Homogenize memory barriers on atomics

Atomic operations with specified barriers have specified barrier semantics.
Set and read operations have undefined barrier semantics. All other atomic
operations implied full memory barriers, except when using the libatomic_ops
library and the tilera atomics api.

Some code in the runtime system assumed that all operations used (except for
set, read and specified) implied full memory barriers. The use of the
libatomic_ops library and the tilera atomics api have therefore been modified
to behave as the other implementations.

Some atomic operations with specified barrier semantics on sparc32 have also
been been relaxed in this commit.

3 files changed, 46 insertions, 42 deletions

diff --git a/erts/include/internal/libatomic_ops/ethr_atomic.h b/erts/include/internal/libatomic_ops/ethr_atomic.h
index d56693dbf8..2fc82c99a8 100644
--- a/erts/include/internal/libatomic_ops/ethr_atomic.h
+++ b/erts/include/internal/libatomic_ops/ethr_atomic.h
@@ -146,13 +146,13 @@ ETHR_NATMC_FUNC__(read)(ETHR_ATMC_T__ *var)
 static ETHR_INLINE ETHR_AINT_T__
 ETHR_NATMC_FUNC__(add_return)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ incr)
 {
-#ifdef AO_HAVE_fetch_and_add
-    return ((ETHR_AINT_T__) AO_fetch_and_add(&var->counter, (AO_t) incr)) + incr;
+#ifdef AO_HAVE_fetch_and_add_full
+    return ((ETHR_AINT_T__) AO_fetch_and_add_full(&var->counter, (AO_t) incr)) + incr;
 #else
     while (1) {
 	AO_t exp = AO_load(&var->counter);
 	AO_t new = exp + (AO_t) incr;
-	if (AO_compare_and_swap(&var->counter, exp, new))
+	if (AO_compare_and_swap_full(&var->counter, exp, new))
 	    return (ETHR_AINT_T__) new;
     }
 #endif
@@ -167,8 +167,8 @@ ETHR_NATMC_FUNC__(add)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ incr)
 static ETHR_INLINE ETHR_AINT_T__
 ETHR_NATMC_FUNC__(inc_return)(ETHR_ATMC_T__ *var)
 {
-#ifdef AO_HAVE_fetch_and_add1
-    return ((ETHR_AINT_T__) AO_fetch_and_add1(&var->counter)) + 1;
+#ifdef AO_HAVE_fetch_and_add1_full
+    return ((ETHR_AINT_T__) AO_fetch_and_add1_full(&var->counter)) + 1;
 #else
     return ETHR_NATMC_FUNC__(add_return)(var, 1);
 #endif
@@ -183,8 +183,8 @@ ETHR_NATMC_FUNC__(inc)(ETHR_ATMC_T__ *var)
 static ETHR_INLINE ETHR_AINT_T__
 ETHR_NATMC_FUNC__(dec_return)(ETHR_ATMC_T__ *var)
 {
-#ifdef AO_HAVE_fetch_and_sub1
-    return ((ETHR_AINT_T__) AO_fetch_and_sub1(&var->counter)) - 1;
+#ifdef AO_HAVE_fetch_and_sub1_full
+    return ((ETHR_AINT_T__) AO_fetch_and_sub1_full(&var->counter)) - 1;
 #else
     return ETHR_NATMC_FUNC__(add_return)(var, -1);
 #endif
@@ -202,7 +202,7 @@ ETHR_NATMC_FUNC__(and_retold)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ mask)
     while (1) {
 	AO_t exp = AO_load(&var->counter);
 	AO_t new = exp & ((AO_t) mask);
-	if (AO_compare_and_swap(&var->counter, exp, new))
+	if (AO_compare_and_swap_full(&var->counter, exp, new))
 	    return (ETHR_AINT_T__) exp;
     }
 }
@@ -213,7 +213,7 @@ ETHR_NATMC_FUNC__(or_retold)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ mask)
     while (1) {
 	AO_t exp = AO_load(&var->counter);
 	AO_t new = exp | ((AO_t) mask);
-	if (AO_compare_and_swap(&var->counter, exp, new))
+	if (AO_compare_and_swap_full(&var->counter, exp, new))
 	    return (ETHR_AINT_T__) exp;
     }
 }
@@ -225,7 +225,7 @@ ETHR_NATMC_FUNC__(cmpxchg)(ETHR_ATMC_T__ *var,
 {
     ETHR_AINT_T__ act;
     do {
-	if (AO_compare_and_swap(&var->counter, (AO_t) exp, (AO_t) new))
+	if (AO_compare_and_swap_full(&var->counter, (AO_t) exp, (AO_t) new))
 	    return exp;
 	act = (ETHR_AINT_T__) AO_load(&var->counter);
     } while (act == exp);
@@ -237,7 +237,7 @@ ETHR_NATMC_FUNC__(xchg)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ new)
 {
     while (1) {
 	AO_t exp = AO_load(&var->counter);
-	if (AO_compare_and_swap(&var->counter, exp, (AO_t) new))
+	if (AO_compare_and_swap_full(&var->counter, exp, (AO_t) new))
 	    return (ETHR_AINT_T__) exp;
     }
 }
@@ -265,7 +265,6 @@ ETHR_NATMC_FUNC__(inc_return_acqb)(ETHR_ATMC_T__ *var)
     return ((ETHR_AINT_T__) AO_fetch_and_add1_acquire(&var->counter)) + 1;
 #else
     ETHR_AINT_T__ res = ETHR_NATMC_FUNC__(add_return)(var, 1);
-    ETHR_MEMORY_BARRIER;
     return res;
 #endif
 }
@@ -287,7 +286,6 @@ ETHR_NATMC_FUNC__(dec_return_relb)(ETHR_ATMC_T__ *var)
 #ifdef AO_HAVE_fetch_and_sub1_release
     return ((ETHR_AINT_T__) AO_fetch_and_sub1_release(&var->counter)) - 1;
 #else
-    ETHR_MEMORY_BARRIER;
     return ETHR_NATMC_FUNC__(dec_return)(var);
 #endif
 }
@@ -314,7 +312,6 @@ ETHR_NATMC_FUNC__(cmpxchg_acqb)(ETHR_ATMC_T__ *var,
     return act;
 #else
     ETHR_AINT_T__ act = ETHR_NATMC_FUNC__(cmpxchg)(var, new, exp);
-    ETHR_MEMORY_BARRIER;
     return act;
 #endif
 }
@@ -333,7 +330,6 @@ ETHR_NATMC_FUNC__(cmpxchg_relb)(ETHR_ATMC_T__ *var,
     } while (act == exp);
     return act;
 #else
-    ETHR_MEMORY_BARRIER;
     return ETHR_NATMC_FUNC__(cmpxchg)(var, new, exp);
 #endif
 }
diff --git a/erts/include/internal/sparc32/atomic.h b/erts/include/internal/sparc32/atomic.h
index 00380dbf07..16182f8b01 100644
--- a/erts/include/internal/sparc32/atomic.h
+++ b/erts/include/internal/sparc32/atomic.h
@@ -95,7 +95,7 @@ ETHR_NATMC_FUNC__(add_return)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ incr)
 {
     ETHR_AINT_T__ old, tmp;
 
