explicit(&mutex->state, MUTEX_TYPE_BITS_NORMAL, memory_order_relaxed);
+      return 0;
     }
 
     uint16_t state = 0;
@@ -543,13 +543,13 @@
         }
         mutex->pi_mutex_id = id;
 #endif
-        atomic_init(&mutex->state, PI_MUTEX_STATE);
+        atomic_store_explicit(&mutex->state, PI_MUTEX_STATE, memory_order_relaxed);
         PIMutex& pi_mutex = mutex->ToPIMutex();
         pi_mutex.type = *attr & MUTEXATTR_TYPE_MASK;
         pi_mutex.shared = (*attr & MUTEXATTR_SHARED_MASK) != 0;
     } else {
-        atomic_init(&mutex->state, state);
-        atomic_init(&mutex->owner_tid, 0);
+      atomic_store_explicit(&mutex->state, state, memory_order_relaxed);
+      atomic_store_explicit(&mutex->owner_tid, 0, memory_order_relaxed);
     }
     return 0;
 }
diff --git a/libc/bionic/pthread_rwlock.cpp b/libc/bionic/pthread_rwlock.cpp
index 6f3c6fe..92134b4 100644
--- a/libc/bionic/pthread_rwlock.cpp
+++ b/libc/bionic/pthread_rwlock.cpp
@@ -247,7 +247,7 @@
     }
   }
 
-  atomic_init(&rwlock->state, 0);
+  atomic_store_explicit(&rwlock->state, 0, memory_order_relaxed);
   rwlock->pending_lock.init(rwlock->pshared);
   return 0;
 }
diff --git a/libc/bionic/semaphore.cpp b/libc/bionic/semaphore.cpp
index 33552a9..2c9b745 100644
--- a/libc/bionic/semaphore.cpp
+++ b/libc/bionic/semaphore.cpp
@@ -113,7 +113,7 @@
   }
 
   atomic_uint* sem_count_ptr = SEM_TO_ATOMIC_POINTER(sem);
-  atomic_init(sem_count_ptr, count);
+  atomic_store_explicit(sem_count_ptr, count, memory_order_relaxed);
   return 0;
 }
 
diff --git a/libc/kernel/tools/defaults.py b/libc/kernel/tools/defaults.py
index a71318e..2994e5e 100644
--- a/libc/kernel/tools/defaults.py
+++ b/libc/kernel/tools/defaults.py
@@ -25,6 +25,23 @@
     # Otherwise, there will be two struct timeval definitions when
     # __kernel_old_timeval is renamed to timeval.
     "__kernel_old_timeval": "1",
+    # Drop the custom byte swap functions and just use the clang builtins.
+    # https://github.com/android/ndk/issues/2107
+    "__arch_swab16": kCppUndefinedMacro,
+    "__arch_swab16p": kCppUndefinedMacro,
+    "__arch_swab16s": kCppUndefinedMacro,
+    "__arch_swab32": kCppUndefinedMacro,
+    "__arch_swab32p": kCppUndefinedMacro,
+    "__arch_swab32s": kCppUndefinedMacro,
+    "__arch_swab64": kCppUndefinedMacro,
+    "__arch_swab64p": kCppUndefinedMacro,
+    "__arch_swab64s": kCppUndefinedMacro,
+    "__arch_swahb32": kCppUndefinedMacro,
+    "__arch_swahb32p": kCppUndefinedMacro,
+    "__arch_swahb32s": kCppUndefinedMacro,
+    "__arch_swahw32": kCppUndefinedMacro,
+    "__arch_swahw32p": kCppUndefinedMacro,
+    "__arch_swahw32s": kCppUndefinedMacro,
     }
 
 # This is the set of known kernel data structures we want to remove from
@@ -133,9 +150,6 @@
           # These are required to support the above functions.
           "__fswahw32",
           "__fswahb32",
-          # As are these, for ILP32.
-          "__arch_swab32",
-          "__arch_swab64",
           # This is used by various macros in <linux/ioprio.h>.
           "ioprio_value",
 
