Merge branch 'rickard/libatomic_ops_improvements/OTP-12302' into maint

* rickard/libatomic_ops_improvements/OTP-12302:
  Implement support for double word atomics using libatomic_ops
  Improve usage of libatomic_ops for word size atomics

4 files changed, 905 insertions, 10 deletions

diff --git a/erts/include/internal/ethread.h b/erts/include/internal/ethread.h
index 72c054b588..ad5d05704c 100644
--- a/erts/include/internal/ethread.h
+++ b/erts/include/internal/ethread.h
@@ -364,8 +364,8 @@ extern ethr_runtime_t ethr_runtime__;
 #          include "sparc64/ethread.h"
 #        endif
 #      endif
-#      include "gcc/ethread.h"
 #      include "libatomic_ops/ethread.h"
+#      include "gcc/ethread.h"
 #    endif
 #  elif defined(ETHR_HAVE_LIBATOMIC_OPS)
 #    include "libatomic_ops/ethread.h"
diff --git a/erts/include/internal/libatomic_ops/ethr_atomic.h b/erts/include/internal/libatomic_ops/ethr_atomic.h
index fb1288c330..734cdf0890 100644
--- a/erts/include/internal/libatomic_ops/ethr_atomic.h
+++ b/erts/include/internal/libatomic_ops/ethr_atomic.h
@@ -1,7 +1,7 @@
 /*
  * %CopyrightBegin%
  *
- * Copyright Ericsson AB 2010-2011. All Rights Reserved.
+ * Copyright Ericsson AB 2010-2014. All Rights Reserved.
  *
  * The contents of this file are subject to the Erlang Public License,
  * Version 1.1, (the "License"); you may not use this file except in
@@ -32,22 +32,23 @@
  * These operations need to be defined by libatomic_ops;
  * otherwise, we won't compile:
  * - AO_nop_full()
- * - AO_load()
- * - AO_store()
- * - AO_compare_and_swap()
+ * - AO_load() || AO_load_aquire()
+ * - AO_store() || AO_store_release()
+ * - AO_compare_and_swap() || AO_compare_and_swap_acquire()
+ *   || AO_compare_and_swap_release() || AO_compare_and_swap_full()
  *
  */
 
 #if ETHR_SIZEOF_AO_T == 4
 #define ETHR_HAVE_NATIVE_ATOMIC32 1
-#define ETHR_NATIVE_ATOMIC32_IMPL "libatomic_ops"
+#define ETHR_NATIVE_ATOMIC32_IMPL ETHR_NATIVE_IMPL__
 #define ETHR_NATMC_FUNC__(X) ethr_native_atomic32_ ## X
 #define ETHR_ATMC_T__ ethr_native_atomic32_t
 #define ETHR_AINT_T__ ethr_sint32_t
 #define ETHR_AINT_SUFFIX__ "l"
 #elif ETHR_SIZEOF_AO_T == 8
 #define ETHR_HAVE_NATIVE_ATOMIC64 1
-#define ETHR_NATIVE_ATOMIC64_IMPL "libatomic_ops"
+#define ETHR_NATIVE_ATOMIC64_IMPL ETHR_NATIVE_IMPL__
 #define ETHR_NATMC_FUNC__(X) ethr_native_atomic64_ ## X
 #define ETHR_ATMC_T__ ethr_native_atomic64_t
 #define ETHR_AINT_T__ ethr_sint64_t
@@ -74,6 +75,8 @@ ETHR_NATMC_FUNC__(addr)(ETHR_ATMC_T__ *var)
     return (ETHR_AINT_T__ *) &var->counter;
 }
 
+#ifdef AO_HAVE_store
+
 #if ETHR_SIZEOF_AO_T == 4
 #  define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_SET 1
 #else
@@ -86,6 +89,24 @@ ETHR_NATMC_FUNC__(set)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ value)
     AO_store(&var->counter, (AO_t) value);
 }
 
+#endif
+
+#ifdef AO_HAVE_store_write
+
+#if ETHR_SIZEOF_AO_T == 4
+#  define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_SET_WB 1
+#else
+#  define ETHR_HAVE_ETHR_NATIVE_ATOMIC64_SET_WB 1
+#endif
+
+static ETHR_INLINE void
+ETHR_NATMC_FUNC__(set_wb)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ value)
+{
+    AO_store_write(&var->counter, (AO_t) value);
+}
+
+#endif
+
 #ifdef AO_HAVE_store_release
 
 #if ETHR_SIZEOF_AO_T == 4
@@ -102,6 +123,24 @@ ETHR_NATMC_FUNC__(set_relb)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ value)
 
 #endif
 
+#ifdef AO_HAVE_store_full
+
+#if ETHR_SIZEOF_AO_T == 4
+#  define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_SET_MB 1
+#else
+#  define ETHR_HAVE_ETHR_NATIVE_ATOMIC64_SET_MB 1
+#endif
+
+static ETHR_INLINE void
+ETHR_NATMC_FUNC__(set_mb)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ value)
+{
+    AO_store_full(&var->counter, (AO_t) value);
+}
+
+#endif
+
+#ifdef AO_HAVE_load
+
 #if ETHR_SIZEOF_AO_T == 4
 #  define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_READ 1
 #else
@@ -114,6 +153,24 @@ ETHR_NATMC_FUNC__(read)(ETHR_ATMC_T__ *var)
     return (ETHR_AINT_T__) AO_load(&var->counter);
 }
 
+#endif
+
+#ifdef AO_HAVE_load_read
+
+#if ETHR_SIZEOF_AO_T == 4
+#  define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_READ_RB 1
+#else
+#  define ETHR_HAVE_ETHR_NATIVE_ATOMIC64_READ_RB 1
+#endif
+
+static ETHR_INLINE ETHR_AINT_T__
+ETHR_NATMC_FUNC__(read_rb)(ETHR_ATMC_T__ *var)
+{
+    return (ETHR_AINT_T__) AO_load_read(&var->counter);
+}
+
+#endif
+
 #ifdef AO_HAVE_load_acquire
 