-    __asm__ __volatile__("membar #LoadLoad|#StoreLoad\n");
+    __asm__ __volatile__("membar #LoadLoad|#StoreLoad\n" : : : "memory");
     do {
 	old = var->counter;
 	tmp = old+incr;
@@ -105,7 +105,7 @@ ETHR_NATMC_FUNC__(add_return)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ incr)
 	    : "r"(old), "r"(&var->counter), "0"(tmp)
 	    : "memory");
     } while (__builtin_expect(old != tmp, 0));
-    __asm__ __volatile__("membar #StoreLoad|#StoreStore");
+    __asm__ __volatile__("membar #StoreLoad|#StoreStore" : : : "memory");
     return old+incr;
 }
 
@@ -144,7 +144,7 @@ ETHR_NATMC_FUNC__(and_retold)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ mask)
 {
     ETHR_AINT_T__ old, tmp;
 
-    __asm__ __volatile__("membar #LoadLoad|#StoreLoad\n");
+    __asm__ __volatile__("membar #LoadLoad|#StoreLoad\n" : : : "memory");
     do {
 	old = var->counter;
 	tmp = old & mask;
@@ -154,7 +154,7 @@ ETHR_NATMC_FUNC__(and_retold)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ mask)
 	    : "r"(old), "r"(&var->counter), "0"(tmp)
 	    : "memory");
     } while (__builtin_expect(old != tmp, 0));
-    __asm__ __volatile__("membar #StoreLoad|#StoreStore");
+    __asm__ __volatile__("membar #StoreLoad|#StoreStore" : : : "memory");
     return old;
 }
 
@@ -163,7 +163,7 @@ ETHR_NATMC_FUNC__(or_retold)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ mask)
 {
     ETHR_AINT_T__ old, tmp;
 
-    __asm__ __volatile__("membar #LoadLoad|#StoreLoad\n");
+    __asm__ __volatile__("membar #LoadLoad|#StoreLoad\n" : : : "memory");
     do {
 	old = var->counter;
 	tmp = old | mask;
@@ -173,7 +173,7 @@ ETHR_NATMC_FUNC__(or_retold)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ mask)
 	    : "r"(old), "r"(&var->counter), "0"(tmp)
 	    : "memory");
     } while (__builtin_expect(old != tmp, 0));
-    __asm__ __volatile__("membar #StoreLoad|#StoreStore");
+    __asm__ __volatile__("membar #StoreLoad|#StoreStore" : : : "memory");
     return old;
 }
 