diff --git a/libc/kernel/uapi/asm-arm/asm/swab.h b/libc/kernel/uapi/asm-arm/asm/swab.h
index 3fff953..7684c22 100644
--- a/libc/kernel/uapi/asm-arm/asm/swab.h
+++ b/libc/kernel/uapi/asm-arm/asm/swab.h
@@ -11,18 +11,7 @@
 #ifndef __STRICT_ANSI__
 #define __SWAB_64_THRU_32__
 #endif
-static inline __attribute__((__const__)) __u32 __arch_swab32(__u32 x) {
-  __u32 t;
 #ifndef __thumb__
-  if(! __builtin_constant_p(x)) {
-    asm("eor\t%0, %1, %1, ror #16" : "=r" (t) : "r" (x));
-  } else
 #endif
-  t = x ^ ((x << 16) | (x >> 16));
-  x = (x << 24) | (x >> 8);
-  t &= ~0x00FF0000;
-  x ^= (t >> 8);
-  return x;
-}
 #define __arch_swab32 __arch_swab32
 #endif
diff --git a/libc/kernel/uapi/asm-x86/asm/swab.h b/libc/kernel/uapi/asm-x86/asm/swab.h
index ce43658..31c850d 100644
--- a/libc/kernel/uapi/asm-x86/asm/swab.h
+++ b/libc/kernel/uapi/asm-x86/asm/swab.h
@@ -8,27 +8,9 @@
 #define _ASM_X86_SWAB_H
 #include <linux/types.h>
 #include <linux/compiler.h>
-static inline __attribute__((__const__)) __u32 __arch_swab32(__u32 val) {
-  asm("bswapl %0" : "=r" (val) : "0" (val));
-  return val;
-}
 #define __arch_swab32 __arch_swab32
-static inline __attribute__((__const__)) __u64 __arch_swab64(__u64 val) {
 #ifdef __i386__
-  union {
-    struct {
-      __u32 a;
-      __u32 b;
-    } s;
-    __u64 u;
-  } v;
-  v.u = val;
-  asm("bswapl %0; bswapl %1; xchgl %0,%1" : "=r" (v.s.a), "=r" (v.s.b) : "0" (v.s.a), "1" (v.s.b));
-  return v.u;
 #else
-  asm("bswapq %0" : "=r" (val) : "0" (val));
-  return val;
 #endif
-}
 #define __arch_swab64 __arch_swab64
 #endif
diff --git a/libc/kernel/uapi/linux/swab.h b/libc/kernel/uapi/linux/swab.h
index 6225a76..5d240e1 100644
--- a/libc/kernel/uapi/linux/swab.h
+++ b/libc/kernel/uapi/linux/swab.h
@@ -15,29 +15,14 @@
 #define ___constant_swab64(x) ((__u64) ((((__u64) (x) & (__u64) 0x00000000000000ffULL) << 56) | (((__u64) (x) & (__u64) 0x000000000000ff00ULL) << 40) | (((__u64) (x) & (__u64) 0x0000000000ff0000ULL) << 24) | (((__u64) (x) & (__u64) 0x00000000ff000000ULL) << 8) | (((__u64) (x) & (__u64) 0x000000ff00000000ULL) >> 8) | (((__u64) (x) & (__u64) 0x0000ff0000000000ULL) >> 24) | (((__u64) (x) & (__u64) 0x00ff000000000000ULL) >> 40) | (((__u64) (x) & (__u64) 0xff00000000000000ULL) >> 56)))
 #define ___constant_swahw32(x) ((__u32) ((((__u32) (x) & (__u32) 0x0000ffffUL) << 16) | (((__u32) (x) & (__u32) 0xffff0000UL) >> 16)))
 #define ___constant_swahb32(x) ((__u32) ((((__u32) (x) & (__u32) 0x00ff00ffUL) << 8) | (((__u32) (x) & (__u32) 0xff00ff00UL) >> 8)))
-#ifdef __arch_swab16
-#else
-#endif
-#ifdef __arch_swab32
-#else
-#endif
-#ifdef __arch_swab64
-#elif defined(__SWAB_64_THRU_32__)
+#ifdef __SWAB_64_THRU_32__
 #else
 #endif
 static inline __attribute__((__const__)) __u32 __fswahw32(__u32 val) {
-#ifdef __arch_swahw32
-  return __arch_swahw32(val);
-#else
   return ___constant_swahw32(val);
-#endif
 }
 static inline __attribute__((__const__)) __u32 __fswahb32(__u32 val) {
-#ifdef __arch_swahb32
-  return __arch_swahb32(val);
-#else
   return ___constant_swahb32(val);
-#endif
 }
 #define __swab16(x) (__u16) __builtin_bswap16((__u16) (x))
 #define __swab32(x) (__u32) __builtin_bswap32((__u32) (x))
@@ -48,73 +33,33 @@
 #define __swahw32(x) (__builtin_constant_p((__u32) (x)) ? ___constant_swahw32(x) : __fswahw32(x))
 #define __swahb32(x) (__builtin_constant_p((__u32) (x)) ? ___constant_swahb32(x) : __fswahb32(x))
 static __always_inline __u16 __swab16p(const __u16 * p) {
-#ifdef __arch_swab16p
-  return __arch_swab16p(p);
-#else
   return __swab16(* p);
-#endif
 }
 static __always_inline __u32 __swab32p(const __u32 * p) {
-#ifdef __arch_swab32p
-  return __arch_swab32p(p);
-#else
   return __swab32(* p);
-#endif
 }
 static __always_inline __u64 __swab64p(const __u64 * p) {
-#ifdef __arch_swab64p
-  return __arch_swab64p(p);
-#else
   return __swab64(* p);
-#endif
 }
 static inline __u32 __swahw32p(const __u32 * p) {
-#ifdef __arch_swahw32p
-  return __arch_swahw32p(p);
-#else
   return __swahw32(* p);
-#endif
 }
 static inline __u32 __swahb32p(const __u32 * p) {
-#ifdef __arch_swahb32p
-  return __arch_swahb32p(p);
-#else
   return __swahb32(* p);
-#endif
 }
 static inline void __swab16s(__u16 * p) {
-#ifdef __arch_swab16s
-  __arch_swab16s(p);
-#else
   * p = __swab16p(p);
-#endif
 }
 static __always_inline void __swab32s(__u32 * p) {
-#ifdef __arch_swab32s
-  __arch_swab32s(p);
-#else
   * p = __swab32p(p);
-#endif
 }
 static __always_inline void __swab64s(__u64 * p) {
-#ifdef __arch_swab64s
-  __arch_swab64s(p);
-#else
   * p = __swab64p(p);
-#endif
 }
 static inline void __swahw32s(__u32 * p) {
-#ifdef __arch_swahw32s
-  __arch_swahw32s(p);
-#else
   * p = __swahw32p(p);
-#endif
 }
 static inline void __swahb32s(__u32 * p) {
-#ifdef __arch_swahb32s
-  __arch_swahb32s(p);
-#else
   * p = __swahb32p(p);
-#endif
 }
 #endif
diff --git a/libc/private/bionic_lock.h b/libc/private/bionic_lock.h
index 8ed4939..d0c6d5e 100644
--- a/libc/private/bionic_lock.h
+++ b/libc/private/bionic_lock.h
@@ -46,7 +46,7 @@
 