 #if ETHR_SIZEOF_AO_T == 4
@@ -130,6 +187,22 @@ ETHR_NATMC_FUNC__(read_acqb)(ETHR_ATMC_T__ *var)
 
 #endif
 
+#ifdef AO_HAVE_load_full
+
+#if ETHR_SIZEOF_AO_T == 4
+#  define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_READ_MB 1
+#else
+#  define ETHR_HAVE_ETHR_NATIVE_ATOMIC64_READ_MB 1
+#endif
+
+static ETHR_INLINE ETHR_AINT_T__
+ETHR_NATMC_FUNC__(read_mb)(ETHR_ATMC_T__ *var)
+{
+    return (ETHR_AINT_T__) AO_load_full(&var->counter);
+}
+
+#endif
+
 #ifdef AO_HAVE_fetch_and_add
 
 #if ETHR_SIZEOF_AO_T == 4
@@ -146,6 +219,54 @@ ETHR_NATMC_FUNC__(add_return)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ incr)
 
 #endif
 
+#ifdef AO_HAVE_fetch_and_add_acquire
+
+#if ETHR_SIZEOF_AO_T == 4
+#  define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_ADD_RETURN_ACQB 1
+#else
+#  define ETHR_HAVE_ETHR_NATIVE_ATOMIC64_ADD_RETURN_ACQB 1
+#endif
+
+static ETHR_INLINE ETHR_AINT_T__
+ETHR_NATMC_FUNC__(add_return_acqb)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ incr)
+{
+    return ((ETHR_AINT_T__) AO_fetch_and_add_acquire(&var->counter, (AO_t) incr)) + incr;
+}
+
+#endif
+
+#ifdef AO_HAVE_fetch_and_add_release
+
+#if ETHR_SIZEOF_AO_T == 4
+#  define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_ADD_RETURN_RELB 1
+#else
+#  define ETHR_HAVE_ETHR_NATIVE_ATOMIC64_ADD_RETURN_RELB 1
+#endif
+
+static ETHR_INLINE ETHR_AINT_T__
+ETHR_NATMC_FUNC__(add_return_relb)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ incr)
+{
+    return ((ETHR_AINT_T__) AO_fetch_and_add_release(&var->counter, (AO_t) incr)) + incr;
+}
+
+#endif
+
+#ifdef AO_HAVE_fetch_and_add_full
+
+#if ETHR_SIZEOF_AO_T == 4
+#  define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_ADD_RETURN_MB 1
+#else
+#  define ETHR_HAVE_ETHR_NATIVE_ATOMIC64_ADD_RETURN_MB 1
+#endif
+
+static ETHR_INLINE ETHR_AINT_T__
+ETHR_NATMC_FUNC__(add_return_mb)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ incr)
+{
+    return ((ETHR_AINT_T__) AO_fetch_and_add_full(&var->counter, (AO_t) incr)) + incr;
+}
+
+#endif
+
 #ifdef AO_HAVE_fetch_and_add1
 
 #if ETHR_SIZEOF_AO_T == 4
@@ -178,6 +299,38 @@ ETHR_NATMC_FUNC__(inc_return_acqb)(ETHR_ATMC_T__ *var)
 
 #endif
 
+#ifdef AO_HAVE_fetch_and_add1_release
+
+#if ETHR_SIZEOF_AO_T == 4
+#  define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_INC_RETURN_RELB 1
+#else
+#  define ETHR_HAVE_ETHR_NATIVE_ATOMIC64_INC_RETURN_RELB 1
+#endif
+
+static ETHR_INLINE ETHR_AINT_T__
+ETHR_NATMC_FUNC__(inc_return_relb)(ETHR_ATMC_T__ *var)
+{
+    return ((ETHR_AINT_T__) AO_fetch_and_add1_release(&var->counter)) + 1;
+}
+
+#endif
+
+#ifdef AO_HAVE_fetch_and_add1_full
+
+#if ETHR_SIZEOF_AO_T == 4
+#  define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_INC_RETURN_MB 1
+#else
+#  define ETHR_HAVE_ETHR_NATIVE_ATOMIC64_INC_RETURN_MB 1
+#endif
+
+static ETHR_INLINE ETHR_AINT_T__
+ETHR_NATMC_FUNC__(inc_return_mb)(ETHR_ATMC_T__ *var)
+{
+    return ((ETHR_AINT_T__) AO_fetch_and_add1_full(&var->counter)) + 1;
+}
+
+#endif
+
 #ifdef AO_HAVE_fetch_and_sub1
 
 #if ETHR_SIZEOF_AO_T == 4
@@ -194,6 +347,22 @@ ETHR_NATMC_FUNC__(dec_return)(ETHR_ATMC_T__ *var)
 
 #endif
 
+#ifdef AO_HAVE_fetch_and_sub1_acquire
+
+#if ETHR_SIZEOF_AO_T == 4
+#  define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_DEC_RETURN_ACQB 1
+#else
+#  define ETHR_HAVE_ETHR_NATIVE_ATOMIC64_DEC_RETURN_ACQB 1
+#endif
+
+static ETHR_INLINE ETHR_AINT_T__
+ETHR_NATMC_FUNC__(dec_return_acqb)(ETHR_ATMC_T__ *var)
+{
+    return ((ETHR_AINT_T__) AO_fetch_and_sub1_acquire(&var->counter)) - 1;
+}
+
+#endif
+
 #ifdef AO_HAVE_fetch_and_sub1_release
 
 #if ETHR_SIZEOF_AO_T == 4
@@ -210,7 +379,60 @@ ETHR_NATMC_FUNC__(dec_return_relb)(ETHR_ATMC_T__ *var)
 