@@ -182,7 +182,7 @@ ETHR_NATMC_FUNC__(xchg)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ val)
 {
     ETHR_AINT_T__ old, new;
 
-    __asm__ __volatile__("membar #LoadLoad|#StoreLoad");
+    __asm__ __volatile__("membar #LoadLoad|#StoreLoad" : : : "memory");
     do {
 	old = var->counter;
 	new = val;
@@ -192,20 +192,20 @@ ETHR_NATMC_FUNC__(xchg)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ val)
 	    : "r"(old), "r"(&var->counter), "0"(new)
 	    : "memory");
     } while (__builtin_expect(old != new, 0));
-    __asm__ __volatile__("membar #StoreLoad|#StoreStore");
+    __asm__ __volatile__("membar #StoreLoad|#StoreStore" : : : "memory");
     return old;
 }
 
 static ETHR_INLINE ETHR_AINT_T__
 ETHR_NATMC_FUNC__(cmpxchg)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ new, ETHR_AINT_T__ old)
 {
-    __asm__ __volatile__("membar #LoadLoad|#StoreLoad\n");
+    __asm__ __volatile__("membar #LoadLoad|#StoreLoad\n" : : : "memory");
     __asm__ __volatile__(
       ETHR_CAS__ " [%2], %1, %0"
       : "=&r"(new)
       : "r"(old), "r"(&var->counter), "0"(new)
       : "memory");
-    __asm__ __volatile__("membar #StoreLoad|#StoreStore");
+    __asm__ __volatile__("membar #StoreLoad|#StoreStore" : : : "memory");
     return new;
 }
 
@@ -213,13 +213,11 @@ ETHR_NATMC_FUNC__(cmpxchg)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ new, ETHR_AINT_T__
  * Atomic ops with at least specified barriers.
  */
 
-/* TODO: relax acquire barriers */
-
 static ETHR_INLINE ETHR_AINT_T__
 ETHR_NATMC_FUNC__(read_acqb)(ETHR_ATMC_T__ *var)
 {
     ETHR_AINT_T__ res = ETHR_NATMC_FUNC__(read)(var);
-    __asm__ __volatile__("membar #LoadLoad|#LoadStore|#StoreLoad|#StoreStore" : : : "memory");
+    __asm__ __volatile__("membar #LoadLoad|#LoadStore" : : : "memory");
     return res;
 }
 
@@ -234,21 +232,18 @@ static ETHR_INLINE ETHR_AINT_T__
 ETHR_NATMC_FUNC__(inc_return_acqb)(ETHR_ATMC_T__ *var)
 {
     ETHR_AINT_T__ res = ETHR_NATMC_FUNC__(inc_return)(var);
-    __asm__ __volatile__("membar #LoadLoad|#LoadStore" : : : "memory");
     return res;
 }
 
 static ETHR_INLINE void
 ETHR_NATMC_FUNC__(dec_relb)(ETHR_ATMC_T__ *var)
 {
-    __asm__ __volatile__("membar #LoadStore|#StoreStore" : : : "memory");
     ETHR_NATMC_FUNC__(dec)(var);
 }
 
 static ETHR_INLINE ETHR_AINT_T__
 ETHR_NATMC_FUNC__(dec_return_relb)(ETHR_ATMC_T__ *var)
 {
-    __asm__ __volatile__("membar #LoadStore|#StoreStore" : : : "memory");
     return ETHR_NATMC_FUNC__(dec_return)(var);
 }
 
@@ -256,14 +251,12 @@ static ETHR_INLINE ETHR_AINT_T__
 ETHR_NATMC_FUNC__(cmpxchg_acqb)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ new, ETHR_AINT_T__ old)
 {
     ETHR_AINT_T__ res = ETHR_NATMC_FUNC__(cmpxchg)(var, new, old);
-    __asm__ __volatile__("membar #LoadLoad|#LoadStore" : : : "memory");
     return res;
 }
 
 static ETHR_INLINE ETHR_AINT_T__
 ETHR_NATMC_FUNC__(cmpxchg_relb)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ new, ETHR_AINT_T__ old)
 {
-    __asm__ __volatile__("membar #LoadStore|#StoreStore" : : : "memory");
     return ETHR_NATMC_FUNC__(cmpxchg)(var, new, old);
 }
 
