diff --git a/include/spinlock/anderson.h b/include/spinlock/anderson.h
index ae77429..9555564 100644
--- a/include/spinlock/anderson.h
+++ b/include/spinlock/anderson.h
@@ -137,7 +137,7 @@ ck_spinlock_anderson_lock(struct ck_spinlock_anderson *lock,
 
 	/* Prepare slot for potential re-use by another thread. */
 	ck_pr_store_uint(&lock->slots[position].locked, true);
-	ck_pr_fence_memory();
+	ck_pr_fence_acquire();
 
 	*slot = lock->slots + position;
 	return;
@@ -149,7 +149,7 @@ ck_spinlock_anderson_unlock(struct ck_spinlock_anderson *lock,
 {
 	unsigned int position;
 
-	ck_pr_fence_memory();
+	ck_pr_fence_release();
 
 	/* Mark next slot as available. */
 	if (lock->wrap == 0)
diff --git a/include/spinlock/cas.h b/include/spinlock/cas.h
index db893af..b2442c5 100644
--- a/include/spinlock/cas.h
+++ b/include/spinlock/cas.h
@@ -60,7 +60,7 @@ ck_spinlock_cas_trylock(struct ck_spinlock_cas *lock)
 
 	value = ck_pr_fas_uint(&lock->value, true);
 	if (value == false)
-		ck_pr_fence_memory();
+		ck_pr_fence_acquire();
 
 	return !value;
 }
@@ -82,7 +82,7 @@ ck_spinlock_cas_lock(struct ck_spinlock_cas *lock)
 			ck_pr_stall();
 	}
 
-	ck_pr_fence_memory();
+	ck_pr_fence_acquire();
 	return;
 }
 
@@ -94,7 +94,7 @@ ck_spinlock_cas_lock_eb(struct ck_spinlock_cas *lock)
 	while (ck_pr_cas_uint(&lock->value, false, true) == false)
 		ck_backoff_eb(&backoff);
 
-	ck_pr_fence_memory();
+	ck_pr_fence_acquire();
 	return;
 }
 
@@ -103,7 +103,7 @@ ck_spinlock_cas_unlock(struct ck_spinlock_cas *lock)
 {
 
 	/* Set lock state to unlocked. */
-	ck_pr_fence_memory();
+	ck_pr_fence_release();
 	ck_pr_store_uint(&lock->value, false);
 	return;
 }
diff --git a/include/spinlock/clh.h b/include/spinlock/clh.h
index b76e413..cec219c 100644
--- a/include/spinlock/clh.h
+++ b/include/spinlock/clh.h
@@ -82,6 +82,7 @@ ck_spinlock_clh_lock(struct ck_spinlock_clh **queue, struct ck_spinlock_clh *thr
 	while (ck_pr_load_uint(&previous->wait) == true)
 		ck_pr_stall();
 
+	ck_pr_fence_load();
 	return;
 }
 
@@ -100,7 +101,7 @@ ck_spinlock_clh_unlock(struct ck_spinlock_clh **thread)
 	previous = thread[0]->previous;
 
 	/* We have to pay this cost anyways, use it as a compiler barrier too. */
-	ck_pr_fence_memory();
+	ck_pr_fence_release();
 	ck_pr_store_uint(&(*thread)->wait, false);
 
 	/*
diff --git a/include/spinlock/dec.h b/include/spinlock/dec.h
index 16a53d4..7874ae2 100644
--- a/include/spinlock/dec.h
+++ b/include/spinlock/dec.h
@@ -62,7 +62,7 @@ ck_spinlock_dec_trylock(struct ck_spinlock_dec *lock)
 
 	value = ck_pr_fas_uint(&lock->value, 0);
 	if (value == 1) {
-		ck_pr_fence_memory();
+		ck_pr_fence_acquire();
 		return true;
 	}
 
@@ -89,7 +89,6 @@ ck_spinlock_dec_lock(struct ck_spinlock_dec *lock)
 		 * UINT_MAX lock requests can happen while the lock is held.
 		 */
 		ck_pr_dec_uint_zero(&lock->value, &r);
-		ck_pr_fence_memory();
 		if (r == true)
 			break;
 
@@ -98,6 +97,7 @@ ck_spinlock_dec_lock(struct ck_spinlock_dec *lock)
 			ck_pr_stall();
 	}
 
+	ck_pr_fence_acquire();
 	return;
 }
 
@@ -115,7 +115,7 @@ ck_spinlock_dec_lock_eb(struct ck_spinlock_dec *lock)
 		ck_backoff_eb(&backoff);
 	}
 
-	ck_pr_fence_memory();
+	ck_pr_fence_acquire();
 	return;
 }
 
@@ -123,7 +123,7 @@ CK_CC_INLINE static void
 ck_spinlock_dec_unlock(struct ck_spinlock_dec *lock)
 {
 
-	ck_pr_fence_memory();
+	ck_pr_fence_release();
 
 	/* Unconditionally set lock value to 1 so someone can decrement lock to 0. */
 	ck_pr_store_uint(&lock->value, 1);
diff --git a/include/spinlock/fas.h b/include/spinlock/fas.h
index b44a6e4..6803616 100644
--- a/include/spinlock/fas.h
+++ b/include/spinlock/fas.h
@@ -58,7 +58,7 @@ ck_spinlock_fas_trylock(struct ck_spinlock_fas *lock)
 
 	value = ck_pr_fas_uint(&lock->value, true);
 	if (value == false)
-		ck_pr_fence_memory();
+		ck_pr_fence_acquire();
 
 	return !value;
 }
@@ -80,7 +80,7 @@ ck_spinlock_fas_lock(struct ck_spinlock_fas *lock)
 			ck_pr_stall();
 	}
 