 #endif
 
-#ifdef AO_HAVE_compare_and_swap
+#ifdef AO_HAVE_fetch_and_sub1_full
+
+#if ETHR_SIZEOF_AO_T == 4
+#  define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_DEC_RETURN_MB 1
+#else
+#  define ETHR_HAVE_ETHR_NATIVE_ATOMIC64_DEC_RETURN_MB 1
+#endif
+
+static ETHR_INLINE ETHR_AINT_T__
+ETHR_NATMC_FUNC__(dec_return_mb)(ETHR_ATMC_T__ *var)
+{
+    return ((ETHR_AINT_T__) AO_fetch_and_sub1_full(&var->counter)) - 1;
+}
+
+#endif
+
+#if defined(AO_HAVE_compare_and_swap_full) || defined(AO_HAVE_fetch_compare_and_swap_full)
+
+#if ETHR_SIZEOF_AO_T == 4
+#  define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_CMPXCHG_MB 1
+#else
+#  define ETHR_HAVE_ETHR_NATIVE_ATOMIC64_CMPXCHG_MB 1
+#endif
+
+static ETHR_INLINE ETHR_AINT_T__
+ETHR_NATMC_FUNC__(cmpxchg_mb)(ETHR_ATMC_T__ *var,
+			      ETHR_AINT_T__ new,
+			      ETHR_AINT_T__ exp)
+{
+#if defined(AO_HAVE_fetch_compare_and_swap_full)
+    return (ETHR_AINT_T__) AO_fetch_compare_and_swap_full(&var->counter,
+							  (AO_t) exp,
+							  (AO_t) new);
+#else
+    ETHR_AINT_T__ act;
+    do {
+	if (AO_compare_and_swap_full(&var->counter, (AO_t) exp, (AO_t) new))
+	    return exp;
+#ifdef AO_HAVE_load_acquire
+	act = (ETHR_AINT_T__) AO_load_acquire(&var->counter);
+#else
+	act = (ETHR_AINT_T__) AO_load(&var->counter);
+#endif
+    } while (act == exp);
+#ifndef AO_HAVE_load_acquire
+    AO_nop_full();
+#endif
+    return act;
+#endif
+}
+
+#endif
+
+#if defined(AO_HAVE_compare_and_swap) || defined(AO_HAVE_fetch_compare_and_swap)
 
 #if ETHR_SIZEOF_AO_T == 4
 #  define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_CMPXCHG 1
@@ -223,18 +445,28 @@ ETHR_NATMC_FUNC__(cmpxchg)(ETHR_ATMC_T__ *var,
 			   ETHR_AINT_T__ new,
 			   ETHR_AINT_T__ exp)
 {
+#if defined(AO_HAVE_fetch_compare_and_swap)
+    return (ETHR_AINT_T__) AO_fetch_compare_and_swap(&var->counter,
+						     (AO_t) exp,
+						     (AO_t) new);
+#else
     ETHR_AINT_T__ act;
     do {
 	if (AO_compare_and_swap(&var->counter, (AO_t) exp, (AO_t) new))
 	    return exp;
+#ifdef AO_HAVE_load
 	act = (ETHR_AINT_T__) AO_load(&var->counter);
+#else
+	act = (ETHR_AINT_T__) AO_load_aquire(&var->counter);
+#endif
     } while (act == exp);
     return act;
+#endif
 }
 
 #endif
 
-#ifdef AO_HAVE_compare_and_swap_acquire
+#if defined(AO_HAVE_compare_and_swap_acquire) || defined(AO_HAVE_fetch_compare_and_swap_acquire)
 
 #if ETHR_SIZEOF_AO_T == 4
 #  define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_CMPXCHG_ACQB 1
@@ -247,6 +479,11 @@ ETHR_NATMC_FUNC__(cmpxchg_acqb)(ETHR_ATMC_T__ *var,
 				ETHR_AINT_T__ new,
 				ETHR_AINT_T__ exp)
 {
+#if defined(AO_HAVE_fetch_compare_and_swap_acquire)
+    return (ETHR_AINT_T__) AO_fetch_compare_and_swap_acquire(&var->counter,
+							     (AO_t) exp,
+							     (AO_t) new);
+#else
     ETHR_AINT_T__ act;
     do {
 	if (AO_compare_and_swap_acquire(&var->counter, (AO_t) exp, (AO_t) new))
@@ -261,11 +498,55 @@ ETHR_NATMC_FUNC__(cmpxchg_acqb)(ETHR_ATMC_T__ *var,
     AO_nop_full();
 #endif
     return act;
+#endif
+}
+
+#endif
+
+#if defined(AO_HAVE_compare_and_swap_read) || defined(AO_HAVE_fetch_compare_and_swap_read)
+
+#if ETHR_SIZEOF_AO_T == 4
+#  define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_CMPXCHG_RB 1
+#else
+#  define ETHR_HAVE_ETHR_NATIVE_ATOMIC64_CMPXCHG_RB 1
+#endif
+
+static ETHR_INLINE ETHR_AINT_T__
+ETHR_NATMC_FUNC__(cmpxchg_rb)(ETHR_ATMC_T__ *var,
+			      ETHR_AINT_T__ new,
+			      ETHR_AINT_T__ exp)
+{
+#if defined(AO_HAVE_fetch_compare_and_swap_read)
+    return (ETHR_AINT_T__) AO_fetch_compare_and_swap_read(&var->counter,
+							     (AO_t) exp,
+							     (AO_t) new);
+#else
+    ETHR_AINT_T__ act;
+    do {
+	if (AO_compare_and_swap_read(&var->counter, (AO_t) exp, (AO_t) new))
+	    return exp;
+#if defined(AO_HAVE_load_read)
+	act = (ETHR_AINT_T__) AO_load_read(&var->counter);
+#elif defined(AO_HAVE_load)
+	act = (ETHR_AINT_T__) AO_load(&var->counter);
+#else
+	act = (ETHR_AINT_T__) AO_load_acquire(&var->counter);
+#endif
+    } while (act == exp);
+#ifndef AO_HAVE_load_read
+#ifdef AO_HAVE_nop_read
+    AO_nop_read();
+#else
+    AO_nop_full();
+#endif
+#endif
+    return act;
+#endif
 }
 
 #endif
 
-#ifdef AO_HAVE_compare_and_swap_release
+#if defined(AO_HAVE_compare_and_swap_release) || defined(AO_HAVE_fetch_compare_and_swap_release)
 