diff --git a/erts/include/internal/tile/atomic.h b/erts/include/internal/tile/atomic.h
index 48e4c0c6c8..0c7b597a6b 100644
--- a/erts/include/internal/tile/atomic.h
+++ b/erts/include/internal/tile/atomic.h
@@ -65,25 +65,35 @@ ethr_native_atomic32_read(ethr_native_atomic32_t *var)
 static ETHR_INLINE void
 ethr_native_atomic32_add(ethr_native_atomic32_t *var, ethr_sint32_t incr)
 {
+    ETHR_MEMORY_BARRIER;
     atomic_add(&var->counter, incr);
+    ETHR_MEMORY_BARRIER;
 }      
        
 static ETHR_INLINE void
 ethr_native_atomic32_inc(ethr_native_atomic32_t *var)
 {
+    ETHR_MEMORY_BARRIER;
     atomic_increment(&var->counter);
+    ETHR_MEMORY_BARRIER;
 }
 
 static ETHR_INLINE void
 ethr_native_atomic32_dec(ethr_native_atomic32_t *var)
 {
+    ETHR_MEMORY_BARRIER;
     atomic_decrement(&var->counter);
+    ETHR_MEMORY_BARRIER;
 }
 
 static ETHR_INLINE ethr_sint32_t
 ethr_native_atomic32_add_return(ethr_native_atomic32_t *var, ethr_sint32_t incr)
 {
-    return atomic_exchange_and_add(&var->counter, incr) + incr;
+    ethr_sint32_t res;
+    ETHR_MEMORY_BARRIER;
+    res = atomic_exchange_and_add(&var->counter, incr) + incr;
+    ETHR_MEMORY_BARRIER;
+    return res;
 }
 
 static ETHR_INLINE ethr_sint32_t
@@ -101,18 +111,27 @@ ethr_native_atomic32_dec_return(ethr_native_atomic32_t *var)
 static ETHR_INLINE ethr_sint32_t
 ethr_native_atomic32_and_retold(ethr_native_atomic32_t *var, ethr_sint32_t mask)
 {
-    return atomic_and_val(&var->counter, mask);
+    ethr_sint32_t res;
+    ETHR_MEMORY_BARRIER;
+    res = atomic_and_val(&var->counter, mask);
+    ETHR_MEMORY_BARRIER;
+    return res;
 }
 
 static ETHR_INLINE ethr_sint32_t
 ethr_native_atomic32_or_retold(ethr_native_atomic32_t *var, ethr_sint32_t mask)
 {
-    return atomic_or_val(&var->counter, mask);
+    ethr_sint32_t res;
+    ETHR_MEMORY_BARRIER;
+    res = atomic_or_val(&var->counter, mask);
+    ETHR_MEMORY_BARRIER;
+    return res;
 }
 
 static ETHR_INLINE ethr_sint32_t
 ethr_native_atomic32_xchg(ethr_native_atomic32_t *var, ethr_sint32_t val)
 {   
+    ETHR_MEMORY_BARRIER;
     return atomic_exchange_acq(&var->counter, val);
 } 
 
@@ -121,6 +140,7 @@ ethr_native_atomic32_cmpxchg(ethr_native_atomic32_t *var,
 			     ethr_sint32_t new,
 			     ethr_sint32_t expected)
 {
+    ETHR_MEMORY_BARRIER;
     return atomic_compare_and_exchange_val_acq(&var->counter, new, expected);
 }
 
@@ -139,9 +159,7 @@ ethr_native_atomic32_read_acqb(ethr_native_atomic32_t *var)
 static ETHR_INLINE ethr_sint32_t
 ethr_native_atomic32_inc_return_acqb(ethr_native_atomic32_t *var)
 {
-    ethr_sint32_t res = ethr_native_atomic32_inc_return(var);
-    ETHR_MEMORY_BARRIER;
-    return res;
+    return ethr_native_atomic32_inc_return(var);
 }
 
 static ETHR_INLINE void
@@ -154,14 +172,12 @@ ethr_native_atomic32_set_relb(ethr_native_atomic32_t *var, ethr_sint32_t val)
 static ETHR_INLINE void
 ethr_native_atomic32_dec_relb(ethr_native_atomic32_t *var)
 {
-    ETHR_MEMORY_BARRIER;
     ethr_native_atomic32_dec(var);
 }
 
 static ETHR_INLINE ethr_sint32_t
 ethr_native_atomic32_dec_return_relb(ethr_native_atomic32_t *var)
 {
-    ETHR_MEMORY_BARRIER;
     return ethr_native_atomic32_dec_return(var);
 }
 
@@ -178,7 +194,6 @@ ethr_native_atomic32_cmpxchg_relb(ethr_native_atomic32_t *var,
 				  ethr_sint32_t new,
 				  ethr_sint32_t exp)
 {
-    ETHR_MEMORY_BARRIER;
     return ethr_native_atomic32_cmpxchg(var, new, exp);
 }