  public:
   void init(bool process_shared) {
-    atomic_init(&state, Unlocked);
+    atomic_store_explicit(&state, Unlocked, memory_order_relaxed);
     this->process_shared = process_shared;
   }
 
diff --git a/libc/system_properties/include/system_properties/prop_area.h b/libc/system_properties/include/system_properties/prop_area.h
index 187ff75..089cf52 100644
--- a/libc/system_properties/include/system_properties/prop_area.h
+++ b/libc/system_properties/include/system_properties/prop_area.h
@@ -102,7 +102,7 @@
   }
 
   prop_area(const uint32_t magic, const uint32_t version) : magic_(magic), version_(version) {
-    atomic_init(&serial_, 0u);
+    atomic_store_explicit(&serial_, 0u, memory_order_relaxed);
     memset(reserved_, 0, sizeof(reserved_));
     // Allocate enough space for the root node.
     bytes_used_ = sizeof(prop_trie_node);
diff --git a/libc/system_properties/prop_info.cpp b/libc/system_properties/prop_info.cpp
index c3bf177..499b36a 100644
--- a/libc/system_properties/prop_info.cpp
+++ b/libc/system_properties/prop_info.cpp
@@ -38,7 +38,7 @@
 prop_info::prop_info(const char* name, uint32_t namelen, const char* value, uint32_t valuelen) {
   memcpy(this->name, name, namelen);
   this->name[namelen] = '\0';
-  atomic_init(&this->serial, valuelen << 24);
+  atomic_store_explicit(&this->serial, valuelen << 24, memory_order_relaxed);
   memcpy(this->value, value, valuelen);
   this->value[valuelen] = '\0';
 }
@@ -48,7 +48,7 @@
   this->name[namelen] = '\0';
 
   auto error_value_len = sizeof(kLongLegacyError) - 1;
-  atomic_init(&this->serial, error_value_len << 24 | kLongFlag);
+  atomic_store_explicit(&this->serial, error_value_len << 24 | kLongFlag, memory_order_relaxed);
   memcpy(this->long_property.error_message, kLongLegacyError, sizeof(kLongLegacyError));
 
   this->long_property.offset = long_offset;
diff --git a/tests/stdatomic_test.cpp b/tests/stdatomic_test.cpp
index 1c51b11..8a54080 100644
--- a/tests/stdatomic_test.cpp
+++ b/tests/stdatomic_test.cpp
@@ -40,7 +40,7 @@
   atomic_int v = 123;
   ASSERT_EQ(123, atomic_load(&v));
 
-  atomic_init(&v, 456);
+  atomic_store_explicit(&v, 456, memory_order_relaxed);
   ASSERT_EQ(456, atomic_load(&v));
 
   atomic_flag f = ATOMIC_FLAG_INIT;
@@ -258,9 +258,9 @@
   // Run a memory ordering smoke test.
   void* result;
   three_atomics a;
-  atomic_init(&a.x, 0ul);
-  atomic_init(&a.y, 0ul);
-  atomic_init(&a.z, 0ul);
+  atomic_store_explicit(&a.x, 0ul, memory_order_relaxed);
+  atomic_store_explicit(&a.y, 0ul, memory_order_relaxed);
+  atomic_store_explicit(&a.z, 0ul, memory_order_relaxed);
   pthread_t t1,t2;
   ASSERT_EQ(0, pthread_create(&t1, nullptr, reader, &a));
   ASSERT_EQ(0, pthread_create(&t2, nullptr, writer, &a));