 #if ETHR_SIZEOF_AO_T == 4
 #  define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_CMPXCHG_RELB 1
@@ -278,13 +559,57 @@ ETHR_NATMC_FUNC__(cmpxchg_relb)(ETHR_ATMC_T__ *var,
 				ETHR_AINT_T__ new,
 				ETHR_AINT_T__ exp)
 {
+#if defined(AO_HAVE_fetch_compare_and_swap_release)
+    return (ETHR_AINT_T__) AO_fetch_compare_and_swap_release(&var->counter,
+							     (AO_t) exp,
+							     (AO_t) new);
+#else
     ETHR_AINT_T__ act;
     do {
 	if (AO_compare_and_swap_release(&var->counter, (AO_t) exp, (AO_t) new))
 	    return exp;
+#ifdef AO_HAVE_load
 	act = (ETHR_AINT_T__) AO_load(&var->counter);
+#else
+	act = (ETHR_AINT_T__) AO_load_acquire(&var->counter);
+#endif
+    } while (act == exp);
+    return act;
+#endif
+}
+
+#endif
+
+#if defined(AO_HAVE_compare_and_swap_write) || defined(AO_HAVE_fetch_compare_and_swap_write)
+
+#if ETHR_SIZEOF_AO_T == 4
+#  define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_CMPXCHG_WB 1
+#else
+#  define ETHR_HAVE_ETHR_NATIVE_ATOMIC64_CMPXCHG_WB 1
+#endif
+
+static ETHR_INLINE ETHR_AINT_T__
+ETHR_NATMC_FUNC__(cmpxchg_wb)(ETHR_ATMC_T__ *var,
+			      ETHR_AINT_T__ new,
+			      ETHR_AINT_T__ exp)
+{
+#if defined(AO_HAVE_fetch_compare_and_swap_write)
+    return (ETHR_AINT_T__) AO_fetch_compare_and_swap_write(&var->counter,
+							   (AO_t) exp,
+							   (AO_t) new);
+#else
+    ETHR_AINT_T__ act;
+    do {
+	if (AO_compare_and_swap_write(&var->counter, (AO_t) exp, (AO_t) new))
+	    return exp;
+#ifdef AO_HAVE_load
+	act = (ETHR_AINT_T__) AO_load(&var->counter);
+#else
+	act = (ETHR_AINT_T__) AO_load_acquire(&var->counter);
+#endif
     } while (act == exp);
     return act;
+#endif
 }
 