-	ck_pr_fence_memory();
+	ck_pr_fence_acquire();
 	return;
 }
 
@@ -92,7 +92,7 @@ ck_spinlock_fas_lock_eb(struct ck_spinlock_fas *lock)
 	while (ck_pr_fas_uint(&lock->value, true) == true)
 		ck_backoff_eb(&backoff);
 
-	ck_pr_fence_memory();
+	ck_pr_fence_acquire();
 	return;
 }
 
@@ -100,7 +100,7 @@ CK_CC_INLINE static void
 ck_spinlock_fas_unlock(struct ck_spinlock_fas *lock)
 {
 
-	ck_pr_fence_memory();
+	ck_pr_fence_release();
 	ck_pr_store_uint(&lock->value, false);
 	return;
 }
diff --git a/include/spinlock/hclh.h b/include/spinlock/hclh.h
index c82ccd0..8e56c10 100644
--- a/include/spinlock/hclh.h
+++ b/include/spinlock/hclh.h
@@ -110,6 +110,7 @@ ck_spinlock_hclh_lock(struct ck_spinlock_hclh **glob_queue,
 	while (ck_pr_load_uint(&previous->wait) == true)
 		ck_pr_stall();
 
+	ck_pr_fence_load();
 	return;
 }
 
@@ -128,7 +129,7 @@ ck_spinlock_hclh_unlock(struct ck_spinlock_hclh **thread)
 	previous = thread[0]->previous;
 
 	/* We have to pay this cost anyways, use it as a compiler barrier too. */
-	ck_pr_fence_memory();
+	ck_pr_fence_release();
 	ck_pr_store_uint(&(*thread)->wait, false);
 
 	/*
diff --git a/include/spinlock/mcs.h b/include/spinlock/mcs.h
index 4c8e759..5ca51ae 100644
--- a/include/spinlock/mcs.h
+++ b/include/spinlock/mcs.h
@@ -61,7 +61,7 @@ ck_spinlock_mcs_trylock(struct ck_spinlock_mcs **queue, struct ck_spinlock_mcs *
 	ck_pr_fence_store_atomic();
 
 	if (ck_pr_cas_ptr(queue, NULL, node) == true) {
-		ck_pr_fence_load();
+		ck_pr_fence_acquire();
 		return true;
 	}
 
@@ -111,7 +111,7 @@ ck_spinlock_mcs_unlock(struct ck_spinlock_mcs **queue, struct ck_spinlock_mcs *n
 {
 	struct ck_spinlock_mcs *next;
 
-	ck_pr_fence_memory();
+	ck_pr_fence_release();
 
 	next = ck_pr_load_ptr(&node->next);
 	if (next == NULL) {
diff --git a/include/spinlock/ticket.h b/include/spinlock/ticket.h
index 288f4a0..1402a50 100644
--- a/include/spinlock/ticket.h
+++ b/include/spinlock/ticket.h
@@ -115,7 +115,7 @@ ck_spinlock_ticket_lock(struct ck_spinlock_ticket *ticket)
 		    CK_SPINLOCK_TICKET_MASK;
 	}
 
-	ck_pr_fence_memory();
+	ck_pr_fence_acquire();
 	return;
 }
 
@@ -142,7 +142,7 @@ ck_spinlock_ticket_lock_pb(struct ck_spinlock_ticket *ticket, unsigned int c)
 		ck_backoff_eb(&backoff);
 	}
 
-	ck_pr_fence_memory();
+	ck_pr_fence_acquire();
 	return;
 }
 
@@ -163,7 +163,7 @@ ck_spinlock_ticket_trylock(struct ck_spinlock_ticket *ticket)
 		return false;
 	}
 
-	ck_pr_fence_memory();
+	ck_pr_fence_acquire();
 	return true;
 }
 
@@ -171,7 +171,7 @@ CK_CC_INLINE static void
 ck_spinlock_ticket_unlock(struct ck_spinlock_ticket *ticket)
 {
 
-	ck_pr_fence_memory();
+	ck_pr_fence_release();
 	CK_SPINLOCK_TICKET_INC((CK_SPINLOCK_TICKET_TYPE_BASE *)(void *)&ticket->value);
 	return;
 }
@@ -235,7 +235,7 @@ ck_spinlock_ticket_lock(struct ck_spinlock_ticket *ticket)
 	while (ck_pr_load_uint(&ticket->position) != request)
 		ck_pr_stall();
 
-	ck_pr_fence_memory();
+	ck_pr_fence_acquire();
 	return;
 }
 
@@ -264,7 +264,7 @@ ck_spinlock_ticket_lock_pb(struct ck_spinlock_ticket *ticket, unsigned int c)
 		ck_backoff_eb(&backoff);
 	}
 
-	ck_pr_fence_memory();
+	ck_pr_fence_acquire();
 	return;
 }
 
@@ -273,7 +273,7 @@ ck_spinlock_ticket_unlock(struct ck_spinlock_ticket *ticket)
 {
 	unsigned int update;
 
-	ck_pr_fence_memory();
+	ck_pr_fence_release();
 
 	/*
 	 * Update current ticket value so next lock request can proceed.