 #endif
diff --git a/erts/include/internal/libatomic_ops/ethr_dw_atomic.h b/erts/include/internal/libatomic_ops/ethr_dw_atomic.h
new file mode 100644
index 0000000000..4dd9f41e96
--- /dev/null
+++ b/erts/include/internal/libatomic_ops/ethr_dw_atomic.h
@@ -0,0 +1,567 @@
+/*
+ * %CopyrightBegin%
+ *
+ * Copyright Ericsson AB 2014. All Rights Reserved.
+ *
+ * The contents of this file are subject to the Erlang Public License,
+ * Version 1.1, (the "License"); you may not use this file except in
+ * compliance with the License. You should have received a copy of the
+ * Erlang Public License along with this software. If not, it can be
+ * retrieved online at http://www.erlang.org/.
+ *
+ * Software distributed under the License is distributed on an "AS IS"
+ * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See
+ * the License for the specific language governing rights and limitations
+ * under the License.
+ *
+ * %CopyrightEnd%
+ */
+
+/*
+ * Description: Native double word atomics using libatomic_ops
+ * Author: Rickard Green
+ */
+
+#ifndef ETHR_LIBATOMIC_OPS_DW_ATOMIC_H__
+#define ETHR_LIBATOMIC_OPS_DW_ATOMIC_H__
+
+#if defined(AO_HAVE_double_t)						\
+    && (defined(AO_HAVE_double_load_acquire)				\
+	|| defined(AO_HAVE_double_load))				\
+    && (defined(AO_HAVE_compare_double_and_swap_double)			\
+	|| defined(AO_HAVE_compare_double_and_swap_double_full)		\
+	|| defined(AO_HAVE_compare_double_and_swap_double_acquire)	\
+	|| defined(AO_HAVE_compare_double_and_swap_double_release)	\
+	|| defined(AO_HAVE_double_compare_and_swap)			\
+	|| defined(AO_HAVE_double_compare_and_swap_full)		\
+	|| defined(AO_HAVE_double_compare_and_swap_acquire)		\
+	|| defined(AO_HAVE_double_compare_and_swap_release))
+
+#if ETHR_SIZEOF_PTR == 4
+#  define ETHR_NATIVE_SU_DW_SINT_T ethr_sint64_t
+#elif ETHR_SIZEOF_PTR == 8 && defined(ETHR_HAVE_INT128_T)
+#  define ETHR_NATIVE_SU_DW_SINT_T ethr_sint128_t
+#endif
+
+typedef union {
+    volatile AO_double_t dw_mem;
+#if defined(ETHR_NATIVE_SU_DW_SINT_T)
+    ETHR_NATIVE_SU_DW_SINT_T su_dw_sint;
+#endif
+} ethr_native_dw_atomic_t;
+
+#if defined(ETHR_NATIVE_SU_DW_SINT_T)
+#  define ETHR_HAVE_NATIVE_SU_DW_ATOMIC
+#else
+#  define ETHR_HAVE_NATIVE_DW_ATOMIC
+#endif
+
+#define ETHR_NATIVE_DW_ATOMIC_IMPL ETHR_NATIVE_IMPL__
+
+#if defined(ETHR_TRY_INLINE_FUNCS) || defined(ETHR_ATOMIC_IMPL__)
+
+
+#if defined(ETHR_NATIVE_SU_DW_SINT_T)
+#  define ETHR_NDWA_FUNC__(Func) ethr_native_su_dw_atomic_ ## Func
+#  define ETHR_NDWA_RET_3_TYPE__ ETHR_NATIVE_SU_DW_SINT_T
+#  define ETHR_NDWA_RET_2_TYPE__ ETHR_NATIVE_SU_DW_SINT_T
+#  define ETHR_NDWA_VAL_ARG_TYPE__ ETHR_NATIVE_SU_DW_SINT_T
+#  define ETHR_NDWA_DECL_ARG__(Arg)
+#  if defined(AO_HAVE_DOUBLE_PTR_STORAGE)
+#    define ETHR_NDWA_VAL2AOVAL__(AOV, V)			\
+    ((AOV).AO_whole = (double_ptr_storage) (V))
+#    define ETHR_NDWA_AOVAL2VAL__(AOV, V)			\
+    ((V) = (ETHR_NATIVE_SU_DW_SINT_T) (AOV).AO_whole)
+#    define ETHR_NDWA_RETURN_VAL_3__(SUCCESS, AOVAL, VAL)	\
+    do {							\
+	return (ETHR_NATIVE_SU_DW_SINT_T) (AOVAL).AO_whole;	\
+    } while (0)
+#    define ETHR_NDWA_RETURN_VAL_2__(AOVAL, VAL)		\
+    do {							\
+	return (ETHR_NATIVE_SU_DW_SINT_T) (AOVAL).AO_whole;	\
+    } while (0)
+#    define ETHR_NDWA_AOVAL_EQ__(AOV1, AOV2)			\
+    ((AOV1).AO_whole == (AOV2).AO_whole)
+#  else
+typedef union {
+    ethr_sint_t sint[2];
+    ETHR_NATIVE_SU_DW_SINT_T dw_sint;
+}  ethr_dw_splitter_t;
+#    define ETHR_NDWA_VAL2AOVAL__(AOV, V)			\
+    do {							\
+	ethr_dw_splitter_t tmp__;				\
+	tmp__.dw_sint = (V);					\
+	(AOV).AO_val1 = (AO_t) tmp__.sint[0];			\
+	(AOV).AO_val2 = (AO_t) tmp__.sint[1];			\
+    } while (0)
+#    define ETHR_NDWA_AOVAL2VAL__(AOV, V)			\
+    do {							\
+	ethr_dw_splitter_t tmp__;				\
+	tmp__.sint[0] = (ethr_sint_t) (AOV).AO_val1;		\
+	tmp__.sint[1] = (ethr_sint_t) (AOV).AO_val2;		\
+	(V) = tmp__.dw_sint;					\
+    } while (0)
+#    define ETHR_NDWA_RETURN_VAL_3__(SUCCESS, AOVAL, VAL)	\
+    do {							\
+	ethr_dw_splitter_t tmp__;				\
+	tmp__.sint[0] = (ethr_sint_t) (AOVAL).AO_val1;		\
+	tmp__.sint[1] = (ethr_sint_t) (AOVAL).AO_val2;		\
+	return tmp__.dw_sint;					\
+    } while (0)
+#    define ETHR_NDWA_AOVAL_EQ__(AOV1, AOV2)			\
+    ((AOV1).AO_val1 == (AOV2).AO_val1				\
+	 && (AOV1).AO_val2 == (AOV2).AO_val2)
+#  endif
+#else
+#  define ETHR_NDWA_FUNC__(Func) ethr_native_dw_atomic_ ## Func
+#  define ETHR_NDWA_RET_3_TYPE__ int
+#  define ETHR_NDWA_RET_2_TYPE__ void
+#  define ETHR_NDWA_VAL_ARG_TYPE__ ethr_sint_t *
+#  define ETHR_NDWA_DECL_ARG__(Arg) , ETHR_NDWA_VAL_ARG_TYPE__ Arg
+#    define ETHR_NDWA_VAL2AOVAL__(AOV, V)			\
+    do {							\
+	(AOV).AO_val1 = (AO_t) (V)[0];				\
+	(AOV).AO_val2 = (AO_t) (V)[1];				\
+    } while (0)
+#    define ETHR_NDWA_AOVAL2VAL__(AOV, V)			\
+    do {							\
+	ethr_dw_splitter_t tmp__;				\
+	(V)[0] = (ethr_sint_t) (AOV).AO_val1;			\
+	(V)[1] = (ethr_sint_t) (AOV).AO_val2;			\
+    } while (0)
+#    define ETHR_NDWA_RETURN_VAL_3__(SUCCESS, AOVAL, VAL)	\
+    do {							\
+	(VAL)[0] = (ethr_sint_t) (AOVAL).AO_val1;		\
+	(VAL)[1] = (ethr_sint_t) (AOVAL).AO_val2;		\
+	return (SUCCESS);					\
+    } while (0)
+#    define ETHR_NDWA_RETURN_VAL_2__(AOVAL, VAL)		\
+    do {							\
+	(VAL)[0] = (ethr_sint_t) (AOVAL).AO_val1;		\
+	(VAL)[1] = (ethr_sint_t) (AOVAL).AO_val2;		\
+	return;							\
+    } while (0)
+#  if defined(AO_HAVE_DOUBLE_PTR_STORAGE)
+#    define ETHR_NDWA_AOVAL_EQ__(AOV1, AOV2)			\
+    ((AOV1).AO_whole == (AOV2).AO_whole)
+#  else
+#    define ETHR_NDWA_AOVAL_EQ__(AOV1, AOV2)			\
+    ((AOV1).AO_val1 == (AOV2).AO_val1				\
+	 && (AOV1).AO_val2 == (AOV2).AO_val2)
+#  endif
+#endif
+
+#define ETHR_HAVE_ETHR_NATIVE_DW_ATOMIC_ADDR
+static ETHR_INLINE ethr_sint_t *
+ethr_native_dw_atomic_addr(ethr_native_dw_atomic_t *var)
+{
+    return (ethr_sint_t *) &var->dw_mem;
+}
+
+#ifdef AO_HAVE_double_load
+
+#if defined(ETHR_NATIVE_SU_DW_SINT_T)
+#  define ETHR_HAVE_ETHR_NATIVE_SU_DW_ATOMIC_READ
+#else
+#  define ETHR_HAVE_ETHR_NATIVE_DW_ATOMIC_READ
+#endif
+
+static ETHR_INLINE ETHR_NDWA_RET_2_TYPE__
+ETHR_NDWA_FUNC__(read)(ethr_native_dw_atomic_t *var
+		       ETHR_NDWA_DECL_ARG__(val))
+{
+    AO_double_t act = AO_double_load(&var->dw_mem);
+    ETHR_NDWA_RETURN_VAL_2__(act, val);
+}
+
+#endif
+
+#ifdef AO_HAVE_double_load_read
+
+#if defined(ETHR_NATIVE_SU_DW_SINT_T)
+#  define ETHR_HAVE_ETHR_NATIVE_SU_DW_ATOMIC_READ_RB
+#else
+#  define ETHR_HAVE_ETHR_NATIVE_DW_ATOMIC_READ_RB
+#endif
+
+static ETHR_INLINE ETHR_NDWA_RET_2_TYPE__
+ETHR_NDWA_FUNC__(read_rb)(ethr_native_dw_atomic_t *var
+			  ETHR_NDWA_DECL_ARG__(val))
+{
+    AO_double_t act = AO_double_load_read(&var->dw_mem);
+    ETHR_NDWA_RETURN_VAL_2__(act, val);
+}
+
+#endif
+
+#ifdef AO_HAVE_double_load_acquire
+
+#if defined(ETHR_NATIVE_SU_DW_SINT_T)
+#  define ETHR_HAVE_ETHR_NATIVE_SU_DW_ATOMIC_READ_ACQB
+#else
+#  define ETHR_HAVE_ETHR_NATIVE_DW_ATOMIC_READ_ACQB
+#endif
+
+static ETHR_INLINE ETHR_NDWA_RET_2_TYPE__
+ETHR_NDWA_FUNC__(read_acqb)(ethr_native_dw_atomic_t *var
+			    ETHR_NDWA_DECL_ARG__(val))
+{
+    AO_double_t act = AO_double_load_acquire(&var->dw_mem);
+    ETHR_NDWA_RETURN_VAL_2__(act, val);
+}
+
+#endif
+
+#ifdef AO_HAVE_double_store
+
+#if defined(ETHR_NATIVE_SU_DW_SINT_T)
+#  define ETHR_HAVE_ETHR_NATIVE_SU_DW_ATOMIC_SET
+#else
+#  define ETHR_HAVE_ETHR_NATIVE_DW_ATOMIC_SET
+#endif
+
+static ETHR_INLINE void
+ETHR_NDWA_FUNC__(set)(ethr_native_dw_atomic_t *var,
+		      ETHR_NDWA_VAL_ARG_TYPE__ val)
+{
+    AO_double_t new;
+    ETHR_NDWA_VAL2AOVAL__(new, val);
+    AO_double_store(&var->dw_mem, new);
+}
+
+#endif
+
+#ifdef AO_HAVE_double_store_write
+
+#if defined(ETHR_NATIVE_SU_DW_SINT_T)
+#  define ETHR_HAVE_ETHR_NATIVE_SU_DW_ATOMIC_SET_WB
+#else
+#  define ETHR_HAVE_ETHR_NATIVE_DW_ATOMIC_SET_WB
+#endif
+
+static ETHR_INLINE void
+ETHR_NDWA_FUNC__(set_wb)(ethr_native_dw_atomic_t *var,
+			 ETHR_NDWA_VAL_ARG_TYPE__ val)
+{
+    AO_double_t new;
+    ETHR_NDWA_VAL2AOVAL__(new, val);
+    AO_double_store_write(&var->dw_mem, new);
+}
+
+#endif
+
+#ifdef AO_HAVE_double_store_release
+
+#if defined(ETHR_NATIVE_SU_DW_SINT_T)
+#  define ETHR_HAVE_ETHR_NATIVE_SU_DW_ATOMIC_SET_RELB
+#else
+#  define ETHR_HAVE_ETHR_NATIVE_DW_ATOMIC_SET_RELB
+#endif
+
+static ETHR_INLINE void
+ETHR_NDWA_FUNC__(set_relb)(ethr_native_dw_atomic_t *var,
+			   ETHR_NDWA_VAL_ARG_TYPE__ val)
+{
+    AO_double_t new;
+    ETHR_NDWA_VAL2AOVAL__(new, val);
+    AO_double_store_release(&var->dw_mem, new);
+}
+
+#endif
+
+#if defined(AO_HAVE_double_compare_and_swap_full) || defined(AO_HAVE_compare_double_and_swap_double_full)
+
+#if defined(ETHR_NATIVE_SU_DW_SINT_T)
+#  define ETHR_HAVE_ETHR_NATIVE_SU_DW_ATOMIC_CMPXCHG_MB
+#else
+#  define ETHR_HAVE_ETHR_NATIVE_DW_ATOMIC_CMPXCHG_MB
+#endif
+
+static ETHR_INLINE ETHR_NDWA_RET_3_TYPE__
+ETHR_NDWA_FUNC__(cmpxchg_mb)(ethr_native_dw_atomic_t *var,
+			     ETHR_NDWA_VAL_ARG_TYPE__ new,
+			     ETHR_NDWA_VAL_ARG_TYPE__ exp)
+{
+    AO_double_t ao_act, ao_new, ao_exp;
+
+    ETHR_NDWA_VAL2AOVAL__(ao_exp, exp);
+    ETHR_NDWA_VAL2AOVAL__(ao_new, new);
+
+    do {
+	int xchgd;
+#if defined(AO_HAVE_double_compare_and_swap_full)
+	xchgd = AO_double_compare_and_swap_full(&var->dw_mem, ao_exp, ao_new);
+#elif defined(AO_HAVE_compare_double_and_swap_double_full)
+	xchgd = AO_compare_double_and_swap_double_full(&var->dw_mem,
+						       ao_exp.AO_val1,
+						       ao_exp.AO_val2,
+						       ao_new.AO_val1,
+						       ao_new.AO_val2);
+#endif
+
+	if (xchgd)
+	    ETHR_NDWA_RETURN_VAL_3__(1, ao_exp, exp);
+
+#ifdef AO_HAVE_double_load_acquire
+	ao_act = AO_double_load_acquire(&var->dw_mem);
+#else
+	ao_act = AO_double_load(&var->dw_mem);
+#endif
+
+    } while (ETHR_NDWA_AOVAL_EQ__(ao_exp, ao_act));
+
+#ifndef AO_HAVE_double_load_acquire
+    AO_nop_full();
+#endif
+
+    ETHR_NDWA_RETURN_VAL_3__(1, ao_act, exp);
+}
+
+#endif
+
+#if defined(AO_HAVE_double_compare_and_swap) || defined(AO_HAVE_compare_double_and_swap_double)
+
+#if defined(ETHR_NATIVE_SU_DW_SINT_T)
+#  define ETHR_HAVE_ETHR_NATIVE_SU_DW_ATOMIC_CMPXCHG
+#else
+#  define ETHR_HAVE_ETHR_NATIVE_DW_ATOMIC_CMPXCHG
+#endif
+
+static ETHR_INLINE ETHR_NDWA_RET_3_TYPE__
+ETHR_NDWA_FUNC__(cmpxchg)(ethr_native_dw_atomic_t *var,
+			  ETHR_NDWA_VAL_ARG_TYPE__ new,
+			  ETHR_NDWA_VAL_ARG_TYPE__ exp)
+{
+    AO_double_t ao_act, ao_new, ao_exp;
+
+    ETHR_NDWA_VAL2AOVAL__(ao_exp, exp);
+    ETHR_NDWA_VAL2AOVAL__(ao_new, new);
+
+    do {
+	int xchgd;
+#if defined(AO_HAVE_double_compare_and_swap)
+	xchgd = AO_double_compare_and_swap(&var->dw_mem, ao_exp, ao_new);
+#elif defined(AO_HAVE_compare_double_and_swap_double)
+	xchgd = AO_compare_double_and_swap_double(&var->dw_mem,
+						  ao_exp.AO_val1,
+						  ao_exp.AO_val2,
+						  ao_new.AO_val1,
+						  ao_new.AO_val2);
+#endif
+
+	if (xchgd)
+	    ETHR_NDWA_RETURN_VAL_3__(1, ao_exp, exp);
+
+#ifdef AO_HAVE_double_load
+	ao_act = AO_double_load(&var->dw_mem);
+#else
+	ao_act = AO_double_load_acquire(&var->dw_mem);
+#endif
+
+    } while (ETHR_NDWA_AOVAL_EQ__(ao_exp, ao_act));
+
+    ETHR_NDWA_RETURN_VAL_3__(1, ao_act, exp);
+}
+
+#endif
+
+#if defined(AO_HAVE_double_compare_and_swap_read) || defined(AO_HAVE_compare_double_and_swap_double_read)
+
+#if defined(ETHR_NATIVE_SU_DW_SINT_T)
+#  define ETHR_HAVE_ETHR_NATIVE_SU_DW_ATOMIC_CMPXCHG_RB
+#else
+#  define ETHR_HAVE_ETHR_NATIVE_DW_ATOMIC_CMPXCHG_RB
+#endif
+
+static ETHR_INLINE ETHR_NDWA_RET_3_TYPE__
+ETHR_NDWA_FUNC__(cmpxchg_rb)(ethr_native_dw_atomic_t *var,
+			     ETHR_NDWA_VAL_ARG_TYPE__ new,
+			     ETHR_NDWA_VAL_ARG_TYPE__ exp)
+{
+    AO_double_t ao_act, ao_new, ao_exp;
+
+    ETHR_NDWA_VAL2AOVAL__(ao_exp, exp);
+    ETHR_NDWA_VAL2AOVAL__(ao_new, new);
+
+    do {
+	int xchgd;
+#if defined(AO_HAVE_double_compare_and_swap_read)
+	xchgd = AO_double_compare_and_swap_read(&var->dw_mem, ao_exp, ao_new);
+#elif defined(AO_HAVE_compare_double_and_swap_double_read)
+	xchgd = AO_compare_double_and_swap_double_read(&var->dw_mem,
+							  ao_exp.AO_val1,
+							  ao_exp.AO_val2,
+							  ao_new.AO_val1,
+							  ao_new.AO_val2);
+#endif
+
+	if (xchgd)
+	    ETHR_NDWA_RETURN_VAL_3__(1, ao_exp, exp);
+
+#if defined(AO_HAVE_double_load_read)
+	ao_act = AO_double_load_read(&var->dw_mem);
+#elif defined(AO_HAVE_double_load)
+	ao_act = AO_double_load(&var->dw_mem);
+#else
+	ao_act = AO_double_load_acquire(&var->dw_mem);
+#endif
+
+    } while (ETHR_NDWA_AOVAL_EQ__(ao_exp, ao_act));
+
+#ifndef AO_HAVE_double_load_read
+#ifdef AO_HAVE_nop_read
+    AO_nop_read();
+#else
+    AO_nop_full();
+#endif
+#endif
+
+    ETHR_NDWA_RETURN_VAL_3__(1, ao_act, exp);
+}
+
+#endif
+
+#if defined(AO_HAVE_double_compare_and_swap_acquire) || defined(AO_HAVE_compare_double_and_swap_double_acquire)
+
+#if defined(ETHR_NATIVE_SU_DW_SINT_T)
+#  define ETHR_HAVE_ETHR_NATIVE_SU_DW_ATOMIC_CMPXCHG_ACQB
+#else
+#  define ETHR_HAVE_ETHR_NATIVE_DW_ATOMIC_CMPXCHG_ACQB
+#endif
+
+static ETHR_INLINE ETHR_NDWA_RET_3_TYPE__
+ETHR_NDWA_FUNC__(cmpxchg_acqb)(ethr_native_dw_atomic_t *var,
+			       ETHR_NDWA_VAL_ARG_TYPE__ new,
+			       ETHR_NDWA_VAL_ARG_TYPE__ exp)
+{
+    AO_double_t ao_act, ao_new, ao_exp;
+
+    ETHR_NDWA_VAL2AOVAL__(ao_exp, exp);
+    ETHR_NDWA_VAL2AOVAL__(ao_new, new);
+
+    do {
+	int xchgd;
+#if defined(AO_HAVE_double_compare_and_swap_acquire)
+	xchgd = AO_double_compare_and_swap_acquire(&var->dw_mem, ao_exp, ao_new);
+#elif defined(AO_HAVE_compare_double_and_swap_double_acquire)
+	xchgd = AO_compare_double_and_swap_double_acquire(&var->dw_mem,
+							  ao_exp.AO_val1,
+							  ao_exp.AO_val2,
+							  ao_new.AO_val1,
+							  ao_new.AO_val2);
+#endif
+
+	if (xchgd)
+	    ETHR_NDWA_RETURN_VAL_3__(1, ao_exp, exp);
+
+#ifdef AO_HAVE_double_load_acquire
+	ao_act = AO_double_load_acquire(&var->dw_mem);
+#else
+	ao_act = AO_double_load(&var->dw_mem);
+#endif
+
+    } while (ETHR_NDWA_AOVAL_EQ__(ao_exp, ao_act));
+
+#ifndef AO_HAVE_double_load_acquire
+    AO_nop_full();
+#endif
+
+    ETHR_NDWA_RETURN_VAL_3__(1, ao_act, exp);
+}
+
+#endif
+
+#if defined(AO_HAVE_double_compare_and_swap_write) || defined(AO_HAVE_compare_double_and_swap_double_write)
+
+#if defined(ETHR_NATIVE_SU_DW_SINT_T)
+#  define ETHR_HAVE_ETHR_NATIVE_SU_DW_ATOMIC_CMPXCHG_WB
+#else
+#  define ETHR_HAVE_ETHR_NATIVE_DW_ATOMIC_CMPXCHG_WB
+#endif
+
+static ETHR_INLINE ETHR_NDWA_RET_3_TYPE__
+ETHR_NDWA_FUNC__(cmpxchg_wb)(ethr_native_dw_atomic_t *var,
+			     ETHR_NDWA_VAL_ARG_TYPE__ new,
+			     ETHR_NDWA_VAL_ARG_TYPE__ exp)
+{
+    AO_double_t ao_act, ao_new, ao_exp;
+
+    ETHR_NDWA_VAL2AOVAL__(ao_exp, exp);
+    ETHR_NDWA_VAL2AOVAL__(ao_new, new);
+
+    do {
+	int xchgd;
+#if defined(AO_HAVE_double_compare_and_swap_write)
+	xchgd = AO_double_compare_and_swap_write(&var->dw_mem, ao_exp, ao_new);
+#elif defined(AO_HAVE_compare_double_and_swap_double_write)
+	xchgd = AO_compare_double_and_swap_double_write(&var->dw_mem,
+							ao_exp.AO_val1,
+							ao_exp.AO_val2,
+							ao_new.AO_val1,
+							ao_new.AO_val2);
+#endif
+
+	if (xchgd)
+	    ETHR_NDWA_RETURN_VAL_3__(1, ao_exp, exp);
+
+#ifdef AO_HAVE_double_load
+	ao_act = AO_double_load(&var->dw_mem);
+#else
+	ao_act = AO_double_load_acquire(&var->dw_mem);
+#endif
+
+    } while (ETHR_NDWA_AOVAL_EQ__(ao_exp, ao_act));
+
+    ETHR_NDWA_RETURN_VAL_3__(1, ao_act, exp);
+}
+
+#endif
+
+#if defined(AO_HAVE_double_compare_and_swap_release) || defined(AO_HAVE_compare_double_and_swap_double_release)
+
+#if defined(ETHR_NATIVE_SU_DW_SINT_T)
+#  define ETHR_HAVE_ETHR_NATIVE_SU_DW_ATOMIC_CMPXCHG_RELB
+#else
+#  define ETHR_HAVE_ETHR_NATIVE_DW_ATOMIC_CMPXCHG_RELB
+#endif
+
+static ETHR_INLINE ETHR_NDWA_RET_3_TYPE__
+ETHR_NDWA_FUNC__(cmpxchg_relb)(ethr_native_dw_atomic_t *var,
+			       ETHR_NDWA_VAL_ARG_TYPE__ new,
+			       ETHR_NDWA_VAL_ARG_TYPE__ exp)
+{
+    AO_double_t ao_act, ao_new, ao_exp;
+
+    ETHR_NDWA_VAL2AOVAL__(ao_exp, exp);
+    ETHR_NDWA_VAL2AOVAL__(ao_new, new);
+
+    do {
+	int xchgd;
+#if defined(AO_HAVE_double_compare_and_swap_release)
+	xchgd = AO_double_compare_and_swap_release(&var->dw_mem, ao_exp, ao_new);
+#elif defined(AO_HAVE_compare_double_and_swap_double_release)
+	xchgd = AO_compare_double_and_swap_double_release(&var->dw_mem,
+							  ao_exp.AO_val1,
+							  ao_exp.AO_val2,
+							  ao_new.AO_val1,
+							  ao_new.AO_val2);
+#endif
+
+	if (xchgd)
+	    ETHR_NDWA_RETURN_VAL_3__(1, ao_exp, exp);
+
+	ao_act = AO_double_load(&var->dw_mem);
+
+    } while (ETHR_NDWA_AOVAL_EQ__(ao_exp, ao_act));
+
+    ETHR_NDWA_RETURN_VAL_3__(1, ao_act, exp);
+}
+
+#endif
+
+#endif /* defined(ETHR_TRY_INLINE_FUNCS) || defined(ETHR_ATOMIC_IMPL__) */
+
+#endif /* Have AO double functionality ... */
+
+#endif /* ETHR_LIBATOMIC_OPS_DW_ATOMIC_H__ */
+
diff --git a/erts/include/internal/libatomic_ops/ethread.h b/erts/include/internal/libatomic_ops/ethread.h
index e1fdd588bb..d65ee19b04 100644
--- a/erts/include/internal/libatomic_ops/ethread.h
+++ b/erts/include/internal/libatomic_ops/ethread.h
@@ -33,9 +33,12 @@
 #define AO_USE_PENTIUM4_INSTRS
 #endif
 
+#define ETHR_NATIVE_IMPL__ "libatomic_ops"
+
 #include "atomic_ops.h"
 #include "ethr_membar.h"
 #include "ethr_atomic.h"
+#include "ethr_dw_atomic.h"
 
 #endif